Mechanical Behavior of Elastic Rods Under Constraint by James T. Miller B.S., Massachusetts Institute of Technology (2006) S.M., Massachusetts Institute of Technology (2008) Submitted to the Department of Civil and Environmental Engineering in partial fulfillment of the requirements for the degree of Doctor of Philosophy in the Field of Structures and Materials at the MASSACHUSETTS INSTITUTE OF TECHNOLOGY February 2014 c Massachusetts Institute of Technology 2014. All rights reserved. Author . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Department of Civil and Environmental Engineering January 10, 2014 Certified by . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Pedro M. Reis Assistant Professor of Civil and Environmental Engineering and Mechanical Engineering Thesis Supervisor Accepted by . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Heidi M. Nepf Chair, Departmental Committee for Graduate Students 2 Mechanical Behavior of Elastic Rods Under Constraint by James T. Miller Submitted to the Department of Civil and Environmental Engineering on January 10, 2014, in partial fulfillment of the requirements for the degree of Doctor of Philosophy in the Field of Structures and Materials Abstract We present the results of an experimental investigation of the mechanics of thin elastic rods under a variety of loading conditions. Four scenarios are explored, with increasing complexity: i) the shape of a naturally curved rod suspended under self-weight, ii) the buckling and post-buckling behavior of a rod compressed inside a cylindrical constraint, iii) the mechanical instabilities arising when a rod is progressively injected into a horizontal cylinder, and iv) strategies for mitigation of these instabilities by dynamic excitation of the constraint. First, we consider the role of natural curvature in determining the shape of a hanging elastic rod suspended under its own weight. We categorize three distinct configurations: planar hooks, localized helices, and global helices. Experimental results are contrasted with simulations and theory and the phase diagram of the system is rationalized. Secondly, in what we call the classic case experiment, we study the buckling and post-buckling behavior of a rod compressed inside a cylindrical constraint. Under imposed displacement, the initially straight rod buckles into a sinusoidal mode and eventually undergoes a secondary instability into a helical configuration. The critical buckling loads are quantified and found to depend strongly on the aspect ratio of the rod to pipe diameter. Thirdly, we inject a thin elastic rod into a horizontal cylinder under imposed velocity in the real case experiment. Friction between the rod and constraining pipe causes an increasing axial load with continued injection. Consecutive buckling transitions lead to straight, sinusoidal, and helical configurations in a spatially heterogeneous distribution. We quantify critical lengths and loads for the onset of the helical instability. The geometric parameters of the system strongly affect the buckling and post-buckling behavior. Finally, we explore active strategies for delaying the onset of helical buckling in the real case. Distributed vertical vibration is applied to the cylindrical constraint, which destabilizes frictional contacts between the rod and pipe. Injection speed, peak acceleration of vibration, and vibration frequency are all found to affect the postponement of helical initiation. The process is rationalized and design guidelines are provided for optimal parameters to actively extend horizontal reach. Thesis Supervisor: Pedro M. Reis Title: Assistant Professor of Civil and Environmental Engineering and Mechanical Engineering 3 4 Acknowledgments This is the last page I’ll write at MIT, left till the end as a chance to reflect on the last few years before moving on to the next challenge. First, I want to express my gratitude to Pedro for his roles as both mentor and friend. It was an awesome experience being here for the starting of a lab at MIT, and working and laughing with you has been a real joy. May you always keep your passion for science. The people of the EGS.Lab give it a special energy that’s rather cool to be around day-to-day. I’d especially like to mention Arnaud, Alice, and Denis as wisecrackers extraordinaire. None of us work in a vacuum, and a big thanks to everyone at SDR (Nathan, Jahir, François, David, and Liz) and Harvard (Katia and Tianxiang) with whom I collaborated. The project with Basile was completely different than anything I’ve done before, which was fascinating. Amy, Sharon, Carolyn, Jeanette, and Kris all saved my bacon regularly and are a serious credit to MIT CEE. I’d particularly like to thank my committee members (Nathan, Professor Bathe, and Professor Kausel) for all their help getting me to graduation. A lot has happened outside of lab these last three years and quite a few people helped me withstand it with a shred of sanity. José, Levi, and Aunt Mary - wow. To all the folks back home: you inspire me more than I can ever put into words, I just hope you know that you are loved deeply. Tyler and Zac, I honestly don’t know how we survived this long knowing one another - but dang, has it been fun. Ten years ago I left Alaska and never really expected to find another home. Two groups of people have proven me wrong. Mike, Cactus, Malcolm, and Breanna, y’all are my heart and crazy as cats. Trips to Glen Ellyn to see Jim, Jeff, Joe, Cathy, and Jenny were the best medicine a person could hope for. Most importantly, I want to thank my parents. Your unwavering support, love, and friendship created my world, molded me into who I am, and sustain me to this day. To put things more simply: thank you. 5 6 Dedicated to my folks, T’n’T 7 8 Contents 1 Introduction 1.1 13 Describing the Behavior of Rods . . . . . . . . . . . . . . . . . . . . . 15 1.1.1 Kinematics of a Slender Rod . . . . . . . . . . . . . . . . . . . 16 1.1.2 Internal Moment and Elastic Energy . . . . . . . . . . . . . . 17 1.1.3 Equilibrium Model: Kirchhoff Equations . . . . . . . . . . . . 19 1.2 Buckling of a Rod . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 1.3 Outline of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 2 Literature Review 2.1 2.2 2.3 2.4 29 Behavior of Naturally Curved Rods . . . . . . . . . . . . . . . . . . . 30 2.1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30 2.1.2 Previous Work . . . . . . . . . . . . . . . . . . . . . . . . . . 33 Compressing a Rod Inside a Cylinder . . . . . . . . . . . . . . . . . . 45 2.2.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 2.2.2 Previous Analytical Work . . . . . . . . . . . . . . . . . . . . 47 Injecting a Rod Into Cylindrical Constraint . . . . . . . . . . . . . . . 61 2.3.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61 2.3.2 Previous Work . . . . . . . . . . . . . . . . . . . . . . . . . . 63 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67 3 Suspending a Naturally Curved Rod 3.1 69 The Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70 3.1.1 71 Material Selection and Properties . . . . . . . . . . . . . . . . 9 3.1.2 Experimental Protocol . . . . . . . . . . . . . . . . . . . . . . 72 3.1.3 Three Dimensional Experimental Reconstructions . . . . . . . 73 Results and Interpretation . . . . . . . . . . . . . . . . . . . . . . . . 75 3.2.1 Rod Morphologies and Comparison with Numerics . . . . . . . 76 3.2.2 Planar to Non-Planar Configurations . . . . . . . . . . . . . . 83 3.2.3 Helical Shapes . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 3.3 Additional Boundary Conditions . . . . . . . . . . . . . . . . . . . . . 91 3.4 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95 3.2 4 Compressing a Rod in a Cylinder 4.1 4.2 4.3 The Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98 4.1.1 Material Selection and Properties . . . . . . . . . . . . . . . . 99 4.1.2 Compression and Data Acquisition System . . . . . . . . . . . 102 4.1.3 Experimental Protocol . . . . . . . . . . . . . . . . . . . . . . 104 Results and Interpretation . . . . . . . . . . . . . . . . . . . . . . . . 105 4.2.1 Load Displacement Signals . . . . . . . . . . . . . . . . . . . . 106 4.2.2 Critical Loads and Length scales . . . . . . . . . . . . . . . . 113 4.2.3 Effect of Imperfections . . . . . . . . . . . . . . . . . . . . . . 118 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122 5 Injecting a Rod into a Cylinder 5.1 5.2 97 125 The Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126 5.1.1 Material Selection and Properties . . . . . . . . . . . . . . . . 128 5.1.2 Injection Sub-System . . . . . . . . . . . . . . . . . . . . . . . 130 5.1.3 Data Acquisition and Control Sub-System . . . . . . . . . . . 133 5.1.4 Experimental Protocol for Rod Injection . . . . . . . . . . . . 134 Results and Interpretation . . . . . . . . . . . . . . . . . . . . . . . . 135 5.2.1 Reaction Force Signals and Video Analysis . . . . . . . . . . . 136 5.2.2 Critical Lengthscales . . . . . . . . . . . . . . . . . . . . . . . 146 5.2.3 Effect of Imperfections: Natural Curvature . . . . . . . . . . . 151 10 5.3 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153 6 Actively Extending Reach 6.1 6.2 6.3 155 The Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156 6.1.1 Driving and Vibration Measurement System . . . . . . . . . . 157 6.1.2 Experimental Protocol . . . . . . . . . . . . . . . . . . . . . . 159 Results and Interpretation . . . . . . . . . . . . . . . . . . . . . . . . 161 6.2.1 Reaction Force Signals . . . . . . . . . . . . . . . . . . . . . . 161 6.2.2 Effect of Vibration Amplitude: Contact Loss . . . . . . . . . . 164 6.2.3 Bending Waves Inside a Cylindrical Constraint . . . . . . . . . 172 6.2.4 Effect of Vibration Frequency on Helix Initiation . . . . . . . 180 6.2.5 Effect of Injection Speed . . . . . . . . . . . . . . . . . . . . . 185 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187 7 Conclusions and Future Work 7.1 189 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192 A Rod Fabrication 195 A.1 Rod Fabrication Procedure . . . . . . . . . . . . . . . . . . . . . . . . 196 A.2 Material Properties and Measurements . . . . . . . . . . . . . . . . . 199 A.2.1 Cross-Section and Density . . . . . . . . . . . . . . . . . . . . 199 A.2.2 Young’s Modulus . . . . . . . . . . . . . . . . . . . . . . . . . 199 A.2.3 Coefficient of Restitution . . . . . . . . . . . . . . . . . . . . . 203 A.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205 B The Shapes of a Suspended Curly Hair 11 207 12 Chapter 1 Introduction Rods are defined as slender filamentary structures whose lengths are much greater than their diameters. They are prevalent across the length scales of human experience. Due to their slenderness, rods often exhibit geometrically nonlinear behavior involving large displacements and rotations, while the constituent material remains in its linear range due to the small underlying strains. The geometric underpinnings of this nonlinear behavior leads to universal deformation modes observable across a variety of materials in structures ranging from the kilometer to nanometer scales. Some examples include coiled tubing in the petroleum industry [1], subsea cables [2], the shape of human hair [3], tendrils of climbing plants [4], compliant members in stretchable electronics [5], bacteria flagella [6], carbon nanotubes [7], and DNA [8]. The modern study of rods dates back to the experimental investigations of Musschenbroek in 1721 [9] and the analytic work of Euler in 1744 [10, 11] for the case of planar buckling. The elegant analogy proposed by Kirchhoff [12–14] between the behavior of rods and a spinning top, which was solved by Lagrange in 1788 [15], made the nonlinear equations of equilibrium tractable for straight rods, and integrable for rods with circular cross-sections [16]. Most rods, however, are not straight. They posses intrinsic natural curvature, which can be regarded as the shape the rod will take in the absence of external forces. For example, a steel spring is made from a rod whose natural curvature is the inverse of the radius of the helix of the spring. Most rods have a natural curvature, 13 which is a product of their genesis, whether in natural or industrial systems. Processes such as molecular interactions in DNA [2, 17–20], differential growth of filamentary plant structures [4, 21–23], the false-twist fabrication of yarns [24], and the spooling of steel pipes and cables for storage and/or transport [25–28] are all responsible for creating rods with natural curvature. The drive to understand natural phenomena involving rodlike structures and to design technology that take advantage of the large deformations of rods requires embracing the nonlinearities inherent to the mechanical description of the behavior of rods. This is to be contrasted with classic design approaches, which linearize these geometric effects by assuming small deformations and rotations (a valid approach for designing engineered structures whose purpose is to avoid large deformations). As will be shown in more detail in §1.1 and §1.2, these nonlinearities are often universal. The shape a rod takes under loading (either typical deflection under transverse loading or buckled shapes) are, under appropriate assumptions, scale-independent. The magnitude of deflection and the force required to achieve said deflection are material dependent, but the phenomenology typically is not. This observation allows for the investigation of problems at disparate length scales (from DNA to the oilfield) at the desktop scale. This is the general approach taken in this thesis, whereby the advantages of rapid prototyping and relatively inexpensive materials and equipment can be utilized to study problems typically encountered at much larger or smaller scales than the mechanical laboratory setting, in a precise way. This thesis describes a collection of problems chosen for their industrial relevance. In each case, we seek to rationalize the rich nature of rods, highlighting the nonlinear geometric behavior and the influence of natural curvature. Whereas the primary thrust of this work will be experimental, we turn next to providing the analytical framework that is typically used to describe the behavior of rods before discussing the specific problems in depth. 14 1.1 Describing the Behavior of Rods A rod is a material body which is significantly longer in one dimension (the length, L) in comparison to the other two dimensions (which define the cross-section, characterized by typical dimension, h), i.e. L >> h. For a quantitative description of rod behavior, we turn to the field of structural mechanics, where analytical models have been developed specifically for use with rods. In particular, we shall focus on a model developed for finite displacements, meaning that material rotations cannot be neglected. Below, we describe the model developed by Kirchhoff and Clebsch in 1859 [14]. Throughout this thesis, we will assume materials to be linearly elastic (or Hookean), whereby stresses, σ, and strains, , are linearly related by the Young’s modulus, E. Moreover, the transverse strains are related to extensional strains by the Poisson ratio, ν, such that a bar under a traction stress, σ, will be stretched by a factor axial = σ/E and its cross-section will contract by a factor lateral = −νaxial 1 . The assumption of linear elastic response is valid for most solid materials at small strains, . 0.01. Ultimately, we seek to have expressions for mechanical equilibrium through a one dimensional description of the rod. This is accomplished in the following subsections by first (§1.1.1) presenting a description of the rod’s deformed geometry as a function of its centerline and then (§1.1.2) relating these quantities to the internal moments and the total elastic energy of the rod. Finally, the Kirchhoff equations of equilibrium are presented in §1.1.3, and general solution strategies are discussed. This derivation follows that presented in greater detail in [29], to which we refer the interested reader. 1 More technically, the stress tensor, σij , is related to the strain tensor, ij , through the relation Eν E σij = λkk δij + 2µij , where λ = (1+ν)(1−2ν) and µ = 2(1+ν) are the Lamé constants and δij is the Kronecker delta, such that δij = 1 if i = j and zero otherwise. Subscripts denote standard Einstein summation. 15 1.1.1 Kinematics of a Slender Rod We define a rod configuration by the location of its centerline, r(s), as a function of its arc length, s. The rod is assumed to be inextensible, so that the centerline does not stretch or contract upon deformation, it only bends and/or twists. In addition to the centerline, one also needs to introduce the material frame, (d1 (s), d2 (s), d3 (s)), that is attached to the centerline. This material frame is orthonormal, such that d3 (s) is tangent to the rod’s centerline (d3 (s) = dr(s) ds = r0 (s)), while d1 and d2 lie in the plane of the cross-section. This kinematic description with a material frame is shown in Fig. 1-1 (a). The combination of a centerline and a material frame is often referred to in the literature as a Cosserat curve [29]. As the rod is deformed, the material frame rotates and twists to remain adapted (an adapted frame is an orthonormal frame tangent to a curve at all points), with d1 and d2 tracking rotations in the cross-section. In the case of small strains, the material frame remains orthonormal through rod deformation, which is equivalent to assuming unshearable cross-sections (also referred to as the Euler-Bernoulli kinematic hypothesis). In the limit of slender aspect ratio, h L → ∞, this can be proven from 3D elasticity [29]. The material frame, therefore, can be described by rigid translations and rotations as a function of the arc length, s. Therefore, the derivatives of the material frame can be expressed with respect to three scalar quantities, κ1 (s), κ2 (s), and τ (s), d01 (s) = τ (s)d2 (s) − κ2 (s)d3 (s) = Ω(s) × d1 (s) d02 (s) = −τ (s)d1 (s) + κ1 (s)d3 (s) = Ω(s) × d2 (s) (1.1) d03 (s) = κ2 (s)d1 (s) − κ1 (s)d2 (s) = Ω(s) × d3 (s), where Ω(s) is the Darboux vector, defined as Ω(s) = κ1 (s)d1 (s) + κ2 (s)d2 (s) + τ (s)d3 (s). (1.2) where Ω(s) can be physically understood as the rotation velocity of the material 16 M(s) = EI1 κ1 (s)d1 (s) + EI2 κ2 (s)d2 (s) + GJτ (s)d3 (s), (1.3) where E and G are the Young’s modulus and shear modulus of the bulk material, respectively, I1 and I2 are the second moments of inertia of the cross-section, and J is the polar moment of inertial of the section. For solid circular cross-sections (as will be the case for the experiments in this thesis), I1 = I2 = π4 r4 and J = π2 r4 , where r is the radius of the cross-section. The connection between the bulk material properties, internal moments, and material curvatures/twist is known as the constitutive relation for the rod. This would take a different form if a different material behavior was selected. Note that in Eq. (1.3), we make the implicit assumption that EI1 , EI2 , and GJ are constant throughout the length of the rod. This assumption is made in order to simplify expressions and will be carried forward, but the derivations hold without it. We can also introduce an internal force, F(s), which is the total external loads applied to the rod, integrated from one end to the point at arc length s, and is transmitted across cross-sections. It can be calculated for a rod with arbitrary point load applied at both ends of the rod, P(0) = F(s = 0) and P(L) = F(s = L), as well as distributed loads with linear density, p(s), F(s) = Z s [p(s0 )] ds0 + P(0). (1.4) 0 We could also calculate the integral from s to L, using the other point force. The distributed force, p(s), can, for example be present in the form of a distributed weight. Continuing with the small strain assumption, we can also relate the local stress and strain states to the global elastic energy of the rod, Eelastic , Eelastic 1 = 2 Z 0 L EI1 (κ1 (s))2 + EI2 (κ2 (s))2 + GJ (τ (s))2 ds. (1.5) The first two terms of the integrand are referred to as the bending energy and the last term is the twisting energy (sometimes referred to as the torsional energy). Once 18 again, this expression is approximate in the sense that the rod must be slender, h << L, for it to be valid. The elastic energy in Eq. (1.5) can be modified in the case of natural curvature, which is the curvature the rod will assume in the absence of any forces. For example, if the rod had a natural curvature, κ0 aligned with κ1 (the rod would assume a circular arc in the plane of d2 - see Fig. 1-1 (a)), the above expression should be modified by replacing the first material curvature with κ1 (s) − κ0 (s). Note that the elastic energy is dependent on the material curvature different from the natural curvature. 1.1.3 Equilibrium Model: Kirchhoff Equations The elastic energy for a thin elastic rod with an arbitrary configuration was given in Eq. (1.5). To predict the shape that the rod will take, we must search for configurations which satisfy equilibrium, i.e. all forces and moments are balanced at every section of rod. Alternatively, but equivalently, one can derive the equilibrium equations by finding stationary points for the energy, including the work done by external forces and torques. The modeled rod includes external forces – point forces P(0) and P(L) at the ends of the rod or a distributed force with linear density p(s) – and torques – point torques Q(0) and Q(L) at the rod ends or a distributed torque with linear density q(s). The equilibrium equations, or Kirchhoff equations, for this rod are then: F0 (s) + p(s) = 0, (1.6a) M0 (s) + d3 (s) × F(s) + q(s) = 0, (1.6b) where one can interpret Eq. (1.6a) as a balance of internal and external forces and Eq. (1.6b) as a balance of internal and external moments. It is interesting to expand the vector equation Eq. (1.6b) into its projections onto the material frame, using the definitions in Eq. (1.1) along with the constitutive relations in Eq. (1.3), 19 EI1 κ01 (s) − EI2 κ2 (s)τ (s) + GJτ (s)κ2 (s) − F(s) · d2 (s) + q1 (s) = 0 (1.7a) EI2 κ02 (s) − GJτ (s)κ1 (s) + EI1 κ1 (s)τ (s) + F(s) · d1 (s) + q2 (s) = 0 (1.7b) GJτ 0 (s) − EI1 κ1 (s)κ2 (s) + EI2 κ2 (s)κ1 (s) + q3 (s) = 0, (1.7c) which produces coupled differential equations in terms of the unknown material curvatures, κ1 and κ2 , and twist, τ . Despite having assumed a linear response of the material and small strains, these equations are intrinsically nonlinear due to the underlying geometry. Once again, we have assumed for simplicity that EI1 , EI2 , and GJ are constant throughout the rod. Techniques to solve the Kirchhoff equations typically involve first integrating Eq. (1.6a) over the entire rod to compute the internal force, F(s). That solution is then used as an input to Eqs. (1.7), with the material curvatures and twist as the unknowns. Solving the coupled differential equations is difficult, however, as the material frame is not known a priori, and also depends on the material curvatures and twist. Solutions to the equations are not certain to be unique, and explicit solutions are often the exception. Presently, however, numerical tools are sufficiently sophisticated and powerful for solutions that satisfy equilibrium to still be found. The next section discusses a well-known historical example with closed form solutions: Euler’s elastica. 1.2 Buckling of a Rod We now consider the case of the buckling of a planar rod as a special case of the general model presented above, and restrict our discussion to two dimensions. A schematic diagram of this problem is illustrated in Fig. 1-2 (a), where a rod is clamped vertically at one end and is free at the other, with a vertical force, P , applied at the free extremity. We describe an arbitrary configuration of the rod in Fig. 1-2 (b) by the arc length, s (with s = 0 at the clamped end), and the local angle between the 20 a) b) Figure 1-2: The assumed configurations of Euler’s elastica, which has an explicit solution satisfying the Kirchhoff equations of equilibrium for a rod. a) Undeformed, straight configuration of a clamped-free, planar rod with vertical end force, P . b) Deformed configuration of the rod is described by the arc length, s, and the local orientation of the rod to vertical, θ(s). The material frame is shown, with d3 and d1 in the plane of deformation, and d2 perpendicular to it. rod and vertical, θ(s). In this configuration, the material frame is oriented such that d3 (s) and d1 (s) lie in the plane while d2 (s) is oriented perpendicular to the plane. We now seek to write an explicit equation for equilibrium of this configuration. We begin by noting that our restriction on the rod to deform in the plane has significant kinematic implications. The material frame cannot rotate out of plane, which implies that τ = κ1 = 0. Additionally, this restriction requires d2 to be perpendicular to P , which acts in the plane of the rod. Incorporating these kinematic constraints into our constitutive relationship of Eq. (1.3) between moment and material curvatures and twist, we can then write the internal moment, M(s) as, M(s) = EIκ2 (s)d2 (s), (1.8) recovering the classic moment curvature result [9]. We can then calculate the derivative of M(s), combining Eq. (1.8) with our kinematic equations for the material frame 21 from Eq. (1.1), as well as our specific kinematic assumptions for this configuration, M0 (s) = EIκ02 (s)d2 (s) + EIκ2 (s) (−τ d1 (s) + κ1 (s)d3 (s)) = EIκ02 (s)d2 (s). (1.9) Combining Eqs. (1.9) and (1.6b) for the balance of moments in our planar rod (with zero distributed torques, q(s) = 0), we can express equilibrium as, EIκ02 d2 + d3 × P = EIκ02 d2 + (d1 × d2 ) × P = 0, (1.10) where Eq. (1.10) can be simplified further with the vector identity (a × b) × c = −a(b · c) + b(a · c) and the definition of κ2 = θ0 , yielding the equation of equilibrium: EIθ00 + P sin θ = 0, (1.11) subject to the boundary conditions θ(s = 0) = 0 and θ0 (s = L) = 0. Our expression of equilibrium for a planar rod subject to a vertical load is the equation of Euler’s elastica, which was first derived by Euler in 1744 [10]. Levien [11] provides an excellent historical account of the elastica problem and the variety of techniques used to describe and solve Eq. (1.11). We proceed by following the presentation of Audoly [29] for the analysis of Eq. (1.11). A closed form solution can be found by multiplying both sides of Eq. (1.11) by θ0 and integrating, which introduces a new constant of integration. By enforcing the boundary condition at the free end, θ0 (s = L) = 0, we recover an expression for the curvature as a function of arc length: 0 θ (s) = ± r 2P (cos θ(s) − cos θ(L)) EI (1.12) which is a first order nonlinear equation in θ(s). By inspection, the trivial solution θ(s) = 0 satisfies Eq. (1.12) for all s, corresponding to the straight rod configuration, which satisfies equilibrium for all loads P . However, the straight solution is not always unique, and other solutions can be found for certain combinations of the load applied, 22 P , and the rod length, L. It is convenient to introduce the dimensionless length, p L = (2P L)/(EI), which combines the applied load with the rod’s elasticity and length. This variable can be understood as a measure of the applied load, with internal moments affected by increasing either P and/or L. Integration of Eq. (1.12), leads to the buckled solutions, that relates the dimensionless length to the rod orientation, L= Z 0 L ds = Z θ(L) 0 q dθ , (1.13) cos θ − cos θ(L) here for the case of positive curvature. In Fig. 1-3, we plot a numerical solution2 relating the tip angle, θ(L), to the dimensionless length, L. To find the transition from the straight configuration to these buckled configurations, we take the limit of Eq. (1.13) as θ(L) → 0, to find the critical value of dimensionless length, Lc , above which the rod buckles, π Lc = √ . 2 (1.14) This expression, in turn, yields the dimensional critical applied buckling load, Pc = π 2 EI , (2L)2 (1.15) above which, non-vertical configurations (buckled configurations) satisfy equilibrium. We can now characterize the system by plotting a characteristic order parameter of the configuration, such as the tip angle, θ(L), as a function of the loading, L, as shown in Fig. 1-3. In this plot, we see that for small loads, L < Lc , only one configuration satisfies equilibrium: the straight configuration (dark blue line with schematic configuration inset). For loads above the critical load, however, there are three possible configurations. The straight configuration is still a solution, albeit unstable, but there are also two buckled configurations (gold line with schematic configuration inset), which are symmetric about the vertical direction. This reflects 2 Numerical integration performed with Wolfram Mathematica. Closed form, explicit solutions to Eq. (1.13) exist, but are quite complicated. 23 6 Dimensionless Length, 5 4 3 2 1 2 1 0 Tip Angle, 1 2 3 Figure 1-3: Equilibrium configurations for a compressed rod, characterized by the tip angle, θ(L), for dimensionless length, L. For low values of the dimensionless length, only the trivial, straight, configuration satisfies equilibrium. Above a critical dimensionless length, Lc , however, two other, symmetric, equilibrium configurations, known as the buckled configurations, also satisfy equilibrium. that the rod is equally likely to buckle to either side. The buckled configuration can be solved for, assuming small values of L − Lc , by expanding the integral and inverting the resulting polynomial, recovering an approximate solution for a description of the buckled shape, θ(s) ≈ θ(L) sin where θ(L) = q √32 (L 2π πs 2L , (1.16) − Lc ). In the post-buckled regime, for loads well beyond Pc , further expansion or more sophisticated solution techniques are required [11]. For L > Lc , it can be shown that the buckled equilibrium shape will be preferred to the straight configuration. This is accomplished by computing the dimensionless total energy of the rod, E tot (assuming the shape defined by Eq. (1.16)), as the sum of elastic bending energy and the work done by the external force, P . The energy then becomes a function of L and θ(L). In Fig. 1-4, we plot this energy profile for two 24 different applied loads, both 2% away from Lc . For the subcritical load, L = 0.98Lc , the vertical configuration represents the minimum energy. It is referred to as stable, with small perturbations of θ(L) causing an increase in energy. As L increases to L = 1.02L, however, the vertical configuration becomes a local maxima with local minima at θ(L) ≈ 0.56 as predicted from Eq. (1.16). The vertical configuration becomes an unstable equilibrium configuration for L > Lc , as small perturbations in θ(L) will decrease the total energy. In this section, we have successfully applied the Kirchhoff equations of equilibrium to the idealized case of Euler’s elastica, recovering classic results and finding non-trivial buckled shapes past a critical applied load as an illustrative example. As they relate strongly to the remainder of this thesis, the following aspects of the results should be emphasized: the critical buckling load (Eq. (1.15)) and the amplitude of the buckled shape are material dependent, whereas the mode shape (Eq. (1.16)) is 1 -0.5 0.5 -1 -2 Figure 1-4: The total energy of a compressed rod, E, as a function of tip angle, θ(L), for a dimensionless length 2% below the critical buckling load (blue line) and 2% above the critical buckling load (gold line). For loads below the critical load, the straight configuration is the minimum energy. Beyond the critical load, the energy becomes meta-stable, with the two buckled configuration (θ(L) ≈ 0.56) local energy minima - the rod will buckle away from the straight configuration with any slight perturbation. 25 material independent. That is to say, given matching boundary conditions, bodies which are long in one dimension will assume universal shapes - the geometric nonlinearities which resulted from the equilibrium equations of Eqs. (1.7) cause macro-scale similarities in geometry, independent of material properties. Beyond the small strain assumption, behavior for different materials will become unique as the constitutive equations can introduce additional material nonlinearities. 1.3 Outline of the Thesis The general framework for the mechanical description of rod behavior has been developed in this introduction. Chapter 2 summarizes results in the literature and highlights industrial motivations specific to the problems investigated in the remainder of this thesis. Each of the subsequent four chapters proceeds with the progression of describing the experimental apparatus built and the methods developed to explore the phenomena of interest, followed by a presentation and interpretation of the results obtained. The order of the chapters reflects increasingly complex problems, where existing theory becoming less extensive. Chapter 3 investigates the role that natural curvature plays in affecting the qualitative and quantitative behavior of a suspended rod subject to a body force in the form of self-weight. This first model experiment makes use of a novel rod fabrication technique developed to give precise control over natural curvature as an independent parameter. Rods are suspended under their own weight from one end, and assume either planar or non-planar configurations, with non-planar configurations further classified as having localized helical structure or global helices. The observed geometries are then rationalized, with excellent agreement found between experiments, simulations, and theoretical predictions. We also discuss the related writhing problem, which consists of clamping both ends of a naturally curved rod and quantifying its behavior under imposed displacement and twist at the boundaries. Chapter 4 describes an experiment that was designed and built to compress a fixed length of rod inside of a horizontal cylinder, which we refer to as the classic case 26 experiment. Unlike Chapter 3, we consider only naturally straight rods, investigating the effect of changing the diameter of the constraining pipe on the critical buckling loads, with straight, sinusoidal, and helical rod configurations observed in all cases. Imperfection in the geometry of the constraining pipe is found to strongly affect the helical buckling load. Chapter 5 deals with the progressive injection of a rod into a glass pipe; the real case experiment. In this configuration, friction opposes insertion and creates an axial compressive load in the rod, portions of which undergo a primary instability into a sinusoidal mode and a secondary instability into a helical configuration. Once again, the size of the nonlinear constraint compared to the rod diameter has a strong effect on the amount of rod that can be injected prior to helical buckling. The experiments also indicate that natural curvature of the injected rod can play a role in causing helical buckling to occur earlier in the injection process, which is both relevant and undesirable in the industrial application. Chapter 6 presents the investigation of a mitigation technique that allows more rod to be injected into a cylinder before helical buckling occurs. This is accomplished by vertically vibrating the constraint, which we refer to as the dynamic real case. To our knowledge, this mechanism has not previously been explored in the existing literature. For sufficiently high peak accelerations of vibration, the rod is observed to lose contact with the constraint, thereby destabilizing frictional interactions and delaying helical initiation. Results show that injection speed and vibration frequency also play a role in determining the amount of improvement in total injected length possible. Finally, Chapter 7 summarizes the major results of the thesis and discusses potential avenues for future research. 27 28 Chapter 2 Literature Review Chapter 1 introduced how rods are mechanically described, and the case of planar buckling was considered. This chapter proceeds by expanding the review and discussion of three problems that have been previously studied in the existing literature and are relevant to this thesis: • §2.1 Describes relevant results in the behavior of a rod under applied tension and moment, the so-called writhing problem, which is a canonical system in the study of rods. These results will be directly applied in Chapter 3. • §2.2 Summarizes the buckling of a rod inside a cylindrical constraint under imposed end-displacement - referred to as the classic case. This will serve as the foundational work of Chapter 4, in addition to being essential to Chapters 5 and 6. • §2.3 Discusses how results from §2.2 have been applied to the real case, wherein a rod is injected into a cylinder, and frictional loading leads to a similar, albeit more complicated, buckling process as observed in the classic case. The work in this section will be applicable to Chapters 5. The review presented in this chapter is by no means an exhaustive list of all the contributors to these problems, as the study of thin rods has a long history in the mechanics literature. Instead, it is meant to highlight the past work most relevant to the following chapters. 29 2.1 2.1.1 Behavior of Naturally Curved Rods Motivation The introduction mentioned a plethora of motivations for the study of rods, particularly those with natural curvature, both in natural and engineered structures. In this section, we wish to focus on two practical scenarios, exhibiting similar mechanical behaviors: the deposition of subsea cables and pipe [26, 27, 30] and the behavior of DNA [8, 20, 31–34]. Both topics are discussed in the thesis of Goyal [35]. The longest subsea power cable in the world at the time of writing is the NorNed HVDC cable, connecting Feda, Norway to Eemshaven, Netherlands. At its deepest depth, the cable is deposited into 420 m of water, with an elliptic cross-section 21.7 × 13.6 cm (length/width ∼ 19,000) [36]. For comparison, a single human chromosome is on the order of 2 cm long, with a diameter of 2 nm (length/width ∼ 10,000,000) [35]. In either case, the structural element is very long in the axial direction in comparison to its two other cross-sectional dimensions, and can be treated as a thin rod in the mechanical sense so Kirchhoff’s equilibrium equations are applicable. In subsea cable deposition, a cable (such as fiber optic cable) or pipe is lowered from a ship to the subsea floor. During the descent from surface to bottom, hydrodynamic loading can induce torsion or twist into the cable. The rod, due to its buoyant weight, is under a decreasing amount of tension along its arc length, with maximum tension at the ship and minimum tension at the sea bed. Near the ship, the tension prevents the rod from losing its linear shape. As we traverse down the arc length of the rod, however, the decreasing tension can be observed to complicate mechanical behavior, with geometric nonlinearities appearing. This situation is shown in Fig. 2-1, which was extracted from [35]. Below a critical tension, the rod will form kinks, loops, and tangles, which are areas of large deformations due to bending and twisting of the rods. The problem is also present in buoy moorings, where the buoy may undergo large amplitude yaw, imposing twist in the cable between buoy and anchor. The cabling in both cases is frustrated by 30 a rod [32–34, 37, 38]. Two types of typical DNA strand configurations are shown in Fig. 2-2 (extracted from [39]), with overall strand lengths typically ranging from a few microns to millimeters. Both shapes exhibit a feature known as supercoiling, which is the combination of bending and twisting over the length of the DNA molecule. The top configuration in Fig. 2-2 is known as a plectoneme, with multiple points of self-contact, while the bottom configuration is known as a solenoid, and consists of a more helical structure, with no points of contact. The plectoneme-form of DNA is the most commonly observed morphology [39]. The mechanical state of stress of these supercoils has been found to affect the behavior of enzymes within the cell [40, 41]. In the cases of both subsea cables and DNA, mechanical instabilities lead to highly nonlinear geometries. This can be harmful in the case of subsea cable hockling or +$)%",-)',!'(.!-'/#,&+.0'#&1,!,"&*',)'233&)*,4'5'(.-)6'7,"8'!-9&'(**,",-)(.' useful in the case of DNA packing. The following review of existing literature will give -#9(",-):*&"(,.!;' insight into the instabilities present in these problems, for both cases of a perfectly naturally straight rod and a rod with intrinsic natural curvature. ' -' ./012' 02' 3/%&&' 4&2#3/' .564&.)' 78644&.3' .564&' 94&:3;' ./01.' <0$=4&>/&4"?' .3%$53$%&' Figure 2-2: A supercoiled DNA typically on/01' the order of micron to millime3&' 5/6"2.' 62<' =6.&>@6"%.;)' A23&%8&<"63&' .564&'strand, 98"<<4&;' ./01.' .&B&%64' <0$=4&> ter scales. The top two images illustrate plectonemic DNA supercoiling, involving 023"2$0$.'@"&5&'0:'<0$=4&>.3%62<&<'+,-)'C6%#&.3'.564&'9%"#/3;'./01.'/01'3/&'.3%62<' significant amounts of self-contact in a highly0%'twisted configuration. The bottom &.' 62<' 31".3.' "2' :0%8"2#' .$@&%50"4.' 902&' "23&%10$2<' @4&5302&8"5E' 62<' 02&' two images show solenoidal supercoiling, with a single helical structure involving no $%3&.DG'H%62<&2'62<'I00J&'K(LM'62<'C&/2"2#&%'&3'64)'KNM;)' self-contact. Image extracted from [39]. 32 !"#("&!'('>?2'9-.&%$.&'-)'"8#&&'*,++&#&)"'.&)6"8'!%(.&!'(!'#&3#-*$%&*'+#-9' !'@5ABC;'D8&'!9(..&!"'.&)6"8'!%(.&'E+(#'.&+"F'!8-7!'('!&69&)"'-+'"8&'+(9,.,(#' 2.1.2 Previous Work The non-linear nature of the Kirchhoff equations, Eq. (1.6), that describe rod behavior make exact solutions uncommon in the existing literature. The addition of natural curvature makes the closed-form solutions to problems even less common. This section deals with a canonical system whose schematic diagram is shown in Fig. 2-3, that has proven to be remarkably fertile for researchers. It consists of an infinitely long and weightless rod, clamped at both ends, with an applied end moment, M , and tension, P . A variety of solutions exist, each considering a different case: i) intrinsically straight rod under compression (P < 0) [42], ii) Intrinsically straight rod under compression and twist [43], iii) intrinsically straight rod under tension and twist [42], iv) intrinsically curved rod under twist [44], and v) intrinsically curved rod under tension [45]. The different cases can be distinguished by the rod considered, with it assumed to be naturally straight or including natural curvature. Results from both assumptions are summarized below. Intrinsically Straight Rod In the case of an infinitely long, intrinsically straight rod, any non-zero compression or moment will induce buckling. Under combined tension, P , and moment, M , however, as shown in Fig. 2-3, the rod may remain stable in a linear configuration (transparent blue curve in Fig. 2-3), or may buckle into a modulated helical structure (solid red curve in Fig. 2-3). Several authors [27, 30, 42, 43, 46, 47] have derived the critical end moment for the loss of stability, Mc , to be, √ Mc = 2 EIP , (2.1) with several authors noting that the equilibrium equations of the rod undergo a Hopf bifurcation at this buckling load [48, 49]. Thompson and Champneys [50] have found that, at the onset of buckling, the helix assumes an axial pitch length given by Lc = 4πEI/Mc . Champneys et al. [44] showed 33 Figure 2-4: Test progression of a square cross-section rubber rod which was cast with zero natural curvature and tested at neutral buoyancy. The rod remains straight (top photo) until it enters a ∼ 3 twist per wave localized helical solution (bottom photo). Images from [44]. result of Eq. (2.2) can be compared to the result of Greenhill [53], who, in 1883, derived using the theory of small displacements the critical load for the case of a finite rod under compression and end torque, finding, m2 − 4p = 16π 2 , (2.3) where m and p are the same dimensionless variables introduced above. In Fig. 2-5, we compare the three predictions (infinite length rod, finite length according to Greenhill, and finite length according to van der Heijden et al.), and plot the corresponding buckling criteria, where the lines represent critical combinations of moment, m, and tension, p, that are sufficient to cause a straight rod to buckle. In the case of an infinite rod, as predicted by Eq. (2.1), any applied compression or moment will cause the rod to buckle. When considering the rod to be of finite length with Eq. (2.2), however, the critical buckling load in pure compression (m = 0) matches that predicted by Euler 35 4 Infinite Rod (2.4) Finite Rod (2.6) - 1st mode Finite Rod (2.6) - Other Modes Finite Rod (2.7) 3 2 1 Straight 0 −1 Mode=2 Buckled −2 Mode=3 −3 −4 −4 Mode=4 −3 −2 −1 0 1 2 3 4 Figure 2-5: Relation between applied moment, m, and tension, p, for a straight rod to buckle. For the case of an infinite rod (thick dark blue line), any compression or applied moment will cause the rod to buckle, whereas in the case of a finite rod (light blue line), the straight rod will remain straight for compression and moment until a critical buckling load is reached. Thin blue lines represent higher buckling modes of a finite length rod. The gold line represents the first historical prediction for buckling under combined moment and compression. buckling, corresponding to p = −1 on the plot. Greenhill’s prediction (Eq. (2.3)) also agrees with Euler’s buckling load in pure compression, but then his solution asymptotes to the second buckling mode predicted by Eq. (2.2). Maddocks later showed that the first mode of buckling under pure compression was the only stable one [54]. Once the rod buckles out of its straight configuration (into a planar configuration if m = 0 and a helical configuration otherwise), further increases in load lead to a secondary instability. Once this instability occurs, the rod jumps into self-contact 36 with the formation of a loop. For an infinitely long rod, Thompson and Champneys √ [50] found this instability to happen at M = Mc / 2. Furthermore, this loop was predicted to form at the center of the helical structure, where the amplitude of the initial buckled deformation was largest. Prior to Thompson and Champneys’ analytic work, the formation of a loop was also of interest to engineers in the field of submarine cabling, where loop formation was often associated with cable damage and/or failure. To investigate the effect of loop formation, some experimental studies were conducted [26, 30, 55], an example of which is shown in Fig. 2-6 from [30]. Photographs were taken from a test with a multi-stranded steel cable (diameter, d = 3.8 cm, and length, L = 6.4 − 7.2 m) placed into tension with a dead load and then twisted. Fig. 2-6 (a) shows that the rod is straight for low values of imposed twist. For increasing twist, Figs. 2-6 (b) - (d), the cable buckles into the predicted helical structure. Eventually, in Fig. 2-6 (e), the cable buckles into a localized configuration, which is commonly referred to as a kink, hockle, or plectoneme. Van der Heijden and collaborators [42, 43, 56] studied (both analytically and experimentally) the scenario of localized buckling shown in Fig. 2-6 (e) by considering a finite-length rod under either fixed end rotation, Φ, or end displacement, δ. Introducing a “semi-infinite” approximation, the researchers found a critical combination of Φ and δ for the jump to self-contact (formation of a loop) to be governed by, 2 Φ= γ r 8 4γ + d2 d r 1 γd − + 4 arccos 2 4 r 1 γd − , 2 4 (2.4) where d = δ/L, γ = GJ/EI, and 0 ≤ d ≤ 2/γ. For end displacements d > 2γ, a different form of instability (governing the behavior of looped planar elastica) is relevant for the rod’s stability [25], but is not of immediate interest here. The term γ represents the ratio of twisting and bending stiffnesses of the rod. Miyazaki [57] also analytically treated finite length rods, although results were presented as highly implicit functions, making them difficult to compare to those of van der Heijden et al. For finite length rods, van der Heijden et al. [43] found that for small end rotations 37 -J& i* (a) Five turns (b) (e) Twenty-two turns Ten turns (c) Fifteen (d) turns Twenty turns Figure 16. Stages of kink formation on 1x48 cable during negative torquing. Figure 2-6: Photographs of cable kink test from Liu [30]. An initially straight, multistranded steel cable (diameter, d = 3.8 cm, and length, L = 6.4 − 7.2 m) is put in tension (load applied at bottom of figure, clamped at top), and then the bottom is twisted. The initially straight cable (a) is seen to first take on a helical configuration with growing amplitude, (b) through (d), and then a kink (also known as a hockle or a plectoneme) forms at twenty-two turns (e). Note that the strand separation in the kink in (e) indicates damage to the cable. (Φ < 2π),(a)there instability involving a jump to self-contact over the applicable One is andno one half (b) Two and one half (c) Three turns turns turns range of 0 ≤ d ≤ 2/γ. For larger end rotations (Φ > 2π), however, there is the same jump to self-contact as discussed for the infinite length case. This dependence on Φ was not observed for the infinite length case, where the secondary instability involving jumping to self-contact was always predicted. Fig. 2-7, extracted from [43], illustrates the case of a rod a) with high end rotation (Φ = 11.4 radians) jumping to selfcontact (Configurations A3 and A4 are snapshots immediately before and after the (d) Four turns (e) Closeup of link instability, respectively) and b) with low end rotation (Φ = 4.265 radians) exhibiting Figure 17. Various stages of kink formation on 1x48 no secondary instability, with the fixed ends positive passing one another in Configurations double -armored cable during torquing. B3 and B4. In the special case of Φ = 0 (no end rotation), the first buckling mode of the 33 rod remains planar. A secondary buckling mode involves the buckled shape going out of plane, taking on a shape similar to Configuration B2 in Fig. 2-7 (b). The 38 a) A1 A2 A3 A4 b) B1 B2 B3 B4 Figure 2-7: A finite length rod with fixed end rotation Φ undergoing imposed end displacement d. a) Under high end rotation (Φ = 11.4 radians), the rod undergoes a buckling instability in Configuration A1 and experiences a jump to self contact at dj = 0.504, Configurations A3 and A4. b) For low end rotation (Φ = 4.265 radians), the rod undergoes a buckling instability in Configuration B1, but then does not have a jump to self-contact, rather taking on a planar loop elastica configuration in Configurations B3 and B4, for d ≥ 1. Figure extracted from [43]. first buckling mode is the Euler buckling load, p = −4π 2 . Van der Heijden et al. [42] provide closed form solutions for the load-displacement behavior for the planar buckled configuration, as well as a solution for the bifurcation point for the rod to buckle out of plane. The load displacement curve is given parametrically by, p = −16K 2 (k) E(k) , d=2 1− K(k) (2.5) where K(k) and E(k) are the Legendre complete elliptic integrals of the first and second kind, respectively, and k is the elliptic modulus, 0 ≤ k ≤ 1, commonly used as the argument parameter for elliptic integrals. The critical point for bifurcation of the planar shape into the non-planar shape is a function of the ratio γ = GJ/EI between 39 twisting and bending stiffness only. The critical point is γ dependent because the instability is a transition from a purely bending mode (planar) to a mixed bendingtwisting mode (a non-planar loop). For a solid rod with a circular cross-section (the following chapters all deal with such rods), γ = 1/(1 + ν), where ν is the Poisson’s ratio of the material. To find the critical load and displacement, one must solve for the elliptic modulus, k, in, K(k) [2(1 − k 2 )K(k) + (−3 + 4k 2 )E(k)] γ= , 2(1 − k 2 )K 2 (k) + (−5 + 4k 2 )K(k)E(k) + 3E 2 (k) (2.6) and then input the value back into Eq. (2.5). For a solid rod with solid and incompressible cross-section (such as rubber), this corresponds to k = 0.5467, which gives the critical values for compression, p = −5.8263π 2 , and end displacement, d = 0.6003. Thus far the writhing problem has been introduced for both infinitely long and finite length, intrinsically straight, rods. Solutions for the initial buckling (either planar or modulated helix depending on the end rotation Φ) were provided, as were solutions to a secondary instability, consisting of a jump to self-contact or an out of plane buckling. Other authors have considered variations of this problem, most notably those considering a variety of boundary conditions [54, 58], non-symmetric cross-sections [52], as well as those dealing directly with computing self-contact of the rod [16, 31, 34, 59]. We now move on to work relevant for naturally curved rods. Intrinsically Curved Rod Champneys and colleagues [44] noted that in qualitative experiments with rubber rods, the expected progression of an intrinsically straight rod under imposed end rotation at fixed end displacement (straight, localized helix, loop) was not strictly observed. Instead, upon twisting the rod, a global helical shape was observed, as seen in Fig. 2-8 from [44] which presents a sequence of photographs from an imposed end rotation test. These results show that prior to the localized helical structure that is expected at high end rotation (as seen in the bottom four photographs), there is a one-twist-per-wave global helical structure observed. This effect was attributed to 40 the natural curvature of the rod caused by spooled storage. Champneys et al. [44] went on to derive the general equilibrium equations for a rod including natural curvature, showing that the natural curvature acted as a perturbation to the Kirchhoff equations of equilibrium. The straight configuration of a twisted rod before localized buckling was indeed replaced by a one-twist-per-wave global helical structure, one whose amplitude grows under continued end rotation. √ They defined a dimensionless load, m = M/ EIT , and dimensionless natural curvature, κ0 = EIκ0 /M (they also assumed that κ0 was small). Numerically solving their equations of equilibrium for an infinite length rod, they again found a Hopf bifurcation for loads m > mc , remarking that the value of mc was initially unchanged and then decreased with increasing κ0 . Goriely and others [4,60–64] have taken an alternative approach to the modeling of rods with natural curvature from the methods mentioned thus far. These researchers worked within the framework of time-dependent Kirchhoff equations for rod equilibrium, studying the linear stability of the solutions, as well as performing post-buckling analysis. After first applying their method to intrinsically straight rods [60–63], they then addressed problems including naturally curved rods [4, 64]. A result of particular interest for our work was the rationalization of the curvatureto-writhe instability. Here a rod starts with an applied tension and zero imposed end rotation, with both ends fixed in rotation (resulting in zero total twist in the rod throughout the duration of the test). For sufficiently large tension, the rod remains in a linear configuration. As tension is progressively released, however, the rod is observed to go through an instability whereby the rod takes on a helical configuration. However, to maintain the zero total twist condition imposed by the boundaries, the rod assumes the shape of two helices with opposite chirality, connected at the middle of the rod by a chirality inversion, or perversion. This instability is shown in Fig. 2-9 as a cartoon taken from [45]. In the straight configuration, the bending energy is high as the material curvature (zero in the case of a straight rod) is not close to the natural curvature. Upon relaxation of the tension, the twisting energy increases with the formation of a helix, but the bending energy decreases as the material curvature 41 Figure 2-8: Photographic sequence of a test with a naturally curved rod under imposed end rotation (increasing rotation for lower pictures). During early stages of test (top four photographs) a one-twist-per-wave helical configuration is observed before the localized helical buckling shape is observed (bottom four photographs). Images extracted from [44]. 42 242 T. McMillen and A. Goriely Fig. 1. A cartoon of the curvature-to-writhe instability; as the tension is decreased, the instability sets in and two helices with opposite handedness are created: a perversion. Figure 2-9: Cartoon representation of the curvature-to-writhe instability. A naturally curved coiling rod under tension initially assumes a straight and linear configuration. of strings, ropes, or telephone cords. If you take a piece of rubber tubing, hold For decreasing tension rod instability, taking onThis a configuration it between your the fingers, andwill twistundergo its ends, thean filament will soon coil on itself. is an example of a writhing instability where a local change in twist eventually results in a consisting of two helices of opposite chirality connected by a chirality inversion, also global reconfiguration of the filament. In this case we have a twist-to-writhe conversion. known as a perversion. Illustration extracted from [45]. The word writhe refers to global deformation of a filamentary structure. This type of instability has received considerable interest and is known to be important in processes such as coiling and super-coiling of DNA structures [1], [2], [3] and morphogenesis in bacterial filaments [4], [5], [6]. typecurvature. of writhing instability is the curvature-to-writhe instability where approaches theAnother natural changes of curvature trigger global shape reconfigurations [7]. This instability can also be observed in telephone cords. If one completely untwists the helical structure of the Goriely and Tabor [4],ends, and subsequently McMillen and Goriely [45], cord and pulls the a straight cord can be obtained. Now, if one slowly releases the studied the ends, the filament suddenly changes shape to a structure composed of two helical struccurvature-to-writhe problem with the and inversion static Kirchhoff equations of equitures with opposite handedness and dynamic linked by a small (see Fig. 1). We refer to this structure as a perversion. The German mathematician J. B. Listing [8], [9] refers librium, respectively. For the case of an infinitely they tofound that the critical to an inversion of chirality as perversion as used by long D’Arcyrod, Thompson characterize seashells: “the one is a mirror-image of the other; and the passing from one to the other tension, Pcr ,through belowthewhich straight rodhaswas unstable is, plane ofasymmetry (which no ‘handedness’) is an operation which Listing called perversion” [10, p. 820]. Maxwell, in his 1873 treatise on electromagnetism, also uses the word perversion: “They are geometrically alike in all respects, except that one is the perversion of the other, like its image in a looking glass” [11]. The usage of the 2 rare left-handed specimens of word perversum actually originated in the description of Pcr = (1 + ν)EIκ0 , (2.7) where all parameters are defined as before. For a perfectly straight rod, κ0 = 0, no values of tension allow for the curvature-to-writhe instability to occur. Instead, the tension for an instability would be negative, i.e. equal to the Euler compressive load, as was discussed in the previous section. For increasing κ0 , the critical tension increases. The authors also found an expression for the case of a rod with finite length, where the critical tension in Eq. (2.7) was decreased as, 43 Pcr (L) = (1 + ν)EIκ20 − EI . L2 (2.8) Goriely and colleagues assumed that the asymptotic helices that are formed (far away from the chirality inversion) adopt a shape such that their material frame coincides with the Frenet frame (an adapted frame whose normal director lies in the plane of greatest geometric curvature of the centerline [29]) for rods with natural curvature. This implies that the axis of the asymptotic helices (the imaginary axis that the helix is wrapped around) is perfectly aligned with the applied tension, which is a common assumption made in the study of stability of helical filaments [65–67]. This assumption allows for a simplified description of the asymptotic helices that are formed when P < Pcr . First of all, the geometry of a constant helix can be described by the radius (of the helix), r, and pitch length (also known as the step or wavelength), 2πp. These two quantities are related to the Frenet curvature, κF , and twist, τF , through the relations, r + r2 p τF = 2 p + r2 κF = p2 (2.9) McMillen and Goriely [45] also relate κF and τF with the natural curvature, κ0 , of the rod, τF2 = p (1 + ν)κF (κ0 − κF ) (2.10) where κF < κ0 (as the rod is between straight and the natural curvature for 0 < P < Pcr ). To find the exact shape of the asymptotic helices, the applied tension is related to the Frenet curvature and twist through the relation, P = EIτF ( q 1 κ0 −1+ ) κ2F + τF2 . κF 1+ν (2.11) Finally, Eqs. (2.10) and (2.11) can be combined to yield an expression that relates the applied tension to the curvature or twist of the asymptotic helices away from the 44 chirality inversion, P (1 + ν)EI 2 = κ0 − κF (1 − 1 ) 1+ν 3 (κ0 − κF ). (2.12) Using Eq. (2.12) and it’s analogue for tension and twist, one can describe the exact geometry of an asymptotic helix for an infinite rod as a function of applied load. For any load there will be two solutions, corresponding to opposite chiralities. The methods of Goriely et al. [4, 45] allow for the description of the chirality inversion as well, but require numerical solution for specific parameters, instead of closed form solutions. Thus far we have seen that there has been a rich body of work devoted to the study of elastic rods under a variety of loading conditions. In the case of a perfectly straight, twistless, finite length rod, compressive forces can be sustained before buckling occurs. In all other instances (naturally curved, imposed end rotations, or infinitely long rods), an instability occurs for sufficiently low tensile forces. In these derivations, sophisticated continuation techniques were often ported from the field of dynamic stability analysis [43, 68]. This illustrates the sophistication needed to exactly solve the highly geometrically nonlinear problems presented by rods. Moving to more complicated centerline geometries or external loadings (e.g. non-symmetric, spatially varying, or non-conservative forces), numerical tools (e.g. shooting methods, finite difference, and finite element) are generally preferred. Sometimes, as in the next section, the centerline geometry resulting after an instability is assumed to take a specific form, making closed form solutions possible. 2.2 2.2.1 Compressing a Rod Inside a Cylinder Motivation Drilling methods for oil and natural gas have evolved dramatically over the last century [69]. In most cases, a borehole is originally drilled using a drill string composed 45 of 90-ft-long segments of drill pipe screwed together with a bottom hole assembly (mainly consisting of instrumentation, a mud rotary motor, and a drill bit) at the string’s downhole extremity. During this process, the top end of the drill pipe is clamped at the drill rig with fixed rotation (either no rotation or a set rotation speed) and the bottom end of the drill string is pressed against the formation being penetrated. Mechanically, the drill string can be considered a rod as its diameter (typically 10 cm or smaller [70]) is small compared to its length (on the order of 1 − 10 km). Advancement of the borehole during drilling requires that a compressive force be applied to the top end of the drill string. The drilling assembly is often steerable, such that the wellbore can be drilled according to a specified trajectory [71]. Recently, there has been a surge in so-called horizontal drilling, where the drilling inclination (angle from vertical) exceeds 80◦ , as shown in Fig. 2-10, where we plot the proportion of active drilling rigs in North America drilling horizontal wellbores. These wells result in greater contact between the well and the formation, leading to an improvement in production [72]. The horizontal distance, also known as horizontal departure, that can be drilled using current methods is remarkable. The current record for horizontal departure is 11,569 m at the Al-Shaheen field BD-04A well in the Persian Gulf off the coast of Qatar [71]. During drilling, large compressive forces are applied to the drill pipe, a slender rod (diameter on the centimeter scale), inside of a cylindrical constraint. Past a critical load, this leads to buckling of the drill string. This buckling can immediately effect the trajectory of the wellbore in addition to the amount of force transmitted from the drill rig to the drill bit. In the longer term, the buckling can negatively affect the service life of the drill string. Mechanically, we describe this problem as a fixed length of rod compressed inside of a cylinder, a situation we will refer to as the classic case, which will be investigated in detail in Chapter 4. Similar to the unconstrained rods of §2.1, the constrained rods of this problem experience a sequence of two buckling instabilities. The first instability is a rather benign sinusoidal buckling mode, which does not significantly effect the rate of penetra46 100 Horizontal Wells % of Active Rigs (N. America) 90 80 70 60 50 40 30 20 10 0 1993 1998 2004 2009 Year Figure 2-10: Percent of active rotary drill rigs in North America drilling horizontal wells as a function of time. Note that starting in early 2010, horizontal drilling accounted for over half of the active wells being drilled. Data from Baker Hughes [73]. tion of drilling nor excessive damage to the compressed rod. However, the secondary instability, which results in a helical configuration, can lead to damage of the rod or the constraining cylinder. We will see in §2.2.2 that analytically, authors simplify the problem of §2.1 by assuming a post-buckled shape and finding the resulting criteria for instabilities. These assumptions are made out of necessity; the nonlinear geometry of the cylindrical constraint precludes closed form solutions to the equations of equilibrium without them. 2.2.2 Previous Analytical Work Unlike §2.1, where exact solutions to Kirchhoff’s equations of equilibrium (Eq. 1.6) could be found for the shape a rod assumed after a mechanical instability, the classic case does not have exact solutions for the buckling and post-buckling behavior of the rod. Instead, authors tend to assume buckled shapes and then calculate the critical loads the transition into those configurations. 47 Figure 2-11 shows the assumptions common to the theoretical work discussed in this section. In Fig. 2-11 (a), an initially straight (blue cylinder in Fig. 2-11 (a)) rod is compressed inside of a constraining cylinder that is inclined from vertical by an angle α. The rod is compressed by two forces, Pin and Pout , applied at the input and output end, respectively. For cases with no frictional interactions, Pin = Pout . Past a critical applied load, Pcrs , the rod buckles into a sinusoidal configuration (green cylinder in Fig. 2-11) with a characteristic wavelength, λ. With increasing applied load, the rod undergoes a secondary instability at the critical helical buckling load, Pcrh , changing into a helical configuration with constant pitch length, p. In Fig. 2-11 (b), an axial view of the rod in its straight (blue), sinusoidal (green), and helical (red) configurations. In all three configurations, authors assume the rod to be in perfect contact with the cylinder. The constraining geometry is defined by the radial clearance, ∆r, defined as half the difference between the cylinder and rod diameters. The amplitude of the sinusoidal configuration is defined by the angular amplitude, β0 , shown in Fig. 2-11 (b). a) b) Figure 2-11: a) Side view of the classic case. An initially straight (blue) rod is compressed inside of a cylindrical constraint that is inclined an angle α from vertical by forces Pin and Pout (with no friction, Pin = Pout ). The rod first buckled into a sinusoidal configuration (green) with a characteristic wavelength, λ, and then into a helical configuration with a constant pitch length, p. b) Axial view of the classic case. The constraint geometry is defined by the radial clearance, ∆r. The sinusoidal configuration has angular amplitude, β0 . In all cases, the rod is assumed to be in perfect contact with the cylinder. 48 This section summarizes results in the literature relevant to this project, specifically: i) the critical load for sinusoidal buckling, Pcrs , the buckling wavelength, λscr , and the normal contact force, Wn , between the sinusoidally buckled rod and pipe, ii) the critical load for the rod to buckle into a helical configuration, Pcrh , the buckled pitch length, phcr , and the normal contact force, Wn , between the helical rod and pipe, iii) the effect on results considering non-zero frictional interactions between the rod and pipe, and iv) the effect of imperfections on critical loads. Sinusoidal Buckling Lubinski [74] was the first to consider buckling of a drill string in a borehole. He derived the first critical buckling load for a vertical wellbore (α = 0 in Fig. 2-11) filled with drilling fluid using small-displacement equilibrium methods. With drilling fluid, the drillstring has a buoyant weight, w = ρAg, where ρ is the volumetric density of the drill string material minus the density of the drilling fluid, A is the crosssectional area of the drill string, and g is the acceleration due to gravity. Assuming the rod to be straight and not in contact with the borehole, Lubinski derived the first √ 3 critical buckling load, Pcr = 1.94 EIw2 , where E is the Young’s modulus and I is the moment of inertia of the pipe cross-section. Above this critical load, Lubinski claimed that: i) the drill string would come into contact with the borehole wall, and this point of tangency on the drill string would fatigue faster than other sections, and ii) drilling progress would deviate from vertical, as the bit inclination would change with the new buckled configuration. This buckled configuration corresponded to a half of a sinusoidal wavelength at the bottom of the drillstring. Lubinski derived a √ 3 secondary buckling load at Pcr = 3.64 EIw2 , which corresponded to a full buckled wavelength. In this first derivation, the radial clearance of the borehole did not factor into the calculation, but it raised awareness of the potential for buckling in the drilling industry. Paslay and Bogey [75] investigated the case of an inclined borehole with a rod inside of it (0 ≤ α ≤ π/2 in Fig. 2-11). For inclined boreholes, the rod was assumed 49 to be perfect contact with the borehole, and that contact was frictionless. Paslay and Bogey included the effects of end thrust and moment for the constrained drill string. They derived a fourth-order differential equation for the configuration using energy minimization techniques and considered two simplified cases: the weightless rod (w = 0) and the horizontal configuration (α = π/2). In the case of a weightless rod, they found the critical load for the onset of sinusoidal buckling, Pcrs , to be, Pcrs = nπ 2 (1 − ν)2 , EI (1 + ν)(1 − 2ν) L (2.13) where ν is the Poisson’s ratio (≈ 0.3 for steel), n is a positive integer value corresponding to the mode number (number of half wavelengths), and L is the rod length. The prefactor in Eq. (2.13) that is a function of Poisson’s ratio is of order 1, so the critical load reduces to approximately that of Euler buckling (minimized with n = 1) in the case of the weightless rod [9]. This result can be interpreted as the inclination of the constraint as well as the radial clearance, ∆r, do not influence Pcrs unless the constrained rod has non-zero buoyant weight. In the horizontal configuration (α = π/2) of a rod with weight, the researchers derived Pcrs as, Pcrs = EI nπ 2 L w + ∆r L nπ 2 , (2.14) where ∆r once again is the radial clearance between the drill string and wellbore, n is the number of half wavelengths of the buckled shape, n = λ/2. The first term in Eq. (2.14) corresponds to the classic Euler buckling load for a column under compression, and the second term is the effect of the rod (with buoyant weight w) climbing the cylindrical constraint, raising the gravitational potential energy. This form for the buckling load is identical to a beam on an elastic foundation with elastic foundation (often referred to as a Winkler foundation) constant w/∆r [1,9]. The last feature of note of Eq. (2.14) is that the number of half wavelengths in the primary (lowest energy) mode of buckling is length-dependent. This result differs from that for unconstrained buckling of columns, where the first mode always consists of one half wavelength (n = 1). These critical sinusoidal buckling loads for different values of n 50 are presented in Fig. 2-12 using typical mechanical properties found in the oilfield [1]: E = 207 GP a, I = 9.97 × 10−6 m4 , w = 315 N/m, and ∆r = 4.1 cm. For sufficiently “long” cylinders – typically L ≥ 5λ – the critical load becomes practically independent of length [1]. This can be seen in Fig. 2-12 by the curves of Pcrs getting progressively flatter near their optimum rod lengths. This asymptotic, length independent value of Pcrs is given by, Pcrs r =2 EIw , ∆r (2.15) where we see the most basic ingredients of the problem: buckling is resisted by the bending stiffness of the rod and the cost in potential energy to lift the buoyant weight up the curvature of the constraining rod. Eq. (2.15) is the most commonly used and quoted value of the critical force for sinusoidal buckling, and the one we shall refer 5 5 x 10 Critical Buckling Load, [N] 4.5 4 3.5 3 2.5 n=1 n=2 n=3 n=4 n=5 2 1.5 Sinusoidal Mode Asymptotic Sinusoidal Mode 1 0.5 0 0 20 40 60 Rod Length [m] 80 100 Figure 2-12: Critical sinusoidal buckling load, Pcr2 , as a function of rod length. For a given rod length, a particular mode number, n, minimizes Pcrs , although for sufficiently long cylinders (L & 5λ), Pcrs reaches an asymptotic value (dashed line). Material and geometric properties taken from [1] are EI = 2.06 N m2 , w = 315 N/m, and ∆r = 4.1 cm. 51 to as the critical sinusoidal force henceforth. Dawson and Paslay [76] showed that this force can be made more generic for the case of rods with self-weight in inclined boreholes by replacing the buoyant weight by the vertical component of the contact force, w sin α. For the case of sufficiently long cylinders (L ≥ 5λ), the sinusoidal wavelength at the initiation of sinusoidal buckling (for horizontal configurations), λscr = 2L/n, can be expressed as [76], λscr = 2π EI∆r w 1/4 . (2.16) Finally, several authors have noted that the contact force per unit length, Wn , between a sinusoidally buckled rod and a horizontal cylindrical constraint can be expressed as [77, 78], 16π 4 EI∆rβ02 2 2π 2 2π 2 4 2π −β0 cos ( x) + 3 sin ( x) − 4 cos ( x) Wn = λ4 λ λ λ 2 2 2π 4π ∆rβ0 P cos( x) + w cos β, + 2 λ λ (2.17) where β0 is the angular amplitude of the sinusoidal shape (see Fig. 2-11 (b)). Examination of Eq. (2.17) shows that the normal contact per unit length does not vary substantially from the unit weight of the rod. This implies that the contact force between the rod and constraint does not change appreciably between the straight and sinusoidally buckled configurations. We will see that this is not the case in the case of helical rod configurations. Helical Buckling If the axial load is increased sufficiently beyond Pcrs , a secondary instability will occur, and the rod buckles into a helical configuration. This phenomena was investigated first by Lubinski [79]. In his work, Lubinski derived a relationship between the applied load above the helical buckling load, P h > Pcrh , and helical pitch, p, 52 Ph = 8π 2 EI . p2 (2.18) Cheatham and Patillo [80] confirmed the results of Lubinski in the case of increasing axial load using energy minimization and stability analyses. They found, however, during decreasing axial load that the force-pitch relationship of Eq. (2.18) changed by a factor of two: P h = 4π 2 EI/p2 . In the work of Lubinski, as well as Cheatham and Patillo, friction between the rod and constraining cylinder is neglected and the rod is assumed to stay in contact with the wall throughout the loading process. Chen, Lin, and Cheatham [81] extended the work above to find the critical load for the initiation of helical buckling, Pcrh = 4EI mπ 2 L w + 2∆r L mπ 2 , (2.19) where m is the number of full pitch lengths of the buckled shape (note the difference from the previous definition for sinusoidal buckling, where n was the number of half wavelengths). Similar to the previous case of Pcrs (Eq. (2.14)), the first term of Eq. (2.19) corresponds to the typical Euler buckling load of a column and the second term is an additional load caused by the cylindrical constraint, ∆r, and the weight of the rod, w. Also similar to sinusoidal buckling, the optimum number of pitches for buckling (minimizing Pcrh ) is length-dependent (just as the optimum n for Pcrs was length dependent in Fig. 2-12). As the length of the rod and cylinder increase, Pcrh asymptotes to become length independent (similar to Eq. (2.15)) and can be expressed as [1], r √ EIw Pcrh = 2 2 , ∆r (2.20) √ suggesting that the applied load must be increased by a factor of 2 above the critical √ sinusoidal load to achieve helical buckling, i.e. Pcrh = 2Pcrs . The value of the critical helical buckling load given in Eq. (2.20) is disputed in the literature for a variety of reasons. Deli et al. [82] utilized equilibrium methods instead of energy minimization, 53 with very small (less than 1%) disagreement with Eq. (2.20). Wu and Juvkam√ Wold [83] argued that the end load should should instead be (2 2 − 1)Pcrs , or 30 % higher than predicted by Eq. (2.20) to guarantee that the average force in the rod reaches the critical helical buckling load. Miska et al. [77,84] argued that Pcrh should be equivalent to the sinusoidal rod configuration becoming unstable, which they derived to be two times higher than Eq. (2.20). In this thesis, Eq. (2.20) will be the value intended when referring to the critical helical buckling load, but it is important to realize that there is a range of values for this critical load in the literature. The disagreement over the critical helical buckling load did not consider the effect of torsion. Deli et al. [82] showed that torsion entered the equilibrium equations √ √ through the parameter 2T / P EI, where T is the applied torque and P is the applied thrust. For the case of drilling a wellbore, this term is very small, and the effect of torque is negligible on the Pcrs . While it was found that torque could have a large effect on equilibrium shapes, it is very small in actual engineering applications, so the effect is relatively small typically (Deli et al. did not provide a relative magnitude, but their conclusions have been accepted in the ensuing literature [78,84]). There has been some work [85] studying the case of large torques of a rod constrained in a cylinder, resulting in a constrained analog to the writhing problem from §2.1. These results are not applicable to this study, so will not be considered. At the onset of helical buckling, there is agreement in the assumption of a constant pitch helix, and there is general agreement that, in the case of sufficiently long cylinders (with the definition of “long” the same as above: L & 5λ), that pitch length can be expressed as, phcr =π 8EI∆r w 1/4 . (2.21) Finally, Mitchell [86] derived the normal contact force per unit length between a helically buckled rod and cylindrical constraint to be: Wn = ∆r(P h )2 + w cos(β), 4EI 54 (2.22) where β is the angle from the vertical of the buckled rod and P h > Pcrh . Eq. (2.22) was derived using equilibrium equations of a rod undergoing large deformations and assuming small angular changes along the axial direction. This assumption is valid for most helical buckling applications, and can only be questioned well into the posthelically-buckled regime. According to Eq. (2.22), the normal force that the rod exerts onto the constraining pipe rapidly increases with increasing axial load in a quadratic power law fashion. The second term of Eq. (2.22) is related to the fact that the weight of the rod will affect the normal contact force, making it higher at the bottom of the pipe and the contact force lower at the top of the pipe. This second term tends to be neglected [1, 84] as a smaller order term than the first term of Eq. (2.22). Mitchell’s expression [86] for the normal force in relation to the applied load can be combined with Lubinski’s derivation [79] for helical pitch and applied force to give an expression for the normal force between the helically buckled rod and the constraint, Wn = A2 E∆r 2 δ , 4IL2 (2.23) where δ is the imposed end displacement (related to the buckled helical pitch, p) and A is the cross-sectional area. Deli and Wu [82] derive a contact force using equilibrium methods instead of energy methods and find, Wn = ∆r(P h )2 9 32 EIw , − w− 4EI 5 25 ∆r(P h )2 (2.24) which approaches Eq. 2.22 for large P h . Inclusion of Friction The analytical work summarized thus far, for both sinusoidal and helical critical loads, has assumed frictionless interaction between the rod and the borehole. This is widely agreed to be a non-physical simplification, leading to underestimated predictions of critical loads. Recent work by Gao and Miska [87, 88] has attempted to rectify this simplification. For both sinusoidal and helical critical loads, a correction factor is 55 added to Eqs. (2.15) and (2.20), respectively. They assume that the buckling rod slides (and does not roll/rotate) during the dynamic process of buckling, meaning that the coefficient of dynamic friction, µ, determines the magnitude of frictional interactions. Gao and Miska also assume that the rod remains in perfect contact with the constraining cylinder at all times. They use energy methods to develop buckling equations for the critical loads, under the assumption that the rod’s velocity does not change directions (which would cause a reversal of the direction of friction forces). The equations require numerical solutions, and for the critical sinusoidal force [87], Pcrs = s 2ψcr r EIw , ∆r (2.25) where s ψcr = 2 0.5qcrs (1 − 1.5a2crs ) 2/3 qcrs = 1 + 0.193(µ) 1 + 2 2qcrs 1+ , acrs = 0.774(µ) 0.125a2crs 1/3 8µ + πacrs (2.26) − 0.371µ, where µ is taken to be the lateral coefficient of dynamic friction between the rod and pipe. They distinguish between axial and lateral coefficients of friction, claiming friction in the axial direction of the rod has negligible effect in delaying the buckling process, which is predominantly composed of lateral motion. For helical buckling, Gao and Miska [88] give two different prefactors, depending on boundary conditions. For both cases, they find that, Pcrh = h 2ψcr r EIw , ∆r (2.27) h where ψcr is for pinned-pinned boundary conditions is, h ψcr 6 = 3 − πµ (π + 2µ)(5 − πµ) 10π 1/4 , (2.28) whereas for for pinned-free boundary conditions (free at the loading end), they find, 56 h ψcr = s 30(π + 2µ) , π(15 − 7πµ) (2.29) where µ is once again the dynamic coefficient of lateral friction. In Fig. 2-13, we plot Eqs. (2.26), (2.28), and (2.29) over the range 0 ≤ µ ≤ 1. The s amplification factor for critical sinusoidal buckling load, ψcr is seen to monotonically increase with µ, while the amplification factors for helical buckling increase asymptoth ically after µ ≈ 0.4. For pinned-pinned boundary conditions, ψcr becomes unbounded at µ = 3/π and has negative values for increasing µ. Pinned-free boundary conditions result in unbounded amplification in the critical load as µ approaches 15/(7π), with complex results afterwards. For the frictionless case (µ = 0), Eq. (2.26) for s h ψcr and Eq. (2.29) for ψcr for pinned-free boundary conditions match the frictionless predictions presented earlier (Eqs. (2.14) and (2.20), respectively). Results in [87,88] were presented over the range 0 ≤ µ ≤ 0.4, without comment on the range of µ the derivations were applicable. 10 9 8 Buckling Prefactor 7 ψs cr h ψcr Pinned−Pinned h cr ψ Pinned−Free 6 5 4 3 2 1 0 −1 0 0.2 0.4 0.6 0.8 Coefficient of Dynamic Friction, µ 1 Figure 2-13: Buckling prefactors predicted by Gao and Miska [87, 88] for sinusoidal s h (ψcr - Eq. (2.25)) and helical (ψcr - Eqs. (2.28) and (2.29)) buckling loads. 57 Frictional Drag Both Wu and Juvkam-Wold [83] and Mitchell [86] derived expressions for the required input load, Pin , to maintain a certain output load, Pout , with both forces shown in Fig. 2-11. Mitchell’s derivation was performed for a vertically oriented wellbore while Wu and Juvkam-Wold considered a horizontal wellbore. In both cases, the authors assume lateral friction does not change the critical helical buckling load (Eq. (2.20)) nor the contact force-applied load relationship (Eq. (2.22)). Both authors then solve the differential equation dP/ds = µWn (s), where s is the arc-length along the rod’s centerline and P is the axial force. Wu and Juvkam-Wold [83] derive an expression for the axial force distribution along the buckled rod to be, P (s) = r 4EIw tan µws ∆r r ∆r + arctan Pout EIw r ∆r EIw !! , (2.30) where s = L is the input end in their notation. Wu and Juvkam-Wold [83] argue that because of this frictional drag, for a given rod length, L, there is a point in loading when increasing Pin will have no effect on Pout . They define this condition as lockup. Gao and Miska [88] consider the role of lateral friction in their definition of critical load, but their approach of integrating contact force as a frictional drag is the same as suggested by Mitchell [86] and Wu and Juvkam-Wold [83]. McCann and Suryanarayana [89, 90] presented experimental evidence confirming the prediction by of Eq. (2.30). In their experiments, a 0.6 cm diameter aluminum rod was compressed inside of a 3.8 cm inner diameter acrylic tube with maximum length L = 3.7m. Good agreement was seen between experiment and theoretical prediction for the frictional drag up to applied loads (Pin ) up to approximately four times greater than the predicted helical buckling load, Pcrh (Eq. (2.20)). They also performed experiments with a vibrating device imparting “small-amplitude, high frequency transverse vibrations” [90], which were observed to yield results for critical buckling loads consistent with frictionless predictions as well as zero frictional drag for applied loads up to approximately three times the predicted helical buckling load. 58 Imperfections Several authors [89,91–94] have considered the case of buckling within a borehole with constant curvature (typically measured in angular change of the centerline tangent of the constraint per unit length). Equilibrium equations are derived including this constant curvature, which typically results in increasing the critical helical buckling load. He and Kyllingstad [91] predict some configurations which could decrease the buckling load, but are not applicable to the case of oilfield wellbore trajectories. A schematic of a typical buckled configuration from experiments performed by He and Kyllingstand [91] compressing steel wire (diameter, d = 3.4 mm) in a plexiglass tube (inner diameter, I.D. = 11 mm) with a radius of curvature, R0 = 0.6m, is shown in Fig. 2-14, extracted from [91]. In this case, the applied load is beyond the predicted helical buckling load, so the rod assumes a helical configuration in the straight portions of the pipe, but did not form a helix in the curved portion of the cylinder, instead pressing against the outside curve. We have presented the relevant results of the classic case in this section. Unlike the previous section, the post-buckled geometries of the constrained rod could not be solved for directly from Kirchhoff’s equations of equilibrium. Instead, buckled shapes which satisfied equilibrium (but not necessarily the lowest energy) were assumed. The rod was assumed to be straight, sinusoidal, or helical as the applied load increased. The configurations were all assumed to be in perfect contact with the constraining cylinder, and most derivations assumed zero frictional interaction between the rod and pipe. With these assumptions, critical buckling loads were derived, with some disagreement in the literature over the correct value of the helical buckling load. As a final note, other authors [95–98] have considered the problem of drillstring dynamics, and have attempted to model the instabilities present in the case of a rotating rod inside of cylindrical constraint. These results, while applicable to the general application of drilling, are not considered in this thesis, as Chapter 4 deals with quasi-static compression only. 59 Figure 2-14: Schematic of buckled configuration of a steel rod inside a Plexiglass pipe with radius of curvature, R0 = 0.6 m. Note the applied load, Fa , is greater than the helical buckling load for a rod inside a straight cylinder. The rod takes a helical configuration on both straight ends of the constraint, remaining unbuckled within the curved constraint itself. Figure extracted from [91]. 60 2.3 2.3.1 Injecting a Rod Into Cylindrical Constraint Motivation Once a horizontal well has been drilled and production has initiated, it may become necessary to access the downhole environment for a variety of purposes, including cleaning out produced sand, acidizing (remove near-wellbore damage), data logging, or mechanical actuation [99, 100]. To accomplish these tasks, tools must be conveyed down to the location in the wellbore where intervention is required. It is often impractical to bring back the drill rig which originally created the wellbore due to operating costs, the time required to trip in drill string (stopping to add a joint of drill pipe every 90 feet), as well as the operational footprint of a drill rig. A product consisting of a single length of continuous, stainless steel coiled tubing (CT) has been developed [101] for these intervention operations. Originally developed in 1962 and seeing steadily increasing use since the 1970’s and 1980’s [100], CT tools can be easily tripped in and out of an existing borehole when attached to a 31,000ft-long, slender pipe [100], thereby reducing the operational time and costs. A coiled tubing unit can be fit onto a truck and trailer, significantly reducing the operational footprint. A photograph of a typical coiled tubing rig is presented in Fig. 2-15. The rig consists of a length of coiled tubing spooled around a reel. The coiled tubing then passes through an injector head and into a wellbore (underground). The entire rig is controlled by operators in the control cabin. Not pictured but often present are fluid reservoirs and high-pressure pumps that often accompany CT operations. The recent increased use of CT rigs has coincided with a leap in the number of horizontal boreholes being drilled. The combination of CT and horizontal boreholes is therefore becoming increasingly common. Insertion of CT into a vertical borehole is typically a mechanically stable process (no buckling instabilities are observed); gravity aids in insertion (keeping the CT in tension) and helps to ensure minimal contact between the CT and the borehole surface. On the other hand, the case of insertion into horizontal sections can be 61 Reel Injector Head Coiled Tubing Control Cabin Figure 2-15: A photograph of a coiled tubing rig, consisting of a long, continuous length of steel pipe spooled around a reel. The coiled tubing is inserted into the wellbore by the injector head. Note how the entire rig consists of a truck and trailer only. Photo courtesy of Schlumberger-Doll Research. significantly more challenging. The CT lies at the bottom of the borehole as it is inserted, leading to the development of frictional forces opposing insertion. This distributed frictional force acts in the axial direction of the injected pipe, leading to both increasing power at the top of the well needed to continue injecting the CT at a given rate and a buildup of compressive axial load in the pipe. After a critical load builds, the rod undergoes a series of buckling instabilities within the cylindrical constraint [1]. Throughout this thesis we refer to this scenario as the real case. As we might expect from the results of §2.2, the rod first buckles into a sinusoidal configuration. As insertion is progressively increased, a second instability is encountered whereby the rod buckles into a helical configuration. The helical configuration results in a rapidly increasing contact force between the pipe and the constraint, setting a limit to the amount of CT which can be injected into a horizontal well, known as the lockup length [1]. Unlike §2.2, buckling of the pipe is not spatially homogeneous. Instead, straight, sinusoidal, and helical configurations can be present 62 simultaneously along the pipe’s length. Currently, the lockup length for coiled tubing is not sufficient to service all extended reach wells. The problem described above is relatively recent, and existing literature from the mechanics community on the topic is limited (beginning in the 1990’s and 2000’s). The results derived in the following section rely on those obtained in §2.2, for the classic case. Note that in addition to the analytical work summarized here, some efforts to simulate coiled tubing lockup [102,103], as well as presentations of mitigation techniques involving downhole tractors (to pull the coiled tubing instead of pushing it downhole) [104,105] can be found in the literature. However, these methods are typically presented in a case-study format, making detailed comparisons or extrapolation to other cases difficult. Therefore, there is a need for a more predictive mechanical understanding of the real case in the current mechanics literature. 2.3.2 Previous Work Analytical work in the existing literature primarily focuses on the amount of rod injected at lockup, referred to as lockup length, LL , and/or the amount of rod injected before the initiation of the helical configuration, Lhel inj . Analytical models by Wicks et al. [1] and McCourt et al. [101, 106] assume that the axial load in the inserted rod arises from frictional resistance to the insertion velocity. Moreover, the lateral friction between rod and constraint was ignored in both sets of work. As with the analytical derivations of the classic case, the rod was assumed to be inextensible, unshearable and to remain in perfect contact with the cylinder throughout the process. McCourt et al. [101] gave a prediction for the onset of sinusoidal buckling and evolution of the buckled shape with further injected length. This derivation assumed that the critical load was equivalent to the buckling of a beam under a modified self-weight, µw, where µ is the coefficient of axial friction between the rod and pipe and w is the weight per unit length (w = ρAg, where ρ is the volumetric density, A the cross-sectional area, and g the acceleration due to gravity). McCourt et al. also assumed that the curvature of the constraining pipe I.D. was negligible, making the 63 critical sinusoidal buckling load independent of the radial clearance, ∆r, between the rod and pipe, unlike the previous derivations in the classic case (see, for example, Eq. (2.14)). McCourt et al. found that the amount of injected rod corresponding to the first sinusoidal buckling mode, L1 , was, s L1 = 7.83 3 EI , µw (2.31) where E is the Young’s modulus of the rod and I is the second moment of inertia of the cross-section. Similar to the approximate result of the classic case (Eq. (2.17)), however, the contact force between the sinusoidally buckled rod and pipe, Wn , was assumed to be identical to that of the straight sections of rod, Wn = µw. Before helical initiation, both McCourt et al. and Wicks et al. defined the injection force as, Pinj = µwLinj , (2.32) where Linj is the injected length of rod, µ is the dynamic coefficient of friction, and w is the weight per unit length of the rod. This linear relationship was assumed to hold for Pinj ≤ Pcrh , where Pcrh is the critical helical buckling load defined in §2.2 in √ p Eq. (2.20) as Pcrh = 2 2 EIw/∆r. Solving for the injected rod length providing the injection force equal to the critical helical buckling load, Lhel inj , Lhel inj √ r EI 2 2 , = µ w∆r (2.33) such that the injected rod length before the first helix forms is inversely related to the coefficient of friction and has a square root dependence on the other parameters, defined above. For rod injected past helix initiation, Linj >Lhel inj , both McCourt et al. and Wicks et al. claimed the Pinj would no longer be linearly related to Linj by Eq. (2.32). This nonlinearity came from a growing normal contact force, Wn , between the rod and the constraint. The two groups disagreed, however, on the form that this normal contact force should take. McCourt et al. [101] made a geometric argument and arrived at, 64 3/2 π∆rPinj Wn = √ , 2EI (2.34) while Wicks et al. [1] assumed that, 2 ∆rPinj , Wn = 4EI (2.35) using the relationship between axial load and normal contact force given in Eq. (2.22) and derived by [86]. In either case, the injection force, Pinj , required to continue injecting rod can be solved through the differential equation, dP/ds = µWn (s), (2.36) where s is the arc-length along the rods centerline, and P is the axial force along the rod. Using either Eq. (2.34) or (2.35) for Wn , Pinj eventually tends to infinity, which then defines the theoretical lockup length, LL . Wicks et al. [1], using Eq. (2.35), defines that point as, √ r 3 2 EI . LL = µ w∆r (2.37) Comparing Eq. (2.37) to Eq. (2.33), we can see that for Wicks et al., LL = 1.5Lhel inj . McCourt et al. suggest a forward stepping algorithm for finding the LL instead of directly applying the differential equation Eq. (2.36), so a direct comparison is not warranted. McCourt et al. [101, 106] also conducted the only laboratory-scale experiments in the existing literature, showing good agreement with their predicted insertion forces as a function of injected length. In their experiments, a stepper motor was used to inject a rubber rod (diameters, d = 2, 3, 4, 5, and 6 mm) into a glass pipe (length, L = 5 m, inner diameters, I.D. = 8.6, 13, 15, 19, 22.8, 28.6 and 33.4 mm). Over the range of tests presented, the fitted coefficient of dynamic friction between rods and glass pipe was in the range 0.6 ≤ µ ≤ 1.2 while monitoring the reaction force at the injector, Pinj . 65 Two other relevant studies were geared toward studying methods to extend the reach, LL , of a coiled tubing operation in a horizontal well. In the first, McCourt et al. [107] continued their work by exploring the effect of varying the wall thickness of the coiled tubing as a function of position. They found that by controlling the geometry of the rod’s cross-section, they could increase lockup length significantly. Combining three discrete values of I.D., with greater wall thickness closest to the injection point, McCourt et al. predict 67% improvements in LL using the model developed in [101]. The second study of relevance was performed by Zheng and Adnan [108], who considered the effect of intrinsic curvature on lockup length. Referring back to the photograph of a typical CT unit shown in Fig. 2-15, one can see that the coiled tubing is spooled. This spooling does plastify (i.e. deforms irreversibly) the steel pipe, imparting a natural curvature to the rod [28]. CT rigs typically include a pipe straightener, specifically designed to remove some of this residual bend. Nevertheless, the tubing still exits the injector head into the wellbore with a residual bend, with radius (the inverse of natural curvature, R0 = 1/κ0 ) in the range 150 < R0 [in.] < 400 [108]. Zheng and Adnan assumed that the rod adopted a helical shape throughout injection and that the lockup length could be expressed as, ∆rR0 Pcrh + 2EI , (2.38) ∆rR0 Pcrh √ q EIw h is the helical buckling load, Pcr = 2 2 ∆r , given in Eq. (2.20) and Ph 2R0 LL = cr + ln 2µw µ where Pcrh the other parameters have been previously defined. In the case of the intrinsically straight rod (R0 = ∞, κ0 = 0), Eq. (2.38) recovers the result from Wicks et al. from Eq. (2.37). These results of Eq. (2.38) have not been validated with experimental results. Adapting arguments and results from the classic case, researchers in the real case have developed predictions for the amount of rod that can be inserted into a horizontal cylinder before a portion of the rod buckles into a helical portion. Increasing insertion is resisted by a growing normal contact force between the helically buckled 66 rod and the constraint, eventually leading to lockup. McCourt et al. [101] conducted precision experiments where they monitored the reaction force at the injector during rod insertion. However, no experiments have yet been performed in the existing literature that investigate the effect of natural curvature of the inserted rod on Lhel inj . 2.4 Outlook In this chapter we have surveyed the existing literature in three active fields of research: • The writhing problem: the most mature of the three problems, with analytical results for straight and naturally curved rods well-developed. Despite the highly symmetric loading and boundary conditions, the resulting geometries are found to be highly nonlinear, with numerical solution techniques needed for non-trivial centerline configurations or finite length effects. • The classic case: with the assumptions of post-buckled configurations and zero frictional interaction between rod and constraint, several closed form solutions have been obtained. The inclusion of friction and dynamic effects is the focus of current efforts in the literature. • The real case: a relatively new field of study. Analytical work has utilized results from the classic case, so that current research also focuses on dynamic and frictional effects. We now turn toward investigating these three topics more in-depth in the following chapters, presenting the problems in the same order as in this chapter. The next chapter deals with the behavior of an unconstrained, naturally curved rod, with strong connections to the theory summarized in §1.1, although the presence of a distributed body force complicates the overall behavior. As we continue through the problems, fewer theoretical predictions will exist for experiments conducted, until Chapter 6, studying the effect of vertical vibration of the cylindrical constraint on the initiation of helical buckling, where there are no theoretical predictions at all. 67 68 Chapter 3 Suspending a Naturally Curved Rod We begin our study of rods by rationalizing how rods with identical material and geometric properties but different natural curvatures, κ0 , behave under self weight when clamped from one end. A rod with constant natural curvature will assume a circular arc with radius, R0 = 1/κ0 , in the absence of external forces. In most natural and engineered systems, rods have a natural curvature due to fabrication or loading history (such as in the case of spooled coiled tubing [28]). Understanding the interaction between natural curvature, elasticity, and gravity is applicable to the remaining chapters. For rods with sufficiently low natural curvature, the rod adopts a planar shape. Past a critical value of natural curvature, the rod transitions to a non-planar shape. Non-planar shapes can be one of two classifications: a localized helix consists of a helical structure beneath a straight portion of rod and a global helix is composed entirely of a helical structure, with no straight portion. This chapter explores the transition from planar to non-planar configurations, as well as the geometry of the helical structures. Appendix B presents a manuscript resulting from this work. The experiment performed to investigate this problem is described in §3.1. In §3.2, experimental results will be interpreted and supported by simulation and theoretical predictions. A related problem of contorting a naturally curved rod clamped at both ends - the writhing problem - will be briefly discussed in §3.3 (having been published elsewhere [109, 110]). Finally, §3.4 will outline open issues in the research. 69 Aluminum Frame Acrylic Clamp Rod Cameras Figure 3-1: Photograph of the experimental setup, which includes a hanging rod, acrylic clamp, aluminum frame, and two perpendicularly mounted digital cameras. 3.1 The Experiment A photograph of the experimental apparatus built to record the configuration of a naturally curved rod hanging under its own weight is shown in Fig. 3-1. Two identical digital cameras1 photographed a rod from perpendicular angles to allow for three-dimensional (3-D) reconstruction and measurement. A fixed length (referred to as the suspended length, L) was clamped in place with a laser-cut acrylic clamp, which was aligned vertically. The custom fabricated rods used for the experiments will be described in further detail in §3.1.1, the experimental protocol will be given in §3.1.2, and post-processing steps will be explained in §3.1.3. 1 Nikon D90 SLR cameras, each with a Sigma Macro 50mm F2.8 EX DG lens, capturing 2848x4288 pixel color images remotely controlled by Nikon Camera Control Pro 2. 70 3.1.1 Material Selection and Properties A rigid aluminum frame2 was built to provide a stable, level support for the experiment while still allowing visual access of the hanging sample. A clear acrylic plate3 was mounted on the top of the frame to serve as a platform for the rod clamp as well as provide a mounting surface for LED lamps to illuminate the sample. Two sides of the frame were closed with black paper to provide a contrasting background for later image processing, as will be described in §3.1.3. Two cameras were mounted on adjustable heads4 to allow for precise leveling and orientation. The entire experiment was mounted on an aluminum breadboard5 for a rigid mounting and to dampen out environmental vibrations. Property Young’s Modulus, E Density, ρ Diameter, d Poisson’s Ratio, ν Value 1296 ± 31 [kP a] 1210 ± 8 [kg/m3 ] 3.16 ± 0.05 [mm] 0.49 Table 3.1: Relevant material properties of rods manufactured for the experiment. Values taken from Appendix A on rod fabrication. Elastomeric (Vinylpolysiloxane) rods were fabricated according to the protocol given in Appendix A, with the physical properties given in Table 3.1. Fourteen rods were created, each with a different, constant, natural curvature in the range of κ0 = 0 m−1 (straight) to 62.3 m−1 , with specific values listed below6 , and curvatures shown graphically in Fig. 3-2. The maximum value of natural curvature that we were able to manufacture (κ = 62.3 m−1 ) was set by ovaling of the cross-section. For this maximum value, a 10% difference was measured between the major and minor diameter of the fabricated rod. All other rods were measured to be circular in crosssection. Rods were at least 30 cm in overall length. We now proceed with describing 2 Constructed with 80/20 Aluminum T-slotted framing [111]. Optically clear cast acrylic sheet purchased from McMaster-Carr. 4 Manfrotto 410 Junior Geared Head with 3-axis adjustment 5 Newport SA2 Series breadboard. 6 κ0 = 0, 8.9, 16.6, 23.4, 32.9, 38.0, 39.3, 44.7, 45.8, 49.5, 52.8, 55.1, 56.2, and 62.3 m−1 . 3 71 the experimental protocol using these rods. 2 cm Figure 3-2: Unscaled diagram of the 14 natural curvatures fabricated for experiments, ranging from κ0 = 0 to 62.3 m−1 . 3.1.2 Experimental Protocol Keeping the material properties constant, an exploratory experimental program was performed to vary the natural curvature, κ0 , and the suspended length, L, of the rods. Each experimental test consisted of the following steps: i) The rod was inserted through the (loose) acrylic clamp to the desired suspended length, L, with an experimental uncertainty of ±1 mm. ii) The sample was clamped in position, taking care to maintain the vertical orientation of the clamped end of the rod (misalignment could arise due to over-tightening of the clamp). iii) The rod was adjusted (rotated) such that no point of the rod was hidden from view and the free tip was not pointing directly at either camera, which would obfuscate some rod length. For rods adopting a helical hanging configuration, each case was inspected to ensure it took on a single chirality. This aspect is discussed in more detail in §3.2. 72 iv) The rod was allowed to come to static equilibrium (judged by visual inspection), at which point images would be captured with both cameras. For each rod, the suspended length could be adjusted in the range 1 ≤ L [cm] ≤ 52. The maximum suspended length that could be tested for each rod (with a unique κ0 ) was dictated by the elevation of the bottom tangent. Rods taking on helical structures could have longer suspended lengths without making contact with the optical table. The frame had 37 cm of clearance between the aluminum breadboard and acrylic clamp. By varying suspended length and natural curvature, 170 unique experimental configurations were photographed. After testing, image processing was used to measure the rod’s hanging geometry, more details of which are given next. 3.1.3 Three Dimensional Experimental Reconstructions For each rod configuration tested, the coordinates of the centerline were found using image processing software7 to combine two photographs. The photographs were taken to be projections of the rod in two perpendicular planes (x − z and y − z planes), as shown for the representative case of a rod with κ0 = 49.5 m−1 and L = 40 cm in Fig. 3-3 (a) and (b). For each image, the green rod was located against a black, contrasting background to enable automated recognition. Starting from the known clamping point (acting as the origin), the rod centerline was traced in each image, as shown by the solid lines in Fig. 3-3 (a) and (b) superposed on the experimental photographs. To combine the two sets of coordinates (the x − z and y − z data into the x − y − z data), a cubic spline function was fit to the data and then sampled in each plane at identical z coordinates (elevation in our chosen coordinates). The compatible data sets were then combined into the 3-D reconstruction, as shown in Fig. 3-3 (c). The centerline of the digitally reconstructed configurations of the rod were determined using an image processing algorithm called image erosion, whereby the fully located rod’s outer edges are eroded until they meet. This process resulted in the re7 MATLAB Image Processing Toolbox. 73 constructed rod tip being approximately one half rod diameter (d = 3.16 mm) behind the actual rod tip, but located along the rod centerline. This was taken as experimental error. The other source of noise that sometimes entered into rod reconstructions was at the bottom tangent of the rod (the point with the lowest total elevation). At this location, a small kink was occasionally observed in the reconstruction as an artifact of only having two view angles (millimetric portions of the rod could be hidden from one camera for helices with sufficiently small radii). While undesirable, the feature was located away from the regions of interest of this study and did not affect the rest of the reconstruction. a) b) c) 5cm 5cm 0.05 0.1 z z [m] 0.15 0.2 0.25 0.3 x-z plane y-z plane 0.01 0.01 −0.02 −0.01 y [m] x [m] Figure 3-3: a) and b) Perpendicular photographs of a hanging rod with κ0 = 49.5 m−1 and L = 40 cm, with extracted centerline traced. c) 3-D reconstructions were created by combining the two located centerlines. 74 3.2 Results and Interpretation Having described the experimental techniques and procedures, we will now report the results from our investigation, combining experimental, numerical, and theoretical results. Numerical work was performed in collaboration with Arnaud Lazarus. The author selected parameters for the exploration of the problem and gave material properties allowing for direct comparison between simulation and experiment, and Lazarus ran the simulation code, using an original method which is discussed in more depth in [109]. The simulations include both the bending and twisting of an inextensible rod perfectly clamped at one end. The rod was discretized into 200 elements, approximating the differential equations modeling the equilibrium of the simulated rod with finite differences (a mesh study was performed to confirm result convergence for this number of elements). The constitutive material behavior was modeled as isotropic and linear elastic, using material properties experimentally measured and given in Table 3.1. The resulting approximate equations of equilibrium are solved with an asymptotic continuation method [112] implemented in the interactive path-following and bifurcation analysis tool, ManLab [113]8 . This package allows the user to interactively determine the bifurcation diagram of a system, following the stationary points of the energy as a function of a control parameter, typically a geometrical or mechanical variable of the problem (κ0 in this case). This tool can be compared to the widely used path-following software, AUTO [114], which automatically determines the bifurcation diagram of the system, but less efficiently. Equilibrium configurations were determined to be locally stable or unstable based on a positive or negative decay rate in post-processing. The centerline coordinates of stable configurations were provided to the author, who completed further post-processing to compare simulation results to experiments (described in more depth later). This section continues with a description of rod morphologies observed and a direct 8 ManLab is used as a package in MATLAB. 75 For the planar shapes, Configuration A is the trivial case of κ0 = 0 m−1 and as such, hangs with a straight, vertical shape. Configuration B, however, exhibits a planar shape consisting of a curved, hook-like portion near the free end underneath a straight, vertical length. Bending energy is released at the cost of increasing gravitational potential energy in the hook, while the upper length is under sufficient tension (generated from the weight of the arc length below it) to remain straight. In the case of the non-planar configurations, one can observe a helical structure whose pitch length appears to lengthen for increasing arc-length from the free tip. The helical structure decreases bending energy throughout the length of the configuration, but increases both twisting and gravitational potential energies. The pitch length increases away from the free tip due to increasing tension in the rod. In Configuration C, the rod appears to adopt a vertical configuration at a point closer to the free tip than in Configuration D, which as the highest natural curvature shown in Fig. 3-4. To move beyond qualitative agreement between experiments and simulations, a direct comparison was made between the measured vertical elevation, h, of the free rod tip for both methods. Experimentally, h was measured directly from the digital reconstructions. In Fig. 3-5, we plot h as a function of the suspended rod length, L, for three values of the natural curvature, κ0 = 16.6, 38.0 and 56.2 m−1 . Excellent quantitative agreement between experiment and simulation is found, with no fitting parameters used in the simulated results. For the lowest two values of natural curvature in Fig. 3-5, κ0 = 16.6 and 38.0 m−1 , the rods maintained planar configurations for all lengths tested, with h decreasing monotonically with L. The highest value of natural curvature tested for the comparison in Fig. 3-5 (κ0 = 56.2 m−1 ), however, assumed planar configurations for small injected lengths (Configurations A and B, L . 0.1 m) and transitioned to non-planar shapes for L & 0.1 m (Configurations C and D). The helical structures pictured in Figs. 3-4 and 3-5 exhibit constant chirality (all are left-handed in the photographs). This was the equilibrium configuration desired, but not the only stable state for helical structures. Experimentally, care needed to be taken not to impart extra twist to the free tip of the sample while mounting 77 A B C D Vertical Elevation of Tip, h [m] 2 cm 0 −0.05 1 −0.1 1 −0.15 −0.2 0 Experiments Simulations 0.05 0.1 0.15 Suspended Length, L [m] 0.2 Figure 3-5: Vertical elevation of the rod’s free tip, h, as a function of suspended length, L, plotted for three different natural curvatures, κ0 = 16.6, 38.0 and 56.2 m−1 , for both experiments (points) and simulations (solid lines). Photographs of experimental configurations show the rod with κ0 = 56.2 m−1 transitioning from planar (A and B) to non-planar (C and D) shapes. the rod, otherwise another, meta-stable, state could be adopted by the rod. This altered configuration was characterized by a chirality inversion (also referred to as a helical perversion in the literature on helices [45]), as shown in Fig. 3-6. Fig. 3-6 (a) is a photograph of a rod with L = 38 cm and κ0 = 55.1 m−1 and (b) is a top view projection of the corresponding reconstruction in the x − y plane. Following the rod centerline starting at the clamp, the initial helical structure is right-handed, but transitions to a left handed helix. This mode was associated with tip elevations closer to the clamping point and was not considered in the numerical simulations. These configurations were discarded as they represented higher energy states than their constant chirality counterparts. Having confirmed the agreement between experiments and numerics, further sim78 a) b) 2cm 0 y [m] −0.01 −0.02 −0.03 z Chirality Inversion −0.03 Clamp Rod −0.02 x −0.01 x [m] 0 0.01 Figure 3-6: A chirality inversion seen in a) a side view photograph and b) top view of digital reconstruction projected onto the x − y plane. These quasi-stable configurations were experimentally avoided for comparison with simulation. ulations were used to expand the number of rod configurations considered in order to construct a phase diagram expanded around the experimental data points. Toward this end, we computed the equilibrium configurations of 11,110 combinations of Young’s modulus (96 ≤ E [kP a] ≤ 9600), suspended length (5 ≤ L [cm] ≤ 50), and natural curvature (0 ≤ κ0 [m−1 ] ≤ 100). In order to create a phase space diagram, rods were classified according to two non-dimensional numbers: dimensionless cur- vature, L = κ0 L, and dimensionless weight, w = w/(EIκ30 ) where w = ρgA is the weight per unit length of the rod (ρ is the volumetric mass density, g the acceleration due to gravity, and A is the cross-sectional area) and EI is the bending stiffness of the rod. All other lengths were normalized by the natural radius of curvature, κ0 . In addition to defining the rod physical properties, we also need a rational description of the rod’s geometry to distinguish planar, localized helix, and global helix structures. The analytic description of rod geometry as provided in §1.1 (summarized below) is sufficient, and will allow for direct comparison with theoretical predictions. 79 A simulated rod configuration is defined by its centerline, r(s), where s is the rescaled arc-length (with its origin at the free end, s = 0), and an orthonormal director basis, (d1 (s), d2 (s), d3 (s)), referred to as the material frame. The director basis is subject to the condition r0 = d3 , where a prime denotes differentiation with respect to s. The other directors are defined such that d2 lies in the plane of natural curvature, κ0 , and d1 = d2 × d3 . These definitions are shown in Fig. 3-7. Experimental reconstructions did not have access to d1 or d2 . Finally, the angle β(s) between the rod’s tangent and vertical (cos β = d3 · ez ) was also measured along the arc-length of each rod, both simulated and experimental, to aid in differentiating between localized helices (which have a vertical portion of rod), and global helices (which are composed entirely of a helical, non-vertical, structure). Configurations were first classified as planar or non-planar. This was done via principle component analysis [115] for digitally reconstructed rods, which determines how many dimensions are sufficient to describe the rod configuration, finding the s Figure 3-7: Illustration of the material frame defined for each simulated rod configuration (characterized by a centerline r(s), where s was the arc-length from the free end of the rod). The angle β between the rod’s centerline tangent and vertical (cos β = d3 · ez ) was used to classify and describe non-planar configurations. 80 orthogonal axes that best account for the data (for example, planar structures can be completely described with only two axes). Simulated rods were classified by directly evaluating the stability of the planar equilibrium shape of a rod. Non-planar shapes were further sub-classified as either local helical structures or global helical structures. The transition point between straight vertical and helical portions of nonplanar rods was defined as the first point (lowest s) where β ≤ 1.5◦ . Theoretically, vertical portions will have β = 0◦ , but a tolerance (in this case, 1.5◦ ) is necessary for numerical data provided by the experimental reconstruction. Local helical structures were defined as those whose suspended arc-length, L, was less than 95% helical. In Fig. 3-8, we present the phase diagram of a naturally curved rod suspended under self weight in the (L, w) space, with different colors/symbols corresponding to different configuration geometries. Experimental points are plotted above shaded regions defined by simulation results (with boundary points plotted). All three geometries are present in the experimental range of configurations, with excellent agreement simulation and experimental data agreeing on geometry definitions throughout the diagram. Three representative experimental configurations are given: global helix (Configuration A - κ0 = 62.3 m−1 , L = 40 cm), localized helix (Configuration B κ0 = 49.5 m−1 , L = 40 cm) and planar (Configuration C - κ0 = 49.5 m−1 , L = 10 cm). Both experiments and simulations agree in general trends: for sufficiently high dimensionless weight, w, only planar shapes are observed. Below this critical value of weight, any of the three geometries can be observed. For low dimensionless curvature, L, planar configurations are stable. With increasing L (vertical lines in the phase diagram representing changing rod length, L), a planar configuration will transition from planar, to global helix, to localized helix. In §3.2.2 and §3.2.3 we aim to rationalize the transitions between the different regions of the phase diagram, which are also known as phase boundaries. 81 3.2.2 Planar to Non-Planar Configurations In the phase diagram of Fig. 3-8, one can see two different asymptotic boundaries between planar and non-planar shapes. The first, vertical asymptote at w ≈ 0.25, represents a planar shape with a hook near the free end under a long straight section (as was seen in Configuration C of Fig. 3-4). This boundary represents an instability transitioning from a planar shape to a localized helix. The second, w-dependent, asymptote (w << 1), represents a transition between planar and global helix configurations. These very stiff configurations correspond to rods making multiple turns in their 2D configurations. This transition does not have a direct physical counterpart, as these configurations involve self-contact. The mechanism behind the first planar to non-planar transition is analogous to an inverted pendulum, as we consider next. We approximate the planar configuration of a rod with suspended length, L, as a perfectly straight, vertical section over a hook with length, c. Nondimensionalization by the rod’s natural radius of curvature, κ−1 0 , yields the dimensionless length of the rod, L, and dimensionless hook length, c, as shown in Fig. 3-9 (a). We make the simplifying assumption that the hook assumes a constant curvature equal to its natural curvature (taking on a circular arc with radius R0 = 1). The proposed mechanism of instability can only occur for values of c > π/2, equivalent to the hook being larger than a quarter circle, as shown in more detail in Fig. 3-9 (b). With this criterium, a portion of the rod (the red section in Fig. 3-9 (b)) will begin to rise above the bottom tangent of the hook (Point P in Fig. 3-9 (b)). As the hook grows, the center of mass of the red portion will rise higher above P. For any out of plane perturbation, an overturning moment about P will be created, with a magnitude dependent on this difference in elevation between the center of mass and P, η, as shown in Fig. 3-9 (b). To find the critical point at which the hook of size c > π/2 becomes unstable (thereby resulting in the planar to non-planar transition), we must: i) derive the length c for a given set of rod properties; ii) find the elevation of the center of mass of the hook tip above the bottom tangent, η; and iii) find the critical height η c for which the hook tip above the bottom tangent becomes unstable. By equating items ii) and 83 a) b) g g P Figure 3-9: Schematic diagram of the assumed rod configuration to help rationalize the planar to non-planar transition. a) A hook of scaled length, c, forms under a straight region (length L − c), taking on a circular arc with constant curvature equal to the natural curvature of the rod (κ0 = 1). b) For hook lengths such that c > π/2, a portion of the rod (red) rises above the bottom tangent of the hook (point P). The center of mass of this section of rod will rise an elevation, η, above the bottom tangent, leading to an out of plane instability. iii) we should then find an expression for the transition from planar to non-planar configurations. The size c of the hook is governed by the competition between bending energy and gravitational potential energy. The total dimensionless energy of the rod, E total , can be written as, 1 1 2 E total = (L − c) + w(L − c2 ) + w(c2 − 1 + cos c), 2 2 (3.1) where the first term is the elastic energy and the two remaining terms account for the gravitational potential energy of the straight and hook portions of the rod, respectively. Equilibrium is satisfied when Eq. (3.1) is stationary under small changes in c, such that, 1 ∂E = 0 =⇒ = c − sin c, ∂c 2w (3.2) which is a nonlinear equation in c and w without a closed-form solution for c. We 84 know, however, that the configurations under consideration are relatively stiff rods with large values of κ0 such that w << 1. We have also assumed that c > π/2. With these two assumptions, Eq. (3.2) can be approximated by neglecting the sin c term, yielding c ≈ 1/2w. The elevation, η, of the center of mass of the hook tip above the bottom tangent can now be found directly when we assume the shape of a circular arc. Also assuming c ≈ 1/2w, we express η as a function of w alone, η= 1 1 − πw + 2w cos 2w . 1 − πw (3.3) Finally, we can express a critical elevation, η c , above which the overturning moment caused by out of plane perturbations of the tip of the hook can no longer be supported by the torsional stiffness of the rod. This calculation relies on the determination of a participating length of rod which provides the torsional stiffness, given that the resisting torque is inversely proportional to the length of rod twisted. We expect this process to be local, unaffected by overall length changes of the rod as L → ∞. We take this local length scale to be approximately c = 1/2w = EIκ30 /w, or essentially the length scale balancing elastic forces and gravity. With this length scale, the critical elevation η c can be expressed as, ηc = 4 w , 1 + ν 1 − πw (3.4) where ν is the Poisson’s ratio, which represents as a coupling factor between bending and twisting. Eqs. (3.3) and (3.4) provide expressions for the elevation of the center of mass of the hook tip as a function of w alone. For ν = 0.5, equating Eqs. (3.3) and (3.4), we find a numerical critical value of dimensionless weight, below which a hook shape is no longer stable, wc = 0.13. While not recovering the exact prefactor, this value recovers the correct scaling of constant w for the phase boundary in Fig. 3-8 as well as an estimation for the order of magnitude. Moreover, this calculation provides physical insight into the instability of planar shapes for increasing κ0 . Our collaborators, using the exact geometry of the hanging planar configurations 85 (not the approximate configuration of Fig. 3-9), and numerically computing the linear stability with respect to out-of-plane buckling for the case of L = ∞ and ν = 0.5, found wc = 0.24. This value agrees well with the vertical simulated boundary between planar and non-planar configurations in Fig. 3-8. Appendix B provides further details on the linear stability analysis. 3.2.3 Helical Shapes This section briefly summarizes the work done in collaboration with colleagues on this project to rationalize the phase boundary between localized and global helices found with experiments and expanded with simulations. We will also describe the geometry of the suspended, non-planar rods. More details regarding the analysis of the helical shapes can be found in the manuscript in Appendix B. The total energy of an inextensible rod, including bending, twisting, natural curvature, and gravity, E, can be expressed as, E= Z 0 L 1 2 2 2 (κ1 − 1) + κ2 + C κ3 − w s cos β ds, 2 (3.5) where κ1 and κ2 are the scaled material curvatures about the directors d1 and d2 , respectively, κ3 is the scaled material twist about the tangent vector, d3 , β is the angle between the rod tangent and vertical (as shown in the inset of Fig. 3-10) and C is the ratio between the twisting and bending stiffnesses (C(ν) = (1 + ν)−1 for circular, solid, cross-sections). The first term of the integrand expresses the two bending strain energies and one twisting strain energy. In this formulation, natural curvature is assumed to be aligned with one of the material planes, so that it only enters the first bending energy term (in the scaled variables κ0 = 1). The second term of the integrand expresses the gravitational potential energy. Inspired by previous work [45,65] on helices under constant tension and/or torsion, it is assumed that the director d2 is perpendicular to the applied force (weight), such that d2 lies in the x − y plane, as shown in the inset of Fig. 3-10. This is equivalent to the helical axis being aligned with gravity. Using this assumption, Eq. (3.5) can 86 be specialized to an expression for the total energy of the helical configurations, E 3D , E 3D = Z 0 L " 1 2 # ! −1 tan2 β(s) 1 0 2 1+ − ws cos β(s) + β (s) ds, 2 C (3.6) where the first (bracketed) term depends on the rod’s local shape, β(s), and the second term depends on the rate of change of β(s) along s. Two classes of solutions can now be found for equilibrium configurations: those that describe a well-developed helical structure (the local helix approximation) and those describing the transition Dimensionless arc length, s Dimensionless arc−length, s from helical to straight rod (the inner layer theory). 15 15 10 10 55 00 00 Experiments Simulation LH Approximation IL Theory 0.7854 Angle from vertical, β [rad] Angle from vertical, β [rad] 1.5708 Figure 3-10: Localized helical configuration (L = 19.8 and w = 0.12) described by the angle from vertical β as a function of arc-length from the free tip, s. Experimental and numerical results are compared to theoretical predictions from the Local Helix (LH) and Inner Layer (IL) approximations, with s = s∗LH and βIL = 1.5◦ , respectively, the predicted point for transition from straight to helical configuration. 87 Considering the localized helix configuration shown in the inset of Fig. 3-10, we plot the measured rod tangent to vertical, β(s), as a function of dimensionless arc length, s, from the free tip, (s = 0), of both an experimental and a simulated configuration for a rod with κ0 = 49.5 m−1 and L = 40 cm (i.e. L = 19.8 and w = 0.12). We observe that near the clamped end (s & 15) the rod remains relatively straight such that β ≈ 0. Below a critical arc length, however, β begins to increase steadily before taking on an oscillatory nature near the free end (s = 0). If we consider portions of the configuration within the helical position of the structure that are away from both the straight portion and the free tip of the rod, we see that β varies slowly with s. For this section, it was assumed that the derivative term in Eq. (3.6) was negligible, with the resulting equation referred to as the Local Helix (LH) approximation. This approximation recovers equilibrium solutions to a rod with natural curvature put under constant tension such as that given in Eq. (2.12) [45]. The solution for β as a function of s is shown as a purple line in Fig. 3-10. It is known that the solution to the LH approximation has a trivial, straight solution for s ≥ s∗LH = (wC)−1 [45], which is the dimensionless form of Eq. (2.7). This can be interpreted as the upper part of the rod having sufficient tension supplied by the hanging weight of lower portions to remain straight. Portions below this critical point, however, do not have sufficient tension to make the straight configuration stable, instead taking on non-zero β values. This solution recovers the experimental and simulated results near the clamp and inside the helical structure, but do not capture behavior near the free tip nor the transition from the straight to helical configurations within the rod. To describe the transition from straight to helical portions of the rod (12 . s . 18), in the vicinity of s = s∗LH , we see that β varies quickly with s. Returning to the full form of Eq. (3.6) and expanding the integrand near s∗LH , minimization of the total energy of the rod recovers a differential equation known as the second Painlevé equation, whose solution is knows as the Hastings-McLeod solution [116]. The solution gives an expression of β near s = s∗LH , and is referred to as the Inner Layer (IL) solution, 88 2 C w1/3 βIL (s) = p BHML 4 − 3C s − s∗LH w−1/3 2 C w1/3 Ai ∼p 4 − 3C s − s∗LH w−1/3 , (3.7) where Ai(S) is the Airy function of S. The solution of Eq. (3.7) is found to capture the smooth transition between straight and helical portions of the rod in Fig. 3-10, shown as a black line. In this transition region, experimental, simulation, and theoretical results all show excellent agreement with one another. We can now predict the transition from localized to global helices using Eq. (3.7) by solving βIL (0.95L) = 1.5◦ (using the same definition of the transition as applied to experiment and simulation). Fig. 3-11 shows the phase space diagram of the system once again, including the theoretical predictions for the planar to non planar transition (vertical thin black line, where w = wc = 0.24, found by the linear stability analysis) and the localized helix to global helix transition (thick black line, using Eq. (3.7). With this last prediction, we have described the physically relevant boundaries to our phase diagram, rationalizing the configurations of a naturally curved rod hanging under self weight and free at one end. Obtaining quality experimental data, including digital three-dimensional reconstructions, has allowed us to create a phase diagram. Using this data to validate numerical data, we then expanded the phase diagram to gain a better representation of phase boundaries. The planar to non-planar transition was understood by a mechanical scaling argument and a more sophisticated analysis was validated against the variation of β with s in both experimental and simulated reconstructions with zero fitting parameters. The next section will briefly discuss the effect of adding an additional constraint; clamping both ends of a naturally curved rod. 89 3.3 Additional Boundary Conditions Thus far, we have considered the shape of a naturally curved rod hanging under its own weight, clamped at the top end and free at the bottom. This sections investigates the behavior and stability of a naturally curved rod hanging under its own weight, but clamped at both ends, as shown in Fig. 3-12 and known as the writhing experiment. The main contributions of the author in this project included designing precision experiments and connecting results with theoretical predictions. For a more detailed presentation of this particular problem, refer to [109, 110]. Figures in this section are adapted from [110]. In this experiment, a rod of total length L is clamped between two concentrically aligned drill chucks. One of the ends could be displaced a horizontal distance, δ, toward the other clamp or an end could be rotated by an angle, Φ, relative to the other clamp. For both cases, the effect of the rod’s natural curvature, κ0 , was investigated. 5 cm imposed rotation imposed distance Figure 3-12: The writhing experiment, wherein a rod of total length L was suspended under its own weight between two concentrically aligned clamps. A clamp could be displaced a distance δ or rotated an angle Φ relative to the other clamp. Adapted from [110]. 91 Similar to the previous study of a rod clamped at one end, this natural curvature, once large enough, qualitatively and quantitatively changed the behavior of the rod. In parallel to experiments, simulations were run by Arnaud Lazarus, using the same method described in §3.2. The first test discussed is the displacement-controlled experiment. Here the chucks were not allowed to rotate and the rod was clamped in an initially twist-free configuration. The displacement of the clamps was normalized by the rod length, δ/L, and images were captured from the top view of the experiment, as shown in Fig. 3-13. The rod initially hangs in a catenary shape in the yz plane (the plane between the clamps and parallel with gravity), until a critical displacement, at which point the rod undergoes an out of plane deformation. This out of plane behavior is different for rods with different natural curvatures, κ0 . Fig. 3-13 a1-3) shows experimental and simulation results for a rod with low natural curvature, κ0 = 16.5 m−1 . In this case, the rod maintains a catenary shape until a1) a) a2) a3) b2) b3) Experiments X Simulations b1) Z Y b) Figure 3-13: Displacement-control test for rods with a) κ0 = 16.5 m−1 and b) κ0 = 39.3 m−1 , with experiments shown on a black background and simulations on a white background. For the low natural curvature rod, a rotationally symmetric out of plane instability is observed at normalized displacement δ/L ∼ 0.8. The high natural curvature rod exhibits a “horseshoe”-like out of plane instability, at approximately half the normalized displacement (δ/L ∼ 0.4) as the lower natural curvature rod. Adapted from [110]. 92 relatively high end displacements (δ/L ∼ 0.8), at which point the rod buckles out of plane in a rotationally-symmetric manner, as shown in Fig. 3-13 a3). At such high end displacements, portions of the rod are in compression, exhibiting buckling behavior similar to that predicted for straight rods in the absence of gravity and observed in stiff nitinol rods (a nickel titanium alloy) and predicted by Eq. (2.5) [42,56]. For rods undergoing this form of out of plane instability, the critical displacement is relatively unaffected by changing κ0 of the tested rod. The configurations shown in Fig. 3-13 b1-3), for a rod with high natural curvature (κ0 = 39.3 m−1 ), however, show that for displacement past a critical value (δ/L ∼ 0.4), the rod displays a different out of plane behavior. The rod takes on a “horseshoe” shape characterized by deformation almost entirely all in the same x-direction. For rods exhibiting this type of behavior, the instability was triggered at a point when the entire rod was in a tensile state. Considering the horseshoe as a chirality inversion, the instability can be understood as a curvature to writhe problem, as described in §2.1.2 [45]. Specifically, Eq. (2.7) predicts the critical tension below which a naturally curved rod will transition from straight to helical with a chirality inversion (to preserve zero total twist in the rod). Applying Eq. (2.7) recovered the scaling of the observed transition, and changes in natural curvature had a strong effect on the critical displacement. The second test consisted of clamping and fixing the rods at a set displacement (δ/L = 0.27), and quasi-statically rotating one clamp in relation to the other an angle, Φ. At a critical rotation angle, the twist in the rod localized into a single, selfcontacting structure, known as a plectoneme [2]. Fig. 3-14 a1-3) shows photographs and simulated reconstructions of a straight rod being twisted. The first out of plane instability in Fig. 3-14 a2) is a benign, two-lobed mode. This mode, however, is followed by plectoneme formation at Φ = 2040◦ . For a naturally curved rod, as shown in Fig. 3-14 b1-3), the rod starts in an out of plane configuration (the “horseshoe” configuration from the displacement-control test). The rod is able to store more twist in a helical configuration than the straight rod before plectoneme formation at the higher critical twist of Φ ∼ 2900◦ . 93 X a1) Z a) Simulations a2) a3) b2) b3) Y Experiments b1) b) Figure 3-14: Twist-control test for rods with a) κ0 = 0 m−1 (straight) and b) κ0 = 44.8 m−1 , with experiments shown on a black background and simulations on a white background. For the straight rod, a dual-lobed structure (Φ = 720◦ ) precedes plectoneme formation at Φ = 2040◦ . The high natural curvature rod exhibits a “horseshoe”-like out of plane instability at low twist. A helical structure is able to store more twist than the straight rod before plectoneme formation at Φ ∼ 2900◦ . Adapted from [110]. In both test cases, natural curvature was observed to affect behavior in a nontrivial way. Rods with low values of natural curvature behaved as essentially straight rods. Past a critical value of κ0 , however, behavior was fundamentally altered. In the case of the displacement-control experiment, critical displacement to first instability was lowered with addition of natural curvature. The mode became tension – not compression – driven. In the case of twist-control, the addition of κ0 delayed plectoneme formation. Precision experiments enabled the exploration of natural curvature as a control parameter, highlighting that the analysis of rods should consider intrinsic curvature (either manufactured or acquired through storage and/or loading) in possible instabilities. For a more thorough discussion of this problem, the reader is encouraged to refer to [109, 110]. 94 3.4 Outlook The effect of natural curvature on the behavior of rods hanging under self-weight was explored in this chapter. The main focus was on the shape of a naturally curved rod clamped at one end and free at the other end. Depending on the control parameter, it was found that the rod could take on a planar or non-planar configuration, with the non-planar configurations further subdivided into localized helices and global helices. As part of a collaborative effort, experimental, numerical, and theoretical tools were used to rationalize and confirm the transitions between these different configurations in a phase diagram. In the second case considered, natural curvature was also found to have a strong effect on behavior on the writhing problem. The scenarios investigated in this chapter had simple boundary conditions, enforcing position and rotation and the end(s) of the tested rod. While geometrically nonlinear, the resulting behavior could be rationalized in a predictive, and even closed form manner in some cases. Subsequent chapters will focus on situations for a thin elastic rod with more complex boundary conditions and frictional interactions. Emphasis will be placed on experimental results as existing literature becomes progressively more scarce. The next chapter explores a compressed rod inside of a cylindrical constraint. 95 96 Chapter 4 Compressing a Rod in a Cylinder The last chapter focused on the configuration of a naturally curved rod with relatively simple boundary conditions. In this chapter, the rod geometry itself is simplified while the boundary conditions are made more complex. We investigate the first two buckling modes and post-buckling behavior of an intrinsically straight rod under axial compression, pinned on both ends, inside of a cylindrical constraint. We refer to this configuration as the classic case. A compressive axial load is provided by displacement of one end, leading to buckling of the rod as it lies along the bottom of the constraint. Buckling causes the rod to climb up the cylindrical constraint, with an associated penalty due to gravitational potential energy. This penalty is proportional to the weight per unit length of the rod, w, and the radial clearance between the rod and constraint, ∆r. The first buckling mode of the rod satisfies the competition between gravity and bending by adopting a sinusoidal or snaking configuration with a characteristic wavelength. As compression continues, axial load increases, eventually passing the second critical load. At this point the rod buckles into a helical configuration. The normal force between the helically configured rod and cylindrical constraint increases dramatically and the axial load at both ends of the rod begin to diverge. In §4.1 the experimental apparatus built to explore this phenomenon is described. The results of the experiment are presented in §4.2, including interpretation and comparison with existing theory, as well as with simulations performed by a collaborator. Finally, §4.3 discusses open issues in the research. 97 4.1 The Experiment In Fig 4-1, we show a picture of the experimental apparatus built to investigate buckling of a rod compressed inside a cylinder in Fig. 4-1 (a) with a schematic of the entire system in Fig. 4-1 (b). Mechanically, the apparatus consisted of a fixed cylindrical constraint (pipe), a pinned rod lying along the constraint’s bottom surface, and a compression system. The compression system consisted of a pinned end which could be displaced along the constraint’s axial direction (the input end) and a pinned end that was fixed (the output end). Both ends were pinned centered on the constraint’s long axis. The constraint and compressed rod are described in more detail in §4.1.1 and the compression and data acquisition components of the apparatus are discussed in §4.1.2. Finally, the experimental protocol is outlined in §4.1.3. a) 10cm b) Linear Actuator Input Force Sensor Constraining Pipe Output Force Sensor Compressed Rod Pipe Supports 10 8 DAQ 6 4 2 0 Computer w/LabView Figure 4-1: a) The experimental apparatus in the lab. Insets show the input end (white) and pipe supports (black) in more detail. b) Schematic representation of the entire experimental system. 98 4.1.1 Material Selection and Properties The cylindrical constraint was supported on a custom-built rigid aluminum frame1 . The pipe was held in place with five acrylic clamps (pictured in detail in Fig. 4-1), which were separated by 75 cm. The constraining pipe was constructed out of two concentrically aligned acrylic pipes2 , 3.095 m in total length. Seven inner diameters (I.D.) were explored, ranging from I.D. = 3 mm to 19 mm (see Fig. 4-2). The compressed rod was made out of Nitinol, a nickel titanium hyperelastic alloy, and two rod diameters (1.14 mm and 1.6 mm) were used3 . Each rod was also cut to a 3.095 m length and both ends were ground to be round. 2cm Pipe I.D. [mm] 3.0 4.3 6.3 9.4 12.4 15.7 19.0 r [mm] 0.70, 0.94 1.35, 1.59 2.35, 2.59 3.90, 4.14 5.40, 5.64 7.05, 7.29 8.70, 8.94 Figure 4-2: Rod diameter (black circle) compared to the inner diameters (I.D.) of pipes tested, true scale (d = 1.6 mm shown). Rod diameter shown is 1.6 mm. Each pipe I.D and its corresponding radial clearances (∆r) is also listed (∆r for d = 1.6 and 1.14 mm, respectively). The material properties of the rods used are summarized in Table 4.1. Each of the properties listed was experimentally measured. Friction was steel on plastic, and found to be accurately described by a simple dry Coulomb friction, which matches existing theory and field drilling conditions of lubricated steel pipe on steel casing. We will now briefly describe the methods used to measure the coefficients of friction and the Young’s modulus. The static and dynamic coefficients of friction (µs and µ, respectively) between the nitinol rods and the inner surface of the acrylic rods was measured using a tilt 1 Constructed with 80/20 Aluminum T-slotted framing [111]. Optically clear acrylic pipe purchased from McMaster-Carr. 3 Nitinol Devices & Components SE 508 wire with black oxide surface [117]. 2 99 Property Diameter, d Young’s Modulus, E Coefficient of Static Friction, µs Coefficient of Dynamic Friction, µ Density, ρ Rod 1 Rod 2 1.14 mm 1.6 mm 68.67 ± 0.27 GP a 68.05 ± 0.15 GP a 0.45 ± 0.07 0.30 ± 0.03 6539 ± 82 kg/m3 Table 4.1: Material properties of Nitinol rods used for the experiments in this chapter. All properties were experimentally measured, with static and dynamic friction coefficients (µs and µ, respectively) measured in an acrylic pipe with I.D. = 15.7 mm. test, a photograph of which is shown in Fig. 4-3. A sample of acrylic pipe was fixed to an aluminum frame via a hinged clamp, which allowed the pipe to be tilted at an arbitrary angle θ from horizontal. The test consisted of three repeated steps: i) A 27 cm-long sample of Nitinol (d = 1.6 mm) was first placed into the acrylic pipe (I.D. = 15.7 mm) and allowed to come to rest on the pipe’s bottom surface. ii) The pipe was then tilted up (increasing θ) by hand until the Nitinol sample began sliding. iii) Finally, the pipe was tilted down (decreasing θ) until the Nitinol sample stopped sliding. These tests were recorded with a digital video camera4 The angles at which the sample started and stopped sliding, θslip and θstick , respectively, were extracted using the image processing software5 for 23 tilt tests. These angles could be related to the coefficients of friction by µs = arctan (θslip ) and µ = arctan (θstick ). These values of the coefficients of friction are measured for the axial direction, but are assumed to be isotropic (such that lateral sliding and axial sliding are met with the same frictional resistance). Single tests were performed to ensure that the coefficients of friction were equal for different acrylic pipes. The Young’s modulus, E, was measured indirectly by measuring the natural frequency of a cantilevered section of rod. This method agreed well with other measure4 5 Kodak PlaySport Zx5 digital video camera, recording at 29.97 frames per second. ImageJ, provided by the National Institute of Health [118] 100 5cm Acrylic Pipe Sample Hinged Clamp g Figure 4-3: Photograph of tilt test for measuring static and dynamic coefficient of friction (µs and µ, respectively) between Nitinol (d = 1.6 mm) and acrylic pipe (I.D. = 15.7 mm). The acrylic pipe was hinged on the lower end and progressively tilted up (increasing θ) until the sample started to slip, at which point the pipe would be tilted down (decreasing θ) until the sample stopped sliding. ment techniques for other materials, as described in Appendix A. A rod was clamped vertically between two acrylic plates with a free length extending above the plates. Fig. 4-4 shows a photograph of the experimental setup. The free tip was displaced a small distance (∼ 1 cm) and released, exciting the first mode of vibration. The vibration was recorded with a digital video camera6 for three seconds. The natural frequency, fn , was extracted from the video (averaged over 50-100 periods). The Young’s modulus was then calculated from [119], β12 fn = 2πL2 s EI , ρA (4.1) where β1 is a constant dependent on the boundary conditions (β1 = 0.597π for our clamped-free conditions), L is the free length, I is the second moment of inertial (I = πd4 /64 for circular solid cross-sections with diameter d), ρ is the volumetric mass density, and A is the cross-sectional area (A = πd2 /4 for circular cross-sections). The test was repeated five times for each measurement. Measurements were taken 6 Nikon 1 J-3 digital camera recording at 402 frames per second. 101 2.5 cm Figure 4-4: Photograph of a Nitinol rod (d = 1.6 mm) clamped between two acrylic plates with cantilevered length L = 126.18 mm. Displacing and releasing the tip would excite the first mode of vibration, the frequency of which was used to measure the Young’s modulus, E. for two different free lengths for each diameter of nitinol rod to ensure consistent measurements. Finally, in order to maintain consistent values of friction (both µs and µ), the acrylic pipes and Nitinol rods were cleaned before each test with a cotton cloth to remove dust which may have settled on the contact surfaces. We now turn to the experimental components which provided compression of the rod inside the cylindrical constraint. 4.1.2 Compression and Data Acquisition System A photograph of the compression system is shown in Fig 4-5. This system had two primary functions: i) to quasi-statically displace the input end of the rod along the cylindrical constraint’s long axis and ii) to record the reaction force at both (input 102 Linear Actuator Mechanical Coupling/ Slider Input Aluminum Force Pinned B.C. Sensor 5cm Acrylic Pipe Nitinol Rod Figure 4-5: Photograph of the compression system (the input end), which consisted of a linear actuator, which displaces an aluminum pin. The aluminum pin was connected to the actuator through a force sensor and an acrylic coupler on a slider. and output) ends of the rod. Displacement was controlled using a linear actuator7 . The actuator was coupled to an acrylic plate that was mounted on a two rail slider, ensuring horizontal travel. A force sensor8 connected the acrylic plate to an aluminum pin that connected to the Nitinol rod. The aluminum pin was machined to allow the Nitinol rod end to rotate with three degrees of freedom while imposing axial position by inserting the Nitinol rod into a small (0.3 mm deep, 1.6 mm diameter) recess on the face of the pin. The input end had a maximum displacement range of 2 cm due to the acrylic slider. The output end was connected to an identical aluminum pin, which was attached to a rigid constraint via a second force sensor. The force sensors were put in line with the two pinned boundary conditions to 7 Intelligent Motion Systems MDrive 14 Plus Linear Actuator, with 51200 micro-steps per revolution and 2048 edges per revolution resolution (internal encoder). Actuated with a 0.25-inch-diameter 10-inch-long lead screw [120]. 8 Futek miniature S-Beam load cell (Model LRM200), with a 22.2 N capacity and ±0.1% maximum nonlinearity and hysteresis. Signals were amplified with a Futek strain gauge amplifier (model GSG110) [121]. 103 allow for the reaction forces at both ends of the compressed rod to be recorded. A LabView9 virtual instrument (VI) was programmed to control the linear actuator10 as well as record readings from the force sensor11 . The signals from both force sensors were recorded to a text file while a test was being run. The number of displacement steps and step size were sufficient to calculate the displacement associated with each force reading. The LabView VI was programmed such that force measurements were taken only when the linear actuator was not moving, and vice versa. 4.1.3 Experimental Protocol The previous two subsections have discussed the components of the experimental apparatus. Here we outline the protocol for this experiment. A run consisted of three distinct steps: i) Mounting the sample by placing both ends into the recesses in the aluminum pins, ensuring the load on both ends of the rod was negligible (< 0.4N ). ii) Loading the sample was accomplished by 0.01 mm displacement steps, with force readings recorded between each step. The imposed displacement was stopped at prescribed intervals (depending on the system geometry of a given test) to measure wavelength or pitch of a buckled rod. Compression was continued well into the helical regime until either reaction load was within 90% of the force sensor’s capacity or the displacement reached the maximum extent of the slider (∼ 2 cm). iii) Unloading the sample consisted of reversing the loading process by stepping backward with 0.01 mm increments, recording the reaction force back to the start position. Near the start position, the rod would sometimes lose contact with the aluminum pin and fall out, ending the run. 9 National Instruments Laboratory Virtual Engineering Workbench, a proprietary software. LabView communicated through and Intelligent Motion Systems, Inc. USB to RS422 converter (model number MD-CC4). 11 The force sensor communicated with LabView through a National Instruments Data Acquisition card (model NI USB-6210), 8 analog inputs, 250 kilosamples per second. 10 104 Five runs were performed to constitute a test in order to ensure reproducibility. After a test, the Nitinol rod and/or the acrylic pipe was changed out to test a new geometry and the rod and acrylic pipe were cleaned with a cotton cloth. 4.2 Results and Interpretation The previous section (§4.1) introduced the experimental apparatus built to investigate the behavior of a rod compressed inside a cylindrical constraint. We now describe the results of our experimental investigation, beginning with a description of the reaction force recordings (§4.2.1), critical forces and length scales (§4.2.2), and ending with the effect of imperfections in the cylindrical constraint’s geometry (§4.2.3). Selected results will be compared to simulations that were performed by a collaborator for this project, Dr. Tianxiang Su. The dynamic behavior of the rod was modeled using a custom code adapted from the modeling of drillstring dynamics [122,123]. A discretized rod is modeled as a chain of rigid bodies connected through axial, shear, torsion, and bending springs, with spring constants defined using beam theory [9]. Dry friction between the rod and pipe is modeled following Coulomb’s Law, allowing for both static and dynamic friction. Contact between the rod and constraint is modeled as a viscoelastic frictional contact using a modified Hertzian contact model that takes the compliance of the rod cross-section into account, as is described in [122, 124]. In the simulations, as the rod is compressed inside of a cylindrical constraint, the elastic strains and stresses are calculated from the spring constants and the rod configuration. Then, external forces from gravity, the contact force between rod and constraint, and frictional forces are applied to the current rod configuration. The equations of motion are then integrated using a Newton-Raphson iteration scheme [125] and the rod configuration is updated for a small time step (1.6 × 10−6 s). The simulations were performed with zero fitting parameters; all values were experimentally measured to enable direct comparison. We will show that similar load-displacement behavior was observed across the dif105 ferent radial clearances tested and that critical loads for transitions to sinusoidal and helical configurations varies with radial clearance in good agreement with theoretical predictions from existing literature (discussed in §2.2.2). We will then show that imperfections in the cylindrical constraint’s geometry can have a strong effect on the critical loads. 4.2.1 Load Displacement Signals Reaction force signals were recorded throughout the quasi-static compression of a hyperelastic Nitinol rod inside of an acrylic pipe. The progression of configurations (straight, sinusoidal, and helical) described at the beginning of this chapter was observed for every radial clearance tested, and each was recognizable in the reaction force-displacement signal. In Fig. 4-6 (a) and (b) we plot typical input reaction force (Pin ) and output reaction force (Pout ) as a function of imposed displacement (δ) for pipe I.D.= 9.4 mm and rod d = 1.14 mm. In Fig. 4-6 (a), data for the entire test is shown, while Fig. 4-6 (b) shows the first 4 mm of compression. Note that Pin and Pout appear to have approximately identical values until ∼ δ = 2.5 mm, after which point they diverge, as is clear in Fig. 4-6(b). The difference between the two loads is defined as ∆P = Pin − Pout . Fig. 4-6 (b) is a closer view of the beginning of the test (0 < δ [mm] < 4) region indicated in Fig. 4-6 (a) with black lines. In this closer view, we can see that the initial response of the rod is very stiff, as indicated by a steep linear slope between s both reaction forces and displacement. Sinusoidal buckling at δ = δcr is characterized by a dramatic softening of the response. After sinusoidal buckling, Pin ≈ Pout until a h clear load drop in Pin at δ = δcr , which is an indication of helical initiation. The drop in load is associated with a portion of the rod losing contact with the constraining h cylinder. As noted for δ > δcr , we see that Pin > Pout , but we also observe more variation in Pin than in Pout . This variation is caused by the formation of new helical pitches as well as the sliding (in the constraint’s axial direction) of existing pitches. Note that friction prevents the rod from sliding freely to re-arrange into a lower 106 a) Reaction Force, P [N] 10 8 6 4 Input Force, Pin 2 Output Force, Pout 0 0 0.002 0.004 0.006 0.008 Displacement, δ [m] 4 Input Force, Pin b) Output Force, P 3.5 Reaction Force, P [N] 0.01 out 3 2.5 2 1.5 1 0.5 0 0 0.5 1 1.5 2 2.5 Displacement, δ [m] 3 3.5 4 −3 x 10 Figure 4-6: a) Reaction force at the input end (Pin ) and output end (Pout ) as a function of imposed displacement (δ) for a typical test. Pin and Pout start to diverge with a difference ∆P after ∼ δ > 2.5 mm. The black box indicates extent of b), indicating critical displacements (δcr ) and reaction forces (Pcr ) for sinusoidal and helical buckling. 107 energy, constant chirality configuration. As such, in nearly every test observed, the helical configuration was very clearly frustrated by friction, and as such exhibited many inversions of chirality (sometimes known as perversions). This observation is not accounted for in analytical treatments in the existing literature. The growing separation between Pin and Pout shown by the data in Fig. 4-6 (a) is caused by the rapid increase in normal contact force between the rod and constraining cylinder after helical buckling. This increasing normal contact force causes an increasing frictional drag, which, when integrated over the rod length, makes up ∆P = Pin − Pout . Fig. 4-7 shows ∆P as a function of δ with logarithmic h axes. For imposed displacements δ < δcr , the load difference is approximately constant and equal to the sliding frictional resistance expected of ∆P = µmg, where µ is the coefficient of dynamic friction, m is the mass of the rod (m = ρAL, where ρ is the volumetric mass density, A is the cross-sectional area, and L is the length of the rod), and g is the acceleration due to gravity. These values were summarized in Table 4.1 above. h Upon helical initiation (δ > δcr ), ∆P is seen to increase quadratically with δ. This functional form is found to agree with Eq. (2.23), repeated here for convenience, Wn = EA2 ∆r 2 δ , 4IL2 where E is the rod’s Young’s modulus, A is the cross-sectional area, ∆r is the radial clearance, I is the second moment of inertial (I = πd4 /64 for circular solid crosssections with diameter d), and L is the length of the rod. The pre-factor of Eq. (2.23) does not, however, capture the exact behavior of ∆P as a function of δ. This is expected as its derivation assumes a constant helix (with no chirality inversions) and frictionless interaction between the rod and pipe. 108 1 10 1 2 Load Difference, Δ P [N] 0 10 −1 10 −2 10 −3 10 −4 10 −3 10 Displacement, δ [m] −2 10 Figure 4-7: Load difference between input and output ends, ∆P = Pin − Pout , as a function of imposed displacement, δ. For displacement before helical buckling, h δ < δcr , ∆P is observed to be approximately equal to the sliding friction of the rod h in the channel, ∆P ≈ µmg. For δ > δcr , however, ∆P ∝ δ 2 . 109 Typical load-displacement behavior for a rod compressed inside a cylindrical constraint was shown in Fig. 4-6 (a) and (b). Two processes exhibited variation between runs: sinusoidal buckling and post-buckling behavior after the rod took on a helical configuration. Fig. 4-8 shows the reaction forces as a function of displacement for a rod with d = 1.6 mm and constraining pipe with I.D. = 9.4 mm. Note the pronounced peak (labeled “Overshoot” in Fig. 4-8) at sinusoidal buckling s (δcr = 0.39 mm). This is a common feature in buckling problems, wherein loading can be applied past the critical load with no buckling before snapping to the stable, buckled configuration occurs (this type of transition is known as a supercritical pitchfork bifurcation). When this behavior was experimentally encountered, the point immediately following the drop in Pin was taken to be the transition from straight to sinusoidal configurations, as indicated in Fig. 4-8. 18 16 Input Force, Pin Output Force, Pout Reaction Force, P [N] 14 12 10 8 Overshoot 6 4 2 0 0 1 2 3 4 Displacement, δ [m] 5 −3 x 10 Figure 4-8: Reaction forces (Pin and Pout ) as a function of imposed displacement (δ) for a rod with d = 1.6 mm compressed inside a pipe with I.D. = 9.4 mm. The transition from straight to sinusoidal configurations in this particular run is associated with a large drop in reaction forces after the point labeled ”Overshoot.” This is a typical feature of some buckling systems, with the critical load taken immediately after the load drop. 110 The second deviation from typical behavior that was observed in some cases is illustrated in Fig. 4-9, showing the load-displacement signals for a rod with d = 1.14 mm compressed inside a cylindrical constraint with I.D. = 9.4 mm (the same test geometry as illustrated in Figs. 4-6 and 4-7). Here we focus on the helical h configuration of the rod (δ > δcr ), where large drops are seen in both Pin and Pout at δ ≈ 3 and 8 mm. These sudden and significant drops in load were not frequent, but, when they occurred, they were associated with a reconfiguration of the helical structure of the rod. Typically, this involved the joining of two helical sections of opposite chirality into a uniform chirality. While this process caused an immediate drop in ∆P , the loads were seen to separate to previous values of ∆P quickly (e.g. ∆P has recovered to its original slope by δ ≈ 4 and 11 mm for the two events in Fig. 4-9). 18 16 Reaction Force, P [N] 14 12 10 Input Force, Pin Output Force, Pout 8 6 4 2 0 0 0.002 0.004 0.006 0.008 Displacement, δ [m] 0.01 Figure 4-9: Reaction forces (Pin and Pout ) as a function of imposed displacement (δ) for a rod with d = 1.14 mm compressed inside a pipe with I.D. = 9.4 mm. Two significant drops in reaction forces (δ ≈ 3 and 8 mm) were associated with a rearrangement of the helical configuration of the rod. This was typically associated with two sections of opposite chirality joining into a section with uniform chirality. 111 Lastly, the behaviors discussed in this section were observed across all geometries tested. In Fig. 4-10, we plot Pin as a function of imposed displacement for four different test configurations, ranging from a tight clearance to a loose one (1.35 ≤ ∆r[mm] ≤ 5.63). The behavior is again characterized by an initially stiff behavior, followed by a softer response. A drop in Pin is associated with the rod beginning h to buckle into a helical configuration (δcr indicated for each clearance in Fig. 4-10), and then Pin is observed to begin stiffening. Fig. 4-10 shows similar behavior, but it also shows different critical loads for each of the test configurations. The next section explores the variation of these critical loads as a function of radial clearance for both diameters of Nitinol rod tested. 10 9 Reaction Force, Pin [N] 8 7 d=1.6mm, ID=4.3mm d=1.6mm, ID=9.3mm d=1.14mm, ID=9.3mm d=1.14mm, ID=12.4mm 6 5 4 3 2 1 0 0 0.5 1 1.5 Displacement, δ [m] 2 2.5 −3 x 10 Figure 4-10: Input reaction force, Pin , as a function of imposed displacement, δ, for four different test configurations (with 1.35 ≤ ∆r[mm] ≤ 5.63). The same typical behavior is seen in all four cases, with an initially stiff response softening before a drop in Pin , followed by a stiffening behavior. 112 4.2.2 Critical Loads and Length scales We now turn to studying the loads at which the constrained rod configurations transitioned from straight to sinusoidal (Pcrs ) and from sinusoidal to helical (Pcrs ), as well as the wavelength (λscr ) and pitch (phcr ) length scales associated with these transitions. The previous section outlined how critical loads were identified by critical points on the Pin -δ signal. Experiments were also visually observed for the transitions, so λ or p data could be measured at any point (as the tests were performed quasistati√ cally). To simplify plots with two rod diameters, loads will be normalized by 2 EIw (where E is the rod’s Young’s modulus, I is the rod’s second moment of inertia, and w is the rod’s weight per unit length) and critical length scales will be normalized by (EI/w)1/4 . Normalized parameters will be denoted with an overbar (e.g. - the p s normalized sinusoidal buckling force, P cr = Pcrs /(2 EI/w)). These normalizations are suggested by previous analytical work in the literature, as discussed in §2.2.2 and will be discussed more for each quantity measured. s The normalized sinusoidal wavelength at the onset of sinusoidal buckling, λcr , was measured for each test and is plotted as a function of radial clearance, ∆r, in Fig. 4-11 (a). Re-arranging Eq. (2.16) as, s λcr = λscr w 1/4 = 2π∆r1/4 , EI we can also compare experimental data to a theoretical prediction based on previous work. Radial clearance has little to no effect on experimentally measured wavelength. s The theoretical prediction captures the magnitude of λcr . However, recall that Eq. (2.16) assumes frictionless interaction between rod and constraint. This is also the case for the next critical length scale discussed: the helical pitch. 113 1.8 a) 1.6 Normalized Wavelength, 1.4 1.2 1 d=1.14 mm Rod d=1.6 mm Rod Theoretical Prediction 0.8 0.6 0.4 0.2 0 0 1.6 2 4 Radial Clearance, 6 r [m] 8 −3 x 10 b) 1.4 Normalized Pitch, 1.2 d=1.14 mm Rod d=1.6 mm Rod Theoretical Prediction 1 0.8 0.6 0.4 0.2 0 0 2 4 Radial Clearance, 6 r [m] s 8 −3 x 10 Figure 4-11: Normalized measured a) wavelength, λcr , and b) pitch, phcr , at the onset of buckling as a function of radial clearance, ∆r, for both rod diameters. Experimental values are shown compared to theoretical prediction (Eqs. (2.16) and (2.21), respectively). 114 The normalized helical pitch at the onset of helical buckling, phcr , was also measured for each test. The results of these measurements, as a function of ∆r, are plotted in Fig. 4-11 (b). The theoretical prediction for phcr once again assumes no frictional interaction between the rod and constraint. From Eq. (2.21) we have, phcr = phcr w 1/4 = π (8∆r)1/4 , EI where pitch is proportional to ∆r1/4 , with a smaller (∼ 84%) pre-factor than the s prediction for λcr . Once again, experimental measurements show little to no variation with ∆r, and theoretical prediction captures the order of magnitude over the range of radial clearances tested. s The normalized critical force associated with sinusoidal buckling (P cr ) was measured for each test and compared with two different theoretical predictions from existing literature, as shown in Fig. 4-12 (a), as a function of ∆r. The first theoretical prediction assumes no frictional interaction between the constrained rod and pipe. It was originally given as Eq. (2.15) and is repeated here, s P cr (µ Pcrs 1 √ = 0) = =√ , 2 EIw ∆r and is plotted in Fig. 4-12 (a) as a dashed line. The second theoretical prediction based on existing theory assumes lateral friction between the constrained rod and pipe, and applies and amplification factor to Eq. (2.15) based on the coefficient of dynamic friction, µ, or, as stated in Eq. (2.25), s P cr (µ s Pcrs ψcr √ √ = µ) = = , 2 EIw ∆r s where ψcr ≈ 1.67 (Eq. 2.26) for the measured value of the coefficient of dynamic friction, µ = 0.3(±0.03), given in Table 4.1. This prediction is plotted in Fig. 4-12 (a) as a solid line. We can see that the experimental data collapses onto a single curve √ (indicating the normalization factor 2 EIw is appropriate), and can see an inverse s s relationship between P cr and ∆r, with P cr varying approximately 50% over the range 115 60 a) d=1.14 mm Rods d=1.6 mm Rods Theoretical Prediction, μ=0 Theoretical Prediction, μ=0.3 Normalized Sin. Buckling Load, 50 40 30 20 10 0 0 2 90 b) 8 −3 x 10 d=1.14 mm Rods d=1.6 mm Rods Theoretical Prediction, μ=0 Theoretical Prediction, μ=0.3 80 70 Normalized Helical Buckling Load, 4 6 Radial Clearance, Δ r [m] 60 50 40 30 20 10 0 0 2 4 6 Radial Clearance, Δ r [m] 8 −3 x 10 s Figure 4-12: Normalized critical a) sinusoidal and b) helical buckling load (P cr and h P cr , respectively) plotted as a function of radial clearance, ∆r. Experimental values are compared to theoretical predictions from the literature for the cases of frictionless (dashed line) and frictional (solid line) interaction between rod and constraint. 116 of radial clearances tested. At large radial clearances (∆r & 4 mm), experimental values agree with the theoretical prediction based on Eq. (2.25), which includes frictional interaction between the rod and the pipe. At tighter radial clearances (∆r . 4 mm), however, the disagreement is more pronounced. This disagreement is discussed more in depth in §4.2.3, where imperfections in the constraint’s geometry give rise to the divergence between theoretical prediction and experimental values. h The measured normalized critical force associated with helical buckling, P cr , for each test was also compared with two different theoretical predictions from existing h literature. The results of P cr as a function of radial clearance are plotted in Fig. 4-12 (b). Once again, the first theory assumes no frictional interaction between the constrained rod and pipe. It was first given as Eq. (2.20) and is repeated here for convenience, Ph 2 h P cr (µ = 0) = √ cr = √ , 2 EIw ∆r and is plotted as a dashed line in Fig. 4-12 (b). The second theoretical prediction from existing theory assumes lateral friction between the constrained rod and pipe, again applying an amplification factor to Eq. (2.20) based on µ and the boundary conditions. Repeated here from Eq. (2.27), Ph ψh h P cr (µ = µ) = √ cr = √ cr , 2 EIw ∆r h where ψcr ≈ 2.44 for the experimentally measured value of µ = 0.3 ± 0.03 and pinned- pinned boundary conditions (Eq. (2.28)). This prediction is plotted as a solid line in Fig. 4-12 (b). Trends are similar to those seen in Fig. 4-12 (a) and, for larger radial clearances (∆r & 2 mm), we see good agreement between experimental observation and theoretical prediction including µ. In this range, an inverse relationship between s P cr and ∆r is again observed. For the two smallest radial clearances for the two rods tested (∆r . 2 mm), however, there is significant (∼ 100%) disagreement beh tween experimental measurement and predicted values of P cr . This inconsistency is discussed in §4.2.3 next, where the role of imperfections in the constraining geometry 117 s h will be found to have a significant effect on both P cr and P cr . 4.2.3 Effect of Imperfections In §4.1, the apparatus built to explore the buckling and post-buckling behavior of a rod inside a cylinder was described. A photograph of the apparatus was shown in Fig. 4-1 (a), in which it is clear that the acrylic pipes which make up the cylindrical constraint are supported at discrete points. These clamps were spaced at 75 cm, which also corresponded to unsupported spans of acrylic pipe. An end-supported beam under its own weight will sag with a maximum displacement [9], ςmax , ςmax wL2 = , 384EI (4.2) where w is the weight per unit length (w = ρgA where ρ is the volumetric mass density, g is the acceleration due to gravity, and A is the cross-sectional area), L is the unsupported length, E is the Young’s modulus, and I is the second moment of inertia of the pipe’s cross-section. Eq. (4.2) corresponds to the case of a rod span clamped on both ends, which matches the experimental apparatus (see the inset of Fig. 4-1). The value of E for the acrylic pipes was experimentally measured to be E = 2.2 GP a (using the same procedure outlined in §4.1.1) and the density was measured to be ρ = 1, 195 kg/m3 . Considering ςmax as an imperfection in the system, we compare it to the radial clearance of a test through a dimensionless number, ϑ = ςmax /∆r. A perfectly straight constraint (the assumed shape in existing theoretical prediction) is the case of ϑ = 0. Fig. 4-13 plots the dimensionless imperfection of each test as a function of radial clearance. One can see that for the two smallest diameter constraining pipes (∆r . 2 mm) have imperfections which are the same size or larger than the radial clearance. The third smallest clearance (∆r ≈ 2.5 mm) has a significant imperfection ∼ 0.4∆r. Next, we examine the influence of imperfection on the critical buckling loads, Pcrs and Pcrh . Instead of critical load as a function of radial clearance, we now plot experimental deviation from theoretical prediction as a function of imperfection size. 118 1.8 Constraint Imperfection, ϑ 1.6 d=1.14mm Rod d=1.6mm Rod 1.4 1.2 1 0.8 0.6 0.4 0.2 0 1 2 3 4 5 6 Radial Clearance, ∆ r [m] 7 8 −3 x 10 Figure 4-13: Normalized maximum sag in the constraint, ϑ, as a function of radial clearance, ∆r. We define the sinusoidal correction factor, χs ≡ ((Pcrs )th − (Pcrs )exp )/(Pcrs )exp , (4.3) and the helical correction factor, χh ≡ ((Pcrh )th − (Pcrh )exp )/(Pcrh )exp , (4.4) where (Pcr )exp is an experimental critical load and (Pcr )th refers to a theoretically predicted critical load. A value of χ = 0 is equivalent to experimental and theoretical results agreeing perfectly. Fig. 4-14 (a) shows deviation between experimental and theoretical predictions for the critical sinusoidal buckling load, χs , as a function of constraint imperfection size, ϑ. For the case of both rods (d = 1.14 and 1.6 mm), disagreement between experiment and theory appears to be linearly related to imper119 1.2 a) Correction Factor, χ s 1 0.8 0.6 0.4 0.2 d=1.14mm Rod d=1.6mm Rod 0 −0.2 0 1.8 0.2 0.4 0.6 0.8 1 1.2 Imperfection Factor, ϑ 1.4 1.6 b) 1.6 Correction Factor, χ h 1.4 1.2 1 0.8 0.6 0.4 d=1.14mm Rod d=1.6mm Rod 0.2 0 −0.2 0 0.2 0.4 0.6 0.8 1 1.2 Imperfection Factor, ϑ 1.4 1.6 Figure 4-14: (a) Experimental deviation from theoretical prediction for critical sinusoidal (a) and helical (b) buckling load (χs and χh , respectively) as a function of constraint imperfection size, ϑ. 120 fection size, with increasing imperfection leading to increasing deviation. Negative values of χs are equivalent to experimentally measured values which are greater than theoretically predicted values. Similar to Fig. 4-14 (a), Fig. 4-14 (b) plots the deviation between experimental and theoretical predictions, but for critical helical buckling load, χh , as a function of imperfection size, ϑ. There is, however, a difference in the functional form of χh when compared to χs . For imperfections ϑ . 0.5, there appears to be little to no change in χh . After a critical size of imperfection (in this case ϑ & 0.5), deviations between experiment and theory are observed (growing χh ) in the case of both rods. Fig. 4-15 shows a comparison between experimental results, theoretical prediction, and simulation results for the critical sinusoidal buckling load, Pcrs . In order to confirm the effect of geometric imperfections in the cylindrical constraint, simulations were run with a perfectly straight constraint as well as a constraint with imperfection size matching that calculated for the experimental apparatus. We can see that all three methods agree at large radial clearances (∆r & 4 mm). Simulations match theoretical predictions for ∆r . 4 mm in the case of a perfectly straight constraint and match experimental measurements when constraint imperfections are included, keeping all other parameters equal, confirming constraint imperfections as the cause of the divergence between experimental results and theoretical prediction. 121 Sinusoidal Buckling Load, Pscr [N] 14 12 10 Experiment Friction Theory Simulation − Straight Simulation − Sag 8 6 4 2 0 0 0.002 0.004 0.006 Radial Clearance, ∆ r [m] 0.008 0.01 Figure 4-15: Comparison between experimental measurements, theoretical prediction, and simulation results for critical sinusoidal buckling force, Pcrs as a function of ∆r for a rod with d = 1.6 mm. Simulations were performed for straight constraints as well as those with imperfections matching experimental configurations. For small ∆r, ∆r . 2 mm, simulations including sag match experimental results while perfectly straight constraints match theoretical predictions. 4.3 Outlook An experimental apparatus was designed and built to explore the buckling and postbuckling behavior of a rod compressed inside of a cylindrical constraint. The expected sequence of rod configurations (straight, sinusoidal, and helical) was observed as predicted by existing theoretical work in the literature. Once in the helical regime, frictional drag was seen to increase quadratically with imposed displacement for all tested geometries. Little variation was observed in experimentally measured sinusoidal wavelength and helical pitch length. For straight cylindrical constraints, existing theoretical predictions appear to capture experimental behavior in terms of 122 critical buckling loads. Critical loads were observed to be affected strongly by imperfections, but with a different functional dependence observed for sinusoidal and helical buckling loads on imperfections. There is a rich and extensive theoretical framework within the existing literature relevant to the experiments performed in this chapter. The next chapter explores the related problem of injecting a rod into a cylinder (rather than compression of a fixed length), which is a more recent problem of industrial relevance. 123 124 Chapter 5 Injecting a Rod into a Cylinder In this chapter we explore the buckling and post-buckling behavior of a rod injected axially into a cylinder, referred to as the real case. A compressive axial load arises from the frictional resistance to injection rather than imposed compression as in the last chapter (the classic case). This frictional resistance is characterized by the dynamic coefficient of friction, µ. Similar to the classic case, buckling is resisted by the bending rigidity, EI, as well as the cost in gravitational energy for the rod to climb up the wall of the constraining pipe (a function of the linear unit weight of the rod, w and the radial clearance between the rod and constraint, ∆r). As rod injection continues, axial load builds linearly with injected arc-length, eventually exceeding the first of two critical buckling loads. The first mode of buckling is a sinusoidal or snaking configuration. After further insertion, the load builds past a second critical point, and the rod buckles into a helical configuration. Past this point, the normal force between the helically buckled rod and the cylindrical constraint increases dramatically, eventually preventing continued rod injection. This point is referred to as lockup. This chapter proceeds with a description of the experimental apparatus built to explore and characterize this problem in §5.1. The results of the experiments run on this apparatus are then discussed in §5.2 and compared to existing theory (summarized in the review in §2.3.2) and simulations run in parallel by a collaborator. Finally, §5.3 discusses possible future directions for this problem. 125 5.1 The Experiment An experimental apparatus was built to explore the problem of injecting a rod into a horizontal cylindrical constraint. In Fig. 5-1 (a) we show a photograph of the injection mechanism and the constraining pipe parts of the test in the lab while in Fig. 5-1 (b) we present a schematic diagram of the entire test apparatus. The apparatus consisted of three components: i) The cylindrical constraint sub-system, into which the rod is injected and eventually buckles; ii) The injection sub-system, which drives the injection of the rod into the cylinder at a controlled rate; and iii) The control/data acquisition sub-system, which both controls the injection system and records information. Each of these sub-systems is described in further detail in §5.1.1, 5.1.2, and 5.1.3, respectively. The motivation for the experimental apparatus is a coiled tubing unit found in the oil field, pictured in Fig. 5-2 (a), which injects a continuous length of steel tubing a) b) 20cm Injection System Cylindrical Constraint Slave Injector Digital Video Camera Primary Injector Force Sensor Constraining Pipe Linear Air Bearing Injected Rod Pipe Supports 10 DAQ 8 6 4 2 0 Data Acquisition/Control System Computer w/LabView Figure 5-1: a) Side view of experiment (the rod is injected from left to right). b) Schematic of experimental setup. The entire setup is broken into three distinct blocks: the cylindrical constraint, the injection system, and the data acquisition/control system. Each block is described in more detail in §5.1.1, 5.1.2, and 5.1.3, respectively. 126 into a wellbore, a schematic diagram of which is given in Fig. 5-2 (b). In the case of extended reach wellbores, the wellbore trajectory begins vertical until the kickoff point, where the path deviates in the build section up to a constant tangent angle [71]. In the build section, the build rate (typically measured in degrees per 100 ft drilled) can be constant or can accelerate. In horizontal directional drilling, the tangent angle can be as low as 80 degrees and can include angles greater than 90 degrees (resulting in drilling upward) [72]. Each borehole is tailored to the geology it penetrates, and as such, the borehole’s trajectory is unique. For this study, the vertical and build sections a) Reel Injector Head Coiled Tubing Control Cabin b) Kickoff Tangent Angle Build Section Figure 5-2: (a) A coiled tubing rig consists of a control cabin (where engineers/technicians monitor/control progress), a reel that the coiled tubing is spooled on for storage and transportation, and an injector head that pushes/pulls the coiled tubing in/out of the wellbore. Photo courtesy of Schlumberger-Doll Research. (b) A typically horizontal well consists of a vertical section before the kickoff point, a build section where the angle of the well is changed away from vertical until reaching a constant tangent angle. 127 are neglected, and only the tangent section is considered as it is the section with the most contact between coiled tubing and borehole. Furthermore, for simplicity, the tangent angle is kept at horizontal throughout this study. 5.1.1 Material Selection and Properties The entire experiment was supported on a custom-built rigid aluminum frame1 . The injection system position was adjustable to allow for the rod to be injected at the bottom surface of the constraining pipe. For reference, the end of the pipe closest to the injector is referred to as the front end, while the other end is referred to as the back end. The constraining pipe (L = 2.46m) was held and aligned with acrylic clamps at five locations separated by 60 cm. For practical considerations, the constraining pipe consisted of two pipes held end to end (aligned concentrically), as shown in the schematic in Fig. 5-1 (b). The rod injected into the constraining cylinders was made out of an elastomer (Vinylpolysiloxane) using a custom fabrication process that is described in more detail in Appendix A on rod fabrication. For convenience, the resulting measured rod properties are given in Table 5.1. Unless stated otherwise, rods were fabricated as straight rods, not introducing natural curvature. Property Young’s Modulus, E Density, ρ Diameter, d Poisson’s Ratio, ν Value 1296 ± 31 [kP a] 1210 ± 8 [kg/m3 ] 3.16 ± 0.05 [mm] 0.49 Table 5.1: Material properties of rods manufactured for the experiment. Values taken from Appendix A on rod fabrication Manufactured rods have a small amount of silicone oil on their surface immediately after de-molding. As a result, the rods will adhere to surfaces and/or particles (such as dust). This coating led to inconsistent rod surface properties along a rod as well 1 Constructed with 80/20 Aluminum T-slotted framing [111]. 128 as between tests. In addition, the presence of moisture on the rod surface could potentially cause capillary attraction with the constraining pipe. This is problematic in terms of modeling the contact between the rods and the constraining cylinder, where existing theories assume a simple Coulomb friction model. This simple friction model is typically used because in the case of the field, contact between coiled tubing and casing is steel on steel. The borehole is full of a fluid (typically petroleum), so the contact can also be considered as lubricated surfaces [86]. It was therefore desirable to make the experimental contact one which could be modeled with Coulomb friction. These factors required us to develop an additional step in the preparation of rods for this particular test, referred to here as rod conditioning. After manufacturing, a rod was kept consistently dirty by placing it in a covered plastic container with 50-100 grams of loose chalk powder2 and hand shaken for approximately five to ten seconds. Excess chalk was wiped off of the rod by hand with a paper towel, first from the front end of the rod (the first end to be injected into the constraining pipe) towards the back end, then from the back towards the front, and finally from the front to the back, applying hand pressure to the paper towel while wiping. This coating was applied before every series of injections, as discussed in §5.1.4. Friction between the constraining pipes and injected rods could then be assumed to be dry Coulomb friction, where the resistive frictional force equals the normal force multiplied by a coefficient of dynamic friction, µ. This was more representative of the original application and existing theoretical work and made the simulated contact model more tractable. Equally important, it led to a reproducible rod surface for experiments. For consistent values of µ between injected rod and constraining pipe, the pipe’s inner surface also needed to be considered. Glass pipes3 were selected as a commonly available material which allowed for visual observation of buckling phenomena, as well as preventing static charge from building up between the pipe and the rod during injection [101]. Pipe lengths were nominally 1.2 meters, resulting in needing two pipes to be placed end to end (concentrically aligned) for the experimental setup. Eight 2 3 Irwin brand chalk for snap-lines, primarily composed of Calcium Carbonate (chalk). Pyrex standard wall borosilicate tubing. 129 inner diameters (I.D.) were used for the experiments described in this chapter and are shown in comparison to the rod diameter in Fig. 5-3. The glass pipes also needed to be treated in a regular and consistent fashion. Instead of coating the pipe (similar to rod conditioning), the pipe was regularly cleaned by pulling a cotton cloth through the inside of the constraining pipe from the front end to the back end, twice. This removed any dust that may have drifted into the pipe, as well as any chalk particles that had fallen off the elastomeric rod from previous runs. Rod conditioning and pipe cleaning intervals are described in §5.1.4. The two processes led to a repeatable coefficient of friction of µ = 0.54 ± 0.11 between the rod and glass pipe. This measurement is discussed in depth in §5.2.1. 2cm Pipe I.D. [mm] 6.6 9.3 12.0 14.0 15.7 18.5 21.7 33.6 r [mm] 1.72 3.07 4.42 5.42 6.27 7.67 9.27 15.22 Figure 5-3: Rod diameter (black circle) compared to the inner diameters of pipes tested, true scale. Each pipe inner diameter (I.D.) and its corresponding radial clearance (∆r) is also listed. 5.1.2 Injection Sub-System A photograph of the injection sub-system is shown in Fig. 5-4 (a), with a detailed photo of an injector in Fig. 5-4 (b). This sub-system had two primary functions. The first was as a driving mechanism to inject an elastomeric rod at a controlled rate. The second function of the sub-system was to measure and record the reaction force throughout the progress of a test. We now describe how the sub-system performed both of these functions in more detail. Before injection, the rod was temporarily spooled in a plastic container (lower left corner of Fig. 5-4 (a)). The free end was brought over a lower feeder roller, wrapped 130 a) 5cm b) Slave Injector Feeder Rollers Air Bearing Mount Force Sensor Slack Loop Idler Wheel 1cm Drive Wheel Primary Injector Air Bearing Slider Extra Rod Rod Injection Stepper Motor Stress Relief Figure 5-4: a) Side view of injection system. The rod (in green) is pulled over two feeder rollers through a slave injector and then through a primary injector into the constraining glass cylinder. Reaction forces are transmitted over an air bearing slider to the force sensor. Note that both injectors are fabricated from stacked plates of acrylic. b) Close view of the injector design. A rod is gripped between the drive wheel (controlled by a stepper motor) and idler wheel, then inserted by active rotation of the drive wheel. The rod is guided through a channel in the injector before insertion into the constraining pipe. clockwise once around the upper feeder roller and then fed through the slave injector. A small portion of rod was suspended between the slave injector and primary injector, which fed into the constraining glass pipe. This suspended portion is referred to as the slack loop. It was built into the system so that the back tension on the primary injector was kept constant during injection or withdrawal of a rod. This constant back tension was important in order to measure the reaction force from injecting the rod into the glass pipe at the primary injector. The reaction force was transmitted from the primary injector to a load sensor4 across a linear air bearing5 . A photograph of one of the injectors is shown in more detail in Fig. 5-4 (b). The 4 Futek miniature S-Beam load cell (Model LRM200), with a 4.5 N capacity and ± 0.1% maximum nonlinearity and hysteresis. Signals were amplified with a Futek strain gauge amplifier (model GSG110) [121]. 5 Nelson Air, Inc. linear air bearing (model RAB2). It can support a moment of 15 in-lbs, with a 2-inch-wide slider with 6-inches of travel. It was operated at a nominal pressure of 60 psi as specified [126]. 131 two injectors were identical in design, which was adapted from the Makerbot Cupcake filament drive mechanism [127] and further modified by the author. A portion of the injectors were made using stacked laser cut acrylic pieces. Rod is inserted into a channel that is sandwiched between acrylic plates, then gripped between an idler and drive wheel. Injection is activated by rotating the drive wheel at a set speed (controlled by a stepper motor6 ), while the idler wheel is free to rotate. Due to the grip of the two wheels, the rod was prevented from twisting at either injector. Stepper motors were preferred over DC motors for their precise speed control, as well as longevity. A careful calibration was performed to ensure that both injectors injected rod at identical velocities. Variables causing differential injection velocities include the distance between the idler and drive wheels as well as the back tension on the rod being injected. This tension was consistently different between the two injectors (the slack loop had less tension than the rod going over the feeder rollers into the slave injector). Each injector requires power and communication cables, as shown in Fig. 5-4(a). For the primary injector, the cabling was separated into individual strands and a stress relief loop was introduced to ensure the cable would not tug on the primary injector, introducing spurious force readings across the air bearing. The stress relief also acted to add compliance to the cabling, ensuring it would not dampen out or oppose true reaction forces. The entire injection system exhibited low losses for transmitting applied forces, measured to be less than 0.01 N. This value was measured using the setup shown in Fig. 5-5 (a), which consisted of mounting a force sensor along the direction of the injection reaction force in the actual experiment. Loads applied to this sensor and the reaction force measurement were compared to measure the losses across the injection system up to approximately 90% of the force sensor capacity, as shown in Fig. 5-5 (b). 6 Intelligent Motion Systems MDrive 14 Plus Integrated Motor, with 51200 micro-steps per revolution and approximately 0.6 micro-step per second resolution for velocity control [128]. 132 a) 5cm Reaction Load Sensor b) Applied Load Sensor 3 Reaction Force [N] 2 Measured Zero Loss 1 0 −1 −2 −3 −4 −2 0 2 Applied Force [N] 4 Figure 5-5: a) Side view of force-loss measurement. A load sensor was attached to the primary injector, so an applied load could be measured and compared to a reaction load. Note that the feeder rollers and constraining glass pipe have been removed from the experimental setup shown in Fig. 5-4. b) Comparison of applied and measured forces. Load losses were measured to be less than 0.01 N. 5.1.3 Data Acquisition and Control Sub-System A LabView7 virtual instrument (VI) was programmed to control the stepper motors8 and record readings from the force sensor. To allow the stepper motors to have precise control of the injection speed, commands could only be updated at a rate of approximately 2 Hz. While a test was being run, the readings from the force sensor were recorded9 to a text file. Time elapsed and force signals were continuously recorded at a rate of 1 kHz in parallel to running the stepper motors. Knowing the time elapsed and the injection velocity, one can readily calculate the length of rod injected into the constraining cylinder. Synchronization lag between the force recordings and the activation of the stepper motor was less than 0.05 seconds (the time LabView waited between sending a message and receiving an echo from the stepper motors). 7 National Instruments Laboratory Virtual Engineering Workbench, a proprietary software. LabView communicated through and Intelligent Motion Systems, Inc. USB to RS422 converter (model number MD-CC4). 9 The force sensor communicated with LabView through a National Instruments Data Acquisition card (model NI USB-6210), 8 analog inputs, 250 kilosamples per second. 8 133 For every experiment run, a digital video camera10 recorded events in the constraining pipe within approximately 30 cm of the injector. The video could be synchronized within approximately 0.02 seconds of the force recording. This extra data stream allows for a connection between the force recording and physical, observable processes. For example, one could then connect the injected length (and reaction force) to the point in the test when a helical configuration was first visually observed, as will be discussed in more depth in §5.2. 5.1.4 Experimental Protocol for Rod Injection The conditioning and cleaning protocols (and materials) for both the injected elastomeric rod and glass constraining pipes were developed to ensure the contact could be assumed to be dry friction and a consistent value of the dynamic coefficient of friction, µ. This repeatability for µ was important to be able to compare to simulations with zero fitting parameters (see §4.2 for a general description of the simulation tool and §5.2 below for specific adaptation for the real case). Toward this aim, much of the experimental protocol described in this section was aimed at producing a consistent coefficient of friction between rod and pipe. A single test consisted of a series of steps: i) Cleaning the pipe; ii) Conditioning the rod; iii) Preparing the injection sub-system; iv)Injecting the rod; and finally v) Withdrawing the rod. A single test included repeating steps iv) and v) ten times. We now describe each step in more detail: i) Cleaning the pipe as described in §5.1.1 was performed before each test. ii) Conditioning the rod as described in §5.1.1 was performed before each test (but not between individual runs). iii) After preparing the rod and pipe, the rod was inserted into the injection system. The front end of the rod was then passed over the top of the bottom feeder roller and wrapped once clockwise around the top feeder roller before going through the slave and primary injectors (creating the 10 Kodak PlaySport Zx5 digital video camera, recording at 59.94 frames per second. 134 slack loop). Less than 1 cm was inserted into the constraining pipe. The air supply to the air bearing would then be opened. The balance load was taken for the run by recording the force sensor for approximately five seconds so reaction forces could be compared to the rest state. iv) Before injecting, the video camera started recording. Then the rod was injected at a set velocity. Once the rod locked up (described in more detail in §5.2), injection and video recorder were stopped. v) Finally, a new video recording was be started and the rod was withdrawn at the same velocity as injection. Withdrawal and video recording were stopped when the rod front end got within 1 cm of the injector. After cycling through the injection and withdrawal steps 10 times, the pipe was removed from the injection sub-system, cleaning was repeated, and the constraint could then be changed to a different clearance if desired. Before freshly fabricated rods were used for data collection, 50 runs were performed following the experimental protocol (without video or force recordings) for tests to become repeatable. 5.2 Results and Interpretation In §5.1 above, we have described the experimental apparatus built to explore the behavior of a rod injected into a cylindrical constraint. We now present the results from our experimental investigation, discussing the video and reaction force recordings (§5.2.1), critical length scales (§5.2.2), and the effect of the injected rod’s natural curvature (§5.2.3). We will show that similar behaviors were observed across the different radial clearances tested, and that the critical length scale for the initiation of the helical configuration varies with radial clearance in a way that is in good agreement with the theoretical prediction from existing literature (which was discussed in §2.3.2). It will also be shown that imperfection in the form of natural curvature can have a strong effect on the lockup length scale. As stated in the experimental protocol (§5.1.4), a test consisted of 10 runs. Throughout this chapter, when reporting the 135 results for each test, the mean value of a test is reported and error bars correspond to one standard deviation above and below the mean. Simulation results shown are those that were performed by Dr. Tianxiang Su, with whom we collaborated with on this project. The simulations for the real case were run with the same model as described in §4.2 for the classic case, but instead of a fixed number of elements being simulated, elements were added as the rod was inserted at an imposed velocity. Self-contact of the rod was not simulated, but could be detected in post-processing. This direct interaction between and juxtaposition of simulation and experiment was an integral part of this project. While the author did not produce code nor operate the simulation tools, the interpretation of (dis)agreements between experiment and simulation results (performed collaboratively) drove the development of the code. These direct comparisons also aided in the improvement of the experimental apparatus. Lastly, simulation results were obtained with zero fitting parameters. All values were measured independently by the author. We now present behaviors observed while injecting a rod into cylinders within a range of inner diameters (listed in Fig. 5-3). 5.2.1 Reaction Force Signals and Video Analysis The progression of configurations (straight, sinusoidal, helical, and locked up) described at the beginning of this chapter was observed and quantified in the experiments. In Fig. 5-6 we show representative photographs of the rod near the injector during a typical test for pipe I.D.=12.0 mm and injection velocity, vinj =0.1 m/s. The corresponding reaction force recorded, Pinj , as a function of injected length, Linj , is also shown. In the early stage of the injection process in Fig. 5-6, frictional forces are not sufficiently large to cause buckling, and the rod lies straight along the bottom surface of the constraining pipe (Fig. 5-6, Configuration A, Linj = 0.22 m). During this regime, an approximately linear relationship between Linj and the Pinj is observed. 136 A Reaction Force, Pinj [N] 1.2 1 B C 0.8 D 0.6 E E 5cm Run Progress 0.4 0.2 0 0 C B A D 0.2 0.4 0.6 Injected Length, L [m] inj 0.8 1 Figure 5-6: Reaction force at the injector as a function of injected length for a typical experimental run and representative photographs of configurations throughout the test (side view). Rod injecting into a constraining pipe with I.D.=12.0 mm at vinj = 0.1 m/s. Configuration A shows the rod in a straight configuration. Upon further injection, the rod buckles into a sinusoidal shape (Configuration B). The sinusoidal amplitude grows with further injected length (and increasing reaction force) until the rod contacts the top of the constraining pipe (Configuration C - helical initiation) at Lhel inj . After this point, the reaction force increases rapidly with injected length until the pitch length decreases to the rod diameter and the rod is said to be locked up (Configuration D) at Llock inj . Further injection results in contacting helical pitches and the test is stopped (Configuration E). After further injection, however, a sinusoidal buckling mode is observed (Fig. 56, Configuration B, Linj = 0.73 m), where the rod climbs up alternate sides of the pipe. Yet, there is not an appreciable change in the relationship between Linj and Pinj between the straight and sinusoidal configurations. The amplitude of sinusoidally buckled configurations is observed to grow with continued injection. Once the rod climbs to approximately halfway up the constraining pipe, the rod buckles into a helical shape (Fig. 5-6, Configuration C, Linj = 0.79 m). We define the moment of 137 helical initiation as when the rod first contacts the top of the constraining pipe and refer to the length of rod injected at that point as Lhel inj (solid vertical line in Fig. 5-6). After helical initiation, Pinj increases rapidly (and non-linearly) with increasing Linj . Limited injection is possible after helical initiation, with the rod taking on a helical configuration with decreasing pitch length. Eventually, once the pitch size approaches the rod diameter, lockup occurs, after which no further injection is possible (Fig. 5-6, Configuration D, Linj = 0.98 m) and injection is stopped shortly thereafter (Fig. 5-6, Configuration E, Linj = 1.05 m). We represent Linj at lockup as Llock inj (dashed vertical line in Fig. 5-6). As part of the experimental protocol (described in §5.1.4), the injection subsystem was then set in reverse and the withdrawal of the rod from the constraint after lockup was also recorded. Fig. 5-7 shows the reaction force as a function of injected length as well as photographs of representative configurations for the withdrawal of a rod at vinj = −0.1 m/s from a constraining pipe with I.D. =12.0 mm. While the rod is withdrawn, the test begins with the maximum injected length and proceeds toward zero injected length (right to left in the graph). The reaction force is initially compressive because the rod is still locked up (Fig. 5-7, Configuration A, Linj = 1.05 m). It quickly transitions to a tensile load as the helical structure collapses with withdrawal (Fig. 5-7, Configuration B, Linj = 0.97 m) and friction opposes the motion of the rod. We denote the amount of rod in the pipe at this transition from helical to straight as Lstr inj (shown as a solid vertical line in Fig. 5-7). This tensile load linearly decreases in magnitude as the rod is progressively withdrawn until it reaches zero at the end of the test (Fig. 5-7, Configurations C and D, Linj =0.49 and 0 m, respectively). We now examine the injection and withdrawal processes in more detail, specifically the loss of contact between the rod and constraining cylinder, the varying nature of Pinj with Linj after Lhel inj , and an alternative form of lockup. First, we revisit the nature of contact between the rod and the constraining pipe, which previous analytical work (described in §2.3.2) assumes to be constant along the entire arc length of the rod. We observed, however, that this is not always the case during an experimental run. 138 0.4 Reaction Force, Pinj [N] 0.35 A 0.3 B 0.25 C 0.2 0.15 0.1 0.05 A D 5cm Run Progress C D B 0 −0.05 −0.1 0 0.2 0.4 0.6 Injected Length, Linj [m] 0.8 1 Figure 5-7: Recorded reaction force as a function of injected length for a typical withdrawal test and representative configurations throughout the test (side view). Rod withdrawing from a constraining pipe with I.D.=12.0 mm at vinj = −0.1 m/s. Configuration A shows the rod locked up, with a compressive reaction force (corresponding to Configuration D in Fig. 5-6). During the initial stages of withdrawal, the reaction force quickly decreases toward a tensile load with the unraveling of the helical structure (Configuration B). Afterwards, the rod takes on a straight configuration and reaction force increases toward zero with continued withdrawal (Configurations C and D). In some cases, we observed loss of contact between the rod and the glass pipe at the point of helical initiation. We also noted that there is loss of contact in the sinusoidal configurations with high amplitudes, prior to the onset of helical initiation. We also recorded a degree of variability in the extent of this contact loss from run to run. In some cases the rod would smoothly climb up the side of the pipe, and in other cases the rod would snap-through (suddenly jump) to the top of the pipe. Fig. 5-8 (a) and (b) show photographs of the video frame immediately before, at, and immediately after Lhel inj for a rod injected into a constraining pipe (I.D.=12.0 mm) at vinj = 0.1 m/s. Fig. 139 a) b) A1 B1 A2 B2 A3 B3 5cm Figure 5-8: Rod injected at vinj = 0.1 m/s into a constraining cylinder with I.D.=12.0 mm. a) Sequence of photographs of rod climbing up the side of the constraint (Configuration A1), smoothly reaching the top of the pipe at Lhel inj (Configuration A2), and immediately after Lhel (Configuration A3). b) Sequence of photographs immeinj hel diately before Linj (Configuration B1 - note the lower elevation of the sinusoidal peak compared to Configuration A1), snapping into Lhel inj (Configuration B2 - note the blurring of the rod indicative of significant vertical velocity), and immediately after Lhel inj (Configuration B3 - qualitatively similar to Configuration A3). 5-8 (a) shows a sequence of photographs of the rod climbing up the side of the pipe on the way to helix initiation without significant loss of contact (Linj = 0.779, 0.783, and 0.788 m), whereas Fig. 5-8 (b) shows a series of photographs of the rod snapping to the top of the pipe, with significant loss of contact (Linj = 0.696, 0.700, and 0.704 m). Examination of the rod immediately before Lhel inj (Fig. 5-8, Configurations A1 and B1) shows a difference in the elevation of the sinusoidal peaks. Helical initiation associated with significant contact loss (Fig. 5-8 (b), Configuration B2) shows significant vertical velocity, as evidenced by blurring of the rod in the photograph. Qualitatively, the resulting configurations immediately after Lhel inj (Fig. 5-8, Configurations A3 and B3) are similar. Once helical initiation occurs (regardless of whether a smooth transition or a snap through has occurred), additional helices form, exhibiting decreasing pitch length with increasing Linj . Typically, one could observe signatures of the formation of each pitch on the reaction force recording. This is shown in Fig. 5-9 (the same run shown in Fig. 5-6), where Pinj is plotted as a function of Linj . After helical initiation at Lhel inj = 0.79 m, the reaction force increases with injected length as a general trend, but it can be seen to have a locally periodic structure, with growing amplitude of variation and decreasing wavelength. These local maxima and minima correspond 140 A Reaction Force, Pinj [N] 1.2 1 0.8 0.6 B C D C 5cm A Run Progress 0.4 0.2 0 0 B D 0.2 0.4 0.6 Injected Length, L [m] inj 0.8 1 Figure 5-9: Examination of the oscillations in Pinj with Linj after Lhel inj from Fig. 5-6. Representative photographs (side view) of these local maxima (Configurations A and C) and minima (Configurations B and D) are shown. to the rod being at the bottom and the top of the pipe at the injector, respectively, as shown in the side view photographs of configurations in Fig. 5-9. When the rod is located at the bottom of the pipe it is directly aligned with the injector and the injector must support the helical shape directly (Fig. 5-9, Configurations A and C, Linj = 0.856 and 0.891 m, respectively). However, when the rod is at the top of the pipe, it must bend (proportional to the inner diameter of the constraining pipe) down to the injector. In this way, the bent point exerts a stronger normal force onto the constraining pipe, thereby leading to a lower value of the reaction force at the injector (Fig. 5-9, Configurations B and D, Linj = 0.873 and 0.912 m, respectively). The growing amplitude of variation is due to the rapidly increasing normal contact force and the decreasing period of these oscillations is due to the pitch length decreasing. After a number of helical pitches formed, the rod locked up inside the constraining 141 pipe. This process was not always identical between runs. In some runs, the rod formed helical pitches that contacted one another, as was shown in Configuration D of Fig. 5-6. In other runs, the helix collapsed as the rod formed a loop along axial direction of the constraining pipe, an example of which is shown in Fig. 5-10. In this case, the collapse of the helix is associated with a sudden drop in the reaction force during injection. Representative photographs are shown of lockup (Fig. 5-10, Configuration A, Linj = 0.95 m) and the collapse of the helical structure near the injector (Fig. 5-10, Configurations B and C, Linj = 0.97 and 1.03 m, respectively). Qualitatively, this helical collapse occurred more often in constraining pipes with larger I.D. and most commonly when lockup coincided with the rod being injected Reaction Force, Pinj [N] into the bottom part (lowest elevation) of a helix. 0.8 A 0.7 B 0.6 C 0.5 5cm B 0.4 Run Progress 0.3 C 0.2 0.1 0 0 A 0.2 0.4 0.6 Injected Length, Linj [m] 0.8 Figure 5-10: Rod injected into constraining pipe (I.D.=12.0 mm) at vinj = 0.1 m/s. Instead of locking up with a stable helix, the injected rod would sometimes form a loop along the cylindrical constraint’s axial direction, associated with a sudden reduction in the reaction force, as shown with the progression of representative photographs (side view) for Configurations A, B, and C immediately following Llock inj . 142 Thus far, we have focused on the experimental results found for ∆r = 4.42 mm (constraint I.D.=12.0 mm). Similar behavior was also observed for all other radial clearances explored. Fig. 5-11 shows the reaction force as a function of injected length for three different radial clearances (∆r = 1.72, 5.42, and 9.27 mm). The same sequence of configurations (straight, sinusoidal, helical, lockup) was observed for all clearances. However, for small values of ∆r, Lhel inj and lockup length increased. This delay in helical initiation will be discussed more in depth in §5.2.2. Finally, by combining the analysis of the video recordings with the reaction force signals, we were able to accurately measure the coefficient of dynamic friction, µ, between the injected elastomeric rod and borosilicate glass pipes (it is assumed that lateral friction is identical to axial friction). Two measurements were taken to ensure Figure 5-11: Reaction force recordings injecting into three different sized cylindrical constraints with radial clearances of 1.72, 5.42, and 9.27 mm (scaled schematics showing ratio of injected rod diameter to constraining pipe I.D.). Similar behavior was noted for all three clearances, with the rod being injected farther as radial clearance decreased. 143 injection and withdrawal were symmetric with respect to dynamic friction. The first measurement was performed during the injection stage of the experimental runs. First, the moment at which the rod became sinusoidal was noted in the video. This was converted into a corresponding injected length. The slope of the Pinj from this point back to zero injected length was then computed using a linear fit. This slope could be used to measure µ using the relation, Pinj /Linj = µw, (5.1) where Pinj /Linj is the ratio of reaction force (Pinj ) to injected length (Linj ) and w is the effective weight per unit length (w = ρAg, where ρ is the volumetric mass, A is the rod’s cross-sectional area, and g is acceleration due to gravity). Eq. (5.1) was applied to each of the runs to calculate µ at the smallest six radial clearances, with the results shown in Fig. 5-12 (a). The largest two radial clearances were not used due to difficulty in measuring the first buckling mode (resulting in large scatter over 10 runs). We also measured µ during the withdrawal stage for the cylindrical constraint with I.D.=21.7 mm (the second largest constraining pipe), again using Eq. (5.1), but for Linj ≤ 40 cm. This measurement was performed over 50 runs, with measurements shown in Fig. 5-12 (b). With these two methods, we measured µ = 0.54 ± 0.11 during both injection and withdrawal (identical results). This measurement confirms that the rod conditioning and pipe cleaning protocols do not introduce different values of µ in the two axial directions. This value was fed directly to simulations, avoiding the necessity for a fitting parameter for the dynamic coefficient of friction. 144 0.8 Coefficient of Friction, μ 0.7 a) 0.6 0.5 0.4 Experiment Mean Measurement Std Measurement 0.3 0.2 0.1 0 0 0.8 Coefficient of Friction, μ 0.7 1 2 3 4 5 6 Radial Clearance, Δ r [mm] 7 8 −3 x 10 b) 0.6 0.5 0.4 0.3 0.2 Experiment Mean Measurement Std Measurement 0.1 0 0 10 20 30 Run Number 40 50 Figure 5-12: (a) The coefficient of dynamic friction, µ, measured during injection into the smallest six radial clearances was µ = 0.54±0.11. (b) µ was also measured during withdrawal from ∆r = 9.27 mm. 145 5.2.2 Critical Lengthscales In the industrial setting of inserting coiled tubing into a wellbore (discussed in §2.3.1), lockup is a potentially catastrophic event due to the tight curvatures involved, making an understanding of the preceding processes desirable. Consider a locked up rod (configuration D of Fig. 5-6). In this case, the rod has taken a radius of curvature that is comparable to the radial clearance. In the field, this configuration should be avoided, as the following brief calculation can show. Typical coiled tubing has an outer diameter (O.D.) of 14 cm and I.D.=11.6 cm and the constraining cylinder (typically casing) has I.D.= 22.2 cm, giving ∆r = 4.1 cm [1]. The tensile and compressive strains on the mean radius, R = 1/2(O.D. − I.D.), on the outside of the bend of the coiled CT tubing, from simple beam bending theory [129], are = κrR where κ is the radius of curvature of the bent beam (in this simplified case, can be assumed to be on the CT order of κ ∼ 1/∆r) and rR is the distance from the neutral axis of the coiled tubing (center of the tubing for circular cross sections) to its mean radius. The strains at lockup can therefore be estimated to be of the order of ∼ 150%, which is well above the ultimate tensile strength (∼ 1%) of most high-grade steels and local buckling bending strains (∼ 5%) of thin-walled pipes [129, 130]; the pipe would rupture before this level of bending strain could be achieved. From a practical standpoint, the force required to inject the rod to the full locked up state would be prohibitively large, and equipment on the surface will be unable to continue injecting the coiled tubing all the way to lock up. Given that lockup as observed in our experiment corresponds to an impractical strain state in the field case, helical initiation was considered next as a critical length scale to experimentally track. Unlike lockup, a simple calculation can show initial helical configurations to be non-catastrophic in terms of developed strains. We recall Eq. (2.21) for the pitch of helices at the onset of buckling (phcr ) for a rod compressed √ p within a cylinder, phcr = 2 2 EIw/∆r, where EI is the bending stiffness of the coiled tubing (composed of Young’s modulus, E, and cross-sectional moment of inertia, I), w is the effective weight per unit length, and ∆r is radial clearance between the 146 coiled tubing and casing. Using typical coiled tubing values (E = 207 GP a, I = 9.97×10−6 m4 , w = 315 N/m [1]), combined with the typical dimensions given above, phcr ∼ 20 m in the field. Knowing the pitch (p = phcr /2π) and radius (∆r) of a circular helix, one can find the geometric curvature through the relation, κ= p2 ∆r . + ∆r2 This curvature would result in bending strains at helical initiation on the order of ≈ 0.02%. Some torsional strains would also be present to maintain the helix’s stability [45, 66], but we can see that helix initiation results in bending strains much more amenable to the field case. For this reason, we chose to track helical initiation throughout different tests. Next we explore the effect of injection speed and radial clearance on this critical length scale and then present data on reaction loads associated with Lhel inj . Having identified the length scale of interest, we now turn to the effect of injection speed (vinj ) on helical initiation. This test was essential to check the experimental apparatus was not introducing an effective friction through the buildup of an electrostatic charge on the constraining pipe [101]. Fig. 5-13 shows Lhel inj as a function of vinj for ∆r = 4.42 and 9.27 mm (I.D. = 12.0 and 21.7 mm). Injection speed was varied from vinj = 1 − 2 cm/s (I.D. = 21.7 and 12.0 mm clearances, respectively) up to vinj = 15 cm/s. Over this range of vinj , there was little variation in helical initiation length, which indicates a consistent coefficient of friction. We do not expect inertial effects over the tested values of vinj as it is much less than the speed of sound in the p material ( E/ρ ≈ 33 m/s). In Fig. 5-13, we also show the theoretical predictions for Lhel inj (dash-dot line) that was presented in §2.3.2. Specifically, Eq. (2.33), repeated here for convenience, Lhel inj √ r EI 2 2 , = µ w∆r where µ is the dynamic coefficient of friction in the axial direction, E is the rod material’s Young’s modulus, I is the second moment of inertia (I = πr4 /4 for solid circular 147 Helix Initiation, Lhel [m] inj 0.7 0.6 0.5 0.4 ID=12mm ID=21.7mm Theory 0.3 0.2 0.1 0 0 0.05 Injection Speed, v inj 0.1 [m/s] 0.15 Figure 5-13: Experimental values of helical initiation, Lhel inj , as function of injection speed, vinj , for I.D. = 12 and 21.7 mm (∆r = 4.42 and 9.27 mm, respectively). Lhel inj was found to not depend on vinj . Experiments were performed at two different clearances with injection velocities ranging from 1 to 15 cm/s. Dashed lines correspond to theoretical predictions (Eq. 2.33). cross-sections of radius r), w is the effective weight per unit length (w = ρAg, where ρ is the volumetric mass, A is the rod’s cross-sectional area, and g is acceleration due to gravity), and ∆r is the radial clearance between the injected rod and constraint. We see good agreement between theoretical prediction and experimental measurement, with experimental values typically greater than theoretical prediction for all injection speeds tested. Having confirmed that injection speed did not have an effect on Lhel inj , we now investigate the effect of radial clearance on helical initiation. Fig. 5-14 shows experimental, simulation, and theoretical results (from Eq. 2.33, using material properties from Table 5.1 and µ = 0.54 ± 0.11 measured and discussed in §5.2.1) for helical initiation as a function of radial clearance for a rod injected into constraining pipes at 10 148 cm/s. All three methods agree in identifying radial clearance as having a major effect on helical initiation, with helical initiation being delayed for tighter radial clearances. Fig. 5-14 shows Lhel inj as a function of ∆r for experiments, simulations, and theory. As a general trend, it can be seen that simulations typically predict higher values for Lhel inj than observed experimentally, which in turn overlaps with theoretical predictions from Eq. 2.33. This finding can be attributed to the presence of manufacturing imperfections, which are present in fabricated rods but not in simulations. Buckling processes are typically sensitive to imperfections in the rod [9], which result in lower critical loads than for straight elements. This translates to lower critical buckling lengths in this problem, as the reaction force at the injector goes as Pinj = µwLinj for Linj < Lhel inj . inj Helix Initiation, Lhel [m] 1.5 Experiment Simulation Theoretical Prediction, Eq. (2.17) 1 0.5 0 0 0.005 0.01 Radial Clearance, ∆ r [m] 0.015 Figure 5-14: Lhel inj as a function of ∆r for experiments (solid circles with error bars), simulations (hollow squares), and theoretical predictions (lines - Eq. (2.33). Theoretical predictions and simulations are based on the experimental measurement of µ = 0.54 ± 0.11. All three approaches give excellent agreement, with helical initiation being delayed (injected length increasing) for smaller radial clearances. 149 Note that the discrepancy between simulation and experiment seen in Fig. 5-14 is not constant; the difference between simulation and experiment grows with increasing radial clearance. As pointed out in §2.2.2, constrained buckling has a strong connection to buckling on an elastic foundation, with the foundation stiffness being equivalent to w/∆r. In these experiments, w is constant, so Fig. 5-14 shows a growing discrepancy between simulation and experiment with a weakening elastic foundation stiffness. Just as w is constant between tests, so too is the level of manufacturing imperfections. In other words, disagreement between experiment and simulation appears to grow as the effective foundation stiffness constant decreases, which agrees with theories of a beam with imperfections on an elastic foundation [9, 131]. Complementary to tracking Lhel inj as a function of ∆r, in Fig. 5-15 we show the hel experimental reaction force at the injector at Lhel inj , Pinj , as a function of ∆r. Values 0.1 0.07 0.06 Reaction Force at L , P hel inj 0.08 hel inj [N] 0.09 Experiment Theoretical Pred., Eq. (2.8) Theoretical Pred., Eq. (2.14) 0.05 0.04 0.03 0.02 0.01 0 0 0.005 0.01 Radial Clearance, ∆ r [m] 0.015 hel Figure 5-15: Reaction force at the injector at Lhel inj , Pinj , as a function of ∆r. Theoretical predictions are for the coefficient of lateral friction, µlat = 0 (solid line) and µlat = µ = 0.54 (dashed line) from Eq. (2.20) and (2.27), respectively. 150 were measured over injected lengths, |Lhel inj − Linj | < 2 cm, to eliminate the effect of sensor noise. Also plotted is the theoretical critical load for helix initiation from the classic case of a pinned rod inside a cylinder for two cases: lateral friction, µlat = 0 q √ q EIw h h h h (Eq. (2.20): Pcr = 2 2 ∆r ) and µlat = µ (Eq. (2.27): Pcr = 2ψcr EIw , where ψcr ∆r is a function of µlat ). Significant scatter is present in the experimental data, but good functional agreement does appear to exist between experiment and theoretical predictions for µlat = 0, hel with the measured Pinj decreasing with increasing ∆r. Noteworthy is the significant hel over-prediction of Pinj by Eq. (2.27) for µlat = µ, which was successful in captur- ing the load at helix initiation, Pcrh , for the classic case. The boundary conditions (pinned-free) are not directly applicable to the case of injecting a rod into cylindrical constraint (where the loading end is clamped), although this adjustment is predicted hel to increase the prediction for Pinj . The agreement between the theoretical prediction assuming µlat = 0 and experiments indicate µlat does not appear to play as large of a role in helical buckling in the real case as in the classic case. Possible causes of this effect include significant loss of contact between the rod and pipe near the injector at helix initiation (discussed in §5.2.1), lateral motion of the rod achieved by rolling instead of sliding motion, or the rod conditioning process causing µlat << µ. This section discussed the rationale of choosing Lhel inj as a practically relevant critical length scale, presented data on the effect injection speed and radial clearance had on the length scale, and the reaction loads associated with Lhel inj . Manufacturing imperfections present in experiments but not present in simulations were suggested as a cause for some discrepancy between the two methods. The next section explores the effect of natural curvature (a systematic “imperfection”) of the injected rod on lockup length. 5.2.3 Effect of Imperfections: Natural Curvature The previous chapter discussed the buckling of a rod compressed inside of a cylindrical constraint (the classic case). The analytical framework for the classic case has 151 historically been ported to apply to the real case (e.g. - the theoretical prediction for Lhel inj for the real case is based on theory developed for the classic case), suggesting similar trends between the two cases. In §4.2.3, it was experimentally shown that sag in the constraint reduced the critical helical buckling load as the imperfection grew to the size of ∆r. This section presents an exploration of the effect of natural curvature, κ0 , which is considered to be an imperfection, on the lockup length, LL , of the injected rod. Previous authors have explored the effect of κ0 on the lock-up process of coiled tubing [108, 132], finding that κ0 reduces LL . To our knowledge, no experimental program has investigated this effect. All rods for the experiments in this chapter were fabricated by the author to enable precise control of κ0 , as described in Appendix A. Up until this point in the chapter, rods used thus far have been straight (κ0 = 0 m−1 ). We now relax this constraint, fabricating rods over a range of 0 < κ0 [m−1 ] < 62. Fig. 5-16 shows experimental and simulated results for LL as a function of κ0 for a rod injected at vinj = 0.1 m/s into a constraining pipe with ID = 18.5 mm (∆r = 7.67 mm). Both experiments and simulations show relatively unaffected behavior for small ( κ0 < 10 m−1 ) natural curvatures. Above this value, however, natural curvature is seen to significantly reduce lockup length. The dominant imperfection (κ0 ) is modeled, so we see experiments and simulations show agreement without a clear offset. This is in contrast to the previous section, which saw simulation consistently give higher values of Lhel inj than experiments. Part of this improved agreement between the two methods is possibly explained by tracking LL instead of Lhel inj . In the classic case, we found that an imperfection in the geometry of the constraint decreased Pcrh dramatically after a critical value. Similarly, we have found that an imperfection in the injected rod (κ0 ) has a negligible effect on LL below a critical value and reduces LL significantly above the critical value. 152 1 0.9 L Lockup Length, L [m] 0.8 Experiment Simulation 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0 10 20 30 40 50−1 Natural Curvature, κ0 [m ] 60 70 Figure 5-16: Lockup length (LL ) as a function of natural curvature (κ0 ) for experiments and simulation for a constraining pipe ID = 18.5 mm. In both experiments and simulations, κ0 is seen to have a negligible effect when small ( κ < 10 m−1 ), after which it has a strong negative effect on LL . 5.3 Outlook This chapter presented an experimental apparatus to explore the buckling and postbuckling behavior of a rod injected into a cylindrical constraint. Unlike the hanging custom fabricated rods of Chapter 3, a protocol was needed to specifically address the surface condition of rods used in this chapter due to the fundamental importance of the frictional interaction between rod and constraint. The dynamic coefficient of friction, µ, in the axial direction was measured with the experimental apparatus, and found to be repeatable in the injection and withdrawal directions. The expected progression of structural configurations (straight, sinusoidal, helical, lockup) was observed, and their connection to the reaction force signals was reported. Contact loss 153 between the rod and pipe, particularly at Lhel inj , is an open issue. Specifically, it remains to be addressed what factors determine if the transition to a helical configuration will be relatively smooth or will be a more violent snap-through process. Using Lhel inj as a critical length scale of industrial relevance, the effect of injection velocity (negligible) and radial clearance was explored, finding good agreement between experiment, simulation, and theory. Interestingly, the theoretical framework that captures loads for helix formation in the classic case (such that µlat = µ) is different from that which hel captures Lhel inj and Pinj for the real case (µlat = 0). Finally, experimental and simulated results were presented for the effect of natural curvature on lockup length. Similar to the classic case, the real case appears to be highly sensitive to imperfections above a critical value. This chapter has also reached the state-of-the-art of the real case (with the exception of non-horizontal constraints). We have characterized what will happen to a rod injected into a cylindrical constraint and we can confidently predict how much rod can be injected. In the following chapter, we will explore one avenue of extending reach, or increasing Lhel inj . 154 Chapter 6 Actively Extending Reach Having investigated the buckling-induced lockup process inside a cylinder in the previous chapter, we now focus on active strategies to extend reach. This is of particular relevance for coiled tubing service operations in horizontal wellbores. As discussed in §2.3.1, extended reach boreholes (for which our constraining cylinders are analogues) can be drilled several times longer than coiled tubing can reach using current technology. In the previous chapter, we identified several passive mechanisms to delay lockup, including: increasing the bending modulus of the injected rod, decreasing the coefficient of friction between the rod and pipe, decreasing the buoyant weight of the rod inside of the pipe, and decreasing the radial clearance of the system. Ensuring that the injected rod does not have excess natural curvature will also increase the reach of a rod. These methods, however, focus on material properties, system geometries, or surface treatments which offer limited gains on reach. In this chapter, instead of passive measures for extending reach, we turn to an active mechanism: adding energy to the system. The mechanism explored involves vibrating the pipe vertically along its entire length, which we refer to as the dynamic real case. Modifications to the apparatus from Chapter 5 are described in §6.1. A discussion of the experiments run is then presented in §6.2, which also provides a rationale for the underlying mechanism, borrowing from structural vibration. Finally, future directions for the work are discussed in §6.3. 155 6.1 The Experiment The experimental apparatus used to explore the effect of vibrating the constraining pipe on helical initiation and lockup included the same subsystems as described in the previous chapter: the injection subsystem, a data acquisition/control subsystem, and the cylindrical constraint (see §5.1 for a detailed description of each subsystem). In addition, we modified the apparatus so as to vibrate the constraint in a controlled manner. In Fig. 6-1 (a) we present a photograph of the new setup and Fig. 6-1 (b) shows a schematic diagram of the system. The new added components, which are described in more detail in §6.1.1, consist of a series of electromagnetic shakers that provide distributed vibration to the whole constraining pipe, as well as a series of a accelerometers mounted onto the pipe for monitor the applied vibration. The experimental protocol employed in this chapa) b) 20cm Driving and Vibration Measurement System Injected Rod Slave Injector Primary Injector Force Sensor Accelerometer Linear Air Bearing 10 DAQ 1.0 0.8 0.6 0.4 0.2 0.0 -0.2 -0.4 -0.6 -0.8 -1.0 8 6 4 2 0 Function Generator Computer w/LabView DAQ Shaker Signal Conditioner Figure 6-1: a) Side view of experiment. b) Schematic of experiment, with added components making the driving and vibration measurement system, including: electromagnetic shakers beneath the pipe, a function generator to drive the shakers, accelerometers mounted above the pipe, and a signal conditioner to amplify accelerometer measurements. 156 ter remains very similar to that described in §5.1.4, with a few important changes that are detailed in §6.1.2. The rods used are the made with the same elastomeric Vinylpolysiloxane polymer of the experiments described in the previous chapter. However, here we focus on rods that are manufactured to be perfectly straight, κ0 = 0 m−1 . The vibration tests reported in this chapter use two sizes of constraining pipes: with diameters I.D. = 12 mm and 21.7 mm. 6.1.1 Driving and Vibration Measurement System In order to explore the effect of vibration on the initiation of helical buckling and lockup of a rod injected into a cylinder, a series actuators were added to the system, as shown in Fig. 6-1. Four electromagnetic shakers were mounted along the length of the constraining pipes equally spaced and in the middle of the spans supported by the aluminum frame. A detailed photograph of a single shaker is shown in Fig. 6-2 (a). Each shaker1 is attached to the pipe using a Nitinol2 stinger and a shaker pipe clamp. The resulting acceleration is monitored using a single-axis accelerometer3 mounted directly above each shaker. The shakers were attached directly onto the aluminum supporting frame. The Nitinol stinger provided a compliant connection between the shaker and pipe clamps, which could bend (bending stiffness, EI = 0.02 N m2 ) to accommodate any misalignment while still efficiently transmitting axial vibrations. More rigid connections were found to over-constrain the system (requiring impractical alignment tolerances). Shaker pipe clamps were laser cut from acrylic and provided a rigid attachment point for the nitinol stingers to the pipes as well as a mounting surface for the accelerometers. Each shaker was treated as an independent actuator for a human-in-the-loop 1 The Modal Shop SmartShaker with integrated power amplifier (model K2007E01) magnet shaker with 21N peak sine force, 13 mm continuous stroke, and frequencies up to 9 kHz [133]. 2 Nitinol wire from Chapter 4 was used, with E = 68.05 ± 0.15 GP a and d = 1.6 mm. Full properties given in Table 4.1. 3 Two different PCB high sensitivity, quartz shear accelerometers were used. Model 353B34 could measure ±50 g peak accelerations over the frequency range 1-4000 Hz with ±5% accuracy with 100 mV/g sensitivity, with a resonant frequency greater than 22 kHz [134]. Model 353B32 could measure ± 100 g peak accelerations over the frequency range 1-5000 Hz with ±5% accuracy with 50 mV/g sensitivity, with a resonant frequency of greater than 28 kHz [135]. 157 Figure 6-2: a) Detailed photograph of an electromagnetic shaker, a nitinol stinger, a pipe clamp, and an accelerometer. b) Polymer sleeves surround the pipe at the pipe supports to dampen the transmission of vibration to the aluminum frame, while allowing the pipe at the clamp to still vibrate, even if with lower amplitude at that point than at the shaker. feedback system to achieve the desired vibration characteristics for each experimental test. Each of the four shakers were controlled independently of one another, and connected to four function generators4 , as shown in Fig. 6-1 (b). These signal generators were used manually to set the vibration frequency and amplitude. While the experimenter could control the amplitude of each shaker, the phase between shakers was arbitrary. Each accelerometer monitoring the resulting vibration was powered by and interfaced with the data acquisition (DAQ) card (see §5.1.3 for further details this) through a signal conditioner5 . The signal from all four accelerometers was sampled 4 B&K Precision model number 4013B direct digital synthesis (DDS) function generator, capable of generating sine waveforms up to 12 MHz with variable output voltage from 10 mV to 10 V (peak to peak) into a 50 Ohm output impedance with ± 2% ± 20 mV amplitude accuracy [136]. 5 PCB model number 482C05 4-channel signal conditioner with unity gain and variable, constant current output [137]. 158 at 1 kHz by a LabView virtual instrument and, once per second, the frequency, f , and peak acceleration, Γ, measured were recorded to a text file. The measured value from the accelerometers was also displayed on the screen in real time, providing feedback to the experimenter to adjust the function generator outputs. As a matter of convention, peak accelerations were recorded as a ratio to the acceleration due to gravity at sea level, g. Assuming sinusoidal vibration, peak acceleration, Γ, and frequency (in Hz), f , can be combined to calculate the amplitude of vibration, A (in units of distance), through the relation Γ = (2πf )2 A/g. The pipe supports that connected the pipe to the aluminum frame were laser cut from acrylic, and each support included elastomeric sleeve6 between the pipe and the supports as shown in Fig. 6-2 (b). The sleeves were compliant (E = 213 ± 1.1 kP a) and thick (2.5 mm) in comparison to shaker stroke, allowing for some pipe motion at the pipe supports. This accomplished two goals. Firstly, it damped vibrations from transmitting to the aluminum frame, and thus to the rest of the system. Secondly, by allowing some vibration of the pipe at the supports, it kept the acceleration profile of the pipe more even. Fig. 6-3 shows the measured peak acceleration profile for a pipe with all four shakers vibrating at Γ = 1, with f = 100 Hz. The profile was measured using an additional accelerometer that was mounted on the pipe and repositioned to different points along the pipe’s length. Fig. 6-3 shows that there is variability in the peak acceleration along the pipe length, but that the acceleration is highest at the shakers and that the acceleration does not drop to zero near the pipe supports. There, near the clamps, the peak acceleration decreases to approximately half of that measured at the shakers. 6.1.2 Experimental Protocol As in §5.1.4 for the previous chapter, a detailed experimental protocol was developed for the experimental tests in this chapter to ensure repeatability between experiments. Past an initial “warm up” period for the electromagnetic shakers and accelerometers, 6 Zhermack Elite Double8 Vinylpolysiloxane polymer, similar to that used for making the elastomeric rods used in this chapter and described in depth in Appendix A. 159 Measured Peak Acceleration [g] 1.2 1 0.8 0.6 0.4 Profile Shakers Clamps 0.2 0 0 50 100 150 Distance from Injector End [cm] 200 Figure 6-3: Acceleration profile along a constraining pipe (ID=21.7mm) when vibrating the shakers at f = 100Hz and Γ = 1. Note the drop in peak acceleration away from the shakers to non-zero values at the pipe clamps. the signal from the accelerometers stabilized. After adjusting the shakers to the desired vibration parameters, the steps followed were broadly the same as the previous, non-vibrating tests: the pipe was cleaned, the rod was treated and loaded into the injection system, and then injection and withdrawal runs were performed. Before running each experimental test, however, the system needed to be prepared. Cleaning the pipe included a single pass of an ethanol-soaked paper towel through the inside surface before pulling a cotton cloth through twice from the front to the back end. The addition of this procedure to the experimental protocol in Chapter 5 of the ethanol pass was necessary due to an increase in the amount of chalk particles present after a test, caused by the agitation brought about from the pipe shaking. The rod was then prepared in the same manner as described in §5.1.4. The surface treatment consisted of applying a chalk coating followed by three wiping passes with a paper towel to remove excess particles. The rod was then loaded into the injection system, passing over both feeder rollers and through the slave and primary injectors, with a slack loop between the two. After loading the rod, the air supply to the air bearing was opened. Before recording the balance force, the shakers were tuned to the appropriate vibration frequency and peak acceleration and monitored for at least 30 seconds, with measurements written to a text file. At this point, the rod was injected into the constraining pipe at a specified velocity 160 until lockup and then withdrawn at the same rate. The dynamics of the rod were recorded by a video camera mounted above the setup near the injector. This injection and withdrawal made up a single run of a test. Unlike the tests without vibration (in which ten runs made up a test), five runs made up a test for theses tests with vibration. This reduction in the number of runs was due to excess chalk falling off of the rod from the agitation of the vibrating rod, reducing the number of repeatable runs. Peak accelerations of the shakers were also monitored between runs of a test and adjusted as needed. 6.2 Results and Interpretation We now report the results of the dynamic real case experiments in this section. In §6.2.1, we report reaction force characteristics, similar to the real case without vibration. We then turn to describing the effect of various control parameters. In §6.2.2, we first explore the effect of peak acceleration, Γ, on injected length to helix initiation, Lhel inj , and establish an analogy with the simpler system of a ball bouncing on a vertically vibrating plate. In this comparison, high-speed video analysis reveals the existence of bending waves in the rod during vertical vibration, which are then investigated further in §6.2.3. Finally, we discuss the effect on Lhel inj of the two remaining control parameters: vibration frequency (§6.2.4) and injection speed (§6.2.5). 6.2.1 Reaction Force Signals The sequence of buckling instabilities followed the same progression as tests without vibration. Initially, for small injected lengths, the rod lies along the bottom of the pipe in a straight configuration. As the rod is injected further, it buckles into a sinusoidal configuration near the injector. Eventually, the rod undergoes a second instability and assumes a helical configuration. As in Chapter 5, we refer to the amount of rod injected at helix initiation as Lhel inj , and will use it as a critical length scale instead of the actual lockup length, LL (discussed in depth in §5.2.2). The buckling sequence was 161 consistent for all the tests reported in this section. Fig. 6-4 shows the reaction force measured at the injection point, Pinj , as a function of Linj for three different values of the vibration parameters and both pipe sizes tested (I.D. = 12.0 and 21.7 mm). We see that while critical values for the different tests (e.g. Lhel inj = 1.15m, 0.77, and 0.66 m for f = 50, 100, and 400 Hz, respectively), the functional form between the tests is very similar. Reaction force, Pinj , increases linearly with increasing injected length, Linj , until Lhel inj , at which point the reaction force increases dramatically. The fine, periodic fluctuations in Pinj after helix initiation that were noted in the previous chapter are also observed. A distinguishing feature of these tests with vibration is the observation that small Figure 6-4: Reaction force, Pinj , measured at the primary injector for tests varying radial clearance (∆r = 4.4 and 9.3 mm), injection speed (vinj = 3 and 10 cm/s), peak acceleration of vibration (1.5 ≤ Γ ≤ 2), and frequency of vibration (50 ≤ f [Hz] ≤ 400Hz). Similar function form of Pinj variation with increasing Linj is noted for all three curves, which are also comparable to curves presented in the previous chapter. 162 (millimeter to centimeter scale) sections of the injected rod intermittently lose contact with the vibrating constraint. This is accompanied by a low level acoustic emission, and will be discussed more in §6.2.2. Also notable are the increasing levels of the noise-to-signal ratio for higher vibration amplitude tests, where we define noise as the standard deviation of the force signal readings over the balance test performed prior to injecting the rod. The noise in these tests was mainly caused by some degree of the vibration from the shakers transmitting along the system to the force sensor. Fig. 6-5 shows the noise measured during balance tests as a function of peak acceleration, Γ, for three different frequencies of vibration (f = 50, 100, and 200 Hz). Noise increased significantly after Γ = 1 in all three cases. Average Noise [N] 0.1 50 Hz 100 Hz 200 Hz 0.08 0.06 0.04 0.02 0 0 0.5 1 1.5 2 2.5 Peak Acceleration [g] 3 3.5 4 Figure 6-5: Noise measured during balance tests (before injection) as a function of peak acceleration for three different frequencies. Tests shown for a glass pipe with ID=21.7 mm. 163 6.2.2 Effect of Vibration Amplitude: Contact Loss In this section, we investigate the effect of increasing peak acceleration, Γ, on the injected length of rod at helix initiation, Lhel inj . Experimental results were obtained under the following conditions: the injection speed was kept constant at vinj = 0.1 m/s inside of a constraining pipe with I.D. = 21.7 mm and the dimensionless peak acceleration was varied in the range 0 ≤ Γ ≤ 4, for three different frequencies, f = 50, 100, and 200 Hz. In the case of f = 50Hz, the dimensionless peak acceleration was limited to the range 0 ≤ Γ ≤ 2. The results of these tests are shown in Fig. 6-6, where we plot 2 1.8 Normalized Helix Initiation 1.6 1.4 1.2 1 0.8 0.6 0.4 f=50 Hz f=100 Hz f=200 Hz 0.2 0 0 0.5 1 1.5 2 2.5 3 Dimensionless Acceleration, Γ [g] 3.5 4 Figure 6-6: Experimentally measured helical initiation as a function of peak acceleration for three different values of frequency of vibration. Helix initiation is normalized by the value for Γ = 0. For accelerations above Γ = 1, improvements in helix initiation are observed. In all cases, injection speed is vinj = 10 cm/s inside of 21.7mm ID pipe. 164 Lhel inj normalized by the value obtained without vertical vibration (hereafter referred to as the static value) as a function of peak acceleration, Γ. We find that below Γ = 1, there is no increase in Lhel inj . Past this point, however, we note improvements (increasing values of normalized Lhel inj , equivalent to delaying helical buckling) with increasing peak acceleration. We also find a difference in the improvements between high (100 Hz and 200 Hz) and low (50 Hz) frequency vibrations, with greater improvements in Lhel inj observed for the low frequency tested. We rationalize this behavior through an analogous system: a ball bouncing on a vibrating plate, as is discussed next. The increase in Lhel inj due to vibration can be attributed to its role in destabilizing the frictional contact points between the rod and the pipe, thus reducing the resistance over the length of the injected rod. Indeed, when peak acceleration is sufficiently large, sections of the injected rod intermittently lose contact with the constraining pipe. To help rationalize this observation, we first consider the simple analogous system of a ball bouncing on a vertically vibrating plate, as illustrated in the schematic shown in Fig. 6-7. The ball can be seen as a cross-section of our rod and the plate as our vertically vibrating pipe. This analogy assumes that neighboring rod cross-sections of the rod behave independently from each other and that the bottom of the pipe can be modeled as flat. This model scenario of a bouncing ball on a vibrating plate has become a canonical system to study nonlinear dynamics and the route to chaos in seemingly unrelated fields such as cosmic radiation, granular media, and droplets bouncing on soap films [138–140]. Considering the ball on a vibrating plate shown schematically in Fig. 6-7, we wish g m x(t) Figure 6-7: Ball on the vibrating plate; a simplified analogous system to a rod inside a vibrating cylinder. The ball has mass m, g is the acceleration due to gravity, and the plate vibrates according to a sinusoidal function x(t) = A sin(2πf t). 165 to derive the amount of contact loss between the ball and the plate (the flight time of the ball) as a function of the plate’s vibration peak acceleration, Γ. Here we are most interested in the maximum flight time for a completely inelastic ball. A more detailed account of the bouncing ball can be found in the literature, including other allowable flight times of the ball (e.g. resulting from period-doubling) and the effect of elastic collisions between the ball and the plate [141–144]. We assume that the plate vibrates with a sinusoidally varying motion x(t) = A sin (2πf t) as a function of time, t, with frequency, f , and amplitude, A. The normal contact force, N , between the vibrating plate and ball of mass, m, is, N = m g − (2πf )2 A sin (2πf t) . (6.1) where g is the acceleration due to gravity. The ball loses contact (flight time starts) with the vibrating plate when N = 0. The ball will go into flight with the velocity of the plate at that point and follow a parabolic trajectory. The maximum possible dimensionless velocity, Vmax , is, Vmax = √ Γ2 − 1, (6.2) where Vmax = vmax (2πf )/g, vmax is the dimensional velocity, and Γ = (2πf )2 A/g is the dimensionless peak acceleration of the plate. Assuming a parabolic flight path returning to the same elevation, the dimensionless flight time, T , of the ball is, √ T = 2 Γ2 − 1, (6.3) where T = 2πf ∆t and ∆t is the dimensional flight time. Gilet et al. [144] have improved on Eq. (6.3) by taking into account the changing elevation of the vibrating plate (which displaces a distance A from its neutral point during vibration), Γ= s 1+ 1 2 T 2 − 1 + cos(T ) T − sin(T ) 2 . (6.4) In Fig. 6-8, the solutions of both Eq. (6.3) and its improvement, Eq. (6.4), are 166 Dimensionless Flight Time, T Approximate Max. Flight Time, Eq. (6.3) Max. Flight Time (Gilet, 2009), Eq. (6.4) 0 0 1 2 4 6 Dimensionless Acceleration, Γ 8 Figure 6-8: Maximum flight time T for a ball bouncing on a vertically vibrating plate with peak acceleration Γ. The simplified prediction from Eq. (6.3) generally agrees with the exact Eq. (6.4). compared with each other, and good overall agreement is found between the two. The ball is predicted to have non-zero flight time past Γ > 1, or peak accelerations over 1 g (the solutions become non-real below this threshold). For the approximate solution of Eq. (6.3), the flight time is found to increase monotonically with acceleration. This prediction of a monotonically increasing flight time with Γ is in qualitative agreement with observations in our experimental system wherein Lhel inj is improved monotonically with Γ as well, as was shown in Fig. 6-6. We now turn to a quantitative comparison between the experimental setup and the analogous system. To further compare the experimental and model systems, high-speed videos7 were captured for a rod inserted into the entire 243 cm length of the constraining pipe with I.D.=12.0 mm (the videos were taken with zero injection velocity). Ten videos were captured, one for each combination of frequency, 22 ≤ f [Hz] ≤ 100, and peak acceleration, 2.5 ≤ Γ ≤ 8. The combinations of (f, Γ) were chosen such that the rod 7 Nikon 1 J-3 digital camera recording at 402 frames per second with a resolution of 640x240 pixels 167 was clearly losing contact with the cylinder, but not making contact with the top surface of the cylinder. Two representative sequences of video frames of a rod inside a vibrating cylinder are shown in Fig. 6-9 (a) and (b), for f = 25 Hz and Γ = 3 and 4, respectively. Qualitatively, one can see that for the same driving frequency, increasing Γ results in greater contact loss between the rod and the pipe. To quantitatively explore the similarities between a rod in a vertically vibrating pipe and a ball bouncing on a vertically vibrating plate, image analysis was used to compare the flight time of the rod to the analytic prediction of Eq. (6.3) for a bouncing ball. In order to make a direct comparison, a single characteristic flight time for the entire rod was calculated, using the techniques we will now outline. For each combination of f and Γ, video was recorded at 402 frames per second. For each frame, the centerline of the rod was located through a grayscale thresholding using MATLAB. In this way, the vertical location of the rod, y, as a function of both time and arc length could be reconstructed. The refraction of the glass constraining pipe was not corrected for, as exact elevation measurement was not required for this calculation. Once the video used for Fig. 6-9 (b) was analyzed, the time history of the vertical position of the rod, y(t), was tracked to determine the average flight time of rod segments losing contact with the cylinder. Four representative time histories, equally spaced along the rod, are shown in Fig. 6-10 (a), plotting y(t) throughout the video taken. The video analysis tracked y(t) for 461 distinct rod locations in the video analysis, with the number of locations determined by the pixel size (0.5 mm per pixel) and the width of video analyzed. To determine the typical flight time for each location, the frequency characteristics were calculated using the following procedure. For each location, the fast Fourier Transform (FFT) power spectrum of the time history, y(t), was calculated using MATLAB’s FFT function. We represent the FFT of y(t) as Y (f ), applicable over the frequency range 0 ≤ f [Hz] ≤ 201, which is set by the video frame rate. The FFT power spectrum is calculated as |Y (f )|2 /Nf f t , where Nf f t is the number of frames analyzed. The FFT power spectrum (sometimes referred to as a periodogram) for the 168 a) 3cm b) t=0.000s t=0.000s t=0.0050s t=0.0050s t=0.0100s t=0.0100s t=0.0149s t=0.0149s t=0.0199s t=0.0199s t=0.0249s t=0.0249s t=0.0299s t=0.0299s Figure 6-9: Sequence of experimental video frames of a rod inside a vibrating glass pipe with I.D.=12.0 mm. Photos were taken from a video recording at 402 frames per second, with the pipe vibrating at 25 Hz in both time series, with (a)Γ = 3 and (b)Γ = 4. Note larger amounts of contact loss in (b) than in (a) as well as the fact that the rod is not moving as a rigid body, but, instead, has an apparent wavelength (discussed more in §6.2.3). 169 a)4 b) b) 2 80 0.5 1 1.5 2 2.5 3 2 0 0 4 0.5 1 1.5 2 2.5 3 2 0 0 4 0.5 1 1.5 2 2.5 70 FFT Power Magnitude Elevation, y [mm] 0 0 4 3 60 50 40 30 20 10 2 0 0 90 0.5 1 1.5 Time, t [s] 2 2.5 0 0 3 10 20 30 Frequency, f [Hz] 40 50 Figure 6-10: a) Rod elevation as a function of time for four different horizontal positions of the rod shown in Fig. 6-9 (b) (f = 25 Hz and Γ = 4). Each position is separated by 2.7 cm. (b) Power density as a function of frequency, averaged from each horizontal position. Note that the driving frequency of f = 25 Hz is different from the main frequency (highlighted with red) of 12.4 Hz. entire rod was calculated by averaging the FFT power spectra of each location along the rod. The FFT power spectrum measured for the video in Fig. 6-9 (b) (f = 25 Hz and Γ = 4) is shown in Fig. 6-10 (b). Two peaks are clear in the FFT, one at the driving frequency of the constraining cylinder, f = 25 Hz, and the other, peak response frequency, at a subharmonic frequency, fpk = 12.4 Hz (indicated with the red circle in Fig. 6-10 (b) as the peak response). The peak response frequency characterizes the contact loss between the rod and vibrating constraint, with a characteristic flight time, ∆t = (2fpk )−1 . In this way, each of the 10 tests were analyzed to find the peak response frequency and characteristic flight time. For comparison with Eq. (6.3) for flight time of a bouncing ball, we computed the dimensionless characteristic flight time as, T = 2πf ∆t = 2πf , 2fpk (6.5) where f is the driving frequency and fpk is the peak response frequency from the measured FFT power spectrum. In Fig. 6-11, the experimentally measured dimensionless flight time (Eq. (6.5)) is compared to the bouncing ball prediction of Eq. 170 (6.3). The majority of experimental points (with the exception of experiments with f = 35 Hz and Γ = 4 and f = 100 Hz and Γ = 8 - which were both characterized by a lack of a dominant peak response frequency) agree relatively well with the theoretical prediction of Eq. (6.3). There is, however, disagreement at higher values of Γ, with flight time less than predicted from the inelastic ball model. Some discrepancies were expected, however, as the model does not account for interaction between adjacent rod sections. This interaction is thought to occur through bending, which would provide both a restoring force for a section of rod that is in flight and allows for the possible coupling from one section to another through bending waves. We have shown that the description of analogous system consisting of a ball bouncing on a vibrating plate predicts the characteristic flight time of a rod inside a ver- 25 Flight Time, T 20 15 10 Experiment Eq. (6.3) 5 0 0 2 4 6 Peak Acceleration, Γ 8 10 Figure 6-11: Dimensionless flight time, T, versus peak acceleration, Γ, for 10 tests with a rod inside a vibrating pipe with I.D.=12.0mm (3 results are indistinguishable for the low flight time value at Γ = 4). Eq. (6.3) is plotted as a solid line. 171 tically vibrating cylindrical constraint. The measurement method consisted of considering the temporal behavior of each rod section independently, and averaged over the entire length of the rod. The next subsection explores the spatial configuration of the entire rod at each time step. 6.2.3 Bending Waves Inside a Cylindrical Constraint Above, we provided a prediction for contact loss between the vibrated cylinder and injected rod modeling the rod as a series of non-interacting cross-sections, reducing the problem to a planar characterization. However, as was seen in the experimental frames from high-speed video in Fig. 6-9, the rod, when in flight, appears to take on a shape with a well-defined wavelength. This section investigates the presence of bending waves in the case of the rod inside the vibrating channel. Two series of experiments were performed. The first, described in the previous section where we characterized flight times, consisted of exciting the glass pipes in the same manner as reach extension experiments (see §6.1.2 for an in depth discussion), with four discrete points vibrated to create a distributed vibration. A rod was placed in the entire length of the pipe, with one end free and the other clamped at the injector. This test is referred to as distributed shaking, and a schematic diagram of the test is given in Fig. 6-12 (a). The second method involved placing a rod into the entire length of the glass pipes, with one end free, and the other (outside of the pipe) clamped to an electromagnetic shaker. This single point was then vibrated vertically. This configuration is referred to as end shaking, and is shown as a schematic diagram in Fig. 6-12 (b). For the case of end shaking, vibration was performed in the range 22 ≤ f [Hz] ≤ 100, for 9.7 < Γ < 45, with 13 individual test. Peak acceleration, Γ, was calculated from observed displacement of the electromagnetic shaker measured directly from the video frames in image processing. The combination of (f, Γ) was selected to excite definite contact loss between the rod and constraining cylinder while preventing contact between the rod and the top of the constraint. In both cases, a section of rod was recorded with 172 view area a) support shaker b) Figure 6-12: a) Distributed shaking experiment identical to the reach extension experiment, but with the rod inserted the entire length of glass pipe, with approximate camera view shown with a dashed box. b) End shaking experiment with the rod inserted the entire length of glass pipe, but clamped to an electromagnetic shaker at one end. Approximate camera view again shown with a dashed box. a high speed camera at 402 frames per second for three seconds. Unlike the sequence of frames in Fig. 6-9, in which the excited motion did not have a clear direction of propagation, a definite traveling wave was excited in the case of the end shaking test. This is shown in the sequence of frames in Fig. 6-13 (a) and (b) for f = 25 Hz and Γ = 12.7 and f = 25 Hz and Γ = 14.6, respectively, where a wave structure is created in which a segment of rod loses contact with the bottom of the constraint and travels away from the shaker. In Fig. 6-13 (b), a smaller amplitude wave seemingly merges with a large amplitude wave by catching up with it. This behavior suggests possible height-dependent wave speed, creating a similar dispersive effect as observed by [145] on the inertial dynamics of a ruck in a rug on a flat floor (the plate analogue to this problem without the top constraint). While this comparison is not explored further in this thesis, it appears to be a rich avenue for future work. Highlighting the difference in excitation between the two forms of shaking, a direct comparison between distributed and end shaking is shown in the image sequences of 173 a) 3cm b) t=0.000s t=0.000s t=0.0100s t=0.0100s t=0.0199s t=0.0199s t=0.0299s t=0.0299s t=0.0398s t=0.0398s t=0.0498s t=0.0498s t=0.0597s t=0.0597s Figure 6-13: Sequence of experimental video frames of the end shaking experiment consisting of a rod inside a stationary glass pipe with I.D.=12.0 mm with the end outside the pipe vibrating at f = 25 Hz at (a)Γ = 12.7 and (b)Γ = 14.6. Photos were taken from a video recording at 402 frames per second. Of note is the definite traveling wave in both cases, with a fine structure in (b) consisting of a larger amplitude wave followed by a smaller amplitude wave, which appears to catch up with the larger wave in the first three images. 174 a) 3cm b) t=0.000s t=0.000s t=0.0050s t=0.0050s t=0.0100s t=0.0100s t=0.0149s t=0.0149s t=0.0199s t=0.0199s t=0.0249s t=0.0249s t=0.0299s t=0.0299s Figure 6-14: Photo sequence of a rod inside a glass pipe with I.D.=12.0 mm. a) Distributed shaking test at f = 62 Hz and Γ = 9. A large amplitude vertical motion in the first three frames appears to generate two smaller, opposite traveling waves over 0.015 ≤ t [s] ≤ 0.0249. b) End shaking test at f = 62 Hz and Γ = 17.2, exciting a wave propagating right to left, away from the shaker. Photos were taken from a video recorded at 402 frames per second. 175 Fig. 6-14 (a) and (b) for distributed and end shaking, respectively. In both cases, the driving frequency is f = 62 Hz, with Γ = 9 and 17.2, respectively. For the case of distributed vibration, we see large amplitude, vertical motion of rod segments without a clear traveling direction, with small traveling waves departing from these areas of large vertical motion. The main difference, however, is that while the end shaking experiment creates a clear preferred direction (away from the point source at the end), the smaller traveling waves excited in the case of distributed shaking travel in both directions, away from the large vertical displacements. Having found qualitative evidence of traveling waves in the rod constrained within the cylinder for both distributed shaking and end shaking tests, we now present results of peak response frequency, as well as wavelength measurements. In Fig. 6-15 we plot the measured peak response frequency, fpk , for both distributed and end shaking tests, as a function of driving frequency, f . Subharmonic response (fpk < f ) was measured 100 Peak Response Frequency, fpk 90 80 70 Distributed Shaking End Shaking 60 50 40 30 20 10 0 0 20 40 60 Driving Frequency, f 80 100 Figure 6-15: Peak response frequency, fpk , as a function of driving frequency, f , for both distributed and end shaking experiments. For high driving frequencies (f & 35 Hz), the response frequency was observed to be subharmonic and relatively constant at approximately 20 Hz. 176 for the majority of tests. For f & 40 Hz, an approximately constant peak response frequency is noted of fpk ≈ 20 Hz. The image sequences in Fig. 6-14 present qualitative evidence for traveling waves, which we hypothesize are bending waves. We proceed by comparing the measured peak response frequency, fpk , measured for the rods with a measured wavelength, λ. For bending waves, these two quantities are related [119] by the expression, λ= s 2π fpk s 4 EI , ρA (6.6) where EI is the bending stiffness of the rod, ρ is the volumetric density, and A is the cross-sectional area. Wavelength was automatically measured for each frame of the high speed videos for both the distributed and end shaking tests with a process shown graphically in Fig. 6-16 for an end shaking test with f = 35 Hz and Γ = 9.7. The measurement was performed by reconstructing the mean elevation, y, and horizontal position, x, of the instantaneous configuration for a particular frame. We define mean elevation, y, as the elevation of a particular point on the rod compared to the average elevation of the entire rod. Fig. 6-16 (a) shows the portion of the video frame analyzed and Fig. 616 (b) presents the reconstructed rod position (with exaggerated vertical scale), with mean rod elevation, y, plotted as a function of horizontal position, x. The wavelength, λ, was the lag that resulted in the maximum autocorrelation of the reconstruction. The autocorrelation function measures spatial self-similarity of a configuration. The autocorrelation, R, of a function, a, for a lag, m, is calculated according to, N −m−1 X 1 R(m) = an+m an , m ≥ 0, N − m n=0 (6.7) where N is the number of coordinates in the spatial reconstruction. The autocorrelation is a signal multiplied by itself after shifting by a distance (lag), m. This is shown in Fig. 6-16 (c), calculating the autocorrelation, R, for the reconstructed configuration for different values of the lag, m. Here a lag of m = 122 mm maximizes 177 R (and is indicated as the wavelength, λ, measured), which agrees with the distance between the two elevation peaks in the reconstructed rod of Fig. 6-16 (b). Lags under 15% or over 85% of the overall length were not considered valid measurements (and set to zero, as seen in Fig. 6-16) as the sample size for those regions was considered too small for accurate measurements. We now have the peak response frequency, fpk , and measured wavelength, λ, for 3cm b) Elevation, y [mm] a) 3 2 1 0 −1 c) Unbiased Correlation, R 0 50 100 150 x [mm] 200 250 50 100 150 m [mm] 200 250 1 0 −1 0 Figure 6-16: Image processing steps to measure the wavelength, λ from experimental video frames. a) Experimental image from high speed video of end shaking test with f = 35 Hz and Γ = 9.7. b) Rod reconstruction plotting mean elevation, y, as a function of horizontal position, x. Note the amplified vertical scale compared to the x-axis. c) Autocorrelation, R, of the reconstruction for different sizes of lags, m, with the peak value of R corresponding to the measured wavelength, λ. 178 the distributed and end shaking tests. To assess the validity of our hypothesis that bending waves are present in both of the vibration cases, we compare the wavelength observed to the driving frequency. The wavelength observed should be corrected, however, to take into account the fact that the peak response frequency, fpk , should be responsible for the bending waves present in the rod. The corrected wavelength, λcorr is calculated by, λcorr = s fpk λ. f (6.8) In Fig. 6-17, we plot wavelength as a function of driving frequency, f . When experimental measurements of λ are corrected for the peak response frequency measured, good functional agreement is observed for both experimental cases. We take this result to be indicative of bending waves being present in the constrained rod for both the vibrated rod and the vibrated constraint. This is important for the dynamic real case as it introduces a frequency-dependent length scale (λ, given in Eq. (6.6)) as well as providing an understanding of how adjacent sections of the constrained rod will interact in a vibrating constraint which was not provided in the previous section, which focused solely on the flight time of an independent cross-section. The next section provides a direct application of this discovery by investigating Lhel inj in the dynamic real case as a function of vibration frequency. 179 Corrected Wavelength [mm] 160 140 120 100 80 60 40 End Shaking Distributed Shaking Bending Waves 20 0 0 20 40 60 Driving Frequency [Hz] 80 100 Figure 6-17: Comparison of observed wavelengths (corrected according to Eq. (6.8) with theoretical prediction (Eq. (6.6)) for the range of driving frequencies tested in both the distributed shaking and end shaking tests, 22 ≤ f [Hz] ≤ 100. 6.2.4 Effect of Vibration Frequency on Helix Initiation In Fig. 6-6, we saw that the length of rod injected before helix initiation, Lhel inj , was greater for vibrations with f = 50 Hz than for either f = 100 or 200 Hz vibrations, while keeping peak acceleration, Γ, and injection speed, vinj , constant. Fig. 6-18 shows Lhel inj plotted as a function of the vibration frequency for a rod injected into a pipe with I.D. = 21.7 mm with Γ = 2 and vinj = 10 cm/s. Lhel inj is plotted normalized by the static value (Lhel inj measured without vibration of the constraining cylinder). We find that Lhel inj remains approximately constant for f ≥ 200 Hz, while showing greater increases for lower vibration frequencies. This observation is consistent with the proposal of bending waves a part of mechanism of improvement since bending waves will only be excited and propagate for wavelengths larger than the rod diameter [119]. 180 Normalized Helix Initiation 1.5 1 Static Value 0.5 0 0 100 200 300 Frequency [Hz] 400 500 Figure 6-18: Helical initiation (normalized by the static value of Lhel inj ) as a function of excitation frequency, with vinj = 10 cm/s and Γ = 2 inside a rod with I.D. =21.7 mm. For the rods used in these experiments, this corresponds to f ≈ 160 Hz (substituting values into Eq. (6.6), with λ = drod ). This limiting value of frequency lies in the range observed in Fig. 6-18. We now consider vibration frequencies that are low enough to excite bending waves, so that we assume adjacent rod sections can influence one another through bending. Specifically, we address the question of whether there is a frequency to maximize Lhel inj . First, we can calculate the amount of energy injected into the vibrating constraint for different frequencies of vibration. Assuming the mass of the system to be constant, we can express the power injected into the system through the vibration of the constraint to be proportional to Γvpk ∝ 1/f (where vpk is the peak velocity of vibration, vpk = Γg/(2πf )), implying that if improvements in Lhel inj are a function of the power input to the rod, we expect Lhel inj to increase monotonically with decreasing f . The other possibility would be a resonant behavior, whereby a peak value of Lhel inj as a function of frequency would exist. 181 To explore these two possibilities for the variation of Lhel inj as a function of frequency, experiments were run at low frequencies (35 ≤ f [Hz] ≤ 100) for Γ = 1.75 inside of pipes with I.D. = 12.0 and 21.7 mm. The injection speed was lowered to a fraction of the peak vibration velocity, vpk , bringing the experiment closer to the situation of the previous section, when there was zero imposed lateral velocity of the rod. Injection velocity was set to vinj = 0.5vpk and 0.25vpk for I.D. = 12.0 and 21.7 m, respectively. In Fig. 6-19, we plot normalized Lhel inj as a function of excitation frequency for both pipes, noting a peak in Lhel inj in both cases, with the maximum increase in Lhel inj being approximately 2.5 and 3 times higher than the static value for I.D. = 12.0 mm and 21.7 mm, respectively. The peak in Lhel inj is at different frequencies for the two cases considered, with the peak located at f = 62 and 72 Hz for I.D. = 21.7 and 12.0 mm, respectively. We propose a rationalization for this behavior. The presence of an optimum frequency of vibration for a constant value of peak acceleration suggests a mechanism other than one purely explained by power input to the system through vibration, and is reminiscent of resonance. Of particular note is the apparent dependency of the optimum frequency on the size of the constraint. The length scale set by bending waves is not dependent on the constraint, rather only the properties of the constrained rod, as was given in Eq. (6.6). The buckling wavelength, however, does depend on the size of the constraint, as characterized by the radial clearance, ∆r, between the rod and cylinder, as was originally presented for the sinusoidal buckling wavelength in the classic case as Eq. (2.16), λscr = 2π EI∆r w 1/4 , where EI is the bending stiffness and w is the weight per unit length of the constrained rod. The value of λscr was derived assuming perfect contact between the rod and cylinder, an assumption we have observed to be violated in our experiments. In the case of the dynamic real case, wherein peak accelerations greater than gravity, Γ > 1, have been shown to cause contact loss (e.g. the experimental frame sequences in Fig. 182 Normalized Helix Initiation 3 2.5 2 1.5 1 ID=12mm, vinj= vpk/2 0.5 ID=21.7mm, v =v /4 inj 0 0 20 40 60 Frequency [Hz] pk 80 100 Figure 6-19: Helical initiation normalized by the static value as a function of excitation frequency for Γ = 1.75 inside of two different sizes of constraining pipes. An optimum excitation frequency appears to exist for both cases (f = 62 and 72 Hz for I.D. = 21.7 and 12.0 mm, respectively). 6-14 (a)), we assume portions of the sinusoidally buckled rod will lose contact with the constraint. The highest points (peaks of the sine wave) will stay in contact with the pipe to support the rod. The lowest point of the rod will be near or in contact with the bottom of the constraining pipe to minimize the gravitational potential energy. The sections of rod between these points, however, may lose contact with the pipe (releasing the bending energy required to remain in perfect contact). These unsupported lengths between the highest and lowest points of the sinusoidal wavelength would be approximately λscr /4 long. With vibrational excitation, we assume that these unsupported lengths are responsible for the resonant behavior observed. If we take the lengths to have clamped 183 end conditions and the length to be exactly λscr /4, we can calculate the resonant frequency, fn , as [119], 1 fn = 2π 1.5π λscr /4 2 s EI ρA (6.9) making the natural frequency inversely proportional to the radial clearance, ∆r. In Fig. 6-20, we plot the peak frequencies from our experiments reported in Fig. 619 as a function of ∆r, comparing them to the analytical prediction of Eq. (6.9), with no fitting parameters. We note that experimental measurements are higher than predicted by Eq. (6.9), which can be physically interpreted at shorter resonant lengths than the λscr /4 predicted, possibly accounting for finite-length contact points. More experimental data points will be required to confirm this prediction of Eq. Peak Helix Initiation Frequency [Hz] 140 Experiment Theory (resonance for λ/4) 120 100 80 60 40 20 0 0 2 4 6 Radial Clearance [mm] 8 10 Figure 6-20: Experimentally measured frequencies resulting in peak Lhel inj as a function of radial clearance, taken from Fig. 6-19. Also plotted is the theoretical prediction (Eq. (6.9)) based on the resonant frequency for the quarter-wavelength of the sinusoidal buckling wavelength, λscr , which appears to agree with experimental results. 184 (6.9), especially at smaller diameter constraining pipes (i.e. smaller ∆r). However, acquiring this data is time intensive, requiring approximately two weeks for each data point. Moreover, the current experimental apparatus will need to be extended to accommodate the increases in Lhel inj anticipated for tighter clearances. The next section explores the last control parameter considered in the dynamic real case, the injection velocity. 6.2.5 Effect of Injection Speed Thus far, both Γ and f have been shown to have a definite effect on Lhel inj . The injection speed was shown to have no effect on Lhel inj in the static case of Chapter 5 (experiments with no vibration of the constraint). In order to explore the effect, experiments were run varying the injection speed for both clearances considered in this chapter, I.D. = 12.0 and 21.7 mm. In Fig. 6-21, for the pipe with I.D. = 12.0 mm, we plot the static-normalized Lhel inj as a function of vinj , which is normalized by the peak velocity of the imposed vibration, vpk = Γg/(2πf ). Tests were conducted with vibrations at three frequencies, f = 50, 75, and 100 Hz, over a range of injection velocities, 0.3 ≤ vinj /vpk ≤ 3.7. The injection speed is not seen to have a strong effect for vinj > vpk , with Lhel inj remaining approximately constant at 1.5 times the static value. However, for “slow” injection speeds (vinj < vpk ), Lhel inj increases with decreasing vinj , reaching approximately 2.5 times the static value of Lhel inj . In the larger diameter constraint, I.D. = 21.7 mm, the improvement of Lhel inj is again noted to increase with decreasing vinj , below the peak velocity of vibration, vpk . This is shown in Fig. 6-22, where we plot the normalized Lhel inj as a function of normalized injection speed. Three vibration frequencies were used once again for experiments, f = 50, 100, and 200 Hz, with injection speeds in the range of 0.2 ≤ vinj /vpk ≤ 4.8. A large frequency dependence was observed for Lhel inj at slow injection speeds, with f = 50 Hz exhibiting almost twice the improvement in Lhel inj than the other two frequencies. All tested frequencies also appeared to have a plateau, whereby there was a secondary critical injection velocity below which Lhel inj did not 185 3 Static 50 Hz 75Hz 100Hz vinj=vpk Normalized Helix Initiation 2.5 2 1.5 1 0.5 0 0 0.5 1 1.5 2 2.5 3 Normalized Injection Speed, vinj/vpk 3.5 Figure 6-21: Normalized helical initiation as a function of normalized injection speed. Shown for pipe ID = 12.0 mm, Γ = 1.75, and f = 50, 75, and 100 Hz. For injection speed above vpk , Lhel inj is approximately constant, while the it increases with slower injection speeds. 3 Static 50 Hz 100 Hz 200 Hz vpk Normalized Helix Initiation 2.5 2 1.5 1 0.5 0 0 1 2 3 Injection Speed/Vpk 4 5 Figure 6-22: Normalized helical initiation as a function of normalized injection speed. Shown for pipe ID = 21.7 mm, Γ = 2, and f = 50, 75, and 100 Hz. For injection speed above vpk , Lhel inj is approximately constant, while the it increases with slower injection speeds. 186 improve with further decreasing vinj . The mechanism for this improvement is not known at this time and remains a topic for future study. Speculatively, the “slow” injection velocity regime, i.e. vinj < vpk , can be considered as corresponding to a motion of rod sections losing contact with the constraining pipe that are predominately vertical. This vertical motion would encourage the propagation of bending waves in both directions, as seen in the distributed shaking experiments in §6.2.3. With increasing injection velocity, a propagation direction away from the injector (and the buckling instabilities) for bending waves may become apparent. The mechanism of improving Lhel inj with decreasing vinj may also be related to the competition between two timescales. Namely, the relaxation time of the buckled rod configuration as frictional contacts are destabilized against the rod being added (and compression forces building through frictional interaction) through injection. Anecdotally, if rod injection is stopped at any point before lockup in the dynamic real case when peak accelerations are sufficiently high (Γ > 1), the rod will adopt a straight configuration along the bottom of the constraining pipe subject to small amplitude displacements due to the vibration. Regardless of the underlying mechanism that is yet to be fully rationalized, the injection velocity is an important ingredient in determining Lhel inj in the dynamic real case. 6.3 Outlook We have experimentally shown that the application of vertical, distributed vibration to the cylindrical constraint increases the amount of rod that can be injected before helical initiation. This is analogous to extending reach, which is highly desirable in the industrial application of coiled tubing operations in horizontal wellbores. The various control parameters of the system (peak acceleration, vibration frequency, and injection velocity) were all seen to affect helix initiation. It was shown that for sufficiently large peak accelerations, Γ > 1, portions of the rod lost contact with the constraint and Lhel inj increased compared to the static (no 187 vibration, Γ = 0) case. The analogous system of the bouncing ball on a vibrating plate was found to provide a good estimation for the amount of contact loss that could be expected for a given peak acceleration. Through high speed video analysis, however, it was observed that bending waves were present. These bending waves appear to dictate an upper limit of useful frequencies, above which, Lhel inj appears constant with changing f . Below this upper bound on vibration frequency, however, an optimum (resonant) frequency appears to exist. An initial theoretical prediction was provided for this optimum frequency, but more experimental results will be required to confirm it. This chapter concludes the investigations reported in this thesis. The following chapter will summarize the main results of the thesis, as well as provide suggestions for future work. 188 Chapter 7 Conclusions and Future Work We have presented an investigation into the geometrically nonlinear behavior of rods under various loading and constraint conditions. Four problems were presented: i) The hanging shape of a naturally curved rod; ii) The classic case, where we investigated mechanical instabilities of a rod compressed inside of a cylinder; iii) The real case, wherein a rod was injected into a cylinder, with frictional resistance causing the observed buckling and post-buckling behavior; and iv) The dynamic real case, where we studied the effect of vertically vibrating the cylindrical constraint of the real case, with the goal of increasing the length of rod injected before helical buckling was observed. In each case, an experimental apparatus was designed and constructed and the results were interpreted and rationalized. The shapes of suspended, naturally curved rods: The spooling of rods for storage and transport can impart a natural curvature, but the way in which this parameter affects the subsequent behavior of the rods is, to date, not yet well understood. We have created a simple model system consisting of a rod hanging under its own weight. The rods were custom fabricated with precise control over the natural curvature (in addition to other geometric and material properties), which was considered as a control parameter in experiments, simulation, and analysis. Through the variation of this parameter, rods of equal length, density, and stiffness were observed to take on planar and non planar shapes. The transition from planar to non-planar shapes was shown to be a mechanical instability arising from a symmetry breaking bifurcation, 189 analogous to an inverted pendulum. The resulting non-planar shapes were further classified as localized helices or global helices. In the localized helices, the transition between the localized helical structure beneath the straight portion was well described by the mechanical instability of a naturally curved rod under decreasing tension. A novel fabrication method was introduced and a framework for a naturally curved rod under variable tension was developed. Classic case: A rod was constrained inside a horizontal cylinder and pinned at both ends and compressed. This scenario is analogous to drilling extended reach boreholes in the petroleum industry. Under imposed end displacement, the initially straight rod first buckled into a sinusoidal mode, with the critical load set by a combination of elasticity, gravity, and the radial clearance between the constraint and the rod. Further compression led to helical buckling of the rod, with the same parameters affecting the magnitude of the critical buckling load. After helical buckling, however, the normal force between the rod and pipe increased rapidly under continued displacement, as evidenced by a divergence between the input and output reaction forces. Friction between the rod and cylinder resulted in higher critical loads for both sinusoidal and helical buckling than predicted, but with general agreement in functional form. Imperfections in the horizontal constraint, present due to sag of the discretely supported pipes, caused large reductions in the critical loads. These reductions appeared to be linear with imperfection size for sinusoidal buckling but non-linear for helical buckling. A precision experimental data set provides an extensive benchmark for existing and future theoretical work, and also allowed for the imperfection analysis performed. Real case: We also explored the related problem of progressively increasing the length of rod injected into the constraining cylinder, under imposed velocity. This scenario is analogous to the insertion of coiled tubing into existing horizontal wellbores in the petroleum industry. Frictional resistance to insertion between the rod and the pipe leads to an increasing axial compressive load during injection, with local sinusoidal buckling observed preceding helical buckling of portions of the rod near the point of injection. Injection past the initiation of helical buckling caused a rapidly 190 increasing reaction force at the injector, with the rod eventually locking up, thereby preventing further insertion. The length of rod that could be injected before helical buckling was found to vary inversely with radial clearance, with good agreement observed between experiment and existing theory. Natural curvature of the injected rod was found to have a nontrivial effect on the problem. Sufficiently small values of natural curvature did not seem to affect the amount of rod which could be injected before helical buckling, but above a critical value, a reduction in the amount of rod injected was observed. This chapter provided an experimental method for reproducible friction between rubber and glass, an extensive phenomenological description of the processes leading to lockup (for comparison with common theoretical assumptions), and a data set for direct comparison with theoretical developments. The data strongly suggests that theories neglecting lateral friction are accurate in predicting helical initiation. Dynamic real case: Finally, we expanded on the real case by vertically vibrating the cylindrical constraint into which the rod was injected. For peak accelerations above 1g, the rod was observed to lose contact with the vibrating constraint. This resulted in the ability to inject more rod before helical buckling occurred, with stronger vibrations resulting in increased reach. Injecting the rod at velocities lower than the peak vertical velocity of the vibrating constraint also resulted in increased injection lengths before helical buckling. In this slow injection speed regime, the frequency of vibration had an effect on the resulting improvements. Frequencies which could excite flexural waves were found to cause the greatest improvements. Our results suggest that a resonant frequency of the system exists, such that the frequency maximizing the amount of rod that can be injected prior to helical initiation depends on the elastic properties of the rod, gravity, and the constraint geometry, as characterized by the radial clearance. We reported improvements of 300% in helical initiation. Further reach may be accomplished using this mechanism, but a longer experimental apparatus will be needed for future testing. The experimental apparatus for this chapter provided the first data set of its kind, suggesting possible directions for future theoretical work. The experiments also provided guidelines for improvement in terms of 191 the ideal frequency of vibration as well as the maximum injection speed, below which reach is extended even more. 7.1 Future Work As with any research project, our work opens avenues that deserve further exploration. Some possible future directions in the existing experiments are listed here. • The effect of elliptical cross-sections on the hanging shape of naturally curved rods would be a natural extension of Chapter 3. Many of the same processes which produce natural curvature in the manufactured environment also result in slight ovaling of the cross-section, particularly in hollow sections, such as pipes. This ovalization breaks the symmetry of circular cross-sections, creating a different coupling between bending and twisting for the two different principal directions, which would introduce non-trivial changes in the transitions between planar and non-planar shapes, as well as the resulting non-planar shapes. • For both the classic and real cases, an important parameter for future exploration is the coefficient of friction. There is limited theoretical works in the existing literature, and experimental results would be beneficial in quantifying scalings for the effect of friction on critical loads and length scales. This quantification could be accomplished through measuring either critical loads or the frictional drag between the two end constraints. • In both the classic and real cases, further quantifying the effect of imperfections in the geometry of the constraining cylinder on critical loads or injected lengths should be explored further. This would directly inform how the majority of theoretical and experimental work can be applied to the industrial problems. All drilled wellbores have inherent tortuosity, but a question still remains as to whether they can be treated as straight during analysis for drilling or coiled tubing operations. 192 • While natural curvature was shown to affect the lockup length in the real case, a theoretical prediction does not yet exist. To inform scalings, more experiments are merited to explore two facets of the problem in particular: i) How does the radial clearance affect the critical natural curvature, above which reductions in lockup length are observed? and ii) What is the functional form of the decay in lockup length with increasing natural curvature above this critical value? This avenue, in particular, motivates a more formal scaling analysis to connect the experimental model system with the field case. • On a more fundamental level, the exact connection between the desktop experiments and the field case is not a simple analytical task, and deserves exploration. Identifying the key dimensionless parameters as combinations of the elastic properties of the rod, the buoyant weight of the rod, the radial clearance between the rod and constraint, as well as the injection velocity will be important in understanding how the lessons learned in the desktop scale can connect to the oil field. • The dynamic real case explorations discussed in Chapter 6 have raised several possible avenues of further research. A more fundamental understanding of bending wave propagation within a cylindrical constraint (with possible connection to the work performed on a traveling wave in a plate lying on a flat constraint [145, 146]) would be helpful in the exploration of such a large parameter space. A more precise prediction of the optimum frequency appears possible with the addition of experimental data. The rate of increase in reach extension as a function of decreasing injection speed an open arena which is also of industrial importance. This injection speed effect is particularly interesting, as injection speed is not considered in existing theories. • Finally, the dynamic real case suggests that one can extend the amount of rod which can be injected before critical buckling loads are reached. Chapter 6 explores one of these methods, with several others deserving investigation. 193 Variable injection speed matched with vertical vibration has been anecdotally shown to have a strong effect on the buckling process. Some of these methods include lateral vibration of the constraint (perpendicular to gravity and injection), axial vibration of the injected rod during injection, and the introduction of non-circular cross-sections. 194 Appendix A Rod Fabrication This appendix presents a novel method for fabricating rods with precise control over natural curvature, κ0 . This method was first reported in [109, 110]. Fig. A-1 (a) is a photograph of several rods made with this procedure, resting at their natural curvature. We were motivated in the development of this procedure by the fact that commercially available rubber rods are typically extruded and then wrapped around a spool for transport and storage, resulting in a set natural curvature, as shown in Fig. A-1 (b), which presents a photograph of a purchased spool of silicone rubber1 . Natural curvature is present in this spool as evidenced by the small hook (curved) portion near the stress-free tip of the rod. Rods such as those shown in Fig. A-1 (b) were unacceptable for use in our experiments for two primary reasons. Firstly, several studies (in this thesis and in the research group as a whole) investigate the role of natural curvature, in which case it must be a control parameter in samples. Secondly, for those experiments not specifically exploring natural curvature, the spooled rods were inappropriate as the intrinsic curvature varied along the arc length, depending on the radius at which it was spooled. 1 O-ring cord stock purchased from McMaster-Carr 195 a) b) 3 cm 4 cm Figure A-1: a) Photograph of several rods fabricated with this method, with natural curvature in the range, 0 ≤ κ0 ≤ 62 m−1 . Different colors correspond to different stiffnesses, with the light green (two extreme curvatures) used in the tests of this thesis. b) Photograph of a spool of commercially available silicone rubber rod, with natural curvature evidenced by the hook at the hanging free tip. A.1 Rod Fabrication Procedure An injection molding procedure was used to fabricate rods with set values of natural curvature. A flexible PVC tube (inner diameter, DI = 3.16 mm, and outer diameter, DO = 5 mm) was wound around a cylindrical object of diameter, Dm , setting the eventual natural curvature of the rod, κ0 to be κ0 = 2(DO + Dm )−1 . In some cases, the desired rod length required several windings around the object. In these cases, the rod was wrapped with a helical structure with a constant pitch of the PVC tubing outer diameter, which was assumed to have a negligible effect on the resulting value of κ0 . Fig. A-2 presents a photograph of a PVC tube wound around a cylindrical object before injection. After the PVC tube was affixed to the cylindrical object, a flowable polymer is injected into the tube using a syringe. Rods were fabricated using a vinylpolysiloxane (VPS) polymer2 , consisting of a base and catalyst part, which, when mixed, would 2 Elite Double 32, a Vinylpolysiloxane duplicating material manufactured by Zhermack was the preferred polymer for this thesis. Zhermack makes other polymers in the Elite Double series, with the following number characterizing the Shore A hardness. Elite 32 is the stiffest material produced by Zhermack. 196 Figure A-2: Flexible PVC tubing is wrapped around a cylindrical object before the two part polymer is injected with a syringe. polymerize and become solid after a working time, and eventually setting. The manufacturer quotes the working time at 10 minutes and setting time at 30 minutes at room temperature, although typical working time was approximately 4-5 minutes at room temperature. After allowing the injected polymer to set for at least one hour inside the mold, the PVC tubing is cut away to release the inner, slender VPS elastic rod with set natural curvature. Fig. A-3 shows a photograph of a VPS elastic rod beginning to be cut from its mold. We found that the material properties of the rods resulting from this manufacturing method were dependent on the protocol used during fabrication, especially in preparing the liquid polymer components. The following procedure was followed for rods prepared in this thesis: 1. Pour equal parts (by mass) catalyst and base into a plastic or paper cup, to the 197 nearest 0.1g. For the rods used in this thesis, 30 g of each part was sufficient. 2. Hand stir the mixture for 60 seconds, taking care not to entrain air while stirring with a wooden tongue depressor. Stir completely, taking care to mix all the polymer that may be on the bottom or sides of the cup. 3. Transfer to a vacuum chamber for 45-50 seconds of degassing, which removes the largest air bubbles from the mixture. 4. Immediately pour the liquid polymer into a syringe attached to the PVC tubing, taking care to pour from a height of 6-12 inches above the syringe to avoid any excess air entrainment. 5. Insert the plunger into the syringe, tilting the injection point up to purge the air in the syringe out before injecting. 6. Inject the polymer via the syringe into the PVC pipe. The quality of mixing as well as the vacuum step appeared to cause the greatest variability between fabricated rods. Rods, when not in use, should be stored at or Figure A-3: Demolding of a VPS elastic rod was accomplished by cutting away the PVC tubing mold. Photo courtesy of Arnaud Lazarus. 198 very near their fabricated natural curvature, otherwise deviations from the set value will occur due to long term curing. We next summarize material properties measured for the fabricated rods. A.2 A.2.1 Material Properties and Measurements Cross-Section and Density The cross-section of representative rods was measured using a desktop digital photograph scanner. The Image Processing Toolbox in MATLAB was used to measure the cross-sectional area of representative samples of the rods, taken from different rods and at different points along the same rod. The area was used to compute the equivalent diameter of a circle. Cross-sections were inspected with calipers for ellipticity, although only rods fabricated with κ0 = 62 m−1 were found to show ovaling of the cross-section (with a 10% difference measured between the major and minor axis). This ovaling was caused by warping of the PVC tube cross-section, making κ0 = 62m−1 the upper limit of rods which can be fabricated using the current PVC tubes. For rods with circular cross-sections, the diameter, d was measured to be d = 3.16 ± 0.05 mm. With the cross-section measured, the mass of rods was measured to obtain the volumetric density, ρ, of the fabricated rods. Varying lengths of rod were used, with the density measured to be ρ = 1210 ± 8 kg/m3 . A.2.2 Young’s Modulus The Young’s modulus, E, was determined using two methods which we refer to as the natural frequency, flick, test, and the annulus test. Both methods measured the bending stiffness, which was used, in conjunction with the radius measurement from the previous section, to calculate the Young’s modulus, E. These techniques were selected for their ability to measure E of a slender rod relatively easily. This is in contrast to most standard, tension, based methods, which require particular specimen 199 geometries or require complex mounting equipment in order to test slender objects3 . Both methods, however, could only be used to test naturally straight (κ0 = 0 m−1 ) rods, which were assumed representative of the fabricated naturally curved rods in terms of material properties. The flick test consists of clamping a rod sample between two acrylic plates with a free length extending above the plates, as is shown in a photograph in Fig. A-4. The free tip was displaced laterally a small distance (∼ 1 cm maximum) and released, exciting the first mode of vibration. The vibration was recorded with a digital video camera4 . Each sample would be tested five times, and each test would be the average of several (∼ 20 − 50) periods. The Young’s modulus was then calculated from [119], fn = β12 2πL2 s EI , ρA (A.1) where fn is the natural frequency of the first mode, β1 is a constant depending on the boundary conditions (for clamped-free, β1 = 0.597π), L is the free length of the rod, EI is the bending stiffness, ρ is the volumetric mass density, and A is the cross-sectional area of the rod. The flick test also allowed one to calculate the damping ratio, ζ, which measures the dynamic losses of a material (modeled as viscous damping). Tracking the lateral deflection of a single point on the rod as a function of time, one can determine ζ by using the logarithmic decrement method. For ζ < 0.1 (which is true of the vast majority of materials), a good approximation for the damping ratio using this method is [119], ζ= ln 2 , 2πN50 (A.2) where N50 is the number of cycles for a signal to decay by 50%. Using this technique, we measured the damping ratio to be ζ = 0.027 ± 0.003. 3 The value of the Poisson’s ratio used in this thesis, ν = 0.49 was measured using a tensile test of a dogbone sample, with the value courtesy of Denis Terwagne. This value is very nearly the incompressible ν = 0.5 that is typically assumed for rubbers. 4 Kodak PlaySport Zx5 digital video camera, recording at 59.94 frames per second. 200 Figure A-4: Photograph of a fabricated straight elastomeric rod (d = 3.16 mm) clamped between two acrylic plates with cantilevered length L = 23 mm. Displacing and releasing the tip would excite the first mode of vibration, the frequency of which was used to measure the Young’s modulus, E. The annulus test [147] relates the elastic properties of an elastic hoop to its shape when it is hung by a single point, as shown in Fig. A-5. Making the hoop in our case required bonding the ends of a length of rod together, using a small (∼ 1 ml) amount of liquid polymer and letting it set for the 30 minutes. Special care was taken to orient the two ends, ensuring that no significant twist was imparted to the rod (another class of stability problems in the study of rods, with applications to DNA [25]), and that the two ends are aligned well. Measuring the width, W , and height, H, of the resulting hanging elliptical shape, one finds the corresponding ratio of total rod length, L, to the gravito-bending length, which is a length scale expressing the balance of elastic and gravity forces, defined as, Lgb = s 3 EI , ρgA (A.3) where all other terms have been measured or are known. The relationship between W/H and L/Lgb is a numerical solution to the Kirchhoff equilibrium equations (pre201 sented in Chapter 1), and is shown graphically in Fig. A-5. The annulus test was used to evaluate the evolution of Young’s modulus immediately following the fabrication of rods. Upon demolding, the rods were observed to have a sheen of silicone oil on them. The Young’s modulus was tested for a rod every day after demolding for the first 11 days. Each day, the modulus was measured three times to ensure reproducibility of the test. In Fig. A-6, we plot the measured Young’s modulus, E, as a function of age after demolding. We see in Fig. A-6 that the E increases steadily over the first week after demolding. This result was confirmed with the flick test. To have consistent material properties, all rods were demolded and then temporarily stored for at least seven days before use in experiments. The value of Young’s modulus measured for the plateau region is E = 1296 ± 31 kP a. a) g b) 10 9 8 Hanging Point 7 gb 6 L/L H 5 4 Weld 3 2 1cm W 1 0 0.4 0.5 0.6 0.7 W/H 0.8 0.9 1 Figure A-5: a) Annulus test consisting of a rod with overall length L = 29.9 cm, with the ends connected at the weld point. The hoop is then hung from a barb of nitinol (d = 1.1 mm) and the width, W to height, H, ratio is measured. b) Using a numerical solution [147], the gravito-bending length, Lgb , and subsequently the Young’s modulus, E, can be found. 202 1600 Young’s modulus, E [KPa] 1500 1400 1300 1200 1100 1000 0 5 10 15 Age [days] 20 25 Figure A-6: Young’s modulus, E, tracked as a function of time, in days, after the rod was demolded from the PVC tubing. Young’s modulus is seen to steadily increase 30% for the first week, and then plateaus. A.2.3 Coefficient of Restitution The final test performed to characterize the custom fabricated rods for this thesis was to measure the dimensionless coefficient of restitution, CR , which measures how elastic collisions are between two objects and lies in the range 0 ≤ CR ≤ 1. It can be measured from the ratio of velocities of an object onto a rigid surface. We measured CR for a sphere of the VPS polymer (with diameter, d = 3.8 cm) suspended from a ground steel shaft by an insulated copper wire impacting a steel plate, as shown in Fig. A-7. The distance between the pivot point and center of mass of the sphere, L, was measured to be L = 28.7 cm. The steel plate was aligned such that the sphere impacted it at the bottom tangent of the arc defined by the pendulum 203 motion. This alignment allowed for the maximum pendulum height before and after an impact to be measured. Performing an energy balance and assuming that there is no frictional, aerodynamic, or vibrational dissipation, the coefficient of restitution could then be calculated as, CR = s 1 − cos θi , 1 − cos θf (A.4) where θi and θf are the maximum angles made by the pendulum with the vertical before and after an impact. This measurement was made with image processing tools after recording 22 impacts between the ball and the steel plate. This technique gave the value CR = 0.75 ± 0.02. Pivot Point Figure A-7: Pendulum impact test to measure the coefficient of restitution, CR , for a sphere of the VPS polymer the rods were manufactured with and a rigid steel plate. The angle between the center of mass of the sphere and the pivot point was tracked, with the maximum angle immediately preceding, θi , and following, θf , impact being used to measure the relative velocities. 204 A.3 Summary We have developed a new method for fabricating rods with a precise, controllable natural curvature, κ0 , in the range 0 ≤ κ0 ≤ 62 m−1 , with the upper bound set by warping of the mold. An array of material properties were measured for the rods and are summarized in Table A.1 and used in Chapters 3, 5, and 6. It was found that the constituent material reached a steady-state value of Young’s modulus approximately one week after demolding, with an approximately 30% increase in E during that time. Property Young’s Modulus, E Density, ρ Diameter, d Poisson’s Ratio, ν Coefficient of Restitution Damping Ratio, η Value 1296 ± 31 [kP a] 1210 ± 8 [kg/m3 ] 3.16 ± 0.05 [mm] 0.49 0.75 ± 0.02 0.026 ± 0.003 Table A.1: Material properties of rods manufactured with tunable natural curvature. 205 206 Appendix B The Shapes of a Suspended Curly Hair In this Appendix we reprint a submitted manuscript which was the result of a collaboration with Basile Audoly1 and Arnaud Lazarus2 . The aim of this work was to rationalize the mechanical behavior (with particular emphasis on instabilities and equilibrium geometry) of a naturally curved rod hanging under its own weight, and is closely related to the material of Chapter 3. 1 2 Institut Jean le Rond d’Alembert, CNRS and Université Paris 6, France Institut Jean le Rond d’Alembert, Université Paris 6, France 207 F G H I 2 cm Vertical Elevation of Tip, h [m] ity, geometry, and gravity. Our approach is complementary to [1], where collisions within an ensemble of hair were treated using a statistical mechanics approach to describe the overall bulk shape of a ponytail. Taking an alternative point of departure, we identify the transitions between planar and nonplanar shapes for an analogue of a single curly and describe the non-planar shapes in detail. In our experiments, we custom fabricate rods by injecting Vinylpolysiloxane (VPS) into a flexible Polyvinyl chloride (PVC) tube, whose inner diameter sets the radius of the rod, r = 1.55mm. The PVC tube is wound around a cylindrical object (or laid straight), which sets a constant radius (or infinite) of curvature on the rod upon subsequent curing and demolding [9]. Our fabrication procedure allows for the precise control of the natural curvature in the range 0 < κn [m−1 ] < 65, a parameter that we vary systematically. The Young’s modulus of the elastomer is measured to be E = 1290 ± 12kPa, the Poisson ratio is ν ≈ 0.5 and the volumetric mass is ρ = 1200kg/m3 . Each experimental test consists of mounting a single rod with suspended length in the range 1 < L [cm] < 20 onto a clamp that is aligned vertically. The rod is then suspended under its own weight and allowed to reach static equilibrium, as shown in Fig. 1 (green configurations). Threedimensional reconstruction of the rods are produced by taking digital images from two perpendicularly directions and performing image processing to obtain their centerlines. We also perform numerical simulations, representative examples of which are presented in Fig. 1 (red configurations), where all parameters match those of the experiments. Good agreement is found throughout between the two. The simulations compute the equilibria of an inextensible three-dimensional Elastica subjected to its own distributed weight, and account for both bending and twisting. The numerical method was developed using the continuation software package MANlab [10] and is described in detail in [9]. Our first quantitative test is provided by comparing the experimentally measured and simulated vertical elevation of the tip, h, between the clamp and the free end of the rod. In Fig. 2 we plot h as a function of the total arc length, L, for three values of the natural curvature, κn . Quantitative agreement is found between experiments (data points) and simulations (solid lines). For the two lowest values of κn = 16.6 and 38 m−1 , the configurations are planar for all lengths tested and h decreases monotonically with L. For κn = 56.2m−1 , however, planar shapes are observed for L . 0.1 m, see Fig. 2F,G, but non- 0 −0.05 1 −0.15 −0.2 s 1 −0.1 0 Experiments Simulations 0.05 0.1 0.15 Suspended Length, L [m] 0.2 FIG. 2. Vertical elevation of the tip, h, vs. arc length of the rod, L, for three different natural curvatures, κn = (16.6, 38.0, 56.2) m−1 ; experiments (circles) and simulations (solid lines). For κn = 56.2m−1 , the configurations F and G are planar while the configurations H and I are non-planar. planar ones are observed for L & 0.1 m, see Fig. 2H,I. With the aim of rationalizing the behavior observed in both the experiments and simulations, we use an inextensible rod model with natural curvature [11]. All lengths are rescaled by the natural radius of curvature, κ−1 n . For example, s = sκn denotes the rescaled arc length, 0 ≤ s ≤ L, with its origin at the free end, s = 0. The dimensionless length, L = κn L, offers a measure of the rod’s curliness. All energies are rescaled by B κn , where B = E I and I = π r4 /4 are the bending stiffness and area moment of inertia of the rod, respectively. The configurations are defined in terms of the position of its centerline, r(s), and an orthonormal director basis, (d1 (s), d2 (s), d3 (s)), subjected to the condition r0 = d3 , with primes denoting derivation with respect to the rescaled arc length, s. The Cartesian basis, ei , is chosen such that the clamping condition writes r(L) = 0 and (d1 , d2 , d3 )s=L = (ey , −ex , ez ). The material curvatures, κ1 and κ2 , and twist, κ3 , are defined by κi = 12 ijk d0j · dk , where ijk is the skew-symmetric permutation tensor, ijk = (ei × ej ) · ek . The total energy of the rod is then written as, Z L 1 2 2 2 (κ1 − 1) + κ2 + C κ3 − w s cos β ds, E= 2 0 (1) where C is the ratio between the twisting and bending moduli and w = Bwκ3 is the dimensionless n weight. The weight per unit length for a rod with circular cross section is w = ρ π r2 g in physical 209 structions and the numerical configurations, finding good agreement between the two. The rod is straight near the clamp, β ≈ 0, while β increases in an oscillatory manner towards the free end at s = 0. To analyze these localized shapes, we first assume that β(s) varies slowly with s, implying that the squared derivative in Eq. (2) can be neglected. We refer to this as the Local Helix approximation (LH). The minimum energy is obtained by locally optimizing f with respect to β: ∂f /∂β = 0. We recover the equation for the helical solutions of a spring subjected to constant tension [8]. It is known that the solution β of this implicit equation undergoes a pitchfork (symmetry-breaking) bifurcation as w s is varied (purple curve in Fig. 4). The straight, vertical configuration, β = 0, is always an extremum of f , but it is unstable beyond s∗LH = (w C)−1 , −1 where ∂ 2 f /∂β 2 = w s − C becomes negative. For ∗ s ≥ sLH , the upper part of the rod is subjected to a sufficiently large tension due to the weight of the portion underneath, causing it to remain vertical. On the other hand, for 0 ≤ s ≤ s∗LH , the tension is low and the optimum value of β is non-zero resulting in a helical configuration. This prediction captures the overall shape shape of the rod (purple curve in Fig. 4): the LH approximation agrees qualitatively with simulations and experiments even if it does not work well near the transition point, s = s∗LH , nor near the free end, s = 0. by previous analyses of helices subjected to constant tension [8], we analyze these shapes assuming that the director d2 is perpendicular to the applied force (weight), such that d2 · ez ≈ 0 (this approximation is justified in the Supplementary Information for a slowly varying tension). These configurations can be parameterized by the two Euler angles, β(s) and γ(s) (shown schematically in Fig. 4), as: d1 = cos β (− sin γ ex + cos γ ey ) + sin β ez , d2 = − cos γ ex − sin γ ey and d3 = − sin β (− sin γ ex + cos γ ey )+cos β ez . When the corresponding strains, κ1 = γ 0 sin β, κ2 = −β 0 and κ3 = γ 0 cos β, are inserted into the expression for the total energy in Eq. (1), we find that E, depends on γ 0 but not on γ, as a consequence of the cylindrical invariance about ez . Optimizing the resulting E with respect to γ 0 n sin β and, after eliminating γ 0 , yields γ 0 = sin2 κβ+C cos2 β we obtain a reduced expression for the energy of helical shapes, E 3D = Z 0 L 1 0 2 f (w s, β(s)) + β (s) ds, 2 (2) Dimensionless arc length, s Dimensionless arc−length, s −1 2 where f (u, β) = 12 1 + tanC β − u cos β. The equilibrium configurations are stationary points of this energy with respect to β(s). To compute them, we first introduce a local helix approximation, which we later refine by an inner layer theory. In Fig. 4 we quantify a representative example of a localized curl in the limit of L 1 by measuring β(s) from both the 3D experimental recon- In the vicinity of the transition point s = s∗LH , the LH approximation fails because β varies quickly (see Fig. 4). Taking an alternative approach to the LH above, we study this region using an Inner Layer (IL) approximation. The derivative β 0 (s) is now restored in Eq. (2), and f is expanded near 1 f4 β 4 , s∗LH for small β as f ≈ f0 + 21 f2 β 2 + 24 ∗ 2 2 where f2 = ∂ f /∂β = w (s − s ), and f4 = ∂ 4 f /∂β 4 = 3 4−32C . Dropping terms that are inC dependent of β, we then have to minimize the func R w (s−s∗ ) 2 f4 4 1 0 2 tional ds within the β + 24 β + 2 β 2 15 15 10 10 55 00 00 ∗ inner layer. By the change of variable S = ws−s −1/3 and q f4 −1/3 unknown B(S) = β(s), the above func12 w R 2 1 tional be rewritten as 2 (S B 2 +B 4 +B 0 ) dS. The Euler-Lagrange condition of optimum yields the second Painlevé equation, B 00 (S) = S B(S) + 2 B 3 (S). Interestingly, this equation arises in domains such as nonlinear optics, Bose-Einstein condensation and random matrix theory [12]. It has a unique solution BHML (S) connecting the symmetric solutionpB → 0 for S → +∞, to the bifurcated solution B ∼ −S/2 for S → −∞, known as the Hastings-McLeod solution [13]. In terms of the original variables, the Experiments Simulation LH Approximation IL Theory, Eq (3) 0.7854 Angle from vertical, vertical, β [rad] Angle from β [rad] 1.5708 FIG. 4. A localized helical configuration as quantified by the angle from vertical, β(s), for L = 19.8 and w = 0.12. Experimental and numerical results are compared to the predictions from the Local Helix (LH) and Inner Layer (IL) approximations described in the text. The arrows show the predicted position of the transition point for each approximation. 211 solution reads, 2 C w1/3 βIL (s) = p BHML 4 − 3C s − s∗LH w−1/3 . [1] R. Goldstein, P. Warren, and R. Ball, Phys. Rev. Lett. 108, 78101 (2012). [2] K. Ward, F. Bertails, T. Kim, S. Marschner, M. Cani, and M. Lin, IEEE Trans. Vis. Comput. Graphics 13, 213 (2007). [3] R. A. Mehta and J. D. Kahn, J. Mol. Biol. 294, 67 (1999); L. M. Edelman, R. Cheong, and J. D. Kahn, Biophys. J. 84, 1131 (2003); M. A. Morgan, K. Okamoto, J. D. Kahn, and D. S. English, ibid. 89, 2588 (2005). [4] M. Raugh, Some geometry problems suggested by the shapes of tendrils (Ph.D. Thesis, Stanford University, 1978); A. Goriely and M. Tabor, Phys. Rev. Lett. 80, 1564 (1998). [5] E. Zajac, J. Appl. Mech. 29, 136 (1962); T. Yabuta, Bulletin of JSME 27, 1821 (1984). [6] G. Kirchhoff, Über das Gleichgewicht und die Bewegung einer elastischen Scheibe (1850); J. Coyne, IEEE J. Oceanic Eng. 15, 72 (1990). [7] A. Goriely and M. Tabor, Proc. R. Soc. London, Ser. A 453, 2583 (1997); A. Champneys, G. Van der Heijden, and J. Thompson, Philos. Trans. R. Soc. London, Ser. A 355, 2151 (1997). [8] T. McMillen, A. Goriely, et al., J. Nonlinear Sci. 12, 241 (2002); N. Chouaieb, A. Goriely, and J. Maddocks, PNAS 103, 9398 (2006). [9] A. Lazarus, J. Miller, and P. Reis, J. Mech. Phys. Solids 61, 1712 (2013); A. Lazarus, J. Miller, M. Metlitz, and P. Reis, Soft Matter (2013). [10] S. Karkar, R. Arquier, and B. Cochelin, User Guide MANLAB 2.0 (2011). [11] B. Audoly and Y. Pomeau, Elasticity and geometry: from hair curls to the non-linear response of shells (Oxford University Press Oxford and New York, 2010). [12] P. A. Clarkson, J. Comput. Appl. Math. 153, 127 (2003); J. A. Giannini and R. I. Joseph, Phys. Lett. A 141, 417 (1989); T. T. Wu, Phys. Rev. A 58, 1465 (1998); C. A. Tracy and H. Widom, Comm. Math. Phys. 159, 151 (1994). [13] S. Hastings and J. McLeod, Arch. Ration. Mech. Anal. 73, 31 (1980). (3) This inner layer solution successfully describes the smooth transition between the helical and straight portions of rod near s∗LH , as shown in Fig. 4. Returning to the phase diagram of Fig. 3, we can now predict the transition from the localized to global helical configurations. With the same localization criterion used above, this phase boundary is expected to occur for βIL (.95 L) = 1.5◦ (thick grey curve in Fig. 3): this is in excellent agreement with the numerical and experimental results when the inner layer is indeed small (L 1, w 1). This, combined with our results above for the 2D-to-3D transition where a planar configuration becomes unstable, completes our rationalization of the phase diagram of Fig. 3. Beyond a predictive description of the aesthetics of curly hair, our results can be directly applicable to a variety of engineering systems such as naturally curved fibers, wires, cables and pipes. All of these rodlike structures are often manufactured, stored and transported in spooled configurations, thereby imparting an intrinsic curvature, which can now be readily quantified using our framework as an inverse problem. The authors wish to thank Y. Pomeau for arousing our interest in hair mechanics, and B. Berry for her help with preliminary experiments. We are grateful for financial support from the National Science Foundation (CMMI1129894), Schlumberger, and the MIT-France Program. AL acknowledges funding from a Battelle-MIT postdoctoral fellowship. ∗ preis@mit.edu 212 Bibliography [1] N. Wicks, B. Wardle, and D. Pafitis, “Horizontal cylinder-in-cylinder buckling under compression and torsion: Review and application to composite drill pipe,” International Journal of Mechanical Sciences, vol. 50, no. 3, pp. 538–549, 2008. [2] S. Goyal, N. Perkins, and C. L. Lee, “Non-linear dynamic intertwining of rods with self-contact,” International Journal of Non-Linear Mechanics, vol. 43, no. 1, pp. 65–73, 2008. [3] F. Bertails, B. Audoly, B. Querleux, F. Leroy, J. Lévêque, M. Cani et al., “Predicting natural hair shapes by solving the statics of flexible rods,” Eurographics Short Papers, 2005. [4] A. Goriely and M. Tabor, “Spontaneous helix hand reversal and tendril perversion in climbing plants,” Physical Review Letters, vol. 80, no. 7, pp. 1564–1567, 1998. [5] Y. Sun, W. M. Choi, H. Jiang, Y. Y. Huang, and J. A. Rogers, “Controlled buckling of semiconductor nanoribbons for stretchable electronics,” Nature Nanotechnology, vol. 1, no. 3, pp. 201–207, 2006. [6] K. Son, J. S. Guasto, and R. Stocker, “Bacteria can exploit a flagellar buckling instability to change direction,” Nature Physics, vol. 9, no. 8, pp. 494–498, 2013. [7] N. Geblinger, A. Ismach, and E. Joselevich, “Self-organized nanotube serpentines,” Nature Nanotechnology, vol. 3, no. 4, pp. 195–200, 2008. [8] N. Clauvelin, B. Audoly, and S. Neukirch, “Analytical results for the plectonemic response of supercoiled dna,” Journal of Computer-Aided Materials Design, vol. 14, pp. 95–101, 2007. [9] S. Timoshenko and J. Gere, Theory of Elasticty Stability, 2nd ed. Dover, 1961. [10] L. Euler, “Methodus inveniendi lineas curvas maximi minimive proprietate gaudentes, sive solutio problematis isoperimetrici latissimo sensu accepti,” E065, Additamentum 1, 1744. [11] R. Levien, “The elastica: a mathematical history,” Electrical Engineering and Computer Sciences University of California at Berkeley, Tech. Rep., 2008. 213 [12] G. Kirchhoff, “Ueber das gleichgewicht und die bewegung eines unendlich dünnen elastischen stabes.” Journal für die Reine und Angewandte Mathematik, vol. 56, pp. 285–313, 1859. [13] A. Clebsch, Theorie der elasticitat fester korper. BG Teubner, 1862. [14] A. Love, A treatise on the mathematical theory of elasticity, 1st ed. Cambridge University Press, 1892. [15] J. L. Lagrange, Mécanique analytique. Mallet-Bachelier, 1853, vol. 1. [16] M. Nizette and A. Goriely, “Towards a classification of euler–kirchhoff filaments,” Journal of Mathematical Physics, vol. 40, p. 2830, 1999. [17] R. A. Mehta and J. D. Kahn, “Designed hyperstable lac repressor· dna loop topologies suggest alternative loop geometries,” Journal of Molecular Biology, vol. 294, no. 1, pp. 67–77, 1999. [18] L. M. Edelman, R. Cheong, and J. D. Kahn, “Fluorescence resonance energy transfer over 130 basepairs in hyperstable lac repressor-dna loops,” Biophysical Journal, vol. 84, no. 2, pp. 1131–1145, 2003. [19] M. A. Morgan, K. Okamoto, J. D. Kahn, and D. S. English, “Single-molecule spectroscopic determination of lac repressor-dna loop conformation,” Biophysical Journal, vol. 89, no. 4, pp. 2588–2596, 2005. [20] Y. Biton, B. Coleman, and D. Swigon, “On bifurcations of equilibria of intrinsically curved, electrically charged, rod-like structures that model dna molecules in solution,” Journal of Elasticity, vol. 87, no. 2, pp. 187–210, 2007. [21] M. Raugh, Some geometry problems suggested by the shapes of tendrils. Dept. of Mathematics, 1979. [22] J. Keller, “Tendril shape and lichen growth,” Lect. Math. Life Sci, vol. 30, pp. 257–274, 1980. [23] W. Silk, “On the curving and twining of stems,” Environmental and Experimental Botany, vol. 29, no. 1, pp. 95–109, 1989. [24] J. Thwaites, “10—the mechanics of friction-twisting,” Journal of the Textile Institute, vol. 61, no. 3, pp. 116–138, 1970. [25] E. Zajac, “Stability of two planar loop elasticas,” Journal of Applied Mechanics, vol. 29, p. 136, 1962. [26] T. Yabuta, “Submarine cable kink analysis,” Bulletin of JSME, vol. 27, no. 231, pp. 1821–1828, 1984. [27] J. Coyne, “Analysis of the formation and elimination of loops in twisted cable,” IEEE Journal of Oceanic Engineering, vol. 15, no. 2, pp. 72–83, 1990. 214 [28] K. Newman and D. Newburn, “Coiled-tubing-life modeling,” in SPE Annual Technical Conference and Exhibition, 1991. [29] B. Audoly and Y. Pomeau, Elasticity and geometry: from hair curls to the non-linear response of shells. Oxford University Press Oxford and New York, 2010. [30] F. Liu and N. Center, Kink formation and rotational response of single and multistrand electromechanical cables. Defense Technical Information Center, 1975. [31] D. Swigon, “Configurations with self-contact in the theory of the elastic rod model for dna,” Ph.D. dissertation, Rutgers University, 1999. [32] N. Clauvelin, B. Audoly, and S. Neukirch, “Elasticity and electrostatics of plectonemic dna,” Biophysical Journal, vol. 96, no. 9, pp. 3716–3723, 2009. [33] A. Balaeff, L. Mahadevan, and K. Schulten, “Modeling dna loops using the theory of elasticity,” Physical Review E, vol. 73, no. 3, p. 031919, 2006. [34] B. D. Coleman and D. Swigon, “Theory of self–contact in kirchhoff rods with applications to supercoiling of knotted and unknotted dna plasmids,” Philosophical Transactions of the Royal Society of London. Series A: Mathematical, Physical and Engineering Sciences, vol. 362, no. 1820, pp. 1281–1299, 2004. [35] S. Goyal, “A dynamic rod model to simulate mechanics of cables and dna,” Ph.D. dissertation, The University of Michigan, 2006. [36] ABB, “The norned hvdc transmission link,” Brochure, 2013. [37] J. Samuel and S. Sinha, “Tops and writhing dna,” Journal of Statistical Physics, vol. 143, no. 2, pp. 399–412, 2011. [38] Y. Shi and J. Hearst, “The kirchhoff elastic rod, the nonlinear schrödinger equation, and dna supercoiling,” The Journal of Chemical Physics, vol. 101, p. 5186, 1994. [39] D. L. Nelson, A. L. Lehninger, and M. M. Cox, Lehninger principles of biochemistry. Macmillan, 2008. [40] D. A. Koster, V. Croquette, C. Dekker, S. Shuman, and N. H. Dekker, “Friction and torque govern the relaxation of dna supercoils by eukaryotic topoisomerase ib,” Nature, vol. 434, no. 7033, pp. 671–674, 2005. [41] A. Revyakin, R. H. Ebright, and T. R. Strick, “Promoter unwinding and promoter clearance by rna polymerase: detection by single-molecule dna nanomanipulation,” Proceedings of the National Academy of Sciences of the United States of America, vol. 101, no. 14, pp. 4776–4780, 2004. 215 [42] G. Van der Heijden, S. Neukirch, V. Goss, and J. Thompson, “Instability and self-contact phenomena in the writhing of clamped rods,” International Journal of Mechanical Sciences, vol. 45, no. 1, pp. 161–196, 2003. [43] S. Neukirch, G. Van Der Heijden, and J. Thompson, “Writhing instabilities of twisted rods: from infinite to finite length,” Journal of the Mechanics and Physics of Solids, vol. 50, no. 6, pp. 1175–1191, 2002. [44] A. Champneys, G. Van der Heijden, and J. Thompson, “Spatially complex localization after one-twist-per-wave equilibria in twisted circular rods with initial curvature,” Philosophical Transactions of the Royal Society of London, Series A, vol. 355, no. 1732, pp. 2151–2174, 1997. [45] T. McMillen, A. Goriely et al., “Tendril perversion in intrinsically curved rods,” Journal of Nonlinear Science, vol. 12, no. 3, pp. 241–281, 2002. [46] S. S. Antman, Nonlinear problems of elasticity. Springer, 2005, vol. 107. [47] G. Van der Heijden and J. Thompson, “Helical and localised buckling in twisted rods: a unified analysis of the symmetric case,” Nonlinear Dynamics, vol. 21, no. 1, pp. 71–99, 2000. [48] J.-C. Van der Meer, The hamiltonian Hopf bifurcation. Springer, 1985. [49] G. Iooss and M. Peroueme, “Perturbed homoclinic solutions in reversible 1:1 resonance vector fields,” Journal of Differential Equations, vol. 102, no. 1, pp. 62–88, 1993. [50] J. Thompson and A. Champneys, “From helix to localized writhing in the torsional post-buckling of elastic rods,” Proceedings Of The Royal Society Of London Series A-Mathematical Physical and Engineering Sciences, vol. 452, no. 1944, pp. 117–138, 1996. [51] S. S. Antman and K. B. Jordan, “5.—qualitative aspects of the spatial deformation of non-linearly elastic rods.” Proceedings of the Royal Society of Edinburgh: Section A Mathematics, vol. 73, pp. 85–105, 1975. [52] A. Champneys and J. Thompson, “A multiplicity of localized buckling modes for twisted rod equations,” Proceedings of the Royal Society of London. Series A: Mathematical, Physical and Engineering Sciences, vol. 452, no. 1954, pp. 2467–2491, 1996. [53] A. Greenhill, “On the strength of shafting when exposed both to torsion and to end thrust,” Proceedings of the Institution of Mechanical Engineers, vol. 34, no. 1, pp. 182–225, 1883. [54] J. H. Maddocks, “Stability of nonlinearly elastic rods,” Archive for Rational Mechanics and Analysis, vol. 85, no. 4, pp. 311–354, 1984. 216 [55] M. Born, Untersuchungen über die stabilität der elastischen Linie in Ebene und Raum: Unter verschiedenen Grenzbedingungen. Dieterich, 1906. [56] V. Goss, G. van der Heijden, J. Thompson, and S. Neukirch, “Experiments on snap buckling, hysteresis and loop formation in twisted rods,” Experimental Mechanics, vol. 45, no. 2, pp. 101–111, 2005. [57] Y. Miyazaki and K. Kondo, “Analytical solution of spatial elastica and its application to kinking problem,” International Journal of Solids and Structures, vol. 34, no. 27, pp. 3619–3636, 1997. [58] K. Kovari, “Räumliche verzweigungsprobleme des dünnen elastischen stabes mit endlichen verformungen,” Archive of Applied Mechanics, vol. 37, no. 6, pp. 393–416, 1969. [59] O. Gonzalez and J. H. Maddocks, “Global curvature, thickness, and the ideal shapes of knots,” Proceedings of the National Academy of Sciences, vol. 96, no. 9, pp. 4769–4773, 1999. [60] A. Goriely and M. Tabor, “New amplitude equations for thin elastic rods,” Physical Review Letters, vol. 77, no. 17, pp. 3537–3540, 1996. [61] ——, “Nonlinear dynamics of filaments i. dynamical instabilities,” Physica D: Nonlinear Phenomena, vol. 105, no. 1, pp. 20–44, 1997. [62] ——, “Nonlinear dynamics of filaments ii. nonlinear analysis,” Physica D: Nonlinear Phenomena, vol. 105, no. 1, pp. 45–61, 1997. [63] ——, “Nonlinear dynamics of filaments. iv spontaneous looping of twisted elastic rods,” Proceedings of the Royal Society of London, Series A: Mathematical, Physical and Engineering Sciences, vol. 454, no. 1980, pp. 3183–3202, 1998. [64] ——, “Nonlinear dynamics of filaments. iii. instabilities of helical rods,” Proceedings of the Royal Society of London, Series A: Mathematical, Physical and Engineering Sciences, vol. 453, no. 1967, pp. 2583–2601, 1997. [65] N. Chouaieb, A. Goriely, and J. Maddocks, “Helices,” Proceedings of the National Academy of Sciences, vol. 103, no. 25, pp. 9398–9403, 2006. [66] N. Chouaieb and J. Maddocks, “Kirchhoff’s problem of helical equilibria of uniform rods,” Journal of Elasticity, vol. 77, no. 3, pp. 221–247, 2004. [67] N. Chouaı̈eb, “Kirchhoff’s problem of helical solutions of uniform rods and their stability properties,” Ph.D. dissertation, École Polytechnique Fédérale de Lausanne, 2003. [68] B. Buffoni, A. Champneys, and J. Toland, “Bifurcation and coalescence of a plethora of homoclinic orbits for a hamiltonian system,” Journal of Dynamics and Differential Equations, vol. 8, no. 2, pp. 221–279, 1996. 217 [69] G. Brusco, P. Lewis, and M. Williams, “Drilling straight down,” Oilfield Review, vol. 16, no. 3, pp. 14–17, 2004. [70] L. Perneder and E. Detournay, “Anomalous behaviors of a propagating borehole,” in SPE Deepwater Drilling and Completions Conference, 2012. [71] B. Bennetzen, J. Fuller, E. Isevcan, T. Krepp, R. Meehan, N. Mohammed, J.F. Poupeau, and K. Sonowal, “Extended-reach wells,” Oilfield Review, vol. 22, no. 3, pp. 4–15, 2010. [72] [Online]. Available: http://www.glossary.oilfield.slb.com/ [73] B. Hughes, “Rig count overview and summary count,” November 2013. [74] A. Lubinski, “A study of the buckling of rotary drilling strings,” Drilling and Production Practice, pp. 178–214, 1950. [75] P. Paslay and D. Bogy, “The stability of a circular rod laterally constrained to be in contact with an inclined circular cylinder,” Journal of Applied Mechanics, vol. 31, pp. 605–610, 1964. [76] R. Dawson and P. Paslay, “Drill pipe buckling in inclined holes,” Journal of Petroleum Technology, vol. 36, no. 10, pp. 1734–1738, 1984. [77] S. Miska, W. Qiu, L. Volk, and J. Cunha, “An improved analysis of axial force along coiled tubing in inclined/horizontal wellbores,” in International Conference on Horizontal Well Technology, 1996. [78] W. Qiu, “Prediction of contact force for drill pipe/coiled tubing,” in SPE Eastern Regional Meeting, 1998. [79] A. Lubinski and W. Althouse, “Helical buckling of tubing sealed in packers,” Journal of Petroleum Technology, vol. 14, no. 6, pp. 655–670, 1962. [80] J. Cheatham Jr, “Helical postbuckling configuration of a weightless column under the action of m axial load,” Old SPE Journal, vol. 24, no. 4, pp. 467–472, 1984. [81] Y. Chen, Y. Lin, and J. Cheatham, “Tubing and casing buckling in horizontal wells (includes associated papers 21257 and 21308),” Journal of Petroleum Technology, vol. 42, no. 2, pp. 140–141, 1990. [82] G. Deli, F. Liu, and B. Xu, “An analysis of helical buckling of long tubulars in horizontal wells,” in SPE International Oil and Gas Conference and Exhibition in China, 1998. [83] J. Wu and H. Juvkam-Wold, “Study of helical buckling of pipes in horizontal wells,” in SPE Production Operations Symposium, 1993. 218 [84] S. Miska and J. Cunha, “An analysis of helical buckling of tubulars subjected to axial and torsional loading in inclined wellbores,” in SPE Production Operations Symposium, 1995. [85] J.-S. Chen and H.-C. Li, “On an elastic rod inside a slender tube under end twisting moment,” Journal of Applied Mechanics, vol. 78, no. 4, 2011. [86] R. Mitchell, “Simple frictional analysis of helical buckling of tubing,” SPE Drilling Engineering, vol. 1, no. 6, pp. 457–465, 1986. [87] G. Gao and S. Miska, “Effects of boundary conditions and friction on static buckling of pipe in a horizontal well,” SPE Journal, vol. 14, no. 4, pp. 782–796, 2009. [88] ——, “Effects of friction on post-buckling behavior and axial load transfer in a horizontal well,” SPE Journal, vol. 15, no. 4, pp. 1104–1118, 2010. [89] R. McCann and P. Suryanarayana, “Experimental study of curvature and frictional effects on buckling,” in Offshore Technology Conference, 1994. [90] P. Suryanarayana and R. McCann, “Experimental study of buckling and postbuckling of laterally constrained rods: Part 1—frictional effects,” in Presentation given at the ASME Energy-Source Technology Conference and Exhibition, New Orleans, 1994, pp. 23–27. [91] X. He and A. Kyllingstad, “Helical buckling and lock-up conditions for coiled tubing in curved wells,” SPE Drilling & Completion, vol. 10, no. 1, pp. 10–15, 1995. [92] J. Wu and H. C. JuvkamWold, “The effect of wellbore curvature on tubular buckling and lockup,” Journal of Energy Resources Technology-Transactions of The ASME, vol. 117, no. 3, pp. 214–218, 1995. [93] E. Kuru, A. Martinez, S. Miska, and W. Qiu, “The buckling behavior of pipes and its influence on the axial force transfer in directional wells,” in SPE/IADC Drilling Conference, 1999. [94] J.-S. Chen and C.-W. Li, “Planar elastica inside a curved tube with clearance,” International Journal of Solids and Structures, vol. 44, no. 18, pp. 6173–6186, 2007. [95] P. Pastusek and V. Brackin, “A model for borehole oscillations,” SPE Paper 84448, 2003. [96] V. Denoël and E. Detournay, “Eulerian formulation of constrained elastica,” International Journal of Solids and Structures, vol. 48, no. 3, pp. 625–636, 2011. 219 [97] ——, “The problem of a drillstring inside a curved borehole: a perturbed eulerian approach,” in First International Colloquium on Non-Linear Dynamics of Deep Drilling Systems, Liëge, Belgium, 2009, pp. 89–95. [98] G. Gao and S. Miska, “Dynamic buckling and snaking motion of rotating drilling pipe in a horizontal well,” SPE Journal, vol. 15, no. 3, pp. 867–877, 2010. [99] A. Acock, T. ORourke, D. Shirmboh, J. Alexander, G. Andersen, T. Kaneko, A. Venkitaraman, J. López-de Cárdenas, M. Nishi, M. Numasawa et al., “Practical approaches to sand management,” Oilfield Review, Spring, pp. 10–27, 2004. [100] A. Afghoul, S. Amaravadi, A. Boumali, J. Calmeto, J. Lima, J. Lovell, S. Tinkham, K. Zemlak, and T. Staal, “Coiled tubing: the next generation,” Oilfield Review, vol. 6, no. 4, pp. 9–23, 1994. [101] I. McCourt, T. Truslove, and J. Kubie, “On the penetration of tubular drill pipes in horizontal oil wells,” Proceedings of the Institution of Mechanical Engineers, Part C: Journal of Mechanical Engineering Science, vol. 218, no. 9, pp. 1063– 1081, 2004. [102] W. van Adrichem and K. Newman, “Validation of coiled-tubing penetration predictions in horizontal wells,” Journal of Petroleum Technology, vol. 45, no. 2, pp. 155–159, 1993. [103] K. Bhalla, “Implementing residual bend in a tubing forces model,” in SPE Annual Technical Conference and Exhibition, 1994. [104] M. Al-dhufairi, S. Al-Ghamdi, J. Noya, K. Al-Aradi, S. AlSarakbi, A. Aldossary, E. Krueger, and N. Moore, “Expanding the envelope of coiledtubing (ct) reach for stimulation of ultradeep openhole horizontal wells,” in SPE/ICoTA Coiled Tubing and Well Intervention Conference and Exhibition, 2010. [105] ——, “Merging tapered in coiled tubing (ct) and well tractor technologies to effectively stimulate extended reach open hole horizontal wells,” in SPE/ICoTA Coiled Tubing and Well Intervention Conference and Exhibition, 2010. [106] I. McCourt, T. Truslove, and J. Kubie, “Penetration of tubulars into horizontal oil wells,” Proceedings of the Institution of Mechanical Engineers, Part C: Journal of Mechanical Engineering Science, vol. 216, no. 12, pp. 1237–1245, 2002. [107] I. McCourt and J. Kubie, “Limits on the penetration of coiled tubing in horizontal oil wells: effect of the pipe geometry,” Proceedings of the Institution of Mechanical Engineers, Part C: Journal of Mechanical Engineering Science, vol. 219, no. 11, p. 1191, 2005. 220 [108] A. Zheng and A. Sarmad, “The penetration of coiled tubing with residual bend in extended-reach wells,” in SPE Annual Technical Conference and Exhibition, 2005. [109] A. Lazarus, J. Miller, and P. Reis, “Continuation of equilibria and stability of slender elastic rods using an asymptotic numerical method,” Journal of the Mechanics and Physics of Solids, vol. 61, no. 8, p. 1712, August 2013. [110] A. Lazarus, J. Miller, M. Metlitz, and P. Reis, “Contorting a heavy and naturally curved elastic rod,” Soft Matter, vol. 9, no. 34, pp. 8274–8281, September 2013. [111] (2005). [Online]. Available: http://www.8020.net [112] B. Cochelin, “A path-following technique via an asymptotic-numerical method,” Computers & Structures, vol. 53, no. 5, pp. 1181–1192, 1994. [113] S. Karkar, R. Arquier, A. Lazarus, O. Thomas, C. Vergez, and B. Cochelin, “Manlab: An interactive path-following and bifurcation analysis software,” 2010. [114] E. J. Doedel, “Auto: A program for the automatic bifurcation analysis of autonomous systems,” Congress Numerantium, vol. 30, pp. 265–284, 1981. [115] D. C. Lay, “Linear algebra and its applications,” Addison Wesesley,, 2003. [116] S. Hastings and J. McLeod, “A boundary value problem associated with the second painlevé transcendent and the korteweg-de vries equation,” Archive for Rational Mechanics and Analysis, vol. 73, no. 1, pp. 31–51, 1980. [117] N. Devices and Components, “Nitinol se 508 wire,” Material Data Sheet, February 2002. [118] W. S. Rasband, ImageJ, US National Institutes of Health, Bethesda, Maryland, USA, 1997. [119] E. Kausel, “Advanced structural dynamics (mechanical vibration),” Fall 2010, lecture notes, MIT. [120] MDrive Linear Actuator, Rev. 080912 ed., Schneider Electric Motion USA, 2012. [121] Futek, “Futek model lrm200 (l2358),” Specification Sheet, 2013. [122] J. Pabon, N. Wicks, Y. Chang, and R. Harmer, “Modeling transient vibrations while drilling using a finite rigid body approach,” Non-linear Dynamics of Deep Drilling Systems, p. 83, 2009. [123] J. Pabon, N. Wicks, Y. Chang, C. Chapman, and V. Singh, “Modeling the transient behavior of bha/drill string while drilling,” Patent, September 2011. 221 [124] R. J. Roark and W. C. Young, Formulas for stress and strain. 1975. McGraw-Hill, [125] M. A. Crisfield, Nonlinear finite element analysis of solids and structures. Volume 1: Essentials. Wiley, New York, NY (United States), 1991. [126] N. A. Corp., “Non-motorized linear air bearing slides: Rab series,” Data Sheet, 2013. [127] [Online]. Available: http://www.thingiverse.com/thing:964 [128] I. Intelligent Motion Systems, MDrive 14 Plus Motion Control Integrated Motor and Driver, 011909th ed., Schneider Electric, January 2009. [129] J. Gere, Mechanics of Materials, 6th ed. Cengage Learning, 2006. [130] A. C. Palmer and R. A. King, Subsea Pipeline Engineering. PennWell Books, 2008. [131] R. Lagrange and D. Averbuch, “Solution methods for the growth of a repeating imperfection in the line of a strut on a nonlinear foundation,” International Journal of Mechanical Sciences, 2012. [132] W. Qiu, S. Miska, and L. Volk, “Effect of coiled tubing (ct) initial configuration on buckling/bending behavior in straight deviated wells,” in Latin American and Caribbean Petroleum Engineering Conference, 1997. [133] T. M. Shop, “Mini smartshaker with integrated power amplifier,” Data Sheet, 2010. [134] PCB, “Model number 353b34 accelerometer, icp (spec. number 353-2340-80),” Specification Sheet, July 2007. [135] ——, “Model number 353b32 accelerometer, icp (spec. number 353-2320-80),” Data Sheet, July 2007. [136] B and K. Precision, “Dds function generators: Models 4007b and 4013b,” Data Sheet, 2012. [137] PCB, “Model number 482c05 four-channel, icp sensor signal conditioner (spec. number: 35061),” Data Sheet, February 2013. [138] E. Fermi, “On the origin of the cosmic radiation,” Physical Review, vol. 75, pp. 1169–1174, Apr 1949. [139] S. McNamara and W. Young, “Inelastic collapse in two dimensions,” Physical Review E, vol. 50, no. 1, p. R28, 1994. [140] T. Gilet and J. W. Bush, “Chaotic bouncing of a droplet on a soap film,” Physical Review Letters, vol. 102, no. 1, p. 014501, 2009. 222 [141] R. Everson, “Chaotic dynamics of a bouncing ball,” Physica D: Nonlinear Phenomena, vol. 19, no. 3, pp. 355–383, 1986. [142] A. Mehta and J. Luck, “Novel temporal behavior of a nonlinear dynamical system: The completely inelastic bouncing ball,” Physical Review Letters, vol. 65, no. 4, p. 393, 1990. [143] J. Luck and A. Mehta, “Bouncing ball with a finite restitution: chattering, locking, and chaos,” Physical Review E, vol. 48, no. 5, p. 3988, 1993. [144] T. Gilet, N. Vandewalle, and S. Dorbolo, “Completely inelastic ball,” Physical Review E, vol. 79, no. 5, p. 055201, 2009. [145] D. Vella, A. Boudaoud, and M. Adda-Bedia, “Statics and inertial dynamics of a ruck in a rug,” Physical Review Letters, vol. 103, no. 17, p. 174301, 2009. [146] J. M. Kolinski, P. Aussillous, and L. Mahadevan, “Shape and motion of a ruck in a rug,” Physical Review Letters, vol. 103, no. 17, p. 174302, 2009. [147] N. Adami, A. Delbos, B. Roman, J. Bico, and H. Caps, “Elasto-capillary collapse of floating structures-non-linear response of elastic structures under capillary forces,” arXiv preprint arXiv:1310.0329, 2013. 223