Researchontechnologyandmethodsof Recognizing Typical Working Condition Of High Speed EMU

With the development of high speed railway in our country,the relationship between the high-speed railway and daily life is more and more closer.Due to the speed of the EMU and less affected by the weather,more and more people choose to take the EMU.The load spectrum based on China working conditions is the key of EMU component reliability design and the fatigue strength evaluation.Recognition of working condition is the premise of analyzing a large number of load spectrum under different working conditions.In this paper,dynamic stress and load signals of CR400AF designed by CRRC QINGDAO SIFANG CO.LTD are analyzed.The main research and conclusions are as follows:(1)Line test for CR400AF,collect dynamic stress,load and acceleration of the car body and the bogie frame,and the gyroscope signal.Signal pre-processing:non-balance processing,remove signal drift,eliminate abnormal signals and filter processing.Then use wavelet threshold denoising method,choose wavelet basis function and threshold function to get rid of most of the random noise.(2)Choose the time-frequency joint analysis method to show more details of the signal characteristics such as wavelet packet decomposition,continuous wavelet transform and Hilbert-Huang transform,and then propose five kinds of signal feature extraction method.For switch,brake,passing events,tunnel,and speed level this five working conditions,select test points which respond obviously to the above conditions.Then use one or more suitable feature extraction methods and show the characteristics of each working condition.Using three-direction gyroscope signal,recognize curve radius,easement curve length and curve super-high values through time domain characteristics of the signal.Recognition errors below 15%mostly and can satisfy the requirement of engineering application.(3)According to the extracted features,build up the feature vector of each working condition as the neural network input,and select the BP neural network to recognize each working condition.Particle swarm optimization(PSO)is used to optimize the BP neural network to eliminate the influence of the random initial thresholds and weights.It can avoid the local minimum and improve the recognition accuracy.The recognition accuracy of switch,brake,passing events and tunnel is over 85%which can satisfy engineering application requirements.But recognition result of speed level can't reach the requirements.So the feature extraction method of speed level needs to be improved.