| dc.description.abstract | This thesis concerns the behaviour and modelling of multi-layered joints. The aim of the thesis is to provide a fundamental understanding of the behaviour of mechanical joints and their numerical modelling. The work mainly consists of experimental and numerical studies of joints between three aluminium sheets with self-piercing rivets (SPR).
The quasi-static behaviour of SPR connections between three stacked aluminium sheets has been investigated through an experimental campaign. The effects of sheet thickness and stacking on joint strength and ductility were studied. Connections between multiple sheets significantly increase the number of possible loading modes, requiring meaningful and controllable tests. This was achieved by developing an experimental set-up for challenging threesheet connections using cross tests. Numerical modelling techniques for multisheet connections in large-scale analyses are scarcely documented compared to general two-sheet connections. A numerical modelling framework for multisheet connections has been established. By applying sandwiched constraints between the parts, the complex failure behaviour of three-sheet joints could be reproduced and the experimental setup allowed the model to be calibrated with a reasonable number of tests.
A virtual calibration procedure for large-scale models of multi-sheet SPR joints was presented. The framework included riveting process simulations, detailed mesoscopic models of SPR unit tests, called virtual tests, and a complex component test. The virtual tests were calibrated against benchmark tests, namely peeling, single-lap joint and an additional cross test. The large scale multi-sheet joint models were calibrated with the virtual cross tests and their accuracy was compared to the experiments. The virtually calibrated large scale models showed satisfactory results when applied to component test simulations. Given the constitutive behaviour of the materials used and the known SPR joint geometry, the proposed virtual calibration strategy and its experimental validation proved to be a promising approach.
Finally, a neural network (NN) modelling approach for application to largescale joint models was presented. Different NN architectures were trained on data generated from a constraint model formulation found in the literature. The NN models were incorporated into an explicit finite element code and were able to predict the local force-displacement response throughout the simulations. One NN design was found to be flexible enough to represent the mechanical behaviour of different SPR and flow-drill screw connections, given their respective training data. The results encouraged the use of NN modelling techniques for joint representations in explicit FE solvers. | en_US |
