Advanced fault diagnosis of railway switch machines utilizing FroFT-Attention-BiLSTM

Railway switch fault detection is an important aspect to ensure the safe and reliable operation of railway systems. This study aims to improve the effectiveness and accuracy of switch fault diagnosis, thereby reducing potential faults and improving the reliability of railway transportation. To solve the problem of imbalanced fault data, the synthetic minority oversampling technique (SMOTE) is used to increase the representativeness of rare fault conditions. Subsequently, the fractional Fourier transform (FroFT) is applied to extract feature vectors with diagnostic value from the current signal of the switch, effectively capturing the subtle changes caused by the fault. These fine features significantly improve the accuracy of fault detection. The extracted features are then input into a bidirectional long short-term memory (BiLSTM) neural network with an integrated attention mechanism to enhance the model’s ability to identify various fault states. Test results show that the proposed FroFT-Attention-BiLSTM method outperforms the traditional BiLSTM model and achieves significant improvements in key performance indicators: accuracy (+6.59%), precision (+6.01%), recall (+5.61%), and F1 score (+6.19%). These remarkable improvements verify the theoretical and practical effectiveness of the proposed method, providing a powerful and innovative solution for railway turnout machine fault diagnosis.