LKJ Operation Recording Data Analysis And Fault Diagnosis Based On Clustering And K Nearest Neighbor Algorithm

Train operation monitoring and recording device is the speed control system of train, which monitors the operation of train safely and is widely used by railway administrations at present. Besides of monitoring function, LKJ(train operation monitoring and recording device) also has the recording function. The LKJ operation recording data files generated by the train operation monitoring and recording device are the effective judgment bases of analyzing train accidents and fault diagnosis of on-board equipment. Currently, the fault diagnosis of on-board equipment of locomotives based on LKJ data is still manual analysis that could have several disadvantages such as a lot of time is spent in determining the cause of the fault, low fault diagnosis efficiency and error-prone while analyzing. In view of above drawbacks, the fault diagnosis of locomotive speed sensor that combines fault diagnosis algorithm with LKJ data analysis is proposed in this dissertation.Firstly, LKJ system composition and the recording contents of the LKJ operation recording data files are introduced and the locomotive speed sensor is identified as fault diagnosis research object in this thesis. Secondly, the structure and the working principle of locomotive speed sensor are analyzed. At the same time, the composition structure of LKJ speed channels is studied. On this basis, kinds of fault types of locomotive speed sensor are determined though site investigation. Subsequently, after the original LKJ operation recording data files format is converted and LKJ data preprocessing is accomplished, data analysis of 25 LKJ data samples of speed sensor in normal condition, 13 LKJ data samples of virtual connection in speed channels and 5 data samples of speed sensor with broken shaft is conducted by making use of EXCEL statistical analysis software. Furthermore, 4 failure rules are obtained by taking corresponding fault symptoms of different fault types in LKJ data files and expert experience into consideration and finally the fault diagnosis characteristic vector is selected based on these 4 failure rules. Clustering-KNN algorithm is the fault diagnosis algorithm in this dissertation through clustering algorithm is used to improve K Nearest Neighbor(KNN) algorithm in terms of ignorance of the importance of neighboring samples to the class to which belong in KNN. Cluster centers of each class are calculated before classification and KNN algorithm is corrected by computing the important factors of each neighboring sample to its cluster center while classification. At last, locomotive speed sensor fault diagnosis based on clustering-KNN classifier is realized by means of programming with MATLAB and some maintenance advice is given.It can be seen from the results of locomotive speed sensor fault diagnosis based on LKJ data analysis that the clustering-KNN classifier has better performance than KNN classifier in precision such as diagnostic accuracy, classifier recall and F1-measure, but could spend more time to finish the diagnosis. However, the clustering-KNN classifier still can meet the requirement of rapid diagnosis on site. Thus, the efficiency of speed sensor fault diagnosis according to LKJ operation recording data files increases and manpower and material resources costs reduce. Therefore, research in this dissertation has practical significance for improving the railway transport turnover efficiency.