Kernel Density Regularized Bayesian Learning Framework for Machining Process Anomaly Detection

Healthy and stable machining processes are critical for ensuring machining accuracy and guaranteeing machine safety. However, due to complex machining conditions and harsh service environments, machining processes inevitably suffer from abnormalities, which can lead to product defects, increased scrap rates, and even catastrophic accidents. To address this issue, a kernel density regularized Bayesian learning framework is proposed for machining process anomaly detection. In this work, an adaptive kernel density estimate is first constructed to eliminate outlier interferences and provide prior distributions to subsequent Bayesian learning for improving detection accuracy. On this basis, the Bayesian learning framework is innovatively developed for incorporating prior knowledge and multi-classification models, which presents a scientific interpretation for detection results from a probabilistic perspective. Finally, two practical engineering applications are employed to validate the effectiveness of the proposed method. The results show that the proposed method not only improves the anomaly detection accuracy under time-varying operating conditions but also provides confidence levels for detection results. By these advantages, this work may provide a useful tool for independently perceiving the health conditions of machine tools.