Health Management Of Train Operation Data Based On Vector Quantizationand Dynamic Time Warping

With the vigorous development of our country’s high-speed railway business, train operation safety reliability has become increasingly important. Train operation data can reflect the working condition of the train equipment. Due to environmental factors and the lack of accurate judgment on the train operation data and health analysis, on the one hand it will cause train safety and efficiencyproblems, on the other hand it has difficulties for the maintenance of the train.Health management technology has got rapid development after decades of research and application in the aerospace system, fault diagnosis, condition monitoring and so on. As the breakthrough point of condition based maintenance, this paper introduces the concept of health management of train operation data. Condition based maintenance can monitor and recognize potential failure as soon as possible, and also it can carry out health management and fault prediction with working condition of train equipment. Using the health management technology, it will change the traditional condition monitoring and fault diagnosis into the integrated management based on intelligent algorithm, providing the technical foundation for the efficient operation of train maintenance.According to the actual demand of train operation data, integrating historical monitoring data and test data of train, this article focuses on the condition monitoring and fault diagnosis of train operation data, also designs and implements health management platform of high real-time, high accuracy and wide range of application for train operation data. At the same time, combining with the similarity of speech recognition and pattern recognition, this paper introduced two mature algorithms in speech signal processing field, Vector Quantizationand Dynamic Time Warping, into the train operation data of the health management field.As train operation data may consist of abnormal data,after corresponding pretreatment, the platform uses Vector Quantization as the training method and gets the optimal codebook of a large amount of iterative training reference sample. At last, the platform uses Dynamic Time Warping to compare real-time operation data with reference sample to get the condition monitoring and fault diagnosis results based on historical statistics data and the outcoming of similarity comparison.The results show that health management of train operationdata based onVector Quantizationand Dynamic Time Warping can meet the real-time and accuracy demand of high speed train, effectively eliminating the effects of measurement noise of train operation data and jump, which has the very high value of engineering application.