Research On The High-Speed Train Running Gear Fault Diagnosis With Improved SVM And T-SNE

In recent years, with the rapid development of high-speed trains, high-speed rail brings us convenience, but also to promote the economic development along the route. Train running gear is an important part of the body, its mechanical properties directly affects driving safety. When running gear failure, if not timely investigation, resulting in a huge security risk. Therefore, high-speed railway running gear fault diagnosis requiring high precision diagnosis, we must also ensure that the diagnosis of high efficiency and good real-time performance.This paper analyzes the high-speed train running gear as well as its four kinds of conditions. For a variety of conditions pattern recognition introduced high-speed train running gear fault classification diagnosis and gives its processes. Then introducing the sources and characteristics of the experimental data of high-speed train running gear and its extraction section.Secondly, SVM classifier to build high-speed train running gear single-channel fault diagnosis model. For SVM parameter optimization problems cited grid search, genetic algorithm and particle swarm algorithm to optimize and SVM for constructing single-channel train running gear fault diagnosis model optimized.Subsequently, this paper introduces FOA (fruit fly optimization algorithm). Features for FOA easy to fall into local optimal solution, the introduction of improved genetic algorithm is proposed GA-FOA. Through six widely used test functions demonstrate the improved performance, the experimental results show that the the improved algorithm is better than the original optimization algorithm to ensure high efficiency while improving the accuracy of optimization. Given the simple GA-FOA algorithm structure, high convergence accuracy and strong generalization ability, etc., we propose a single channel running gear fault diagnosis model based on GA-FOA-SVM high-speed trains.Finally, for the problems of the data for high-speed train running gear are very huge and single fault diagnosis does not comprehensive.This paper introduces a t-SNE dimensionality reduction algorithms and correlation dimension estimate of intrinsic dimensionality binding GA-FOA-SVM to build high-speed train running gear of the multi-sensor fusion fault diagnosis model. Experimental results show that the model in diagnostic accuracy was 100% of the cases, while shorten the time to diagnosis and improve diagnostic efficiency.