Edge-IoT Signal Acquisition and Incremental Learning-Enabled Real-Time Fault Diagnosis for Railway Traction Gearboxes

Abstract

Reliable condition monitoring of traction drive units in rail transit is crucial for the safe and stable operation of rail vehicles. However, vibration signals acquired under actual operating conditions often exhibit significant randomness, load fluctuations, and speed fluctuations, leading to unclear fault characteristics and high non-stationarity, thus reducing the effectiveness of traditional autonomous diagnostic models. To overcome these limitations, this study proposes a real-time fault diagnosis framework based on Edge-IoT. This framework combines high-precision onboard signal acquisition, resource-efficient incremental learning, and joint optimization of edge and cloud systems. The edge layer efficiently extracts features and performs low-latency inference, while the cloud layer provides long-term knowledge integration to continuously improve the model’s adaptability. The incremental learning strategy enables the diagnostic model to adapt to new operating conditions and constantly changing fault types without requiring large amounts of labeled data or complete retraining. Experiments on traction drive unit datasets under various operating conditions demonstrate that, compared with existing methods, the proposed method achieves higher diagnostic accuracy, lower latency, and significantly reduced communication overhead. This work provides a practical and scalable solution for the intelligent maintenance of railway drive systems.