Abstract:Addressing the issues of limited information from a single data source and difficulty in ensuring diagnostic accuracy in railway switch machine fault diagnosis, this paper proposes an intelligent fault diagnosis and decision-making model for ZYJ7 switch machines based on multi-source data fusion. The model first standardizes the data collected from multiple sensors of ZYJ7 switch machines, including current, power, and oil pressure, through data preprocessing and feature extraction techniques. It then extracts multi-dimensional signal features from the time domain, frequency domain, and time-frequency domain, fully exploiting the feature information from each data source. Finally, an intelligent voting strategy based on confidence weighting is introduced. Through an adaptive weight allocation mechanism, it integrates the advantages of three different sub-diagnosticians: convolutional neural network-long short-term memory (CNN-LSTM), particle swarm optimization-support vector machine (PSO-SVM), and simulated annealing-particle swarm optimization-probabilistic neural network (SA-PSO-PNN), achieving dual-layer confidence weighting integration at both the sub-model and modal levels to obtain accurate fault discrimination results. The experiments utilize an actual multi-source dataset of ZYJ7 switch machines, including a total of 800 samples of normal and typical faults. The experimental results show that the proposed model improves the average accuracy, precision, recall, and F1 score of fault identification by 2.77%, 2.54%, 2.70%, and 2.88%, respectively, compared to traditional single-source algorithm models.