Frequency-Domain Analysis of Electrode Force Signals in Resistance Spot Welding for Quality Evaluation

Resistance Spot Welding (RSW) is one of the joining techniques widely used in different industries, mainly in the automotive industry. It is popular because of its ease of use, speed, reliability, cost-effectiveness, and opportunity for automation. Traditionally, destructive testing methods are used to assess the quality of the welds, which are time-consuming, often impractical for real-time monitoring, and expensive. To overcome these challenges, this thesis proposed a novel way to predict nugget size based on Machine Learning (ML) methodology, using features derived from electrode force signals processed through the Fast Fourier Transform (FFT). The research begins by analyzing 50 records of electrode force signals in frequency domains. Fast Fourier Transformation (FFT) was used to transfer the signal from Time-Domain to Frequency-Domain. Two feature selection methods, Pearson Correlation and Recursive Feature Elimination (RFE), were employed to choose the most representative features from extracted features related to electrode force in Frequency-Domain. Then, six ML models were employed: Linear Regression (LR), Decision Tree (DT), Random Forest (RF), Support Vector Regression (SVR), and K-Nearest Neighbor (KNN). Moreover, the validation was done based on Train-Test Split, 5-Fold Cross-Validation (5-FCV), and Leave One-Out Cross-Validation (LOOCV), where Mean Absolute Percentage Error (MAPE) and Mean Square Error (MSE) were calculated. The results demonstrated that tree-based models and RFE-based approaches validated with LOOCV provided the best performance, considering the dataset size. Also, a comparative analysis between Time-Domain features, Frequency-Domain features, and the combination of both revealed that Frequency-Domain based features are better representatives with more power of prediction for weld nugget size estimation. This research highlights the potential of the FFT signal analyses combined with ML tools for real-time monitoring of RSW quality, which will lead to more intelligent and efficient manufacturing.