Research on AGV fault diagnosis based on 2D-swin transformer

Abstract In industrial automation scenarios, sudden failures and high maintenance costs of Automated Guided Vehicles (AGVs) severely limit productivity. Aiming at the shortcomings of passive lag and high false alarm rate of traditional diagnostic methods, this study proposes an intelligent diagnostic model based on 2D Swin Transformer: build a multi-source data testing platform for AGVs, collect real-time battery voltage, current and vibration signals from bearings, and then generate time-frequency maps to enhance the expression of fault features through Z-score normalisation and Short Time Fourier Transform (STFT). The model integrates the local feature extraction capability of CNN and the attention mechanism of Swin Transformer’s hierarchical window and captures the time-frequency dynamic modes through the non-overlapping window division and cross-layer shift strategy to accurately classify faults at multiple scales. Experiments show that the model has an accuracy rate of 94.2% in six types of typical fault diagnosis, and this study achieves accurate fault diagnosis of AGVs, which has significant application value in engineering