Turnout Switch Machine Fault Diagnosis Based on 1DCBA

Abstract Turnout switch machine fault data is characterized by limited quantity and imbalanced class distribution. This limits the use of deep learning in fault diagnosis applications. Therefore, this paper presents a fault diagnosis method for turnout switch machines based on the 1DCBA model. Firstly, to mitigate the problem of insufficient fault data, the KMeans-SMOTE is employed to oversample the current data, effectively expanding the dataset. Secondly, a 1D-CNN is utilized to capture local features, and it is combined with a BiLSTM to exploit the temporal dependencies within the data. Finally, an attention mechanism is introduced to improve the model’s classification ability and fault diagnosis performance by allowing it to focus on the most critical features. The results of experiments indicate that the suggested approach attains a fault diagnosis accuracy of 100%.