TESTING, MATERIAL DATA AND IDENTIFICATION OF MODEL PARAMETERS FOR TOUGHENED STRUCTURAL ADHESIVES Dominik Teutenberg1), Gerson Meschut1), Michael Schlimmer2), Ulrich Kroll3), Anton Matzenmiller3) 1) University of Paderborn, 2) WWV Consulting GmbH, 3) University of Kassel Paderborn, D-33098, Germany dominik.teutenberg@lwf.upb.de Introduction Structural adhesives are widely used in industrial applications to join components of different materials in lightweight design. Therefore, it is mandatory to provide reliable numerical simulations of the bondings in such multi-material-designed components under various loading conditions. When calculating adhesive bonds, two basic tasks arise: determining stresses in the adhesive layer in relation to external loads and other constraints, and determining stress limits that lead to functional or structural failure of the joint. Generally, a non-homogenous, multiaxial state of stress exists in the adhesive layer. Therefore, an appropriate material equation, yield stress hypothesis and failure conditions are required for a FE-analysis. Material equations and failure conditions for multiaxial loaded adhesive layers can only be determined by testing adhesively joined specimens under homogenous stress conditions. This means a great deal of experimental effort. A nearly homogenous shear stress distribution exists in the so-called double hollow cylinder sample as described in ASTM E 229 and ISO 11003-1. First known experiments with these types of adhesively bonded samples can be traced back to MÜLLER (1959) [1], KUENZI and STEVENS (1963) [2] and BOSSLER among others (1968) [3]. ENGASSER and PUCK conducted combined tensiontorsion experiments (1980) [4], and SCHLIMMER published a yield stress hypothesis for adhesively bonded joints based on the invariants of stress tensors (1982) [5]. Based on this work, the first research project in Germany [6] began with the cooperation of the automobile industry in 2000. These research projects of the research association Forschungsvereinigung Stahlanwendung (FOSTA) were founded by the AiF under the program for the promotion of joint industrial research and development (IGF) by the federal ministry of economics and technology and Stiftung Stahlanwendungsforschung. Affiliated research continued [7,8] with the goal of FE-simulating the mechanical behaviour of large adhesively bonded structures in automobile manufacturing under a variety of loads, from crash to fatigue to temperature change conditions. Experimental System The stress–strain behaviour of an adhesively bonded joint under quasi-static load can be determined on bluntly bonded cylinder samples under shear and tension load according to ISO 11003-1. When compared with the thick adherend shear specimen (ISO 11003-2), the cylinder sample can be tested under torsion load. This loading results in an almost homogenous shear stress distribution within the adhesive layer between the cylinder halves. That is why this experiment provides the best and most reproducible results for testing the mechanical behaviour of adhesive bonds. In order to compile a material equation for adhesives, a shear stress-shear strain correlation as well as a tensile stress-strain correlation is needed. In contrast to the tension test of bulk adhesive specimen, the thin adhesive layer is in a uniaxial strain state, which results in a high hydrostatic tensile stress. SCHLIMMER developed testing equipment in the 1990’s to determine the strain behaviour of thin adhesive layers in a cylinder specimen with uniaxial and combined shear-/normal stress (Figure 1a). It was possible to determine the complex mechanical behaviour of an adhesively bonded joint. The adhesive layer’s reaction to exterior load is measured using a biaxial deformation sensor (Figure 1b). Adherend deformations are eliminated by measurement software in order to measure and control tension and shear strains within the adhesive layer. Constant adhesive layer strain rates under tension and/or shear loads are controlled with a PID feedback control system. Figure 1: a) Testing machine for tension / torsion loads b) Cylinder specimen with biaxial deformation sensor [9] Results and Discussion 50 A flexibilized and toughened heat curing epoxy adhesive for car body structure is investigated: Betamate 1496V from DOW. The experimental results of tests with four different strain proportions are displayed in Figure 2: pure shear strain, pure tensile strain as well as two combined strains. The slopes of the curves within the linear elastic course are similar, regardless of the strain combination. This permits the determination of elastic material parameters for shear modulus, tensile modulus and POISSON’s ratio, whereas the yield points and maximum stresses depend on the strain combination. 40 40 40 30 30 xy [MPa] x [MPa] 20 10 Zug- / Torsionsversuche tension and shear tests d 10-3-3s-1s-1 d11 // dt dt==11·10 epoxy adhesive Klebstoff: EP 13 adhesive layer thickness: 0.2 mm Fügeteile: Stahl tensile strain: Klebschichtdicke: 0,2 mm xy / 2x = 0 Zugbeanspruchung: 20 xy / 2x = 0 combined strain: Kombinierte xy / 2x Beanspruchung = 0.5 2xx ==20,5 xy // 2 xy 10 xy / 2x = 2 0 0,00 0.00 0,25 0.25 0,50 0.50 x [-] 0 0,0 0.0 0,5 0.5 1,0 1.0 xy / 2x = Figure 2: Stress-strain behaviour of a toughened EPadhesive with uniaxial and combined shear/normal strain. [6] Figure 2 shows the results of the uniaxial and combined shear-/normal strain tube specimens’ yield stress and maximum strength and their dependence on one another. A curve drawn through these points is labelled as yield strength curve or maximum stress curve and is in general a state of stress function. 𝐹 = 𝐹(𝜎𝑥 , 𝜎𝑦 , 𝜎𝑧 , 𝜏𝑥𝑦 , 𝜏𝑦𝑧 , 𝜏𝑥𝑧 ) (1) Because the yield condition at isotropy is invariant against coordination transformation, the following applies: 𝐹(𝐽1 , 𝐽2 , 𝐽3 ) (2) 𝐽1 , 𝐽2 and 𝐽3 are the stress tensor’s invariants. The tensor’s split into deviator and spherical tensor leads to the invariants 𝐽2′ , 𝐽3′ of the stress deviator. For slightly plastic compressible materials the yield potential can be written as 𝐹(𝐽1 , 𝐽2′ , 𝐽3′ ). The invariants 𝐽1 , 𝐽2′ are determined for the experimental yield strengths and maximum strengths from Figure 2 by 𝐽1 = 1+𝑣 1−𝑣 1 1−2𝑣 2 𝜎𝑥 , 𝐽2′ = ( 3 1−𝑣 ) 𝜎𝑥 2 + 𝜏𝑥𝑦 2 Spannungsmaximum maximum stress 1/2 (J2') 20 yield strength Fließgrenze 10 0 0 25 50 75 (3a,b) by taking into consideration the prevention of transverse strain within the experiment. This leads to Figure 3. 100 125 J1 [MPa] Figure 3: Description of yield strength and maximum strength as invariants of the toughened EP. [6] Non-linear relationships between the stress tensor’s first invariant 𝐽1 and the stress deviator’s second invariant 𝐽2′ are shown for yield strength and stress maximum. These can be described using the following equation taken from [10]: 1 1 3 3 (4) 𝑎1 and 𝑎2 are free parameters to be experimentally determined, 𝜎𝐹 is the yield stress. The shear-based equivalent stress condition is obtained by redefining the yield limit as 1 1 3 3 𝜏𝐹 2 = 𝐽2′ + 𝑎1 𝜎𝐹 𝐽1 + 𝑎2 𝐽1 2 Yield condition and strength diagram 13 30 𝐹 = 𝐽2′ + 𝑎1 𝜎𝐹 𝐽1 + 𝑎2 𝐽1 2 shear strain xy / 2x = ∞ Schubbeanspruchung xy [-] [MPa] Klebstoff: EP epoxy adhesive (5) A physical explanation for this yield condition is outlined in [6]. Based on these results, a rate-dependent I1-J2plasticity theory with continuum damage is chosen to describe the inelastic mechanical behaviour of the adhesive. The constitutive equations comprise in addition a non-associated flow rule, a non-linear hardening assumption and an isotropic damage model of the KACHANOV-RABOTNOV type with plasticity driven damage evolution [06, 07, 11] to account for stiffness and strength degradation of the adhesive layer. For use in crashworthiness analysis, the rate-dependency of the yield strength is modeled with the JOHNSON-COOK approach. The material equations for toughened adhesive polymers (TAPO) are implemented into the commercial finite element (FE) software LS-DYNA [12], verified as shown below in Figure 6, and validated at various experimentally based test examples. The TAPO-model is available for large-scale simulations in research and industrial applications. With regard to the identification of the model parameters pi, the best characteristic values pi,opt may be even determined by hand or numerically optimized with the help of the commercial optimization tool LS-OPT [13] by minimizing the mean square error between the computed results from the FE-simulation and the test data for some chosen quantities of the bluntly glued, thinwalled specimen at hand. Figure 4: FE-model of two bluntly glued, thin- walled cylinders Due to the high stiffness of the bonded steel parts compared to the adhesive layer, only an excerpt is modeled for the FE-analysis. While the adhesive (blue) is represented by a single layer of 32 eight-node-solid elements along the circumference and one hexahedron in radial direction, 160 fully integrated linear bricks are used for the adherends (red). The loading is applied by a rigid ring (green) connected to the upper adherend’s free end, whereas the lower boundary of the bonded cylinder is fixed. The shear and normal strain in the bond 𝜀 = ln (1 + ∆𝑁 𝑑𝑘 ) and 𝛾 = arctan ( ∆𝑇 𝑑𝑘 +∆𝑁 ) (6) depend on either the measured ( )exp or computed ( )fem values of the displacement jumps ΔT,N across the adhesive layer in tangential and normal direction. In order to regulate the jumps ΔT,c and ΔN,c for the feedback control testing system close to the target value αtar, the error eexp is minimized between a given, desired strain proportion αtar= 0.0, 0.5, 2.0, or ∞ and its experimental output αexp for the strain ratio α=γ ∕ 2ε. Figure 6: Verification of model and identified param. pi,opt Conclusions The bluntly glued, thin-walled cylindrical specimen is presented to investigate the mechanical behavior of structural adhesives. A nearly homogenous state of stress due to defined torsion, tension or compression and any combination may be achieved in the bonding layer. The double hollow cylinder provides valid test data for the determination of the adhesive’s model parameters and for the numerical simulation of joint failure in bonded components. As a result of verification and validation of the computational model, it may be concluded that the specimen at hand allows for a robust computer aided design of adhesively assembled components to predict deformation and failure of their bonds. Acknowledgements The authors gratefully acknowledge financial support of AiF under the program for the promotion of joint industrial research and development (IGF) by the federal ministry of economics and technology through FOSTA together with Stiftung Stahlanwendungsforschung. Figure 5: Feedback control system and loop dashed line for identification of model parameters pi References The output variables yi={ΔT ,ΔN, F, M, α} comprise either the experimental yi,exp or the computed values yi,fem. From the axial force F and the torsional moment M, the engineering shear and normal stress of the FE-analysis and the experiment are computed as τxy=M/W and σx=F∕A0, where A0 denotes the initial cross-sectional area and W the section modulus of the specimen. The model error efem between the FEM simulation yi,fem and the output yi,exp of the testing system may be reduced up to a minimum by feeding improved estimates pi for the material parameters into the FEM model with the help of LS-OPT. The procedure may be repeated until the identification method provides the final model parameters pi,opt for the best fit. Note, the feedback in the control system uses purely experimental data. The model is only unilaterally coupled to the loop of the controlled testing system. The normal and tangential stress-displacement output of the FEsimulation and the testing system are displayed in figure 6 - taken from [14] - for the various strain proportions α given above. [1] G. Müller, PhD Thesis, TU Berlin, 1959 [2] E.E. Kuenzi and G.H. Stevens, Forest Products Laboratory Report FPL-011, 1963 [3] F.C. Bossler, M.C. Franzblau, J.L. Rutherford, J. Scientific Instruments (J. Physics E), 1968, 2, 1, 829-833 [4] I. Engasser and A. Puck, Kunststoffe, 1980, 70, 423493 [5] M. Schlimmer, Z. Werkstofftech., 1982, 13, 215-221 [6] M. Schlimmer: IGF-Report Nr. 76 ZN/P 593, 2005 [7] M. Brede: FOSTA-Report P 676, 2008 [8] M. Brede: Ongoing project IGF-Nr. 338 ZN/P 828 [9] M. Schlimmer, IfW, Univ. Kassel, 2007 [10] M. Schlimmer, PhD Thesis, RWTH Aachen, 1974 [11] R. Mahnken and M. Schlimmer, IJNME, 2005, 63, 1461-1477 [12] J. O. Hallquist: LS-DYNA Theory Manual. LSTC, Livermore, CA, 2006 [13] N. Stander et al.: LS-OPT User's Manual. LSTC, Livermore, CA, 2010 [14] U. Kroll: MSc Thesis, Dept. Mech. Eng., Univ. Kassel, 2011