ABSTRACT of the Doctoral Thesis
The use of materials in modern lightweight car bodies is becoming increasingly complex. As a consequence, studying the behavior of such materials, of components made of them, and of the relevant production, assembly and fastening processes has taken on prime importance. Traditionally, much of this work has been based on experimental procedures. In the past decades, numerical simulations, usually based on finite element methods, have become well accepted and extensively used engineering tools in the development process of structural design and optimization in order to reduce both the costs and the complications in experiments.
Numerical simulations have given excellent results in predicting the behavior of car bodies, but much work remains to be done, e.g., on complex materials such as composites and ultra-high strength steels and on fastening, e.g., on spot welds, adhesive joints, rivets, etc., which continue to pose major challenges to researchers and scientists especially in reproducing their complex failure behaviors. When modeling the damage and failure of components, details of the models can be of major importance. One aspect limiting the accuracy of the numerical simulations are the underlying material models and the parameters used in them, the latter being typically difficult to measure and often being of a stochastic nature.
This thesis aims at introducing and developing some non-traditional computational methods into modeling the mechanical behavior of spot welds, an important joining technique that is currently used in fabricating virtually all car bodies. The studies are based on the capabilities for the mesh-independent modeling of fasteners in the explicit FE code ABAQUS/Explicit that are provided by CONNECTOR elements in combination with the FASTENER feature. Spot welds joining sheets of a number of steel grades and various different loading conditions are investigated by a number of approaches, all of which make use of artificial neural networks (ANN) and other soft computing methods, and the simulation results are compared with the force-displacement responses of tested specimens.
One part of this thesis explores the use of ANN for identifying material parameters for ABAQUS/Explicit models employing the CONNECTOR/FASTENER option. The work is based, on the one hand, on results from series of experiments on specimens that use single weld spots to join two steel sheets. On the other hand, the range of behaviors attainable by the CONNECTOR/FASTENER model was probed using a stochastic simulation. On the basis of this data ANN were trained to give material parameters for the ABAQUS model, and the suitability of the resulting parameter sets for practical use was assessed.
In the second part of the work, a beam-like user-defined element (VUEL) for ABAQUS was implemented. In an initial version it used Timoshenko kinematics, whereas in later version the beam's non-linear axial stiffness was controlled by a global ANN. This ANN had been generated and trained to describe the behavior of spot weld specimens under a normal tensile loading condition. Even though special provisions had to be made to describe the local behavior of the weld spot, which is required in the VUEL, by a global ANN, an initial proof of concept of controlling a finite element describing the nonlinear behavior of a weld spot was successfully provided.