Research On Improved BP Neural Network And DS Evidence Theory For The High-Speed Train Running Gear Fault Diagnosis

With the rapid development of China high-speed train technology,serious railway accidents also alarm that we need concerning about the safety issues,High-speed trains running gear as an important component of the train,which will be directly related to the traffic safety.Currently,there are no health monitoring and diagnostic systems for high-speed trains running gears.Based on this,Southwest Jiaotong University Traction Power Laboratory are doing experiments on this,the high-speed train bogies fitted with a large number of sensors to capture the process of train vibration signal.By monitoring the vibration signal we can determine the operational status of running gear. However,due to the large number and different types of sensors,how information from different sources for data fusion become an urgent problem to be solved.This thesis will combine the GA-TS-BPNN and DS evidence theory to carry out the high-speed train running gear fault diagnosis.In this thesis,we conducted research on the dynamics of knowledge associated with the running gear briefly,the experimental data and the vibration signal processing methods also were introduced.On this basis,a BPNN high-speed train running gear fault diagnosis model and the BPNN weights and thresholds update formula was presented.For BPNN in fault diagnosis of convergence is slow and easy to fall into local minimum problem,a high-speed train running gear fault diagnosis model based on GA-BPNN was built.Finally,combining their strengths with genetic algorithm (GA) and tabu search algorithm (TS),a high-speed train running gear fault diagnosis model based on GA-TS-BPNN was built.The improved neural network was used for high-speed trains running gear fault diagnosis,the results show the effect of improved diagnostic methods is better.Because the types of sensors,precision and environment are different,the results of such diagnosis is uncertain even contradictory.In this thesis, DS evidence theory was introduced to process this problem.Considering the classical DS evidence theory can not process the conflict of evidence fusion problem, by analysing the current improved method,a new evidence algorithm based on distance-weighted was presented.The example shows that the improved method can solve the problem of conflict evidence fusion and relatively reduced computational complexity.At the end of this thesis,considering the single-sensor-based fault diagnosis did not have a good result,the GA-TS-BPNN and DS evidence theory were combined for high-speed trains running gear fault diagnosis, and the fault diagnosis experiment for high-speed trains running gear show that this method can effectively improve the accuracy of fault diagnosis.