Development of a Machine-Learning Model for monitoring the Friction-Stir-Welding Tool using Vibration Signatures

Friction Stir Welding (FSW) forms high-strength welds with outstanding mechanical characteristics. FSW differs from traditional welding techniques by operating at temperatures below the melting point of the materials. Therefore, condition monitoring of the tool is essential to seamlessly utilize FSW in manufacturing and its applications. For this purpose, a machine learning model is employed. Machine learning models play a pivotal role in identifying any wear and tear in the tool by predicting failures before they occur. The process starts with collecting raw data at three different speeds (in RPM) involving real-time vibrational data. This paper is based on a histogram approach to data analysis, using machine-learning (ML) algorithms such as Random Forest, Random Subspace, Naive Bayes, Naive Bayes Multinomial, Logistic Model Tree (LMT), KStar, Logistic Regression, IBK, J48, and PART. Among all classifiers, Random Forest provides the best results.