An Intelligent Milling Tool Wear Monitoring Methodology Based on Convolutional Neural Network with Derived Wavelet Frames Coefficient

Abstract

Tool wear and breakage are inevitable due to the severe stress and high temperature in the cutting zone. A highly reliable tool condition monitoring system is necessary to increase productivity and quality, reduce tool costs and equipment downtime. Although many studies have been conducted, most of them focused on single-step process or continuous cutting. In this paper, a high robust milling tool wear monitoring methodology based on 2-D convolutional neural network (CNN) and derived wavelet frames (DWFs) is presented. The frequency band of high signal-to-noise ratio is extracted via derived wavelet frames, and the spectrum is further folded into a 2-D matrix to train 2-D CNN. The feature extraction ability of the 2-D CNN is fully utilized, bypassing the complex and low-portability feature engineering. The full life test of the end mill was carried out with S45C steel work piece and multiple sets of cutting conditions. The recognition accuracy of the proposed methodology reaches 98.5%, and the performance of 1-D CNN as well as the beneficial effects of the DWFs are verified.

Keywords

• tool wear
• fault diagnosis
• wavelet
• convolutional neural network (CNN)
• deep learning