Anomaly Transformer for Vibration Data Analysis: Case Study on CWRU Dataset

Time series anomaly detection has a wide range of applications in various fields of society, one of which is the detection of anomalies in vibration data. Abnormal vibration data often indicates the existence of potential disasters or abnormal operation of equipment, so timely detection and identification is important for the prevention of earthquakes and improvement of industrial production efficiency. In this paper, an unsupervised time series anomaly detection algorithm, Anomaly Transformer, is used to detect anomalies in CWRU vibration datasets.Anomaly Transformer leverages the powerful characterization capabilities of the Transformer network and an improved anomaly attention mechanism to capture anomalies in the time series. Experimental results show that Anomaly Transformer exhibits excellent performance in the CWRU vibration dataset, achieving high precision and recall. In addition, Anomaly Transformer demonstrated a low false alarm rate, indicating its high accuracy in recognizing real anomalies. Future research can further explore the application of Anomaly Transformer in other domains with deeper optimization and extensions.