A Network Structure Search Method Based on Reinforcement Learning for Rolling Bearing Fault Diagnosis

Accurate fault diagnosis is very important to the operation and maintenance of rotating machinery. Therefore, a neural network structure search method is proposed for rolling bearing fault diagnosis. Firstly, the child model is constructed by convolutional neural network (CNN) and the search space of child model is preset. Secondly, the controller is constructed by the recurrent neural network (RNN), the current state and reward of the child model are taken as inputs to generate the selected child model structure. Finally, reinforcement learning method is used to update the controller until the optimal child model is selected. The proposed method is used to the rolling bearing of electric locomotive. The experimental result shows the optimal child model selected in this paper is superior to other child model structures and some mainstream deep learning models. This method can successfully achieve the automatic search for neural network structure based on the given dataset.