Abstract
Fault diagnosis is among the most crucial steps in maintenance strategies to sustain the health of machine tools. Traditionally, fault diagnosis was performed based on engineers' vast expertise and technical understanding. However, advances in Machine Learning (ML) theories have decreased the role of human specialists in machine fault diagnosis, introducing Intelligent Fault Diagnosis (IFD). IFD approaches have obtained significant attention in academic and industrial applications due to their accuracy and velocity in recognizing machines' health states automatically. This research presents a novel fault identification process that uses an Extra Tree classification algorithm to classify manufacturing process defects with a feature selection approach based on feature importance. This approach is evaluated and compared against multiple machine learning algorithms, including tree-based methods, artificial neural networks, and traditional algorithms such as the Support Vector Machines (SVM). The assessment results confirm that the proposed algorithm can achieve an accuracy of above 99% in the classification task and significantly improve training time and computational resource efficiency. The proposed algorithm also enables researchers to analyze the causality of each fault based on the influential features. Further instructions to continue this line of research are correspondingly presented to enhance the proposed approach by using novel transfer learning and generative approaches.