Detection of tool wear using gradient adaptive lattice and pattern recognition analysis

The approach, which employs gradient adaptive lattice analysis and pattern recognition techniques, is fast and yields accurate recognition of tool wear states in a wide range of cutting conditions.

Abstract

A method of recognising tool wear states in a turning operation from the integrated information of cutting force and acoustic emission signals is presented. The approach, which employs gradient adaptive lattice analysis and pattern recognition techniques, is fast and yields accurate recognition of tool wear states in a wide range of cutting conditions. The gradient adaptive lattice algorithm is applied to recursively compute the autoregressive as well as partial correlation coefficients for both signals with high computational efficiency. Those coefficients are chosen to characterise the sensing signals and used as the feature inputs for pattern recognition. A stepwise search procedure was applied to select the most useful features for the recognition process. Both unsupervised learning technique (fuzzy C-means algorithm) and supervised learning technique (linear discriminant analysis) are used in the pattern recognition analysis. Characteristics of each method are discussed and performance in recognising the states of cutting tool (fresh or worn) is evaluated. The results show that this approach was successfully applied to signify a fresh or worn tool state with a high percentage of correct classification (over 90%).