Cutting force reconstruction in milling by multi-sensor fusion with hybrid aid of process and data-driven models

Abstract

This paper proposes an approach to cutting force reconstruction in milling combining a machine learning model with cutting process simulations. The machine learning model is developed in the feature space of amplitude spectra associated with the tooth-passing components extracted through the moving Fourier transform of time series data. Subsequently, the cutting force time waveform is reconstructed by integrating the estimated amplitude spectra with phase information provided by a cutting simulator. This approach facilitates efficient model construction compared to directly using time waveforms for training. It also enables straightforward multiple-sensor fusion using the ML model, thereby enhancing estimation performance.