Optimizing spot welding monitoring using sample entropy and advanced AI methods

Abstract

In the welding field, non-destructive testing (NDT) is essential to ensure welding consistency. However, decision-making based on NDT results is not straightforward, necessitating the use of tools based on artificial intelligence, which can save time and reduce costs. This study proposes a feature extraction method based on sample entropy from acoustic emission (AE) signals captured by three piezoelectric sensors. To avoid redundancy of information and minimize irrelevant features, the maximum relevance minimum redundancy (mRMR) algorithm is applied to create a more effective feature vector. Finally, the extracted discriminative features are input into the decision tree (DT) and k-Nearest Neighbors (KNN) models for classification. The AE signals were obtained during the refill friction stir spot welding (RFSSW) process. The results demonstrate that mRMR algorithm effectively reduces the number of extracted features while maintaining or even improving improving classification accuracy. The obtained results highlight the sensitivity of the proposed method to different welding conditions, which can help improve welding processes.