An intrinsic timescale decomposition-based kernel extreme learning machine method to detect tool wear conditions in the milling process

Abstract

Detecting tool wear conditions in milling process is of significance to enhance the reliability of machining equipment. However, traditional methods have run into difficulties due to interference from strong noise and other unknown vibration sources. To solve this problem, an intrinsic timescale decomposition (ITD) technique is combined with a kernel extreme learning machine (KELM) technique. In this method, ITD is firstly employed to decompose multiple sensor signals into several sets of proper rotation (PR) components. Next, the optimal PR component of each set is selected by correlation coefficient analysis. A series of feature sets are then constructed according to the data indicators extracted from the selected PR components in time and frequency domains. Finally, the feature sets are fed into the KELM, which classifies the tool wear conditions. Experimental investigations are conducted to determine three stages of tool wear in the milling process; the ITD-KELM method achieved 93.28% classification accuracy, which verifies its feasibility and effectiveness for detecting tool wear. The superior performance of the proposed method is further demonstrated by comparing it with four other methods: ITD-based SVM, EEMD-based KELM, VMD-based KELM, and KELM.