Random-Sampling-Based Performance Evaluation Method of Fault Detection and Diagnosis for Railway Traction System

A random-sampling-based performance evaluation method to comprehensively and objectively reflect fault detection and diagnosis (FDD) is proposed. This method is used to test and evaluate the FDD algorithms to be used in traction control system(TCS). Firstly, the needed fault information of fault injection models to simulate fault scenarios is divided into fault location, fault type, and fault parameter. Then, various sampling methods are used to hierarchically sample fault information to generate a specific fault injection model. As a consequence, some FDD algorithms can be executed and tested in these simulated fault scenarios and their test results will also be recorded. Finally, according to a large number of tests and their generating results, three-level evaluation indexes are constructed to evaluate the mentioned FDD algorithms. The proposed method is implemented in MATLAB/Simulink and embedded into the developed fault-injection software for TCS, and the simulation results more objectively and comprehensively evaluate the performance of the tested FDD algorithms, which will be useful for researches to select and improve satisfied algorithms.