Fault Detection and Diagnosis Using Statistic Feature and Improved Broad Learning for Traction Systems in High-Speed Trains
Li Guo, Runze Li, Bin Jiang
Abstract
Sensors equipped in the high-speed trains can collect a lot of data with the normal working condition and different faults that occurred. In recent years, many data-driven methods were developed for fault detection and diagnosis (FDD). However, inaccurate diagnosis and costly computation are still the great challenges that exist. In order to address those issues, this article developed an FDD architecture using statistic feature and the improved broad learning system (BLS) to promote performance. It uses statistic feature to capture the inherit discrimination of normal data and fault data, and then adopts the improved BLS model to achieve the accurate and fast fault diagnosis without time-consuming training and mathematical models of high-speed trains. In validation, the proposed FDD scheme is first conducted on a software-based fault-injection simulation platform; it can give guiding significance for the subsequent hardware-in-the-loop simulation platform. All results show that the presented FDD framework achieves state-of-the-art performance than the other mainstream FDD methods.