Effects of Hard-to-Measure Material Parameters on Clinching Joint Geometries Using Combined Finite Element Method and Machine Learning

Abstract

In this article, we investigated the effects of material parameters on the clinching joint geometry using finite element model (FEM) simulation and machine learning-based metamodels. The FEM described in this study was first developed to reproduce the shape of clinching joints between two AA5052 aluminum alloy sheets. Neural network metamodels were then used to investigate the relation between material parameters and joint geometry as predicted by FEM. By interpreting the data-driven metamodels using explainable machine learning techniques, the effects of the hard-to-measure material parameters during the clinching are studied. It is demonstrated that the friction between the two metal sheets and the flow stress of the material at high (up to 100%) plastic strain are the most influential factors on the interlock and the neck thickness of the clinching joints. However, their dependence on the material parameters is found to be opposite. First, while the friction between the two metal sheets promotes the formation of the interlock, it reduces the neck thickness and thus increases the risk of breaking in this region. Second, it is easier to form the interlock if the deformed material exhibits small flow stress at high plastic strain, but the neck thickness tends to be thinner in this case. The identified material parameters help to significantly reduce the relative error between the simulated results and the experimental results, not only in the configurations from which they are identified but also in a new configuration. This methodology shows its potential in the cases where material parameters are not available or difficult to measure.