Key points are not available for this paper at this time.
The use of destructive testing can lead to bottlenecks in manufacturing due to the high costs associated with scrap. To address this issue, smart factory methodologies are being developed, which utilize artificial intelligence (AI) alongside traditional manufacturing processes to improve productivity and produce defect-free products. In this study, we examine the predictive accuracy of various machine learning (ML) algorithms in forecasting the durability of resistance spot welding (RSW) welds, a crucial aspect for the advancement of intelligent manufacturing environments. It explored a range of ML methods – such as multi-linear regression (MLR), Gaussian process regression (GPR), artificial neural networks (ANNs), support vector regression (SVR), and random forest (RF) – by analyzing RSW experiments. We used a dataset of 435 cases comprising diverse materials, design parameters, and process conditions. The dataset covered a wide range of metal thicknesses, welding currents, forces, times, and electrode diameters, providing a robust foundation for developing a data-driven AI chatbot that can assist in property-specific process parameter design. The findings reveal that ML algorithms can accurately estimate the strength of spot welds. Among the algorithms tested, RF proved to be the most effective, outperforming MLR, GPR, ANNs, and SVR, with an MSE and R2 of 0.29474 and 96.35%, respectively, with the largest training dataset (90% of the total data). On the other hand, the MLR model performs poorly, with an MSE of 1.93182 in 90% of the training data. This study demonstrates ML’s ability to predict metal strength, enabling quality assessment without destructive testing.
Dang et al. (Sat,) studied this question.