A Fault Diagnosis Model Based on Kernel Auto-encoder and Improved Chaos Firefly Optimization Algorithm

Automatically extracting features from large scale raw data for fault diagnosis is important in the current era of big data. In this paper, a deep neural network based on the kernel function and denoising auto-encoder is proposed. The kernel denoising auto-encoder (KDAE) neural network consists of one KDAE layer and multiple auto-encoder (AE) layers to automatically extract the fault features from raw data. Then, the softmax classifier is added as classifier layer. The improved chaos firefly algorithm is used to optimize the undetermined parameters of the kernel function and the network to obtain the diagnosis model. The proposed method is then verified by the typical failure test data of the aero-engine intermediate bearing, which has achieved higher classification accuracy than the traditional denoising auto-encoder network.