Effects of Extrinsic Noise Factors on Machine Learning–Based Chatter Detection in Machining

Abstract

Abstract Unmitigated chatter can result in poor part quality, accelerated tool wear, and possible damage to spindle and machine. While various methods have been shown to effectively detect chatter, implementation of these methods in noisy environments, such as factory floors, has not been well studied. The present study seeks to explore the effects of extrinsic noise sources on threshold-based and machine learning–based chatter detection methods using audio signals of the machining process. To accomplish this, stable and unstable cuts were made on a milling machine and the audio signal was collected. Data augmentation using Gaussian white noise and periodic noise was conducted to simulate a range of noise levels and types. The performance of these techniques were then compared with respect to the increasing levels of noise. It was found that machine learning–based approaches achieved satisfactory accuracies up to 98.6 % under the presence of extrinsic noise. Conventional static threshold techniques, however, failed under most noise conditions and resulted in false positives depending on the threshold values used. Furthermore, support vector machine approaches demonstrated an ability to classify noisy data despite limited training.