A Generalized Multi-Stage Deep Machine Learning Framework for Tool Wear Level Prediction in Milling Operations

Abstract

This work proposes a tool condition monitoring system based on a generalized multi-stage deep machine learning framework for real-time measurement of flank wear in milling processes. At low computational cost, a deep wavelet scattering convolution neural network framework was developed and optimized to generate low-variant features. Discriminative features were automatically selected using a neighborhood component analysis. A gaussian process regression (GPR) model was developed for feature fusion and tool wear measurement. The model was benchmarked against different machine learning algorithms. Extensive experimental validation tests demonstrated the GPR model superiority under various cutting conditions, which facilitates its implementation in industrial facilities.