A Review on Rolling Bearing Fault Diagnosis Techniques based on Deep Learning

Abstract

Rolling bearings, as core transmission components in rotating machinery, directly determine the reliability of equipment operation and the safety of industrial production. They play an irreplaceable role in key sectors such as aerospace, rail transportation, petrochemical industries, and intelligent manufacturing. In complex working conditions, bearing vibration signals exhibit strong non-stationarity and significant nonlinearity. Traditional fault diagnosis methods face challenges such as a high dependency on manual feature extraction, poor robustness under complex conditions, and insufficient ability to recognize multiple faults simultaneously. Deep learning techniques, with their advantages of end-to-end automatic feature learning and complex pattern recognition, have become the core research direction in the field of rolling bearing fault diagnosis. This paper systematically reviews recent research on rolling bearing fault diagnosis based on Convolutional Neural Networks (CNN) and their fusion models. It focuses on analyzing the technical principles, model architecture design, and performance optimization mechanisms of typical methods such as COA-CNN, FFT-CNN-Transformer, MSCNN-GRU, and 2D-CNN-GRU. The paper also compares the feature enhancement effects of preprocessing techniques such as Discrete Continuous Wavelet Transform (DCWT), Fast Fourier Transform (FFT), and Variational Mode Decomposition (VMD). Additionally, it explores optimization strategies for diagnostic robustness under noise interference, varying operating conditions, and small sample scenarios. Engineering application cases, such as coal chemical circulation fans and CNC machine tool spindle boxes, are used to verify the practical effectiveness of these methods. Finally, the paper looks ahead to future developments in areas such as small sample learning, dynamic condition adaptation, model lightweighting, and multi-modal fusion, providing theoretical references and technical support for the development and industrial application of intelligent diagnostic systems for rolling bearings.