Remaining Useful Life Prediction of Cutting Tools with an Optimal Feature Construction Method of DiCCA

Remaining useful life (RUL) prediction plays an essential role in the prognostic health management of cutting tools, which aims to ensure workpiece surface quality and extend tool life. Although deep learning methods have exhibited great performance in the field of tool RUL prediction, establishing an effective deep learning model usually requires extensive training data, and collecting substantial tool failure data in the actual milling process is time-consuming and expensive. Alternatively, constructing robust tool degradation features and developing a lightweight prediction model might be more practical and deployment-friendly. Therefore, this paper introduces an optimal tool degradation feature extraction method with a tailored dimension reduction algorithm named dynamic-inner canonical correlation analysis (DiCCA), where the temporal dependency and dynamic information hidden in the sequential tool degradation process are captured for the RUL prediction of cutting tools. Experimental studies are performed with a milling dataset, and the results indicate that the optimal tool degradation feature extracted by the proposed DiCCA can explicitly reflect the degradation process of cutting tools, and accurate RUL prediction can be achieved through a lightweight BP model.