Data-driven Prediction Method for High-speed Railway EMU Train Front and Rear Car Wheel Tread Defects

Abstract

Objective Tread defects are a primary manifestation of wheel failures in high-speed railway EMU (electrical multiple unites) trains, significantly impacting both EMU train operation safety and passenger ride comfort. Wheel tread defects are predominantly concentrated on front and rear cars, which may result from a combination of various factors, requiring predictive method that comprehensively integrate various influencing factors. Method Based on the wheel reprofiling maintenance data of EMU train operated by a railway bureau, the dataset sample consists of 10 features (including 4 nominal and 6 continuous features) and the data are preprocessed. By treating the imbalanced dataset through synthetic minority over-sampling technique (SMOTE), a standardized dataset is constructed. A DNN (deep neural network) model is established to combine the underlying features and form a high-level abstract representation of the features. The optimal learning performance of the model is achieved through network structure adjustment and hyperparameter optimization. The model is trained and tested to verify its prediction effect. Result & Conclusion The data-driven prediction method for wheel tread defects of front and rear cars demonstrates high-predictive accuracy and relatively excellent comprehensive performance, achieving a precision rate of 92.5%. Thus the probability of wheel tread damage of front and rear cars can be effectively predicted.

Keywords

• high-speed railway
• emu
• data-driven
• tread defect
• over-sampling technology
• deep neural network
• hyperparameter optimization