In-process chatter detection in micro-milling using acoustic emission via machine learning classifiers

Predicting chatter stability in a micro-milling operation is challenging since the experimental identification of the tool-tip dynamics is a complicated task. In micro-milling operations, in-process chatter monitoring strategies can use acoustic emission signals, which present an expressive rise during unstable cutting. Several authors propose different time and frequency domain metrics for chatter detection during micro-milling operations. Nevertheless, some of them cannot be exploited during cutting since they require long acquisition periods. This work proposes an in-process chatter detection method for micro-milling operation. A sliding window algorithm is responsible for extracting datasets from the acoustic emissions using optimal window and step packet sizes. Nine statistical-based features are derived from these datasets and used during training/testing phases of machine-learning classifiers. Once trained, machine learning classifiers can be used in-process chatter detection. The results assessed the trade-off between the number of features and the complexity of the classifier. On the one hand, a Perceptron-based classifier converged when trained and tested with the complete set of features. On the other hand, a support vector classifier achieved good accuracy values, false positive and negative rates, considering the two most relevant features. A classifier’s output is derived at every step; therefore, both proposals are suitable for in-process chatter detection.