AUTHOR=Chen Jin 

TITLE=Fault diagnosis of large-scale electric pumping and irrigation electromechanical equipment by integrating CWT and swin transformer

JOURNAL=Frontiers in Mechanical Engineering

VOLUME=Volume 11 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/mechanical-engineering/articles/10.3389/fmech.2025.1708745

DOI=10.3389/fmech.2025.1708745

ISSN=2297-3079

ABSTRACT=This paper proposes a large-scale electromechanical equipment fault diagnosis framework that integrates continuous wavelet transform (CWT), Swin transformer, and cross attention. Firstly, CWT uses optimized complex Morlet wavelets (impact sharpness ratio factor 0.82, frequency focus ratio factor 0.67) to map vibration signals to high-resolution time-frequency images. Secondly, the four level Swin Transformer layer extracts multi-scale features (56 × 56 × 96 to 7 × 7 × 768), while stride convolution aligns shallow features for cross attention fusion, where shallow features serve as queries and deep features serve as keys/values, dynamically weighting and integrating cross stage information. The experimental results show that the classification accuracy is 98.7% (macro F1 is 98.5%), which can perfectly identify the motor rotor eccentricity (MRE). T-SNE visualization validates the enhanced intra class compactness and inter class separability, particularly between MRE and gear wear (GTW). The model maintains robustness under noise, with an accuracy increase from 82.3% (signal-to-noise ratio = 10 dB) to 98.7% (signal-to-noise ratio = 90 dB), verifying its effectiveness and reliability in intelligent maintenance in complex industrial environments.