Online Tool Wear Monitoring Via Hidden Semi-Markov Model With Dependent Durations

Abstract

The tool wear monitoring (TWM) system that could estimate tool wear conditions and predict remaining useful life (RUL) is important to meet the high precision requirement and improve productivity in automated machining. Due to its good properties in representing non-stationary and complex physical process, hidden semi-Markov Model (HSMM) is adapted to model the progressive tool wear in this paper. In order to describe the time-variant transition probability of tool wear states and the state duration dependency, the HSMM is improved by learning the duration parameters and RUL distribution database. The Forward algorithm is utilized for on-line tool wear estimation and remaining life prognosis, and an on-line implementation approach is developed to reduce computational cost. Experimental results show that the approach is effective and the proposed method of duration dependency modeling leads to more accurate TWM in high speed milling.