Combining translation-invariant wavelet frames and convolutional neural network for intelligent tool wear state identification

Abstract

On-machine monitoring of tool wear in machining processes has found its importance to reduce equipment downtime and reduce tooling costs. As the tool wears out gradually, the contact state of the cutting edge and the workpiece changes, which has a significant influence on the vibration state of the spindle. The performance of traditional intelligent fault diagnosis methods depend on feature extraction of dynamic signals, which requires expert knowledge and human labor. Recently, deep learning algorithms have been applied widely in machine health monitoring. In this paper, we present a novel intelligent technique for tool wear state recognition using machine spindle vibration signals. The proposed technique combines derived wavelet frames (DWFs) and convolutional neural network (CNN). Constructed based on dual tree wavelets, DWF are equipped with merits of centralized multiresolution and nearly translation-invariance. In this method, DWFs are employed to decompose the original signal into frequency bands of different bandwidths and different center frequencies, which are more pronounced as the tool wears. Further, the reconstructed sub-signals are stacked into a 2-D signal matrix to match the structure of 2-D CNN while retaining more dynamic information. The 2-D convolutional neural network is utilized to automatically recognize features from the multiscale 2-D signal matrix. End-milling experiments were performed on a S45C steel workpiece at different machining parameters. The experiment results of the recognition for tool wear state show the feasibility and effectiveness of the proposed method.