Issues and Tips:A Set of Integrated Experiments of Applying Auto-Encoder and Convolutional Neural Network in Feature Extraction and Fault Diagnosis

In recent years, deep learning technology has made a breakthrough and rapid development, and it provides a new direction for the research of Prognostics Health and Management (PHM). In this paper, we propose two deep learning models to solve feature extraction problem and faults diagnosis problem. First model is based on Auto-Encoder (AE) and Support Vector Machines (SVM). AE is used to reduce the dimensions of original signal and efficiently extract features. Then the extracted features are classified as the input of SVM. Second model is based on Convolution Neural Network (CNN), we propose a 1D-CNN model to process the original bearing vibration signal and directly output the type of fault. These models have yielded good results on the milling datasets and CWRU bearing dataset respectively. This paper verified the feasibility of these methods, summarized the application experiences and obtained their performance indicators as a benchmark for research.