A reinforcement neural architecture search method for rolling bearing fault diagnosis

Abstract

The fault diagnosis of rolling bearing has always been a research hotspot, and it is an urgent task to develop the effective method for rolling bearing fault identification. Most traditional methods cannot automatically build appropriately models for different datasets. In this paper, a neural network architecture automatic search method based on reinforcement learning is proposed for fault diagnosis of rolling bearings. The framework of proposed method contains of two components: a controller model and child models. The controller is recurrent neural network (RNN) and generates a series of actions, each action specifies a design choice to construct the child models for fault diagnosis. Then, the controller parameters are updated using the policy gradient method of reinforcement learning by maximizing the accuracy of the child models. The results confirm that the proposed method can realize the automatic design of neural network architecture and overcome the limitation of traditional methods.