A chatter detection method in milling of thin-walled TC4 alloy workpiece based on auto-encoding and hybrid clustering

Chatter is a typical fault in milling, which is a self-excited vibration. Chatter can be diagnosed by using several methods, such as Delay Differential Equation (DDE) modelling methods, Artificial Neural Network (ANN) supervised learning methods, etc. These methods are effective. But they have some limitations, such as susceptible to measurement errors, require all data to be labeled, etc. In this paper, an unsupervised method to diagnose the chatter stability in milling according to massive unlabeled measured dynamic signals is proposed. The proposed method is not susceptible to measurement errors, don't require labels, and is robust. The dynamic signals are acquired from several milling trials. In the proposed method, the measured signals are compressed by using auto-encoding based method. Then, a hybrid clustering method based on both density metric and distance metric is used to cluster the compressed signals. The proposed method achieves a detection accuracy of 95.6033% on the experimental measured dynamic signals. (c) 2021 Elsevier Ltd. All rights reserved.

Automated summary

In this study, a new unsupervised method is proposed to diagnose chatter stability in milling based on unlabeled dynamic signals, which is not susceptible to measurement errors, does not require labels, and demonstrates robustness.