Vibration-based estimation of tool major flank wear in a turning process using ARMA models

This paper investigates the correlation between vibrational features of tool/holder assembly and tool major flank wear in a turning process. During cutting, 3D tool acceleration signals are recorded by an accelerometer. Then, based on the dynamics of the tool/holder system that manifests itself in natural frequencies/modes, wear sensitive features are determined and derived from autoregressive moving average (ARMA) model of the recorded signals. After a comprehensive investigation of extracted features, dispersion ratio of natural modes, a correlation between the wear and features is found. Moreover, a metric calculated based on the system eigenvalues is introduced to find the distance between ARMA models of tool with different wear levels and the baseline model, as another wear-sensitive feature. Analysis of experimental results reveals that in the accelerating stage of wear, a change of vibration mode from second bending mode in the main cutting direction to the first bending mode in feed direction occurs, and simultaneously, the ARMA distance reaches a minimum value. This analysis provides a reliable algorithm for tool wear estimation since it directly originates from tool/holder system natural modes/frequencies and interprets the physical behaviour of the system in connection with the tool wear. Based on the results from dispersion analysis, the outputs of ARMA metric can also be used to provide reliable predictions for tool replacement policy. The method obtained here can be utilized for development of an online real-time tool wear estimation algorithm in turning.