MECH3300/3302 COMPUTER TUTORIAL 2 – Modelling 2D problems 1. Plane strain crack model This model will be constructed in STRAND7. The dimensions are those of the test specimen used in the experiment in fracture mechanics. STRAND7 can import CAD geometry, but cannot create geometry, hence for regular simple geometry, it is best to create a coarse mesh that captures the geometry, that can then be refined (subdivided). First enter x,y coordinates of the nodes of the coarse mesh below. 205 mm 148 mm 118 mm 88 mm 68 mm 48 mm 36 mm 26 mm y y fixed – beyond crack tip x and y fixed x Pick Quad4 elements in Create, Element and join the nodes to create the coarse mesh. The mesh can then be refined using Tools, Subdivide, to create a finer mesh, such as that above. The mesh shown is too fine for the demonstration version of the program. Any location to be treated as a crack tip is further refined using the mesh grading tool (Grade Plates and Bricks) on the same tool menu. To replace an element by the subdivision selected, click near the two sides of the element to be given the finer mesh – the sides shown cross-hatched on the menu in STRAND7. The mesh needs to be restrained to stop rigid body motion. First select Global, Load and Freedom Cases and pick the 2D plane case under Freedom Cases – note that only x and y displacements are used as unknowns. We wish to fix displacement normal to the plane of symmetry ahead of the crack - the lower edge of the mesh, which is a half-model. This will be revised to consider crack lengths from the load of 18, 30, 50, 70, 100 and 130 mm – the shortest of these is not used in the practical. Use Attribute, Node, Restraint to achieve this. Note one node must be fully fixed, to prevent sliding x-wards. The load can be applied most simply to a single node. This is unrealistic locally, but should still give a reasonable deflection. We need a unit load to find the compliance or flexibility. The Property setting is Plate property, then 2D Plane Strain. The properties of Perspex are needed. Try Young’s modulus 2.6GPa and Poisson’s ratio 0.26. Solve and note the displacement at the load for the various crack lengths, changing the restraints each time. We will compare these values to the expectations of beam theory. Also plot the xdirection stress (Results, Contour, Stress). Use the global XX stress, as this is definitely in the X direction. The local xx stress is in a direction determined from the first node picked to the second node picked, when each element of the coarse mesh was created. An advantage of the “combined” stresses, that are the principal stresses or functions of them, are that they are independent of the choice of coordinates. Have a look at some of these as well. Note that Von Mises stress, and Tresca stress – both numbers to compare to yield strength - are always positive and do not distinguish tension from compression. Examine the transverse shear stress. In theory for a cantilever, this is distributed quadratically through the depth of the beam, peaking a mid-depth, the peak value being Pc2/(2I), where c is half beam depth. Note, as the stress is really discontinuous at element boundaries, the shear stress does not fall to zero at the surface, as it should. To test the plane strain assumption, rerun the two longest longest crack lengths with the actual 5.42 mm thickness of the specimen and a Plane Stress setting in the Property menu. To see how the accuracy is affected by an increase in the polynomial order in each element (p-refinement), construct the mesh below with Quad8 elements. If in doubt about the state of the model at any time, run Tools, Clean. This removes duplicate or free nodes or duplicate elements, and checks the connections between elements. The hole is created using the Tools, Grade Plates and Bricks menu, using the default setting of 0.5 on radius to set the size of the hole. Note the unit load is now spread over three nodes for greater realism. The mid-side nodes now present also need restraining ahead of the crack tip. Solve for the different crack lengths again. To look at the effect of using more, smaller linear elements (h-refinement), construct a mesh with six elements through the depth and appropriate numbers along the crack specimen. Estimating fracture mechanics parameters The parameters G and KI both measure the severity of the stress conditions leading to crack growth. G (strain energy released per crack surface area formed) can be estimated from the change in compliance (or flexibility) with crack length a. Using Matlab or some other tool, plot deflection at the unit load (twice that of this half model) versus crack length, measured from the load. Do this for both linear meshes and the quadratic one. Superimpose the deflection predicted by beam theory: (2PL3/(3EI)) for two cantilevers. We could either think of P as unit load per width we applied, times the width of 5.42 mm, or think of I here as the value for unit width, leaving P as unity, as that is the assumption in a plane strain model. The former is more appropriate here. Also find the deflection for the plane stress case of the coarser linear mesh for the two longest cracks, and compare this to beam theory. It is worth looking at a log-log plot of the deflection v crack length. Beam theory makes this a straight line with a slope of 3, due to the L3 term. What are the finite element estimates of this slope from the plane stress and plane strain data for the two longest crack lengths? This slope gives the change in compliance with crack length d/da. In the case of the shortest length, the depth of the beam is becoming comparable to its length. In this situation, a correction of the deflection due to shear deformation is worth considering. The extra deflection is 2Pc2L(1+)/(EI) where c is half the depth of the beam, and is Poisson’s ratio. Estimate G as a function of P G = (P2/(2B)) d/da B here is really 3 mm, not 5.42 mm, as there is a groove cut ahead of the crack to guide its growth. Ignore this for the current computational exercise. The mean stress ahead of the crack tip can be used to make an estimate of stress intensity KI that should improve as the crack tip is approached. The quantity {(x+y)/2}(2x) does in theory tend to KI, as x, the distance from the crack tip, ahead of the crack along the plane of symmetry, tends to zero, as the stress components in polar coordinates all depend on KI/(2r) f(), where r is radius from the crack tip, and f( is some function of the angle. Setting 0 and r = x gives x = y = KI/(2x). The quantity KI {(x+y)/2}(2x) can be estimated from the stresses in each element to the right of the crack tip, along the plane of symmetry. Collect the data on these stresses and note it on paper for the finer linear mesh. Then use (say) a spreadsheet to plot the above estimate of KI, extrapolating to the crack tip. The (crack opening) displacements at nodes prior to the crack tip (to the left of it) can also be used to estimate KI by extrapolation. Again use the finer linear mesh. This time, the quantity to be calculated at each node is L u y Eu y 2 KI 2 4(1 ) L(1 L ) 2a a As L tends to zero, this estimate should again improve, the change being smoother than that made from the stress, as displacement is the integral of strain. For plane strain conditions, KI = (EG/(1-2)). How do the above two estimates of KI compare to the estimate of G? 2. An axisymmetric model This model will be created in the demonstration version of Nastran for windows, with a 300 node limit. Nastran can create a geometric (CAD) model itself, but one for the purpose of meshing with finite elements. Loads and boundary conditions can be applied to the geometry, as well as to the finite element model. There are many options in this user interface. The F6 key brings up a menu that allows you to change the default settings of what you see. In addition, on the top “View bar” there are options to change the display (eg turning elements or geometry on or off, or switching from wireframe to solid appearance of a mesh). Initially scales are visible, indicating the size of the model. These can be switched off in “Tools and View Style”, “Workplane and Rulers”. Change “Rulers” to “1. Skip Rulers”. The Quick Options button is useful for turning geometry on and off. Examine the buttons on across the top of the screen and what they do. Loads, materials, properties and restraints (constraints) in Nastran all have set numbers. Sets can lie around in the data inactive, unless selected when solving. You can give a set a name to help you remember what is in it (eg the type of loading). An axisymmetric analysis involves modelling half a cross-section, each element representing a ring of material. In Nastran, the choice of elements is limited for axisymmetric analysis. We will use 6 node triangles to automatically mesh the halfsection below, because, if we do not, the elements will be changed to this type anyway. In an axisymmetric analysis, it is assumed that the deflections are just in the plane modelled (ie radial and axial), and the global X axis is taken as the radius, global Z being the axis of symmetry – here the axis of the shaft on which the flywheel spins. Hence you should rotate to the XZ plane before creating any geometry. Axis Z 5R Surface 1 7.5R Dimensions in mm Surface 3 Surface 2 30 X 50 5 Plane of symmetry 15 Note: top half modelled. 10 20 80R 90R To build the largest model possible within the 300 node limit, draw the boundary of the part of the half-section of the flywheel above its plane of symmetry as a series of curves (Geometry, Curve). Note there are many ways of constructing a circular arc (Geometry, Curve, Arc). A simple way is to first place a point at the centre of the arc, and then use the Center, Start, End option. You need to start at the end that makes the arc anticlockwise. The arcs above can be approximated to 90 degrees. Having constructed the boundary, we need to create surfaces to mesh. The mesh generation software needs simpler surfaces to do a good job, than the entire cross-section. Hence the geometric model is divided into three surfaces as indicated above. The surfaces are specified using Geometry, Boundary Surface, From Curves. Pick each curve in turn in sequence around the boundary, to define the boundary of each of the three surfaces, and hence the surface enclosed by it. As usual, it is necessary to specify a material – this time use a high strength alloy steel (in SI units) – note that E does not vary much with the alloy content of steels. To find SI units, you need to scroll down the list of materials available – the default is imperial pound force inch units. A property must be specified with Elem/Property Type set to Axisymmetric, and to Parabolic elements. Make the property list point to the material. Note Nastran expects you to create the material data first, then the property data, and then finally the elements with that property. The thickness data is redundant. Set the element default size to 4 mm (Mesh, Mesh Control). Then mesh the surface with triangles (not the default quadrilaterals) in Mesh, Geometry, Surface. The number of nodes created should be below the demostration version limit of 300. To be more precise, in the Automesh Surfaces window, set “All Triangles” and then “Midside Nodes on Geometry”. This does not in fact force the midside nodes to be exactly on the boundary, as there is also a tolerance on element distortion to be observed. It is important to check for duplicate nodes on common edges of surfaces meshed. (Tools, Check, Coincident nodes). The default tolerance needs increasing to pick them up. Use Model, Constraint Set to create a set of restraints. Name the set for later reference. Open the menu Model, Constraint, Nodal to add restraints to this set. Restraints are displayed as numbers – 1,2,3 are x,y,z displacements and 4,5,6 are the corresponding rotations. There is no need to constrain this model radially (X-wards), as radial motion is not a rigid-body motion – it stretches each ring element circumferentially. Nodes on the central plane of symmetry must be fixed in the axial (Z) direction, making axial displacements relative to this plane. Only TX and TZ translations are needed at each node to write the equations that are solved. To load the model, we wish to apply inertia loading corresponding to spinning at 10000 rev min-1. First create a load set (Model, Load, Set), then apply a body force (Model, Load, Body). Make Velocity active and set Wz, the angular velocity about the Z axis in rev s-1. This will cause inertia forces to be found by integration over each element. Check the model: Tools, Check, Coincident Nodes. If this is not done, the flywheel model will remain in three separate sections – one for each surface meshed. Run the analysis (File, Analyse). The right hand menu – an alternative to the top menu, is most convenient for examining results. Click on the bottom icon to open the postprocessing menu – and to select the latest output. The deflected shape can be switched on. The scale may need to be set in the View Options (F6) menu. The Post Data {} icon is used to select what quantities to plot as contours. Of interest in the output, is the maximum principal stress, to assess if fracture could occur, with an assumed flaw size that would not be easily detected, and the maximum Von Mises stress to check for yielding. Locate these maxima. Also find the maximum radial deflection at the hub. This is of interest in assessing what shrink fit is needed to avoid the flywheel coming loose on its shaft. Advise on the safety of the flywheel and on a suitable shrink fit. What would be the effect on the stressing of the flywheel of a shrink fit? In order to make a rough check the accuracy of the stressing of the flywheel, the rim can be analysed as an unsupported spinning ring. This will exaggerate the stresses, as the web acts to limit radial expansion in reality. The expression for circumferential (or hoop) stress at radius r in a spinning ring of outside radius b and inside radius a is given by 3 a 2 b 2 1 3 2 2 (b 2 a 2 2 r ) 8 3 r again is Poisson’s ratio. is density. is angular velocity (rad s-1). V1 L1 C1 Y X Z What to submit (1) Crack specimens Submit the plots used to estimate compliance. Report the two estimates of stress intensity, and the comparison of these to the estimate of G from change of compliance with crack length. Report on the accuracy of beam theory in this problem, and on the difference between the deflection results from the meshes of linear and quadratic elements. As well as comparing deflections, for the longest crack size compare the most compressive stress on the top surface of each of the models to that expected for a beam. Also compare the maximum transverse shear stress to its value from beam theory. Show how the stresses change with h or p-refinement. (2) Axisymmetric problem Submit the comparison of the hoop stress in the rim with that in an unsupported rim. Show the locations and magnitudes of critical stresses as requested in the discussion above. Suggest a suitable interference fit at the hub.