Artificial neural network-based resistance spot welding quality assessment system

On-line quality assessment has become one of the most critical requirements for improving the efficiency and the autonomy of automatic resistance spot welding (RSW) processes. An accurate and efficient model to perform non-destructive quality estimation is an essential part of the assessment process. This paper presents a structured and systematic approach developed to design an effective ANN-based model for on-line quality assessment in RSW. The proposed approach examines welding parameters and conditions known to have an influence on weld quality, and builds a quality estimation model step by step. The modeling procedure begins by examining, through a structured experimental design, the effect of welding parameters (welding time, welding current, electrode force and sheet metal thickness) and welding conditions represented by typical characteristics of the dynamic resistance curves on multiple welding quality indicators (indentation depth, nugget diameter and nugget penetration) and by analyzing their interactions and their sensitivity to the variation of the dynamic process conditions. Using these results and by combining an efficient modeling planning method, neural network paradigm, multi-criteria optimization and various statistical tools, the identification of the model form and the variables to be included in the model is achieved by executing a systematic model optimization procedure. The results demonstrate that the proposed approach can lead to a general ANN-based model able to accurately and reliably provide an appropriate assessment of the weld quality under diverse and variable welding conditions.