Research On Running Gear Vibration Data Processing Of The High-Speed Rail Based On Spark Parallel Computing Framework

The security problem of high-speed rail has gained more and more attention. The vibration signals of the train operation are collected by the sensors installed on the high speed rail. The faults of high speed rail can be discovered by analyzing and processing the collected vibration data. In order to ensure the safety of train operation, the sensor is installed on the train more and more. These sensors can collect a lot of vibration data during the train running. How to quickly extract features and classification and recognition from these massive vibration data is a difficult problem needs to be solved.As a large data processing framework based on distributed memory computing, Spark has a unique advantage. It can read files from the HDFS file and can read files from the local file system. It is based on the idea of programming similar to MapReduce programming. It dispatches tasks to different Workers by applying the divide-and-conquer principle. Finally the results of the implementation of the summary are collected to the Master. The Empirical Mode Decomposition (EMD) is a method which decomposes nonlinear and non-stationary signals to a sum of several intrinsic mode functions, so it plays a vital role in the domains of signal analysis and processing. Ensemble Empirical Mode Decomposition (EEMD) is an improvement on EMD, which is added to the original decomposition signal Gauss white noise. This eliminates the EMD in the decomposition process occurs in the mixed linear mode.This thesis presents a parallelized EEMD algorithm, parallelized building energy moment, parallelized k-Nearest Neighbor (KNN) classification algorithm under Spark, a framework based on the distributed memory computing and Resilient Distributed Datasets (RDD). The real data is employed to evaluate the parallelized algorithm and compare it with the result of stand-alone to verify its correctness. The experimental results are analyzed by three indexes, e.g., Speedup, Sizeup and Scaleup. It is shown that the parallelized method has good effect on the three indexes. These parallel algorithms can improve the processing efficiency of high speed rail vibration data and can provide a reliable solution for the decomposition of a large number of vibration signals.