Research on bearing fault diagnosis based on the fusion of CNN and LSTM algorithms

Abstract With the evolution of industrial equipment toward high - speed, heavy - load, and intelligent operation, bearing fault diagnosis encounters multiple challenges—such as strong background-noise interference and the extraction of weak fault features under non - stationary conditions. Traditional one - dimensional analysis methods based solely on time-domain or frequency-domain features struggle to capture the local time-frequency characteristics of non-stationary signals, resulting in limited diagnostic accuracy; likewise, a single deep-learning model is constrained in both its completeness of feature representation and its ability to model temporal dependencies. To address these issues, this chapter proposes a CNN–LSTM diagnostic method driven by fused wavelet time-frequency map features. By combining the CNN’s prowess in spatial feature extraction with the LSTM’s capacity to capture temporal dependencies in sequence data, the method effectively enhances the accuracy and robustness of bearing fault diagnosis.