Rotating machinery anomaly detection using data reconstruction generative adversarial networks with vibration energy analysis

Abstract

Rotating machines, such as engines, turbines, or gearboxes, are widely used in modern society. Their mechanical components, such as rotors, bearings, or gears, are the main parts, and any failure in them can lead to a complete shutdown of the rotating machinery. Anomaly detection in such critical systems is essential for the healthy operation of rotating machinery. As the requirement of obtaining sufficient fault data of rotating machinery is challenging to satisfy, a new anomaly detection model is proposed for rotating machinery, which can achieve anomaly detection without fault samples. The model combines vibration energy features, adversarial learning mechanism, and long short-term memory to develop a novel anomaly detection model based on data reconstruction. The model was validated using two different datasets. The proposed model achieves the highest accuracy of 98.90% for anomaly detection under different working conditions, outperforming similar models.