A Gearbox Anomaly Detection Method for Rotating Pump Equipment Based on TadGAN Algorithm

Rotating pump equipment may experience wear and corrosion of critical components during long-Term operation, which can subsequently lead to various pump failures, resulting in significant economic losses and potential operational hazards. Traditional anomaly detection methods for pump equipment primarily rely on threshold judgments based on industry experience, which are often characterized by vagueness and latency. When confronted with massive, imbalanced data samples generated during actual operation, methods that rely solely on big data analysis without algorithmic support tend to yield suboptimal results. Therefore, this study adopts the TadGAN algorithm structure, based on generative adversarial networks (GAN), and designs an anomaly detection method grounded in reconstruction techniques. This approach provides an effective unsupervised learning framework for early fault detection in pump systems, enhancing operational reliability and reducing maintenance costs.