Multi-domain feature-enhanced fault diagnosis method for abrasive waterjet cutting head

Abstract

As the only cold high energy beam processing technology in the world, abrasive water jet (AWJ) has been widely used in various fields. Compared with other processing equipment, AWJ machine tools are more prone to fail. This leads to a great uncertainty in production process, which further limits this potential technology to be applied extensively. By watching the vacuum in the mixing chamber of abrasive cutting head, it is feasible to support fault diagnosis for most types of cutting head faults. However, it is difficult to identify each type of fault because of the overlapping vacuum values across different faults. To address this gap, this study proposes a novel multi-domain feature enhancement framework that integrates time-domain, frequency-domain, and time-frequency features to distinguish overlapping vacuum features. Based on it, an ensemble model, including Random Forest and LightGBM, has been used for fault type classification. And classic algorithms such as KNN and SVR were used for comparative analysis. The four algorithms achieved relatively good classification accuracy, with most of them reaching over 90%. But the result shows that the ensemble model can achieve robust prediction performance, outperforming traditional algorithms. It enables real-time, intelligent fault detection in AWJ systems.