Neural Network-Based Resistance Spot Welding Control and Quality Prediction

This paper describes the development and evaluation of neural network-based systems for industrial welding process control and weld quality assessment. The developed systems utilize recurrent neural networks for process control and both recurrent networks and static networks for quality prediction. The first section describes a system capable of both welding process control and real-time weld quality assessment. The second describes the development and evaluation of a static neural network-based weld quality assessment system that relied on experimental design to limit the influence of environmental variability. Relevant data analysis methods are also discussed. The weld classifier resulting from the analysis successfully balances predictive power and simplicity of interpretation. The results presented for both systems demonstrate clearly that neural networks can be employed to address two significant problems common to the resistance spot welding industry, control of the process itself, and non-destructive determination of resulting weld quality. INTRODUCTION Several factors may influence the quality of a forming resistance spot weld. Among the more important of these are failures in weld tip geometry, improper alignment of welder electrodes and metal surfaces to be joined, dirt and corrosion on the electrodes andor metal surfaces, and uncompensated variations in AC supply voltage. Each of these may influence the principal variables (the welder output current and weld tip voltage drop) to which the weld controllquality assessment system is permitted to have access. Of primary importance is the signature defined by the temporal variations of these two quantities. The first task of the work reported here was to develop a system capable of controlling the resistance spot welding process in real time. Realization of such a capability depended critically on development of methods for detecting, and compensating for, influential factors on a short enough time scale to support modulation of an evolving weld. The recurrent neural network system developed for this purpose is discussed in the next section. The second task was to develop a system suitable for performing a posteriori weld quality assessment. In one approach, the recurrent neural network system developed as part of the first task was adapted to produce evolving evaluations of welding signatures and to extract from them a measure of weld quality. In an alternative approach, a static neural network was incorporated in an adaptive system that proved capable of determining weld quality by predicting key characteristics of the weld nugget size and indentation and by the subsequent mapping of these characteristics onto a pass/fail classification. EXPERIMENTS IN WELD EVOLUTION CAPTURE FOR PROCESS CONTROL AND QUALITY ASSESSMENT Critically important to the production of quality spot welds is the application of appropriate welding current for a time sufficient to ensure the formation of the weld nugget and short enough to avoid undesirable effects (e.g., excessive denting of the material surfaces and thinning of the weld region due to the prolonged application of electrode pressure). A reliable indicator of the onset of these and other processes is the expulsion of weld material that occurs when the forming weld nugget begins to exceed in volume that which can be effectively contained within the electrode boundaries. Welding current should be (indeed, should already have been) removed when this onset is