Adaptive Broad Learning System for High-Efficiency Fault Diagnosis of Rotating Machinery

Rotating machinery fault diagnosis is vital to enhance the reliability and safety of modern equipment. Recently, deep learning (DL) models have achieved breakthrough achievements in fault diagnosis. However, most of DL models are restricted to the time-consuming training task caused by a huge number of connecting parameters in the deep architecture. Besides, most of DL models require an entire retraining process if new architecture hyper-parameters are chosen, which means to obtain a good diagnosis model, a great deal of time is wasted to train useless models with inapposite hyper-parameters. These problems affect the efficiency of diagnosis tasks badly. Therefore, this article proposes an adaptive broad learning system (ABLS) for fault diagnosis of rotating machinery. In the proposed ABLS, the original vibration signals are transformed into feature nodes and enhancement nodes, and all hidden nodes are directly connected to fault labels. Then, two adaptive incremental learning strategies are developed to fast adjust the network architecture without the entire retraining task. Finally, three case studies are implemented to illustrate the effectiveness of the proposed ABLS. The results demonstrate that the proposed ABLS offers a high-efficiency solution for rotating machinery fault diagnosis.