Machine Tool Fault Diagnosis Model Based on Discrete Wavelet Transform and Transfer Learning

The machining industry has strict precision requirements. Machine tools are often equipped with sensors to collect signal data, which can be analyzed to control and improve various machining parameters. Chatter, a major issue in the machining process, can substantially diminish production capacity. Thus, chatter monitoring is crucial. This study proposed a method for detecting chatter that combines discrete wavelet transform (DWT) with transfer learning (TL). DWT provides multiscale analysis and exhibits excellent performance in the local analysis of signals, making it effective for detecting chatter phenomena. DWT is also computationally light, making it useful in time-sensitive applications. TL offers strong generalization capabilities and uses training time efficiently by building upon previously acquired knowledge from trained models instead of starting from scratch. However, TL requires a certain similarity between the original and target data. In this study, model training was conducted on machine tools (lathes and milling machines) to confirm the suitability of TL in this context. The accuracy of the proposed model reached up to 99% for turning-to-turning scenarios, 90% for turning-to-milling scenarios, and 97% for milling-to-turning scenarios, validating the effectiveness of TL on different machine tools.