Data-driven and simulation models for resistance spot welding

Resistance Spot Welding (RSW) process is a technique used to join overlapping and thin metal sheets, and it is widely employed in the automotive and railway sectors. In this process, a current flow passes through a pair of electrodes pressed against each other, utilizing the heat generated by Joule effect to perform welding. In such a process, maintenance costs significantly impact the overall expenses. Therefore, thanks to the dissemination of Industry 4.0 paradigms and the use of sensors collecting data from process signals, it has become possible to develop data-driven models for implementing predictive maintenance. In this thesis work, based on real data collected in an experimental campaign at the J-Tech laboratory of the Polytechnic University of Turin, regression models were developed to predict welding outcomes related to two target variables, namely the welding diameter and the tensile-shear load. These two target variables estimated through data-driven models are among the most important characteristics that are monitored to define the quality of a resistance spot welding. Predictive variables were extracted from mechanical signals such as electrode force and electrode displacement. Regression algorithms were chosen to make predictions on the values of the target variables. Results of the data-driven approach were then compared with those obtained by performing Finite Element Method (FEM) numerical simulations using the Sorpas® simulator. Sorpas® has been used with the aim of identifying the parameter range where data-driven models perform best and the range of parameters where Sorpas® performs better. Through this simulator, the same welds performed in the laboratory were simulated, exactly replicating all input parameters.