Inverse problem of tool wear estimation using vibration signals is considered via non-stationary functional series time-dependent autoregressive moving average (FS-TARMA) model in this paper. The estimation procedure of FS-TARMA models is presented and through the obtained models, dynamics of the tool-holder system is identified. For finding a relationship between wear and the models, two wear sensitive features are used. First, the models are clustered considering autoregressive (AR) distance as a feature and then, damping ratios of tool-holder bending modes are used as another feature for correlating tool wear with the vibrations. The AR metric provides a parsimonious parametric way for comparison of the structures generating the time series. The obtained wear-AR distance curves possess extremums at critical wear stage. Moreover, in wear-damping ratio curves, which are obtained first time in this paper, extremums appear in the vicinity of critical wear point. These extremums can be used as a measure for tool change policy. The results of the study demonstrate the good accuracy of FS-TARMA models in prediction of tool non-stationary signals and the effectiveness of the selected features for estimation of tool major flank wear.