Research On Track Irregularity Estimation Method Based On Data-Driven

Railway transportation is one of the most important traffic mode in our country, and is also the lifeblood of the national economic and social development. With the improvement of speed and traffic density, the dynamic effect between the vehicle and track is becoming harder, and the damage to the line structure is becoming more and more serious, which will affect the vehicle operation safety and ride comfort in return. Therefore, it is of great meaning to master the track state for ensuring the train operation safety and developing the maintenance plan.In this thesis, the track irregularity is the object we studied, combining with the model of vehicle-track system. The method of estimating the track irregularity is studied, which is based on data-driven. The neural network as well as support vector machine are the main methods of data-driven used in thesis, which realized deducing the state of the track and the type of the defect according to the axle vibration response.Firstly, the random wave length and the random amplitude are used to construct the sample of several common short wave irregularity, which are described by the harmonic model. Then the random sample of short wave irregularities is obtained by combining the sample of short wave irregularities with the sample of the fourth stage US track spectrum, which can provide the data for the study of the axle box vibration under the track irregularity.Then, the dynamic model of rail vehicle is established by using SIMPACK. The maximum vibration acceleration as well as the root locus of the system are calculated to verify the model’s correctness and stability. Besides, the reliability of the simulated acceleration data is verified by comparing measured data with model output. Simulation of the axle box vibration under different types of track irregularity provided a large number of data for the track irregularity estimation.To the calculation method, a series of data preprocessing are done to make preparation for the estimation of track irregularity by data-driven, such as feature extraction, normalization and principal component analysis. Then, the neural network as well as support vector machine are used to estimate the track irregularity. To improve the accuracy of track irregularity estimation, genetic algorithm and particle swarm optimization are studied to improve the support vector machine by optimizing the selection of parameter.To the software design, a software of track irregularity estimation based on data-driven is designed by LabVIEW, which is verified by the measured data.The measured data is collected from the 952 train, which run around on the test line of the Longyang Road. The validation results of the measured data show that the method studied in paper realized the estimation of track irregularity function by using the data of measured axle box acceleration.