A Fault Diagnosis Method for Rotating Machinery Based on Compressed Sensing and Deep Convolutional Neural Network with SE Block

Abstract

Long-term condition monitoring of rotating machinery at high sampling rate generates large amounts of operational data, causing serious problems for data storage, transmission and diagnosis. And traditional deep learning-based fault diagnosis algorithms lack a mechanism to distinguish the importance of big data features. To solve the above problems, inspired by compressed sensing (CS) and attention mechanisms, this paper proposes a fault diagnosis method for rotating machinery based on compressed sensing and deep convolutional neural networks (DCNN) with squeeze-and-excitation (SE) block, called CS-SEDCNN. Compressed sensing is used to reduce the amount of data and improve diagnostic efficiency. The SEDCNN model is constructed for fault identification. The SE block can selectively focus on important features and suppress less useful features, enhancing the feature learning ability on compressed data. The proposed method achieves high diagnostic accuracy and faster diagnostic speed on the acoustic emission dataset of the wind power condition monitoring and diagnosis test rig.