Gaussian process regression to predict the morphology of friction-stir-welded aluminum/copper lap joints

Abstract

The joining of materials with different or even competing properties is of high industrial interest regarding resource-efficient production. Friction stir welding (FSW) has been employed to create high-quality joints of dissimilar material combinations. Many studies report both metallurgical bonding and form-fit to be the relevant joining mechanisms. While metallurgical bonding is driven by interdiffusion and occurs in almost every case, form-fit can only appear if the interface is deformed. The hooks of the deformed interface cause interlocking; however, they also result in an increased stress concentration. Hence, the hooking can either enhance or reduce the joint strength depending on their geometries. This study demonstrates an approach to predict the morphology of the cross-sectional interfacial area of friction-stir-welded multi-material joints. Image processing was used to convert cross sections of aluminum/copper lap joints into binary b/w images. Using Gaussian process regression, a data-driven model of the interfacial area’s morphology was constructed based on 13 data sets. The applicability of the resulting Gaussian process model was tested for seven data sets by comparing the algorithm’s morphological predictions with cross sections welded with test parameters that were not used for training. This allows to estimate, which joining mechanism is relevant or dominant for the overall joint strength. The predicted results agreed well with the actual cross sections. Recesses as well as hooks at the interfacial area were successfully predicted even for a limited number of training data. To enhance the space of possible uses of the model for subsequent applications (e.g., simulation of fracture mechanics), more input parameters can be implemented into the model.