Poster Presentations I - the home page of EM
Transcription
Poster Presentations I - the home page of EM
4 SURVEY of POSTER PRESENTATIONS This section contains a survey of poster presentations of actual PhD-projects within the Graduate School Engineering Mechanics. Individual poster presentations are collected in a separate report, which will be supplied at the start of the symposium and can be obtained from the Secretariat of the Graduate School. Furthermore, poster presentations are available through: http://www.em.tue.nl Survey of Poster Presentations Nr 1 2 3 4 5 6 7 8 9 Name N.P. van der Aa B.J. Aalderink A. Andreykiv A. Balmachnov R.A. van de Berg A.W. Blom S.H.A. Boers M.H.A. Bonte M.J. v.d.Bosch Uni TU/e UT TUD TU/e TU/e TUD TU/e UT TU/e 10 11 12 13 14 15 16 17 18 19 20 P. Broomans I.A. Burchitz M.V. Cid Alfaro N.E. Conza W. Dijkstra N.T. Dung M.E. Erinç R.P.H. Faassen M.W. de Graaf W.B.J. Hakvoort M.H.C. Hannink TUD UT TUD TUD TU/e TUD TU/e TU/e UT UT UT 22 23 24 25 26 27 28 29 30 31 32 33 34 35 P.J. Heres B.D. Heru Utomo T.S. Hille T. Hofman M. Hrapko P.J.M. Janssen A.J.J. Koopman M.G.M.M. v. Kraaij F. Kraaijeveld M. Langelaar G.K. Lau G. van der Linde O. Lloberas J.M. López de la Cruz X. Ma N.J. Mallon R.M.C. Mestrom M.J.J. Nijhof E.J. Oosterhuis P. Owczarek I. Özdemir M.J.P.D. Patricio R.R. Pedersen E.S. Perdahcioglu T. Rahman M. van Riel M.K. Saraswat R. Scholte E. Shcherbakov TU/e TUD TUD TU/e TU/e TU/e UT TU/e TU/e TUD TUD UT TUD TUD Title Poster Sensitivity theory of the Rigorous Coupled-Wave Analysis Shedding Light on Laser Welding Numerical modelling of electrostatic-structural coupling Tailoring of processable metastable steels (1) Convengence of switched systems Mechanics of Thin-Walled, Variable-Stiffness Shells made by Tow-Placed Composites Optimum path and discrete 3D forming Optimisation of Forming Processes A combined experimental-numerical approach to characterize delamination in polymer coated steel Orientation and Position of the Glenoid Component Improving numerical predictability of Springback Mode I Crack Tunneling in Fibre-Metal Laminates Frequency dependent properties in modal parameter identification Condition number of the boundary element method matrices Discontinuous Galerkin methods for structural systems Fatigue Analysis of Lead-free Solder Interconnect Identification of Spindle Dynamics in High Speed Milling Real-time trajectory generation for sensor-guided robotic laser welding Improved tip tracking for an industrial robot using Iterative Learning Control Reduction of sound transmission through panels by means of tuned acoustic resonators Robust Model Order Reduction by Krylov subspace methods High-speed impact modelling and testing of composite structures Design and development of high temperature coating system for engines Design of a Hybrid Vehicle: From generic to specific design The Mechanical Behaviour of Brain Tissue Grain statistics: an experimental investigation Flow front tracking in aluminium extrusion dies by means of particle trajectories A more Rigorous Coupled-Wave Analysis A singularity solution of shear faulting in swelling ionised porous media Topology Optimization of Shape Memory Alloy Thermal Actuators Electrostatic Actuators Using Elastomers Galling mechanism in deep drawing processes First-order multi-scale framework for discontinous modelling of failure Simulation of Environmentally assisted failures of metals TUD TU/e TU/e UT UT UT TU/e TU/e TUD UT TUD UT TUD TU/e TU/e Effect of Moisture on the Viscoeleasticity of Thermosetting IC Packaging Polymers Dynamic stability of a base excited thin beam with top mass Dynamics and multiphysics interaction in Microsystems Optimization of folded resonators for broadband reduction of computer fan noise Inverse dynamics for durability testing Design rules for close tolerance and lubricant free piston compressors Multiscale Modeling of Thermal Shock: From Microstructure to Failure Elastic stationary analysis of a cracked plate Computational study of impact fracture of concrete structures Constitutive modeling of metastable austenitic stainless steel Fast tools for Multi-Fidelity Nonlinear Finite Element Analysis of Structures Strain path dependent Material models for forming and crash Cure Shrinkage Monitoring in Thermoset Resins High Resolution Planar Near-field Acoustic Holography Development of a Maxwell's Solver 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 Eighth Engineering Mechanics Symposium, Nov. 2005 39 Nr 51 Name J. Shi 52 53 P.J. Sloetjes Q.H.C. Snippe 54 55 56 N.J. Suman P. Tiso S. Tosserams 57 58 59 60 61 62 63 64 A.A. van Veggel I. Vegt A. Verhoeven C.V. Verhoosel L. Wang J.W. Wind A.J. de Wit D.G. Yang 65 K.G. van der Zee Uni Title poster TUD The competition between solid-state phase transformations and plactic deformation: discrete interfaces and discrete dislocations UT Balancing and stabilizing flexible shafts with piezoelectric materials UT Design and optimisation of vertex detector foils as a function of the deformation process, in particular superplastic forming TUD Effect of resin chemistry on thermoset visoelasticity TUD Finite Element Based Asymptotic Postbuckling Analysis of General Structures TU/e An Augmented Lagrangian Relaxation for Analytical Target Cascading using the Alternating Directions Method of Multipliers TU/e Silicon Carbide optical bench for stability measurements TUD Experimental study to the impact behaviour of concrete TU/e Hierarchical multirate BDF methods for IC transient simulation TUD Spectral solution of the random eigenvalue problem TUD Flexibility Study of Ultra-thin Substrate UT Inverse acoustics IBEM vs. PNAH TUD Multilevel Optimization of Composite Structures TUD Micromechanical Modeling of Cure-Induced Stresses in a Particle-Filled Electronic Packaging Polymer TUD A Posteriori Error Estimation for Free-Boundary Problems Eighth Engineering Mechanics Symposium, Nov. 2005 40 Sensitivity theory of the Rigorous Coupled-Wave Analysis N.P. van der Aa Eindhoven University of Technology Faculty of Mathematics and Computing Science P.O. Box 513, NL 5600 MB Eindhoven phone +31-(0)40-2475518, email [email protected] Introduction Sensitivity theory Diffraction of light plays an important role in various metrology applications. Periodic structures, called gratings, cause the monochromatic incident light to be diffracted (see figure). This diffracted field gives an accurate measurement of the position of the grating, which in general has a period in the same order of magnitude as the field’s wavelength. Since the estimate of the grating shape can never be 100 % accurate, it is crucial to know how the diffracted field behaves when small deviations occur in the grating shape. Two methods are under consideration: • Finite Difference Technique (FDT) The FDT is a general method to compute sensitivity. It recomputes the field for a small change in the shape parameter under consideration, because the difference between new and old field gives the derivative information. • Analytical approach Straightforward differentiation of the RCWA equations requires eigenvalue and eigenvector derivatives. Fortunately, these derivatives can be computed without solving additional eigensystems according to Murthy and Haftka [2]. The analytical approach has two advantages: no additional eigenvalue systems need to be solved (faster) and no additional approximations are made besides the ones already made by RCWA (more accurate). Figure 1 : Diffraction by a one-dimensional grating. Objective RCWA Before the field behaviour with respect to grating shape parameters can be computed, the field itself should be available for a given grating profile. A powerful method to find to field is the Rigorous CoupledWave Analysis (RCWA) [1]. Its key features are: • Division of the grating structure in thin layers; • Approximation of the material properties by a piecewise constant function; • Introduction of Fourier expansions of both field and material properties inside each layer; As a result, Maxwell’s equations inside each layer are reduced to one eigenvalue problem. The fields inside each layer are matched by means of the boundary conditions. The accuracy of RCWA depends on the number of layers and the number of terms in the Fourier expansions. Results The field derivatives for a binary grating can be computed by both methods. The difference between them will not be the accuracy, but the computation speed. As predicted and illustrated in the figure, the analytical approach is much faster than FDT. FDT analytical approach time The goal of this project is to find the shape parameter sensitivity of the diffracted field in an accurate, but fast way for any type of diffraction grating. truncation number Figure 3 : Computation speed of the sensitivity theories. Conclusion A faster theory than FDT is developed to compute the field sensitivity with respect to the grating shape. References 1. Figure 2 : Actual (left) and approximated grating (right) Eighth Engineering Mechanics Symposium, Nov. 2005 Moharam M.G. et al. (1995) Formulation for stable and efficient implementation of the rigorous coupled-wave analysis of binary gratings, J. Opt. Soc. Am. A 12(5). 2. Murthy, D.V. and Haftka, R.T. (1988) Derivatives of Eigenvalues and Eigenvectors of a General Complex Matrix, Int. J. Num. Methods in Engineering 26. 1 Shedding Light on Laser Welding B.J. Aalderink*, R.G.K.M. Aarts, J.B. Jonker and J. Meijer *Netherlands Institute for Metals Research University of Twente, Laboratory of Mechanical Automation Institute of Mechanics, Processes and Control - Twente P.O. Box 217, 7500 AE Enschede, The Netherlands +31-(0)53-4894846, [email protected] Introduction Performance Nd:YAG laser welding is often used in industry to obtain high quality joints. This however does not mean that monitoring or control of this process is common practice. A few commercial products are available but none of these systems can be used for monitoring the laser welding process of aluminium. Within the NIMR project Multivariable melt pool control for double spot laser welding a monitoring system is developed based on a CMOS camera which is suited for the observation of Nd:YAG laser welding of various materials, including aluminium, under different conditions. The current system uses a 10 W, 810 nm diode laser and a narrowband optical filter to achieve a sufficiently large S/N ratio. In figure 2 some examples of the resulting images using this monitoring system are displayed, together with an image without illumination. The images illustrate that the system is applicable for the observation of single and double spot laser welding of aluminium, with and without cold wire feeding, using both a standard CMOS and a high speed camera. Spectroscopic measurements show that an optimized system will also function for laser welding of steel although this has not yet been tested. Current investigations also focus on the applicability for laser/MIG hybrid welding. High power laser beam Optical fiber Coaxial CMOS camera Image quality standard CMOS camera without illumination (single spot laser welding AA5182). Welding head Narrow band optical filter Image quality new monitoring system with high speed camera (single spot laser welding AA5182). Illumination diode laser light Weld plume Image quality new monitoring system with standard CMOS camera Workpiece (twin spot laser welding AA5182, with cold wire feeding). Figure 1: Overview of the new monitoring system. Traditional Monitoring System Image is mixture of reflected laser light and temperature radiation, making it difficult to interpret. Only indirect information on welding process and Figure 2: Image quality of the monitoring system using different configurations compared to typical traditional monitoring system images. quality. Information depends strongly on material and process parameters. New Monitoring System Image contains only geometrical information, simplifying interpretation and FE model verification. References 1. 2. B.J. Aalderink, R.G.K.M. Aarts, J.B. Jonker, J. Meijer. Weld Plume Emissions During Nd:YAG Laser Welding, Proceedings of LIM '05, 2005. B.J. Aalderink, R.G.K.M. Aarts, J.B. Jonker, J. Meijer. Experimental Observations of the Laser Keyhole Welding Process of AA5182, Proceedings of ICALEO '05, 2005. Direct information on the welding process and quality. Information independent from material and welding parameters. Eighth Engineering Mechanics Symposium, Nov. 2005 2 Numerical modelling of electrostaticstructural coupling A. Andreykiv, D. J. Rixen, F. van Keulen Department of Precision and Microsystems Engineering Faculty 3ME Delft University of Technology Mekelweg 2, 2628 CD Delft, The Netherlands Phone +31 (0)15 2786818, E-mail [email protected] Introduction Results Fig. 1. Example of an electrostatic actuation Fig. 2 Electric potential around two beams with applied voltage. Voltage, V Coulomb force is a basic principle of many MEMS actuators. Previously, calculation of electrostatic structural coupling was mainly performed using a staggered method. However, it can not be applied in case of a strong coupling or a calculation of a snap back behaviour. An earlier introduced monolithic formulation [1] was able to resolve these issues, but poor conditioning of the resulting system required an additional step for the electrostatic domain (“air”) mesh update. This potentially could result in severe distortions of the elements on the structure-“air” interface. Alternative formulation Similarly to [1], electrostatic and mechanical fields are solved simultaneously in the same variational formulation. Strong form of the formulation is obtained by taking a derivative of the energy of the system. The energy of the system is taken as the sum of the mechanical and electrical energies. The source of instability in [1] was a non-linear effect that appeared when the electrostatic force was applied to the nodes of the “air” domain. In this work we propose an alternative formulation, where the structure is considered independent of the electrostatic field, while the electrostatic field is updated with the motion of the structure. The calculated electrostatic force is then applied only to the interface between the structure and the “air”. The electrostatic force is considered as a follower force, so the appropriate modifications to the stiffness matrix of the interface “air” elements are made, making it fully consistent. Eighth Engineering Mechanics Symposium, Nov. 2005 0 0 Displacement, m 0.00359 Fig. 3 Displacement of the left beam tip versus the applied voltage. A snap “through behaviour” calculated with an Arc-length method allows determination of a critical (“pull-in”) voltage. Discussion The proposed formulation allows a single step solution of the electrostatic-structural problems which can improve the stability and convergence of the problem comparatively to the two-step approach. The formulation can also be easily implemented in a commercial FE code. References 1. V. Rochus, D. J. Rixen, J.-C. Golinval. Int. J. Numer. Meth. Engng (2005) In Press 3 Tailoring of processable metastable steels (1) A. Balmachnov♮ , V.G. Kouznetsova, and M.G.D. Geers Eindhoven University of Technology Faculty of Mechanical Engineering P.O. Box 513, NL 5600 MB Eindhoven ♮ phone: +31-(0)40-2472054, ♮ e-mail: [email protected] Results The lamellae model was successfully implemented within 2D non-linear FE-framework and tested on various plain strain homogeneous loading cases (see Fig. 2, 3: Plain strain tensile test). 0 MACRO MICRO austenite (grains) Engineering level Lamellae model martensite FE F Solving boundary value problem 1 ξ ~ N FM,PM Hydrostatic pressure, GPa −0.2 −0.3 −0.4 −0.5 0.08 0.06 0.04 0.02 Equivalent Green−Lagrange strain 0 (b) 0.4 MESO RVE level 0.4 0.2 −0.1 0.08 0.06 0.04 0.02 Equivalent Green−Lagrange strain Figure 2 : (a) Equivalent Cauchy stress and (b) hydrostatic pressure vs. equivalent Green-Lagrange strain Martensite volume fraction This project focuses on multi-scale computational modeling and microstructural optimization of metastable austenitic steels within the multi-scale computational framework (Fig. 1). 0.8 0.6 0 0 (a) Method 1 (a) 0.4 Martensite volume fraction Thanks to their properties, advanced materials like steels with transforming metastable phases are used for great number of technological applications. Design of products with certain properties/specifications relies on properties of employed materials. However, for metastable steels material design can not be done independently from the processing route during which the material will be formed and shaped towards a product with optimal performance. Employed interdisciplinary approach aims to bridge materials science (micromechanics) with materials engineering (process & product engineering). • Averaging rules • Constitutive equations for each phase • Interface interaction relations • Transformation criterion Further on, averaging over all 24 martensitic variants is performed to capture behavior of a transforming austenitic grain. Equiv. Cauchy stress, GPa Introduction 0.3 0.2 0.1 0 0 0.08 0.06 0.04 0.02 Equivalent Green−Lagrange strain (b) 0.3 1, 2, 5, 6 3, 4, 7, 8 9, 10, 13, 14 11, 12, 15, 16 17, 18, 23, 24 19, 20, 21, 22 0.2 0.1 0 0 0.08 0.06 0.04 0.02 Equivalent Green−Lagrange strain martensite austenite F A, P A PE P, x Figure 3 : Martensite volume fraction vs. equivalent Green-Lagrange strain: (a) averaged (b) per variant Future Work Figure 1 : General multi-scale framework for metastable austenitic steel Micromechanical Model On the micro-level, a lamellae model is employed to compute the evolution of martensitic volume fraction ξ and the behavior of one transforming martensitic vari¯. ant for a given total gradient deformation tensor F The model consists of Eighth Engineering Mechanics Symposium, Nov. 2005 • Extension of the implementation to the 3D nonlinear FE setting • Comparison of simulations with experiments References 1. Chen, S. P., Kouznetsova, V., and Geers, M. G. D. (2005), Modelling martensitic transformation induced plasticity at finite strains, submitted 4 Convergence of switched systems R.A. van den Berg, A.Yu. Pogromsky, J.E. Rooda Eindhoven University of Technology Department of Mechanical Engineering P.O. Box 513, NL 5600 MB Eindhoven phone +31-(0)40-2473360, email [email protected] Introduction Example It is well known that LTI systems with bounded input have a unique limit solution that only depends on its input. Nonlinear or hybrid systems that also show this desirable property are referred to as convergent. Since the solution of a convergent system does not depend on its initial conditions (after a transient period), simulation can be used for performance evaluation. We consider the class of switched linear systems: Consider the switched system in Fig. 1 that can be written in the form of (1), with ẋ = Ai x + Bi w, i = 1, . . . , k y = Cx, (1) where x(t) ∈ Rn is the state, input w(t) is piecewise continuous and bounded for all t ∈ R, and i denotes the operation mode. Furthermore, all system matrices Ai are Hurwitz. 0.6 −1.6 1 0 A1 = −2.5 0 1 , B1 = 1.5 , 0.7 −0.7 0 0 1.6 −2.6 1 0 A2 = −3.3 0 1 , B2 = 2.3 , C = [ 1 0 0 ] . 1.2 −1.2 0 0 For this system a common Lyapunov matrix P can be found such that the conditions of Theorem 1 are satisfied. For the convergence of the transient part of the solution the following bound can be determined: |x1 (t) − x2 (t)| ≤ 2.666 |x1 (0) − x2 (0)| e−0.205t Objective The general problem is to find a switching rule such that the closed-loop system (1) is convergent. Here we focus on a switching rule that is based on static state feedback, i.e. i = σ(x). For the limit solution a Bode-like plot is drawn (Fig. 2). Here we used w(t) = sin(bt) with b ∈ [10−2 , 102 ], and we compared the switched system with the separate PID controllers. PID1 SWITCH - linear system y(t) PID2 ||w − y||2 /||w||2 w(t) + 1 0.8 PID1 PID2 Switched 0.6 0.4 0.2 Figure 1 : Switched system Exponential convergence 0 10−2 10−1 100 b 101 102 Figure 2 : Performance evaluation Theorem 1 Let P = P T > 0 satisfy ATi P + P Ai < 0, P (Ai − Aj ) − (Ai − Aj )T P = 0 ∀i, j ≤ k. Then, the following switching rule σ(x, w) = arg min{xT (ATi P + P Ai )x + 4xT P Bi w} i makes system (1) exponentially convergent. Eighth Engineering Mechanics Symposium, Nov. 2005 Discussion Once convergence is obtained for a certain (sub)class of nonlinear/hybrid systems, e.g. by defining a specific switching rule, simulation can be used for performance evaluation and optimization of the system. 5 Mechanics of Thin-Walled, Variable-Stiffness Shells made by Tow-Placed Composites A.W. Blom Delft University of Technology Faculty of Aerospace Engineering Kluyverweg 1, 2629 HS Delft phone +31-(0)15-2785145, email [email protected] Introduction Fiber-reinforced composites are gaining ground on metals in aerospace industry, largely due to intelligent use of the directional properties of fibers. With towplacement machines these directional properties can be used even more efficiently, due to the capability to steer the fibers within the plane of a laminate. By fiber steering, the local stiffness properties of a laminate can be varied. Earlier research on variable stiffness flat plates by Gürdal and Tatting demonstrated that buckling load may be increased by a factor of 2 to 3, while keeping the panel weight approximately same as the traditional laminates 1 . manufacturing constraints. Arbitrary combinations of paths can be used to construct a laminate. The variable stiffness properties are implemented in a finite element program, such that any finite element analysis can be carried out. Figure 2 : Surface plot of the number of layers of a variable-stiffness cone made of a ± overlap laminate Discussion Figure 1 : A variable-stiffness overlap panel Objective The objective of this research is to improve the structural response of thin-walled cylindrical and conical shells through local stiffness tailoring achieved by judicious selection of fiber paths. Different cone geometries under different loading conditions are being studied to investigate the influence of path parameters on buckling and vibration performance. Furthermore, configurations are being designed for experimental validation of the concept. Finally, cost effective analytical solution techniques are being considered for three-dimensional shell configurations. Methods General path definitions along which the fibers can be steered are developed, taking into account the limitations of the tow-placement machine. Using these definitions and the shell geometry, a program is written that provides a finite element analysis with the local stiffness properties. The capability allows parametric study of improved lay-ups as well as determination of best design using optimization. Results Four consistent path definitions are developed for general conical shells, while taking into account the Eighth Engineering Mechanics Symposium, Nov. 2005 Figure 3 : Apache tailboom, possible application of a variable-stiffness shell References 1. Tatting, B.F., and Gürdal, Z. ”Design and Manufacture of Elastically Tailored Tow Placed Plates,” NASA/CR 2002211919, August 2002 6 Optimum path and discrete 3D forming S.H.A. Boers, P.J.G. Schreurs, M.G.D. Geers Eindhoven University of Technology Faculty of Mechanical Engineering P.O. Box 513, NL 5600 MB Eindhoven phone +31-(0)40-2474022, email [email protected] Introduction Results Discrete die forming is a useful concept in a small-lot production environment because different products can be made with the same reconfigurable die and many resources are saved. From a scientific point of view: a discrete die allows a changing strain path during forming; products with an optimized internal strain distribution can be produced, e.g. the lifetime of parts carrying cyclic (thermal) loading can be enhanced considerably. Using the discrete die and photogrammetry technique, two geometrically equal products are produced with different strain distributions. Conventional die Discrete die rubber pad Pyramid−parabolic vs. Parabolic geometry, major strain % 20 10 10 0 0 −10 −10 −20 −20 Discrete die with 'interpolator' rubber pad sheet metal Initial set-up 20 rubber interpolator discrete die die −30 −30 discrete die −20 −10 −4 −2 0 10 0 −30 −30 20 2 4 6 −20 −10 8 0 10 10 20 12 Forming step Final part smooth surface dimpled surface smooth surface Figure 1 : Rubber pad forming with conventional die, reconfigurable die and interpolator Figure 3 : Distribution of major plastic strain for two geometrically identical products Using the flexible mould, a product without defects is produced with an intermediate forming step. Objective: A numerical tool to calculate the boundary conditions of a deformation process that yields geometrically identical products with different internal strain distributions. Validation will be done by using an experimental set-up. The numerical tool controls the internal strain distribution by means of non-proportional forming-limit diagrams. Figure 4 : Multi-step forming experiment with discrete die Discussion Figure 2 : Prototype of discrete die, surface dimensions: 20x30 mm, forming pressures up to 5 ton Methods ¯ Experimental deformation process with discrete die prototype in which the strain path is variable. ¯ Determination of strain distribution and product geometry with photogrammetry technique. ¯ Numerical algorithm, providing boundary conditions for an optimal deformation process. Eighth Engineering Mechanics Symposium, Nov. 2005 ¯ Non-proportional loading tests must be done for model input. ¯ Implement numerical tool within FEM framework. ¯ Criterion must be formulated to determine optimal strain distribution. ¯ Automated control of discrete die is preferable. References 1. 2. S. Li, E. Hoferlin, A. v. Bael, P. v. Houtte and C. Teodosiu (2001), Finite element modeling of plastic anisotropy induced by texture and strain-path change,International Journal of Plasticity 19 (2003), 647674 C. Zhongyi and L.Mingzhe, Optimum path forming technique for sheet metal and its realization in multi-point forming, Journal of Materials Processing Technology, Volume 110, Issue 2, 19 March 2001, Pages 136-141 7 Optimisation of Forming Processes M.H.A. Bonte University of Twente P.O. Box 217, 7500 AE Enschede, The Netherlands phone +31-(0)53-4894069, email [email protected] Introduction Cost saving and product improvement have always been important goals in the metal forming industry. To achieve these goals, metal forming processes need to be optimised. During the last decades, simulation software based on the Finite Element Method (FEM) has contributed significantly to designing feasible processes more easily. More recently, coupling FEM to mathematical optimisation algorithms is offering a very promising opportunity to design optimal metal forming processes instead of only feasible ones. Objective The aim of the project is the design of an optimisation strategy for metal forming processes. An optimisation strategy consists of the modelling and the solving of a mathematical optimisation problem [1]. This poster addresses the solving part: an optimisation algorithm is proposed and its high potential is demonstrated by optimising a forging process. Figure 2 : A forging process: (a) Preform; (b) Gear Optimising a forging process The SAO algorithm is applied to optimise the forging process of a gear. Figure 2 shows the two stages of the process: producing a preform and forging this preform into the final product. The objective is to minimise (i) defects in the final part; and (ii) the energy consumption during forging. This is achieved by influencing the design variables that describe the preform shape as shown in Figure 3(a). Figure 3(b) shows the preform shape proposed by the gear manufacturer. Results and Discussion The optimised preform shape is presented in Figure 3(c). The optimised forging process resulted in a 10% reduction of both part defects and energy consumption. These good results demonstrate the potential of the SAO algorithm for forming processes. References 1. Papalambros, P.Y. et al. (2000) Principles of Optimal Design, 2nd ed., Cambridge (UK), Cambridge University Press. Figure 1 : Sequential Approximate Optimisation Methods The optimisation algorithm A Sequential Approximate Optimisation (SAO) algorithm is proposed for optimising metal forming processes using time-consuming FEM simulations. The algorithm is presented in Figure 1. It includes a Design Of Experiments (DOE) strategy, running a number of parallel FEM calculations, fitting a metamodel and optimising this metamodel. The algorithm allows for sequential improvement to obtain more accurate results. Eighth Engineering Mechanics Symposium, Nov. 2005 Figure 3 : (a) Design variables; (b) Initial preform shape; (c) Optimised preform shape 8 A combined experimental-numerical approach to characterize delamination in polymer coated steel M.J. van den Bosch, P.J.G. Schreurs, M.G.D. Geers Netherlands Institute for Metals Research Eindhoven University of Technology Faculty of Mechanical Engineering P.O. Box 513, NL 5600 MB Eindhoven phone +31-(0)40-2472054, email [email protected] Introduction Results Polymer coated metal sheet is developed by Corus to reduce production costs of e.g. aerosol and food cans. Delamination experiments are conducted inside a Scanning Electron Microscope (SEM). The delamination front is observed (see figure 3) and the force is measured. Figure 1 : Examples of applications for polymer coated metal sheets. During forming processes the coated sheets may be damaged, as shown in figure 2: Figure 3 : Micrograph of a delamination experiment in the SEM. The interface parameters are determined by simulating the experiments and fit the results (see figure 4). This approach requires accurate material models and parameter values for both the polymer and steel. Figure 2 : Delamination, cracking and surface roughening of the polymer layer after a forming process. Objective Predict the occurrence of coating delamination during industrial forming processes. Figure 4 : Schematics of a zero degree delamination experiment with the result of a simulation. The colors indicate the equivalent von Mises stress [Nmm−2 ]. Methods A Finite Element model is used to simulate industrial forming processes, such as bending and deepdrawing. Between the polymer layer and the metal substrate interface elements are present to simulate the delamination of the polymer layers. Experiments are needed to determine parameters for the simulations. Eighth Engineering Mechanics Symposium, Nov. 2005 Discussion The interface parameters can be determined with the combined experimental/numerical procedure. In the future the influence of pre-deformation on the interface parameters will be investigated. 9 Orientation and Position of the Glenoid Component P. Broomans, C.W. Oosterlee, F. van Keulen and F.J. Vermolen Delft University of Technology Faculty of Mechanical Engineering & Marine Technology Mekelweg 2, 2628 CD Delft phone +31-(0)15-2786818, email [email protected] Introduction energy density (SED) U remains below In contrast to total hip and total knee arthroplasty, replacement of the shoulder joint is faced with much more problems post-operatively and revisions. Component loosening has been reported in 44% to 59% of the implanted glenoids. It has been reported that malalignment of an implant can be a cause of or contribute to implant failure1 . Alignment of the glenoid component has shown to be of influence on component loosening, humeral head subluxation, cement mantle stresses and shoulder muscle forces, but investigations focused on the mechanical environment in the bone adjacent to the implant have not been so extensive. (1 − s)Uref , where s, the remodeling threshold, is set to 0.1 and Uref is the reference SED value. Results Initial results show that bone resorption mainly occurs on the lower side of the bone-implant interface (Figure 2). The highest percentages of bone resorption occur in the configuration with the backward inclined implant. For the forward and downward (not shown!) inclined implants nearly no resorption occurs. Figure 1 : Cross section of the glenoid part of the 3D model showing the different materials. (red polyethylene; blue - CoCrMo metal; yellow - porous CoCrMo metal; green - bone). The interface is defined as the line between metal and bone. Objective The goal is to determine the influence of the orientation of the glenoid component on bone resorption and the possibility of bone ingrowth. Methods A 3D material model of the scapula, acquired through data from a CT scan, is available for use in a FE analysis. Five alignments of the implant are simulated, one in a ’central’ alignment and four each 10◦ inclined in either forward, downward, backward and upward direction with respect to the central alignment From the abduction of the upper arm six static load cases have been derived with the aid of the Delft Shoulder and Elbow Model2 . A configuration without prosthesis is used as a reference to determine in which parts of the bone-implant interface bone resorption occurs. At a specific location bone resorption within the model occurs if for all load cases the strain Eighth Engineering Mechanics Symposium, Nov. 2005 Figure 2 : Percentages of the particular areas of the boneimplant interface where bone resorption will occur (left is back side) for three alignments. The hatched area is the location of the screw. Hopkins et al.3 reported that the lowest cement mantle stresses are calculated for the central alignment and the highest values appear for downward and upward directed inclinations. For the latter configuration, however, our results suggest that also areas with significant lower stress values are observed. Future Work An extension of the model to incorporate the process of bone ingrowth is the next goal of this project. Modelling the influence of locally applied growth factors, e.g. within coatings of metal-backed implants, can be of interest for our model. References 1. 2. 3. Hasan, S.S. et al. (2002) Characteristics of unsatisfactory shoulder arthroplasties, J Shoulder Elbow Surg 11: 431-441. Van der Helm, F.C.T. (1994) Analysis of the kinematic and dynamic behavior of the shoulder mechanism, J Biomech, 27: 527-550. Hopkins, A.R. et al. (2004) The effects of glenoid component alignment variations on cement mantle stresses in total shoulder arthroplasty, J Shoulder Elbow Surg 13: 668-675. 10 Improving numerical predictability of Springback I.Burchitz, T. Meinders, J. Huétink Faculty of Engineering Technology, NIMR - University of Twente P.O. Box 217, 7500 AE Enschede, The Netherlands phone: +31-(0)53-4894069, email: [email protected] Introduction Finite element software is used in the design process of new sheet metal parts (figure 1). During the process the amount of springback (elastically-driven change of product shape) is numerically predicted. This information, being used in tools design phase, ensures that the desired product shape will be reached. Current accuracy of numerical prediction of springback is insufficient. Required surfaces of tools can only be obtained after employing the extensive experimental trial and error process. Figure 2 : Characteristic component 3 Solver type. Sensitivity analysis on component 3 (figure 2) and simulations of U-bending test showed that an iterative solver can deliver an inaccurate solution. Depending on its parameters and values of global convergence criteria the product shape after springback may be completely unrealistic (figure 3). Guidelines are needed that define a selection of iterative solver parameters for springback analysis. Figure 1 : Schematic of the design process Objective The major goal of the project is to improve the numerical predictability of springback to meet industrial requirements. Methods Additional analysis of sensitivity of springback to various physical and numerical parameters showed that: • increasing coefficient of friction does not necessarily decrease springback, i.e. in a situation when a change of shape is dominated by a relaxation of membrane stresses; • springback is highly influenced by a decrease of the apparent unloading modulus. Results Mesh density. Recommendations, available in the literature, were tested using the U-bending problem. Results showed that an optimal discretisation level strongly depends on in-plane tension, R/t ratio and material properties. To understand this dependency a simple model of a beam under combined bending moment and tension was built. The model is used to develop practical guidelines that define an appropriate mesh density to assure the required accuracy of springback prediction. Eighth Engineering Mechanics Symposium, Nov. 2005 Figure 3 : Influence of solver type on springback Discussion Authors in [1] showed that depending on material and process parameters 30-68 integration points through the thickness are required to reach 1% of springback accuracy. Accurate analysis of a plastically deforming sheet material requires that an integration point lies on each surface where yielding begins. An attractive approach is to perform a through-thickness integration by an algorithm that adapts sampling points to the stress situation. The adaptability may include changing a location of integrations points and/or changing their absolute number. This adaptive integration algorithm can help to achieve 1% accuracy of springback prediction at minimal costs. References 1. Wagoner, R. H. and Li, M. ”Advances in springback.” in proc. Numisheet 2005. Detroit, MI, U.S.A., p.209-214. 11 Mode I Crack Tunneling in Fibre-Metal Laminates M. Cid Alfaro Delft University of Technology Delft University of Technology Faculty of Aerospace Engineering P.O.Box 5058, NL 2600 GB Delft phone +31-(0)15-2786380, email [email protected] Introduction Recently, steady-state crack tunneling and planestrain delamination have been studied in laminates of alternating layers of two dissimilar but isotropic elastic, brittle solids, subjected to a remote uniform tensile strain within a two-dimensional finite element framework [1]. In this work three fracture mechanisms were distinguished, which are (i) the tunneling of a mode I crack without delamination, (ii) the tunneling of an H-shape crack with constant delamination length, and (iii) the unstable delamination growth in all directions, see Figure 1. Mechanism 1 Mechanism 2 and fibre-reinforced prepeg layers are modelled using continuum solid like shell elements, combined with an isotropic elastic model. The assumption of elastic isotropic layers may be acceptable for laminates with fibres in multiple directions. Preliminary Results The traction profile in the tunneling direction is shown in Figure 2. The crack initiates from a small imperfection and induces a traction profile that is approximately uniform over the specimen width in the tunneling direction. This is, because the fracture process zone is of a similar size as the specimen width. Mechanism3 Figure 1 : Three possible failure mechanisms for a laminate of two dissimilar, isotropic materials. Figure 2 : Traction profile in the tunneling direction. Left: hardening branch; Right: softening branch. Objective Analyse Mode I crack propagation in a Fibre-Metal Laminate comprising two layers of aluminium and one fibre-reinforced epoxy layer. Numerical Models Three-dimensional numerical simulations of a tunneling mode I crack without delamination have been performed. An interface damage model [2] has been implemented to describe the onset and growth of mode I cracking in the aluminium sheets. Interface elements furnished with the interface damage model were placed where cracking is expected to occur. The interface layer is endowed with a high stiffness in order to suppress elastic deformations prior to the onset of cracking. The individual aluminium Eighth Engineering Mechanics Symposium, Nov. 2005 Future Work Examine size effects due to ratio between fracture process zone and specimen width, study effect of plasticity in aluminium layers, investigate the effects of delamination at the layer interfaces and anisotropy in the fibre-reinforced epoxy layers. References [1] A.S.J. Suiker & N.A. Fleck, Crack tunneling and planestrain delamination in layered solids. International Journal of Fracture 125: 1-32, 2004. [2] A. Turon, P.P. Camanho, J. Costa & C.G. Davila An interface damage model for the simulation of delamination under variable-mode ratio in composite materials. NASA/TM2004-213277. 12 Frequency dependent properties in modal parameter identification N. E. Conza and D. J. Rixen Faculty of Mechanical, Maritime and Materials Engineering Engineering Dynamics TU Delft, Mekelweg 2, 2628 CD Delft, The Netherlands Tel: ++31 15 278 6508 e-mail: [email protected] Introduction The assumption underlying our research is that abnormal biomechanical properties of the sacroiliac joints in the pelvis might provoke low back pain. Up to now constant mechanical properties have been assumed in the pelvis; however there are indications that stiffness and damping might depend on the excitation frequency. This fact might impair the identification starting from measured Frequency Response Functions. Objective The objective of this study is to investigate the impact that frequency dependency in the system matrices might have on the quality of the system identification. Two commercially available identification tools have been tested (SDT and ME'scope). Method Data generation The model use to generate data consists of three rigid bodies, twelve linear springs, and damping matrix constructed from preset damping ratios. Frequency dependency has been introduced in the stiffness and damping matrices according to the formula: ( −ω2 M + i ωC + iD + K + ωK' ) z = s ω with D = f C max 3 and K' = f K 2 ω max The factor f has been set at different values between 0 and 1 to investigate the effects at different level of dependency. Eighth Engineering Mechanics Symposium, Nov. 2005 Identification tools evaluation The 18 identified resonance frequencies, modal damping ratios and mode shapes are compared to the theoretical values computed by means of the model. Results Resonance frequencies: very well for both tools (errors < 1% for all factors f). Damping ratio: ok for both tools (errors ranging between -5% to +20% at f = 1). Mode shapes: very well for SDT (all good except mode 18), a little worse for ME'scope (difficulties in modes 9, 15, 16, 17, 18). Discussion The better mode shape estimation of SDT is probably due to its fitting algorithm, which optimizes poles and residues at the same time [1]. ME'scope, on the other hand, optimizes first the poles, and then the residues [2]. Conclusion Results show that standard identification tools assuming constant system matrices are still able to correctly identify the frequencies, while damping and mode shapes estimation present more difficulties. References [1] Balmès E, Frequency Domain Identification of Structural Dynamics Using the Pole/Residue Parametrization, International Modal Analysis Conference XIV, Dearborn, Michigan, 2006. [2] Richardson MH, Formenti DL, Global Curve Fitting of Frequency Response Measurements using the Rational Fraction Polynomial Method, International Modal Analysis Conference III, Orlando, Florida, 1985. 13 Condition number of the boundary element method matrices tu W. Dijkstra Eindhoven University of Technology Faculty of Mathematics P.O. Box 513, NL 5600 MB Eindhoven phone +31-(0)40-2474328, email [email protected] Introduction Dirichlet problem (m = N ) We focus on the Laplace problem on a circular domain with mixed boundary conditions. Partial tion ∆u u q differential equa- = 0, x ∈ Ω, = ũ, x ∈ Γu , ∂u := ∂n = q̃, x ∈ Γq , Unknowns: u and q . Γq R W G u G π . N| In the figures below we show the behavior of the condition numbers. 100 50 N=4 N=8 N=12 N=16 45 40 10 35 30 25 10 q 5 1 1 0 2 5 R ~u H ~ = ~q Neuman problem (m = 0) π π + tan N cond(A) = |π − (N − 1) tan 20 Algebraic equations First m elements have Dirichlet boundary conditions. 1 Gq = I + H u =: H̃u, 2 To construct the system matrix A we select a block from G and from H̃. G 15 Γu q 1 2 , | log R| . 1 N , | log R| cond(A) Equations max cond(A) = min cond(G) The boundary element method (BEM) transforms a partial differential equation into a boundary integral equation, and after discretisation of the boundary of the domain into a set of linear algebraic equations. The matrix of this set of equations has a special structure, and we investigate the condition number of the matrix. Dirichlet problem 10 15 20 25 30 N 35 40 45 50 Neuman problem Mixed boundary conditions (0 < m < N ) We use a decomposition A = F∗ DQU, with F and Q unitary, D diagonal and U block diagonal. The condition number of A is estimated by cond(A) ≤ cond(D)cond(U). 5 u 10 estimate exact 4 x B ~u 3 10 2 ~q f 10 1 10 0.5 This yields Ax = f. We investigate the condition number of the matrix A. Results For m = N and m = 0 explicit expressions for the condition number: Eighth Engineering Mechanics Symposium, Nov. 2005 1 1.5 2 2.5 3 3.5 4 R { A = u { A q cond(A) 10 Mixed problem Discussion We want to extend the research to a square domain. After analysing the circular and square domain we hope to draw conclusions for more general domains. 14 Discontinuous Galerkin methods for structural systems N.T. Dung and G.N. Wells Delft University of Technology Faculty of Civil Engineering and Geosciences P.O. Box 5048, 2600 GA Delft phone +31 15 278 85710, email: [email protected] phone +31 15 278 87922, email: [email protected] Introduction When solving structural systems such as thin plates or shells with the Continuous Galerkin (CG) method using C1 continuous interpolation functions, challenges derive from the fact that it is not easy to fulfill all the requirements of continuity. A preferred approach is using C 0 elements combined with ReissnerMindlin theory, but other difficulties appear, such as shear-locking problems. When the Discontinuous Galerkin (DG) method is used, only C 0 elements are needed, no shear-locking difficulty appears, and the above problems can be avoided. fracture problems and delamination of structures, the methods have obvious advantages since the discontinuities on the interior edges between elements are naturally existent. Although FFC is still a program in the early-stage of its development, in the near future it might be very powerful for solving large structural mechanical problems. The code shown in Box 1 is an example for future applications of FFC. (a) Objective In this work, various Discontinuous Galerkin (DG) and Continuous/Discontinuous Galerkin (C/DG) formulations are implemented to analyse thin plate and shell structures. The two important aspects that have been studied are the stability condition and the convergence rate. (b) Methods For solving second-order problems (e.g. membrane plate), DG formulations are used; while for fourthorder problems (e.g. bending plate), C/DG formulations seem to be more convenient. The FEniCS Form Compiler (FFC) program (see www.fenics.org for more details) has been used in order to make multi-forms in finite element processes automatically by describing only simple notations. Results The numerical analyses for the elastic PoissonKirchhoff plate have been made. In Fig 1, one can see the discontinuities of displacements (Fig 1.b) or of the slopes (Fig 1.c) across interior boundaries. When penalty parameters are chosen large enough, the stability condition is fulfilled and the results are converging to the exact solution rapidly. Future research Applying Discontinuous Galerkin methods for nonlinear problems should be considered. For solving Eighth Engineering Mechanics Symposium, Nov. 2005 (c) Fig 1: Illustration of solving plate structure: (a) Model, (b) Deformation for membrane load case, (c) Deformation for bending load case. e = FiniteElement("Discontinuous Lagrange", "tetrahedron", 2) v = BasisFunction(e) # test function u = BasisFunction(e) # trial function f = Function(e) # source term b = Constant() # penalty term a = dot(grad(v),grad(u))*dx - dot(jump(v),ave(gradn(u)))*ds \ - dot(ave(gradn(v)),jump(u))*ds + (b/h)*dot(jum(v),jump(u))*ds L = dot(v,f)*dx Box 1: Illustration of FFC code, in which the Interior Penalty formulation will be applied for the Poisson equation using 3D quadratic elements Acknowledgments Financial support from the Vietnamese Ministry of Education and Training (VMOET) and the Netherlands Technology Foundation (STW) is gratefully acknowledged. 15 Fatigue Analysis of Lead-free Solder Interconnect M. Erinç, P.Schreurs, GQ. Zhang and M.Geers Eindhoven University of Technology Faculty of Mechanical Engineering P.O. Box 513, NL 5600 MB Eindhoven phone +31-(0)40-2472245, email [email protected] Introduction Fatigue Tests and Modeling In the industrial search for a lead-free solder, Sn4Ag-0.5Cu alloy is currently the most promising alternative. However, there is an urgent need for characterization of material and mechanical properties, the most important being thermo-mechanical creep&fatigue behavior. In BGA assemblies, failure is caused by thermo-mechanical fatigue and generally occurs at the bump/pad connection. Thus the solder/metallization interface is of crucial importance in solder joint reliability, which is determined by the metallurgical reactions between the solder, SnAgCu, and the metallization, Ni/Au. Pure tensile and pure shear experiments are set up, which are used to fit the normal and tangential springs of the cohesive zones. Specimens are cyclically loaded at various peak strain levels. Figure 3 : Soldered joints of Au coated Ni plates. In the cohesive zone formulation, traction is a linear function of the separation. A damage variable is introduced to account for the accumulation of damage throughout the cycling process(Abdul-Baqi et al): Tα = kα (1 − Dα )∆α ˙ α | (1 − Dα + r)m Ḋα = cα |∆ Figure 1 : BGA layout and a failed solder bump. Objective Under mechanical loading, the crack path between Ni/Au substrate and SnAgCu follows pre-known interfaces and the propagation can be split into normal and tangential components (Erinç et al). Thus the cohesive zone approach fits well to the experimental observations. Figure 2 : (a)Crack paths under normal and tangential loading, (b)Cohesive zone element. The objective of this study is to conduct dedicated fatigue tests with Ni/Au-SnAgCu soldered specimens and use the experimental data to develop a cohesive zone based solder joint fatigue model. Eighth Engineering Mechanics Symposium, Nov. 2005 (1) |Tα | − σf 1 − Dα (2) Individual fatigue tests are simulated with finite elements under plane stress conditions. Material parameters are gathered experimentally. A thin layer of cohesive zones is placed between the solder and the intermetallic layer, since this is the region where failure is observed. After a number of cycles, damage evolves at the bonding interfaces, thus the reaction force decreases at successive cycles. Figure 4 : (a)Damage at cohesive zones after 500 cycles, (b)Reaction force drop due fatigue damage. Conclusions The solder/metallization interface is characterized under fatigue loading. Using a cohesive zone approach to simulate fatigue damage in solder interconnects is promising. 16 Identification of Spindle Dynamics in High Speed Milling R.P.H. Faassen, N. van de Wouw, J.A.J. Oosterling, H. Nijmeijer Eindhoven University of Technology Department of Mechanical Engineering P.O. Box 513, NL 5600 MB Eindhoven phone +31-(0)40-247 5730, email [email protected] Introduction Results To obtain the maximum material removal rate in highspeed milling [2], an accurate model is required, see figure 1. This model will be used to suggest a proper set of machining conditions (e.g. spindle speed) to the machineoperator. Identification of the machine-tool dynamics is performed with cutting tests. Measurements of the force by the dynamometer (F2 ) and the displacement by the eddy current sensor (x2 ) are analysed with N 4 SID [3] and by using Welch’s averaged periodogram method (WAPM), see figure 3. For comparison, also the FRF using the hammer tests are shown Fx11 . Here, besides using WAPM, a model is fitted using the RFP method [1]. Delay Static chip thickness + Cutting Force Machine Displacement − + + −6 10 Dynamic chip thickness Trigonometric relations In this model, the machine-tool dynamics are represented by a linear state-space model (the block Machine). Usually, the necessary parameters for this model are obtained by performing impulse hammer tests on a non-rotating tool. However, in practice, it would be more beneficial if it is possible to identify the machine-tool dynamics by performing cutting tests. Magnitude [m/N] Figure 1: Block diagram of the milling process. −7 10 −8 10 Hammer Fx11 WAPM Cutting Fx22 WAPM Hammer Fx11 RFP Cutting Fx22 N 4 SID Objective −9 The goal of this research is to obtain the machine-tool dynamics by performing cutting tests. 10 2000 3000 4000 5000 6000 7000 8000 9000 10000 Figure 3: Frequency Response Functions. In order to imitate the impact of the hammer by performing cutting tests, a small strip is cut, see figure 2. This results in a short impulse-like force to act on the cutter. After the very short contact time, the tool can freely vibrate until the next tooth enters the cut. For comparison, also standard impulse hammer tests have been performed. Spindle Accelerometer Toolholder Fy , xy Mounting device x2 Feed x1 , F1 Tool Eddy current sensor Workpiece Dynamometer F2 (a) Top view 1000 Frequency [Hz] Methods Ω 0 Accelerometer The benefits (X) and drawbacks (✘) of the method are presented below: X In-process excitation X Easy to do in a machine shop environment X Spindle-speed dependency can be modelled X Broadband frequency excitation X Resonance frequencies can be found ✘ No mode shape information ✘ Poor sensor placement ✘ Unknown number of modes Future work Future topics include improving the sensor position and/or the data processing and validating the model. References [1] [2] (b) Front view Figure 2: Experimental set-up. Eighth Engineering Mechanics Symposium, Nov. 2005 [3] D.J. Ewins. 2000, Modal Testing, Research Studies Press. R.P.H. Faassen, N. van de Wouw, J.A.J. Oosterling, H. Nijmeijer. 2003, Prediction of regenerative chatter by modelling and analysis of high-speed milling Int. J. Mach. Tool. & Manu., 43:1437-1446. L. Ljung. 1999, System Identification , Prentice Hall. 17 Real-time trajectory generation for sensor-guided robotic laser welding Menno de Graaf, Ronald Aarts, Ben Jonker University of Twente, Laboratory of Mechanical Automation Institute of Mechanics, Processes and Control - Twente P.O. Box 217, 7500 AE Enschede, The Netherlands +31-(0)53-4895442/2502, [email protected] Introduction Results Robotic laser welding imposes high demands on the used manipulator as high accuracies (down to 0.1 mm) have to be reached at high velocities (up to 250 mm/s). To meet these specifications with industrial robots, a sensor measuring at the robot tip needs to be applied. A corner trajectory (Figure 2) has been replayed with a maximum linear velocity of 100 mm/s and rotational velocity of 60 deg/s. Motion descriptor Motion Location Buffer Reference joint position Cartesian setpoints Setpoint Generator Inverse Kinematics + Robot Joint Motion Controller - Joint velocity [deg/s] Integration of an optical seam-tracking sensor in a robotic laser welding cell for: • Sensor-guided teaching of the 3D seam trajectory. • Increasing the positional accuracy during welding. Joint position [deg] Objectives 150 150 100 50 0 −50 −100 0 100 50 0 −50 −100 1 2 Time [s] −150 0 3 1 2 Time [s] 3 Figure 3 : Reference joint position and velocity (Joint 1, Joint 2, Joint 3, Joint 4, Joint 5, Joint 6). Both the reference joint position and velocity are smooth. Measured joint position 100 80 80 Tip velocity To use sensor information obtained during the robot motion a real-time trajectory generator (Figure 1) based on cubic interpolation of both position and orientation has been implemented on an industrial Stäubli robot. It calculates setpoints for the robot joint motion controller every 4ms. Features of the trajectory generator are: • The setpoints are calculated on-the-fly, which allows the addition of cartesian locations to the Motion Location Buffer during the robot motion. • Seam locations obtained from the robot and seamtracking sensor can be added to the Motion Location Buffer after proper filtering. Such seam locations are e.g. computed from synchronised data of robot joint angles and sensor measurement 100 60 40 20 0 0 1 2 Time [s] Trajectory 1 r=50 mm 5 4 2 3 0.1 0 −0.1 0 x 2 Time [s] −0.2 6 y 1 0.2 d=50 mm Current location 0 0 3 Figure 4 : Reference and measured tip velocity (Linear [mm/s], Rotational [deg/s]). The linear velocity decreases at the corner, because of the bounds in rotational velocity. 7 Setpoint 60 40 20 Tip position error [mm] Methods Tip velocity Figure 1 : Real-time control architecture 3 Figure 2 : Corner trajectory Eighth Engineering Mechanics Symposium, Nov. 2005 1 2 Time [s] 3 Figure 5 : Measured tip position error (X, Y, Z) based on joint measurements. The tip position error is largest at the acceleration/deceleration phases and when the joint velocity changes sign (joint friction). References 1. M.W. de Graaf et al, Real-time trajectory generation for sensorguided robotic lase welding, Submitted to the IFAC Syroco 2006 conference, Bologna, 2006 18 Improved tip tracking for an industrial robot using Iterative Learning Control W.B.J. Hakvoort*, R.G.K.M. Aarts, J.B. Jonker * Netherlands Institute for Metals Research University of Twente, Laboratory of Mechanical Automation Institute of Mechanics, Processes and Control - Twente P.O. Box 217, 7500 AE Enschede, The Netherlands phone: +31-(0)53-4895442, email: [email protected] Introduction The laser welding process puts high demands on the manipulator that moves the laser beam with respect to the weld seam. Typically an accuracy of about 0.1 mm is required at speeds beyond 100 mm/s. From an industrial perspective the use of six-axes industrial robots is attractive as these can access complex three dimensional seam geometries. However, using standard industrial controllers the tracking accuracy of these robots is insufficient for laser welding. Objective The goal of this project is to improve the tracking accuracy at the tip of an industrial robot (Figure 1). welding head camera laser diode weld seam product Figure 2 : Seam tracking sensor Results -660 seam 0.1mm boundary [x25] trial 0 [x25] trial 5 [x25] y-axis [mm] -680 -700 -720 -740 -760 450 500 550 600 x-axis [mm] Figure 1 : The six-axes industrial Stäubli RX130 robot Methods The tip tracking accuracy is improved with Iterative Learning Control [1]. • An initial reference trajectory for the robot joints is obtained from CAD-data of the seam and a kinematic model of the robot. • The robot joints track the reference trajectory, while an optical seam tracking sensor (Figure 2) measures the tip tracking error. • Using a straightforward model of the robot the reference trajectory for the joints is updated to compensate for the measured tip tracking error. • The last two steps are repeated until the tip tracking error converges to a steady value. Eighth Engineering Mechanics Symposium, Nov. 2005 Figure 3 : Tracking error perpendicular to weld seam Discussion Iterative Learning Control can reduce the tracking error at the tip of an industrial robot, measured with an optical seam tracking sensor, close to 0.1 mm. The remaining error is mainly due to resonance vibrations of the robot. Further reduction of the error requires either a better robot model or smooth trajectory generation. References 1. Hakvoort, W.B.J., R.G.K.M. Aarts, J. van Dijk, J.B. Jonker, Iterative Learning Control for Improved End-effector Accuracy of an Industrial Robot., submitted for the 8th IFAC Symposium on Robot Control, September 2006, Bologna, Italy. 19 Reduction of sound transmission through panels by means of tuned acoustic resonators M.H.C. Hannink, Y.H. Wijnant and A. de Boer Institute of Mechanics, Processes and Control - Twente Chair of Structural Dynamics and Acoustics, University of Twente P.O. Box 217, 7500 AE Enschede, The Netherlands phone +31-(0)53-4895618, email [email protected] Introduction When a panel is acoustically excited by a sound source, it will vibrate (Figure 1). Due to this vibration, sound is radiated to the other side of the panel. This phenomenon is called sound transmission and is mostly unwanted. The radiated sound can be reduced by means of acoustic resonators (Figure 2), which are tuned in such a way that the volume flow at the entrance of the resonators is opposite to the volume flow at the surface of the panel (Figure 3). Sound source Vibrating panel Incident Radiated Reflected Resonator Panel Figure 3 : Normal incidence transmission of sound through a characteristic area Results The sound transmission loss T L is a measure for the sound reduction, and is defined as the ratio of the incident and the radiated sound power: T L = 10 log10 |pA /pC |2 Receiving room Figure 4 shows the sound transmission loss for panels with different porosities Ω and a resonator length of 0.11 m. Transmission loss [dB] 120 Figure 1 : Sound transmission between two rooms Objective The reduction of sound transmission through panels by the application of tuned acoustic resonators. Characteristic area Ω=0 Ω = 0.25 Ω = 0.40 Ω = 0.48 100 80 60 40 20 0 5⋅102 Resonator Figure 2 : Part of a panel with acoustic resonators Methods The harmonically vibrating panel is divided into a number of identical characteristic areas (Figure 2). The effect of the acoustic resonators is studied with a one-dimensional model of such a characteristic area (Figure 3). By solving the one-dimensional wave equations, the radiated sound pressure pC is calculated. The radiated sound is minimised by tuning the resonator length and the porosity of the panel. These parameters determine the frequency range in which the radiated sound is reduced and the shape of the spectrum, respectively. Eighth Engineering Mechanics Symposium, Nov. 2005 2⋅103 3⋅103 Frequency [Hz] Panel 3 10 Figure 4 : Normal incidence sound transmission loss for different porosities Discussion Panels with acoustic resonators show a large reduction of the radiated sound over a broad frequency range, compared to a panel of the same mass without acoustic resonators (–). The next step is to study the effect of acoustic resonators on a large scale, both numerically and experimentally. Reference 1. Hannink, M.H.C. et al. (2005) Application of acoustically tuned resonators for the improvement of sound insulation in aircraft, Internoise, Rio de Janeiro, Brazil. 20 Multivariable Frequency Response Functions Estimation for Industrial Robots T. Hardeman, R.G.K.M. Aarts and J.B. Jonker University of Twente, Laboratory of Mechanical Automation Institute of Mechanics, Processes and Control - Twente P.O. Box 217, 7500 AE Enschede, The Netherlands +31-(0)53-4892567, [email protected] Introduction shows that the square wave is effective in preventing unwanted velocity reversals, leading to complicated friction behavior. ė3 (m) (rad/s) The accuracy of industrial robots limits its applicability for high demanding processes, like robotised laser welding. We are working on a nonlinear exible model of 0.1 the robot manipulator to predict these inaccuracies. This 0 poster presents the experimental results on estimating −0.1 PSfrag replacements the Multivariable Frequency Response Functions (MFRF) 5 4 3 2 1 0 time (s) ¤ of the Staubli RX90 robot depicted in gure 1. Future work will be the parametrisation of the frequency response funcFigure 3 : Resulting velocity joint 3 tions based on physical models. 9 8 7 6 MFRF estimation Figure 1 : Stäubli RX90 robot with 3D acceleration sensor Closed loop robot system For stability and safety reasons, the experiments will be carried out in closed loop; see gure 2. The robot controller will drive the robot such that the joint angles e(m) and velocities ė(m) are in agreement with the reference trajectory r and ṙ. The driving torques τ (m) are perturbed with feedforward torques τ (f f ) having a frequency spectrum above the bandwidth of the closed loop system. The (m) outputs lacements(m)of the system are the joint angles e and velocities ė . Furthermore a 3D acceleration sensor measures the accelerations of the tip in horizontal s(h) and vertical s(v) direction. + - Controller + e(m), ė(m) Figure 2 : Closed loop system where G is the MFRF. To be able to extract G from data m different (independent) experiments are needed, were m is the number of inputs. The data vectors from different experiments can than be collected into matrices U and Y, where each column corresponds to one experiment. An estimate of G can be formed as provided matrix U has full rank. Results (m) s(h,v) Robot The feedforward signal τ (f f ) is a multi sine containing 100 frequencies in the range from 10 to 100 Hz. Crest factor optimisation of this signal has improved the signal to noise ratio of the measurements by a factor 2. The reference trajectory ṙ for the joint velocity is a square wave. In gure 3 the resulting velocity of joint three is given. The gure Eighth Engineering Mechanics Symposium, Nov. 2005 (m) The estimated MFRF from τ2 and τ3 to the outputs (m) (m) (v) (h) ë2 , ë3 , s and s are given in gure 4. The gure shows that although the excitation is only in the vertical plane (see gure 1), the resulting motion is a complicated 3D motion. PSfrag replacements time (s) (m) ė3 (rad/s) Experiment Design (2) Ĝ(ωk ) = Y(ωk )U−1 (ωk ), 1 10 Ĝτ (m) ,ë(m) 2 amplitude r, ṙ τ (ff ) + τ (m) Let U (ωk ) be the Discreet Fourier Transform (DFT) of the input signals consisting of the driving torques τ m at frequency ωk and let Y (ωk ) be the output vector consisting of the DFT of the joint accelerations ë(m) and the accelerations s(h,v) . For periodic signals the following linear mapping holds Y (ωk ) = G(ωk )U (ωk ), (1) 2 Ĝτ (m) ,ë(m) 2 3 Ĝτ (m) ,s(h) 2 Ĝτ (m) ,s(v) −1 10 2 Ĝτ (m) ,ë(m) 3 2 Ĝτ (m) ,ë(m) 3 3 Ĝτ (m) ,s(h) 3 Ĝτ (m) ,s(v) −3 10 0 3 25 50 75 frequency (Hz) 100 Figure 4 : Frequency Response Functions in (rad/Nms2 ) and (m/Nms2 ) 21 Robust Model Order Reduction by Krylov subspace methods Pieter Heres, Wil Schilders Eindhoven University of Technology Center for Analysis, Scientific computing and Applications P.O. Box 513, NL 5600 MB Eindhoven phone +31-(0)40-2475546, email [email protected] Large linear time-invariant models can very well be replaced by smaller equivalents. Examples of such models can be found in many areas, as for instance in CFD, Structural Mechanics, circuit simulation and electromagnetism. The smaller models have the same behaviour as the large models and the reduction techniques preserve the stability of the models. In this research we investigated a robust and efficient algorithm for Krylov subspace methods for Model Order Reduction (MOR). We here show a model from the SLICOT benchmark library [1], in our opinion the one which is most hard to approximate. The original model (size 480) and its apprximation of size 80 are shown in the picture below. In the next picture we zoomed in. 0 −50 Magnitude transfer function Introduction Objective From the host of MOR methods, Krylov space methods are popular since they are relatively cheap and can be generally applied. The methods can be applied to any system of ODE’s or DAE’s: C −100 −150 −200 −250 −300 −1 10 dx(t) = −Gx(t) + Bu(t) dt y(t) = BT x(t) 1 0 10 Frequency (Hz) 10 3 2 10 10 −30 To ensure robustness, the formulation of the system and the numerical process of orthogonalisation should be handled with special care. Preferable C and G are positive definite and positive real, respectively. Magnitude transfer function −40 Methods The basic idea is to project the large system onto a smaller space. Accordingly, the following Krylov space is generated: Kq (B, A) = [B, AB, . . . , Aq B] −50 −60 −70 −80 −90 −0.6 10 −0.4 10 −0.2 10 Frequency (Hz) 0 10 Let V be the orthonormal basis of this space, then the matrices of the reduced system are formed as: e = VT GV G e = VT CV C e i = V T Bi B e o = V T Bo B Results With a robust implementation of the aforementioned methods models can be fairly reduced, while preserving their behavior. The largest model we reduced was a dense model of size 5200, which had 4 ports. It was reduced to size 40, which gave an excellent approximation of the model up to the maximum frequency of 30 Ghz. Eighth Engineering Mechanics Symposium, Nov. 2005 Discussion Krylov subspace methods are able to preserve the stability of the model. Moreover, a stable reduced model can be formulated in terms of positive RLC-components. Consequently, the compact model can be analysed quickly, by a tool tailored for circuits. References 1. 2. 3. Y. Chahlaoui and P. Van Dooren, 2002. Available at: www.win.tue.nl/niconet/NIC2/benchmodred.html Check www.win.tue.nl/smurf for more references This research was granted by NWO, grant number 635.000.010. 22 High-speed impact modelling and testing of composite structures B.D. Heru Utomo Faculty of Mechanical, Maritime and Material Engineering Mechanics of Materials TU Delft, Mekelweg 2, NL 2628 CD Delft Tel: +31-(0)15-278 6512, e-mail: [email protected] Introduction Methods Composite structures are nowadays more and more applied in many defence applications, such as in vehicle and personal protection (figure 1). To obtain a thorough understanding of the failure mechanisms that occur, techniques like microscopy (figure 2) are used. Ballistic experiments will be performed. During these experiments, the material behaviour will be recorded with equipment such as high-speed (video)cameras (figure 3) and Doppler radars. The development of a predictive tool will be done in parallel to this process. Figure 1: Personal protective equipment. Helmet (left), bullet proof vests (right) Therefore, the availability of (computer) tools that are able to predict their behaviour under ballistic loading has become indispensable to guarantee a continuous improvement of quality and performance. Figure 3: High speed camera images of ballistic impact Results Objectives The objectives of the present research are to obtain a thorough understanding of the failure events that occur in composite structures that are loaded by ballistic impact and to create a tool that is able to predict the behaviour of composite structures subjected to such loading. Predicting the behaviour of ballistically loaded composites seems promising (figure 4). For a good prediction however, more information on the failure mechanisms is still required. Figure 4: Simulation of projectile impact on composite target plate Discussion Figure 2: SEM image of composite plate loaded by a fragment Eighth Engineering Mechanics Symposium, Nov. 2005 The (prediction of) failure mechanisms of composites under ballistic loading are complicated. The question is, on which level (macro, meso or micro) the predictions should be done such that they are still accurate enough. 23 Design and development of high temperature coating system for engines Thomas S. Hille Delft University of Technology Faculty of Aerospace Engineering P.O.Box 5058, NL 2600 GB Delft phone +31-(0)15-2781528, email [email protected] Introduction The efficiency of high temperature engines, such as airplane jet-turbines, increases with service temperature, whereas the mechanical strength of the employed structural components (substrate) generally decreases. To allow for higher service temperatures, the substrate is protected against melting and oxidation by a thermal barrier coating (TBC) system. This system provides a temperature difference of about 200 ◦ C between the external surface and the substrate surface. s o m e mm R = some mm Figure 2 : A schematization of the TBC-substrate system corresponding to the blade geometry. Preliminary Results Hot gas 1200°C Te m pe ra tu re gr ad ie nt 300 μm T B C L a y e r TGO B C Substrate cooling system 1000°C Figure 1 : A micrograph section of a TBC-system showing the coloumnar structure of the TBC-layer on top, the bond coating (BC) adjacent to the substrate and the thin thermally grown oxid (TGO) in between. Due to a mismatch in the thermal expansion coefficients of the coating components, stresses evolve when the temperature distribution through the TBC departs from the stress-free reference state. A thermo-mechanical model that has been coded in C++ computes these stresses for a deviation of about 1200 ◦ C from the reference state. Locally, the von Mises stress exceeds the yield strength of the BC significantly, which motivates the implementation of a plasticity model for this component. von Mises stress [Pa] 2.87e9 Objective Development of a numerical tool to simulate the thermal, chemical and mechanical processes occuring in the TBC-system during life time. Special attention is dedicated to the simulation of spallation mechanisms, which lead to failure of the coating system. 1.0e8 Figure 3 : A contour plot of the von Mises stress. The yield strength of the BC relates to light yellow. Future Work Required Models • • • • TBC-layer: Transversely isotropic. TGO: Thermodynamic kinetic growth law. BC: J2 -plasticity / single-crystal plasticity. Fracture mechanics: Cohesive zones combined with the partition of unity method. • Multi-scale approach for different length scales. • Fatigue model for cyclic loading characteristics. Eighth Engineering Mechanics Symposium, Nov. 2005 A J2 -plasticity model for the BC, using a closest point return-mapping algorithm, has already been coded and will be integrated in the simulation. Attention will be given to the transversely isotropic TBC-layer. Acknowledgement The Netherlands Institute for Metals Research is gratefully acknowledged for financing this project. 24 Design of a Hybrid Vehicle: From generic to specific design Theo Hofman Eindhoven University of Technology Department of Mechanical Engineering P.O. Box 513, NL 5600 MB Eindhoven phone +31-(0)40-2474132, email [email protected] Introduction A schematic overview is given of different applicable Drivetrain hybridization implies adding a Secondary power technologies for the S (Battery, Flywheel and Super source (S) to a Primary power source (P) in order to Capacitor storage systems with energy conversion and improve driving functions, i.e., fuel economy, emissions, transmission components) and the T in Figure 3. driveability, comfort and safety. Designing a hybrid $$ drivetrain fulfilling the required vehicle driving functions is a $- . * + complex task, due to unknown sensitivity of vehicle !" !" ! performance to system component specifications, the / 01 2 34 * + /$ ! defined interaction between system components and the ability to operate the system components at different !" set-points at any time (see Figure 1). !" / / , / & Hybrid functions, e.g., Start-Stop, Brake Energy Recovery (BER), Power Boost, Hybrid and fully Electric propulsion of the vehicle increase with system voltage specification of t h e S (se e Fi g u re 2 ). Th e sh o rt a n d lo n g -t e r m e n e r g y/ p o we r r e q u i r e m e n t s , e . g . , a c c e l e r a t i o n performance, gradeability, towing capacity, electric driving, determine the size of the P and S. The ability to operate the components at optimal operation points and the maximum power throughput determine the design specifications for the transmission technology (T). " ! ! " ! ## ! $ $ !" % & ' , 6"1 7 " #/ $ 6 /$ 7 ! " '( )) * + * + 1 6!" 7 Development of the design method % , ! 5 " ! # !" Research steps for developing a method are: (i) Defining order of design steps, (ii) Sensitivity analysis of design specs for P and S to targets. Selecting and designing of new technologies for P, S and T. (iii) Sequentially optimization of system components (S, T) and the vehicle system control (C) to targets. (iv) Finally, designing an optimal hybrid drivetrain according to specs. A cascaded design process is proposed (see Figure 4). The design order is defined by the sensitivity of the total system efficiency to component efficiency and the decrease in functional constraints of the sub-system components. !! !! *+ // $$ - → $$ 33 $%& ! % '( // .*! " '( *+ → , !! ! - !! " *" + $ , // $& '( $$ && '( 3 33 → )& '( 33 !! // 33 Conclusion $$ *+ , // 8 ! // 33 ' 3 # / ! A design process for designing a hybrid vehicle is discussed. Currently, the proposed design method and the developed design optimization tools are evaluated on the design of a hybrid drivetrain for a mid-sized passenger car (1134 kg) focused on significant fuel use and emissions reduction (>50%) on a representative drive cycle. Eighth Engineering Mechanics Symposium, Nov. 2005 25 The Mechanical Behaviour of Brain Tissue M. Hrapko, J.A.W. van Dommelen, G.W.M. Peters, and J.S.H.M. Wismans Eindhoven University of Technology Faculty of Mechanical Engineering P.O. Box 513, 5600 MB Eindhoven phone +31-(0)40-2475701, email [email protected] Introduction Model Brain injury is the major cause of fatalities in traffic accidents. Finite Element (FE) models are being developed (Figure 1, right), in order to predict the mechanical response of the contents of the head during impact. However, they still lack an accurate descriptions of the mechanical behaviour of brain tissue. Objective The purpose of this study was to experimentally characterise the mechanical behaviour of brain tissue and to develop a constitutive model that is able to describe the behaviour of brain tissue in: • large strain behaviour (up to a strain of 20%), • complex loading paths (loading/unloading), • different deformation modes (shear - compression). Based on the stress relaxation measurements, a nonlinear viscoelastic model was developed to describe the mechanical behavior of brain tissue (Figure 1, middle). The viscoelastic modes σ dve were modeled by an elastic Mooney-Rivlin model and a Ellis model was chosen to describe the stress-dependence of the viscosity. The elastic part σ de was described by a NeoHookean model modified with a damping function. h i p σ de = Ge (1 − A)exp −C I1 − 3 + A B d Results In the following figure experimental results (black line), the model fit (red line) as well as model predictions (blue line) are shown. 200 80 protocol A 150 40 τ [Pa] 100 τ [Pa] protocol B 60 50 20 0 0 −20 −50 −40 0.2 0.1 0 γ [−] 0.3 0.4 0.5 20 15 10 5 0 25 time [s] 80 150 Figure 1 Schematical representation of FE modeling. −σ [Pa] 40 20 0 An improved eccentric sample placement was used in shear experiments on an ARES II rotational rheometer. Shear measurements consist of: • loading/unloading cycles with a constant shear rate and increasing strain levels (Figure 2 left), • stress relaxation tests with constant shear rate and increasing strain levels (Figure 2 right). protocol A strain strain 100 τ [Pa] Methods time protocol B −σ [Pa] protocol B 60 −20 50 100 0 −100 0 0.05 0.1 −ε [−] 0 −50 −40 −100 0 0.05 0.1 γ [−] 0.15 0.2 0 10 30 20 40 time [s] Figure 3 Top & bottom left: shear tests; bottom right: compression tests. No significant immediate change in mechanical behaviour due to previous deformation was found (protocol A). Conclusions time Figure 2 Testing protocols. Compression experiments were performed on an MTS device and consist of stress relaxation tests with a constant shear rate. Eighth Engineering Mechanics Symposium, Nov. 2005 The model showed: • a good prediction in the loading phase of shear deformation, • a lack of accuracy in the unloading phase, • a partially qualitative description of the behaviour in compression. 26 Grain statistics: an experimental investigation P.J.M. Janssen1,2 , Th.H. de Keijser1 , J.P.M. Hoefnagels2 , M.G.D. Geers2 1 Netherlands Institute for Metals Research 2 Eindhoven University of Technology Faculty of Mechanical Engineering P.O. Box 513, NL 5600 MB Eindhoven phone +31-(0)40-2472857, email [email protected] Introduction Due to miniaturisation ever thinner metal sheets are being processed (figure 1). Consequently, only a few crystals may be present in the sheet. As a result, the properties of the individual grains become more important; grain statistics effects occur. Figure 1: (a) Industrial applications (b) Micro-parts. The obtained grain structure is rather homogeneous and the specimens have a pronounced Cube texture (figure 3). The average grain size measured on the specimen surface is about 800 µm. Figure 3: Microstructure and {001} pole figure of a typical recrystallised specimen. Objective Investigation of grain statistics effects, by analysis of the mechanical behaviour of thin Al sheet (320 µm) with through thickness grains. The number of grains in the specimens is altered by changing the specimen geometry. Mechanical behaviour The mechanical properties are analysed in uniaxial tension. Two specimen widths and lengths have been analysed, preliminary results are shown in figure 4. 40 Microstructure b 35 30 30 25 25 σtrue (MPa) σtrue (MPa) A strain-anneal protocol is followed to produce material with a reproducible microstructure. From the asreceived sheets, strips are cut parallel to the rolling direction. These strips are recovered at 200 o C for one hour to relieve internal stresses. As expected, there are no significant differences in the grain structure and orientation (figure 2). 40 a 35 20 15 10 10 5 0 0 20 15 w = 9.8 mm w = 2.1 mm 0.05 0.1 0.2 0.15 εtrue 0.25 0.3 0.35 5 0 0 l = 10 mm l = 16 mm 0.05 0.1 0.2 0.15 εtrue 0.25 0.3 0.35 Figure 4: Stress-strain curves, (a) varying width and constant length (b) varying length and constant width. Decreasing the specimen width: • the spread in stress-strain behaviour increases • small steps in the tensile curves become visible • explanation: increasing individuality of the grains Increasing the specimen length: • the average fracture strain decreases • the spread in fracture strain increases • explanation: increasing probability of the presence of weaker grains Figure 2: Microstructures and {001} pole figures of (a) asreceived material (b) recovered material (grain colour indicates crystal direction parallel to the specimen normal). The recovered strips are strained 5.5 % in uniaxial tension and recrystallised for 30 minutes at 600 o C. Eighth Engineering Mechanics Symposium, Nov. 2005 Future work Further exploration of the mechanical behaviour and the local deformation (using digital image correlation) of specimens with known microstructure. 27 Flow front tracking in aluminium extrusion dies by means of particle trajectories A.J. Koopman, H.J.M. Geijselaers, J. Hu étink Institute of Mechanics, Processes and Control - Twente University of Twente P.O. Box 217, 7500 AE Enschede, The Netherlands phone +31-(0)53-4893405, email [email protected] Introduction Methods Increasingly tighter requirements on complexity and geometric tolerances raise the demand for more insight in the aluminium extrusion process. While an optimized extrusion process should create not more then 15% scrap, it is in everyday practice not uncommon to have a scrap rates over 25%. A consistent application of design rules developed either by simulations or by experience shows a dramatical increase in the performance of extrusion dies. In 1986 Thompson [1] introduced a pseudoconcentration function C(X, t) to track the flow front. Objective Billet to billet extrusion is not a continuous process. Temperature, stresses and deformation rates change during the process. Figure 2 : Extrusion force from Benchmark Zurich The discontiniuty of the pseudo-concentration function around the flow front introduces inaccuracy during the convection through the FEM mesh. To overcome this problem we introduced a smooth function, the original coordinate function (figure 3) The original coordinates are transported through the mesh and every step the stiffness matrix is adjusted based on the original coordinates. . Figure 1 : Filling of the extrusion dies However, the aluminium flow through the die after initial start-up can be regarded as a semi-stationary process. Shown in figure 2, after the first 50 mm the extrusion force is decreasing linearly with the ram displacement, in literature this effect is attributed to the decreasing friction between the billet and the container. Due to the stationary nature of the aluminium flow an Eulerian FEM formulation is used to obtain the results. Using an Eulerian description the high deformations in the extrusion dies can be simulated without the need for constant remeshing. In the extrusion practice the suspicion is raised that the die deflection during the filling affects the aluminium flow in the semi-stationary part. To simulate this effect the aluminium flow front has to be tracked in the Eulerian domain. Eighth Engineering Mechanics Symposium, Nov. 2005 Figure 3 : Pseudo-concentration function versus original coordinate function Discussion Simulations using this method will help with the development of new design rules for extrusion dies. References 1. Thompson, E. (1986). ”Use of pseudo-concentrations to follow creeping viscous flows during transient analysis.” Int. J. Num. Meth. Fluids 6(10): 749-761. 28 A more Rigorous Coupled-Wave Analysis M.G.M.M. van Kraaij Technische Universiteit Eindhoven Department of Mathematics and Computer Science P.O. Box 513, NL 5600 MB Eindhoven phone +31-(0)40-2472685, email [email protected] Introduction Analysis The semiconductor industry uses lithography systems for manufacturing complex integrated circuits (also called ICs) onto wafers. Because the complexity of producing ICs with more functionality increases, good models and simulation tools are needed. Because the grating is infinitely periodic, the computational domain can be reduced to one unit-cell. For each layer i the following equations are solved: Here we focus on a small part of the chip making process that deals with measuring the position of a wafer (’alignment’). Today, small gratings Lithography system on the wafer are used for this alignment step. Gratings are periodic structures printed on the wafer and are even smaller than ICs. These gratings are illuminated with a laser beam and by measuring the diffracted light one can get information on the position of a wafer. Here Maxwell’s equations are the starting point and an algorithm known as Rigorous Coupled-Wave Analysis (RCWA) is then used to calculate the diffracted field [1]. TE polarization ∂2 ∂x2 Ei,y + ∂2 ∂z 2 Ei,y + k02 εri (x)Ei,y = 0, TM polarization ∂ 1 ∂ r εi (x) ∂x εr (x) ∂x Hi,y + i ∂2 ∂z 2 Hi,y + k02 εri (x)Hi,y = 0. Here Ei,y and Hi,y represent the y-components of the electromagnetic field. The complex relative permittivity is denoted with εri (x) and k0 is the wave number of free space. The following steps are taken to calculate the scattered field: • Fourier expansions for permittivity [2]. • Pseudo-periodic Fourier expansions for field inside grating layers. • Rayleigh expansions for scattered field. • Stable enhanced transmittance matrix approach. Objective Computational results Get a better understanding of the optical diffraction model and RCWA algorithm. Improving the convergence and robustness of the algorithm. For a grating an angle-resolved spectrum is calculated with the RCWA-algorithm and compared with a measurement. Both figures show the relative amount of energy that is reflected and captured by a sensor. Assumptions 0.3 0.3 0.25 0.25 Simplifications of Maxwell’s equations are based on the following assumptions: 0.2 0.2 0.15 0.15 • Incident field is an arbitrary linearly polarized monochromatic plane wave. • Electromagnetic fields are time-harmonic. • Media are linear, homogeneous and isotropic. • Gratings are infinitely periodic and are approximated with a layered structure: 0.1 0.1 0.05 0.05 0 Measured image 0 Calculated image References 1. re - y x Real trapezoidal grating ? z Approximated layered grating Eighth Engineering Mechanics Symposium, Nov. 2005 Moharam, M.G. et al. Formulation for stable and efficient implementation of the rigorous coupled-wave analysis of binary gratings. J. Opt. Soc. Am. A, 12(5):1068-1076, May 1995. 2. Li, L. Use of Fourier series in the analysis of discontinuous periodic structures. J. Opt. Soc. Am. A, 13(9):1870-1876, September 1996. 29 A singularity solution of shear faulting in swelling ionised porous media F. Kraaijeveld Eindhoven University of Technology Faculty of Mechanical Engineering P.O. Box 513, NL 5600 MB Eindhoven phone +31-(0)40-2475415, email [email protected] Introduction Results and discussion Shales, clays, gels and biological tissues all endure swelling associated with ionisation of the porous medium [2.]. Phenomena like crack formation and crack propagation during shrinkage are studied through numerical simulations [4.,6.]. Analytical solutions are a must for validation of numerical codes. Shear stress The shear stress shows an high order singularity in x (O ( r12 )). Shear stress at t = 0 is the result of initial incompressible behaviour of the medium. Fluid flow results in relaxation of the stresses. This is consolidation. Shear stress at equilibrium decreases with decreasing osmotic pressure. (fig 2) 0 Objective Shearstress for x = 2.40e−001 with cfc =−2.00e−004 −6 −3.2 x 10 −3.4 A 2D analytical solution for a dislocation in a swelling ionised porous medium. shearstress (Gpa) −3.6 Methods −3.8 −4 −4.2 −4.4 −4.6 −4.8 Model Lanir’s small deformation theory [3] describes the saturated porous medium by assuming incompressible constituents and infinitely fast ion flow leaving a strongly coupled, biphasic model. Perturbation on a homogeneous situation ( = tr ()RT C1 ), pre-stressed or not, gives: ~ tr() r ~ f = 0 ; r2~u + ( + + RT C1 )r tr() K r2 f = 0: t c = 1.50e−004 ex c = 3.50e−004 ex c = 5.50e−004 −5 ex −5.2 0 100 300 200 600 500 400 700 800 900 1000 time (s) Figure 2: Shear stress in crack: consolidation. Chemical potential At compression areas (xz > 0) chemical potential is high and fluid flow is initiated to lower areas (xz < 0) . Chemical potential rapidly decreases further from the crack. (fig 3) =−2.00e−004 at t=7.40e+001 muf(x,z) for cex = 1.50e−004 with cfc 0 −6 x 10 2 Shear faulting A dislocation on an existing crack is described by a displacement jump between z = 0+ and z = 0 [1.] (fig 1). 3 1.5 z (mm) 2 1 1 0.5 0 0 −0.5 −1 −1 −2 −1.5 −3 −2 −2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5 x (mm) Figure 1 : Shear fault: u = f (x ) H(t) Derivation solution Decoupling the systems of equations by means of stress functions [5.] results in r r S = 0; K r E = 0: 2 2 E t Figure 3: Chemical potential: fluid transport Black line denotes crack References 4 The new system is solved using linear transformations (Fourier and Laplace). The shear stress is calculated at the crack surface. Using Simpson rule chemical potential away from the surface is computed. Eighth Engineering Mechanics Symposium, Nov. 2005 1. 2. 3. 4. 5. 6. J.R. Booker, J. of Geophys. Res., 1974 J.M. Huyghe et al., Int. J. of Eng. Science, 1997 Y. Lanir, Biorheology, 1987 R. van Loon et al., Int. J. for Num. Meth. in Eng., 2003 J. McNamee and R.E.Gibson, Quart. J. Mech & Appl. Math., 1960 S. Wognum et al., Spine, in press 30 Topology Optimization of Shape Memory Alloy Thermal Actuators M. Langelaar1 , G.H. Yoon2 , Y.Y. Kim2 and F. van Keulen1 1 Delft University of Technology, Faculty of Mechanical, Maritime and Materials Engineering, Mekelweg 2, 2628 CD Delft phone +31-(0)15-2786506, e-mail [email protected] 2 Seoul National University, National Creative Research Initiatives Center for Multiscale Design, Seoul, Korea Introduction Results Shape memory alloys (SMA’s) are excellent materials for miniature actuation applications involving large forces and displacements, because of their unique properties. However, the complexity of their constitutive behavior complicates the design of effective two- and three-dimensional SMA actuators. Improvements are possible through the use of structured design techniques combined with computational modeling of SMA behavior. Topology optimization has successfully been performed on a number of thermal actuator design problems, using the MMA optimizer [4]. Objective of the optimization is to maximize the stroke of the SMA actuator over a given temperature range. No volume constraint is used. Starting from a uniform material distribution, functional SMA actuators were obtained for various loadcases (see e.g. Fig. 2). Objective The aim of this research is to develop robust and effective topology optimization techniques for the design of SMA actuators. Initially, the focus is on planar SMA structures in plane stress configurations, acting against a constant load, and actuated by uniform changes in temperature. Methods The conventional approach in topology optimization is the so-called density-based method, where material properties of finite elements are functions of the associated design variables [1]. This approach is popular for linear problems, but it has several disadvantages for physically and geometrically nonlinear problems, such as the present SMA actuator case: Figure 2 : Topology optimization problem definition, final structure and evolution during optimization. Furthermore, also the redesign of existing structures has been studied. Fig. 3 shows the result of applying the developed ECP-based topology optimization technique to an SMA thermal actuator problem starting from a given baseline design. An improvement of the output stroke of more than 100% has been obtained. 1. excessive distortion of low-density elements, resulting in lack of convergence, 2. influential arbitrariness in choosing material property interpolation functions, 3. and the need to perform full sensitivity analysis of complex material models. These disadvantages have lead to the development of an alternative formulation of the topology optimization problem, the Element Connectivity Parameterization [2,3]. In this ECP approach not the material properties of elements, but rather the stiffness of zero-length links connecting the elements are controlled by the design variables (Fig. 1). Figure 3 : Baseline and optimized SMA thermal actuator design, and a performance comparison. Discussion Topology optimization of SMA thermal actuators has been performed for the first time. Crucial to this result is the use of the ECP formulation instead of the conventional densitybased approach. Next to the automatic generation of efficient actuator designs, the developed technique is also capable of significantly improving existing SMA designs. References Figure 1 : Mesh layout, element-link patch and zero-length link used in the ECP approach. Eighth Engineering Mechanics Symposium, Nov. 2005 [1] Bendsoe, M.P. and Sigmund, O., Topology Optimization - Theory, Methods and Applications, Springer-Verlag, Berlin, 2003. [2] Yoon, G.H. and Kim, Y.Y., Element connectivity parameterization for topology optimization of geometrically nonlinear structures, International Journal of Solids and Structures 42(7) 1983–2009, 2005. [3] Langelaar, M., Yoon, G.H., Kim, Y.Y. and Van Keulen, F., Topology optimization of shape memory alloy actuators using element connectivity parameterization, 6th World Congress on Structural and Multidisciplinary Optimization, Rio de Janeiro, Brazil, May 30 - June 3, 2005. [4] Svanberg, K., MMA - Method of moving asymptotes - a new method for structural optimization, International Journal for Numerical Methods in Engineering 24 359–373, 1987. 31