Machine Learning Modeling of Dynamic Strength of Resistance Spot Welds in High Strength Steels

Abstract

Abstract Electric resistance spot welding technology has been extensively applied to join sheet metal parts in the automobile industry to decrease vehicle weight and increase its integrity. Due to the dynamic nature of motor vehicles, it is necessary to measure the dynamic strength of spot welds. Both the traditional regression approach and the state-of-the-art machine learning method are utilized in this paper to quantify the dynamic strength of spot welds based on a large database for different steels, where the dynamic strength is nominally defined as a peak load normalized by specimen thickness and weld nugget size at a specific loading rate. Experimental results showed that the dynamic strength of spot welds is a function of the loading rate, specimen thickness, spot weld nugget size, specimen type and steel grade. To formulate the dynamic strength, the traditional regression needs to simplify the strength as a function of loading rate only, and then determine it using the best curve-fitting approach. In contrast, the machine learning method, based on an artificial neural network with built-in learning functions and algorithms, can determine a multi-variable function of dynamic strength to describe the effect of the loading rate, specimen thickness, and spot weld nugget size. To demonstrate the capability of the machine learning technology, a simple and a more complex neural network architecture are adopted to predict the dynamic strength of spot welds. The final machine learning results, advantages and disadvantages are evaluated in comparison with the regression approach.