Hybrid functional machine learning for modeling the effect of Al/steel resistance spot weld process parameters on microhardness curve shape and quality

Quantifying the effect of resistance spot welding (RSW) process parameters, such as clamp force, current levels, and cooling times, on weld microstructure is vital for improving weld quality. One such microstructural feature of interest is the microhardness of the joint interface. When the microhardness of the entire joint interface is plotted as a function of the location, a microhardness curve is obtained. Because microhardness curves can be considered continuous, smooth curves, functional data analysis (FDA) is an appropriate statistical paradigm for modeling the effect of RSW process parameters on microhardness curve shape and quality. This work presents a hybrid functional machine learning approach to exploring and modeling the relationship between RSW process parameters and microhardness curves. Process-microstructure data from two aluminum/steel RSW stack-ups provided by General Motors are considered in this study. For each stack-up, microhardness curves corresponding to different process windows are smoothed, registered, and decomposed into constituent functional principal components. A series of random forest models is applied to characterize the effect of the RSW process parameters on the microhardness curve structures. RSW process parameters are ranked by importance across the different curve regions and constituent shapes. Based on the hybrid functional machine learning approach, ranges and combinations of RSW process parameters that correspond to high-quality microhardness curve shapes are suggested, providing insight to the complex relationships between RSW process parameters, and microhardness curve shapes.