A Practical Bearing Failure Detection Method Using a New Efficient Deep Network With the Knowledge Self-Adaptive Evolution

Abstract

Intelligent fault diagnosis technology based on the deep neural network has shown significant advancements in recent years. However, it is difficult and expensive to deploy a fault diagnosis neural network with a huge number of parameters to an embedded computing platform with limited hardware resources. To address this issue, a practical bearing failure detection method using a new efficient deep network with the knowledge self-adaptive evolution, named autonomous compression method based on network pruning and knowledge distillation (AMC-NPKD), is proposed in this paper. In the proposed method, the reinforcement learning technique based on the deep deterministic policy gradient (DDPG) is employed to iteratively prune the network’s structure. The knowledge distillation (K-D) process is employed to fine-tune the pruned network after each pruning iteration. The results based on two datasets demonstrate that the proposed method effectively optimizes the structure of fault diagnosis networks. The proposed AMC-NPKD method is meaningful for promoting the engineering development of the intelligent fault diagnosis technology.