Cutting anomaly detection in end-milling by multimodal variational autoencoder

Anomaly detection for predictive maintenance in the cutting process is one of challenging problems shop-floor management. A modern machine learning approach, including deep learning, has been widely studied last decade. This study focuses on multimodality various time-series data extracting features status and proposes a multimodal variational autoencoder (MVAE) method. We collect time series vibration acceleration tool main spindle motor load. Various collected by conducting experiments under diverse conditions. Normal abnormal are collected, only normal used to train MVAE. MVAE learns so-called generative model, which implicit but stochastic, capable reproducing original data. Because an unsupervised method, it does not require during training. Therefore, considered suitable tools management where difficult Euclidean distance employed evaluate normality given latent space acquired demonstrate applicability proposed method anomaly end-milling comparing with conventional methods such as autoencoder.