The Application Of Manifold Learning In Data Analysis Of High-speed Train Safety State Evaluation

Modern high-speed train technology is developing rapidly, so security is a precondition to adapt to the economic and social development of railway transportation. Research on high-speed train safety warning and health maintenance, eliminating or reducing safety accidents, is the key to the sustainable and healthy development of the high-speed railway in our country, and has become the pressing demand of the day for high-speed rail development. The effective monitoring and evaluation for high-speed train service condition is an important means of high-speed train safety warning and health maintenance. Running gear is the main part of railway locomotive vehicle and the most critical component of train safe operation. The failure of running gear will severely affect train safe operation, and train vibration monitor signal collected by Sensors on the bogie can reflect the running state. So the feature analysis of high-speed train vibration signal is particularly important.High-speed train Vibration is complex, the monitoring data is broader and rich and the influence factors of the data are many, so vibration signal is a kind of nonlinear and non-stationary signal. Manifold Learning as a kind of nonlinear dimension reduction method is a new method in the field of international machine learning, and it has the advantages that the algorithm is simple and suitable for nonlinear non-stationary signal analysis, etc. Manifold learning is mainly divided into local analysis method and global analysis method. This paper mainly adhibited the two representational methods of the local embedding method (LLE) and isometric feature mapping (ISOMAP). Through the simulation analysis, the classic isometric feature mapping algorithms is better than LLE. ISOMAP well keeps the intrinsic geometric structure of the original data by calculating geodesic distance matrix of sample points of data collection, and using multidimensional scaling analysis method(MDS) drops the dimension. This paper will use the method to conduct the high-speed train characteristics analysis for simulation data and test monitoring data.The high-speed train failures studied by this paper mainly include three kinds of typical failures:air spring loss of air pressure, resist sinusoidal vibration damper failure and lateral damper failure. In the simulation data we calculated the more faults working condition of two kinds of fault happened at the same time and three kinds of working conditions whose fault parameters are gradually varied corresponding air spring vertical stiffness change and resist sinusoidal shock absorber damping values change and the secondary lateral damping value change. The appropriate classifier is designed to conduct classification recognition for two-dimensional manifold features of single fault. Through manifold feature analysis of the simulation data and test monitoring data, we can see that air spring loss of air pressure characteristics are obvious for vertical acceleration, resist sinusoidal vibration damper failure and lateral damper failure characteristics are obvious for transverse vibration signal. The feasibility and effectiveness of Manifold.learning algorithm for high-speed train safety state evaluation is verified.This study is derived from the high-speed train service security state characteristics analysis of the key project of national natural science fund "the key problem research based on monitoring data of high-speed train service security state assessment".