Abstract: Health monitoring and fault diagnosis of rolling bearings can ensure continuous and effective work of mechanical equipment. When using deep learning to model the massive and complex data in the context of industrial big data, it needs a lot of computational resources, resulting in problems such as training stagnation or difficulty in training. This paper attempts to use Broad Learning Systems to replace deep learning for bearing fault diagnosis. To address the problem that the classification effect of a broad learning system is limited by the choice of its own hyperparameters, the sparrow search algorithm in metaheuristic algorithms is used to optimize the hyperparameters of a Broad Learning System to improve the accuracy of the broad learning system. The optimized model is applied to the bearing dataset from Western Reserve University and compared with various neural network models to verify the fault diagnosis capability of the proposed method.

Key words: rolling bearing, fault diagnosis, sparrow search algorithm (SSA), broad learning system