In this work, the quality of resistance spot welding (RSW) joints of 304 austenitic stainless steel (SS) is assessed from its tensile shear load bearing capacity (TSLBC). A predictive model using a polynomial expansion of the relevant welding parameters, i.e. welding current (WC), welding time (WT) and electrode force (EF) and elastic net regularization is proposed. The predictive power of the elastic net approach has been compared to artificial neural networks (ANNs), previously used to predict TSLBC, and smoothing splines in the framework of a generalized additive model. The results show that the predictive and classification error of the elastic net model are statistically comparable to benchmarks of the best pattern recognition tools whereas it overcomes correlation problems and performs variable selection at the same time, resulting in a simpler and more interpretable model. These features make the elastic net model amenable to be used in the design of welding conditions and in the control of manufacturing processes.
