A Data-Based Approach for Sensor Fault Detection and Diagnosis of Electro-Pneumatic Brake

Abstract

Sensor faults in the train electro-pneumatic brake system can cause severe performance degradation which may lead to accidents. Therefore, sensor fault detection and diagnosis of the brake are crucial for the train safety. However, the brake system is a complex integrated system with multiple operating modes, which makes the sensor fault detection and diagnosis more difficult. In this paper, we propose a data-based analytical redundancy method to detect and isolate sensor faults of the electro-pneumatic brake in the dynamic process of different operating modes. Firstly, we apply the Gradient Boosting Decision Trees regression analysis model to determine the normal quantitative correlation between two sensor measurements by training non-fault sensor measurements. Secondly, a measurement estimation is calculated by fusing multi-sensor measurements with a data fusion method that is insensitive to faulty sensor measurements. Finally, the sensor fault is detected by comparing the fusion estimation and each sensor measurement. The feasibility and effectiveness of the data-based approach are verified by experiments and simulations.