Functional Series TARMA Models for Non-stationary Tool Vibration Signals Representation and Wear Estimation

Aghdam, Behrang Hosseini
Ciğeroğlu, Ender
This article addresses the problem of tool wear estimation using vibration signals. Time dependent time series models are suitable for extraction of time varying dynamics embedded in the non-stationary signals. A version of non-stationary time series known as Functional Series Time dependent AutoRegressive Moving Average (FS-TARMA) is employed for estimation of tool vibration signals and identification of the dynamics of tool/holder system. The obtained models associated with different levels of tool wear are compared by using characteristic quantities calculated based on model parameters. In this method, called model parameter-based method wear is estimated using a feature that is a function of model parameter vector obtained from FS-TARMA models. The advantage of this method over the ARMA metric employed in a previous study is that it does not violate the non-stationarity assumption of signals. The results of this study demonstrate that the FS-TARMA models with model parameter-based method provides higher accuracy in wear estimation compared with ARMA counterpart and also FS-TARMA with ARMA metric.