Study On Text Mining Based Fault Classification For Turnout

With the rapid development of railway transportation, increasing demands is proposed from the railway transportation for the signaling system. Meanwhile, a more efficient maintenance of equipment is needed for the signaling system. Turnout is one of the important equipment in the railway signaling systems to ensure the efficient operation of railway transportation. Hence, turnout is the key equipment in the maintenance of the Communication and Signal Division. Workers in the Communication and Signal Division spend a lot of time and money to maintain the turnouts, but the turnout fault is still in the tops list of the faulted parts. Therefore, it is of great significance to study on the categorization of turnout fault and to improve the efficiency of turnout maintenance.In the process of operation and maintenance for turnout, a lot of maintenance data has been accumulated with the form of text. These maintenance data can be used as important reference to analysis equipment fault and to make the corresponding maintenance decision. Based on the fault statistics list provided by Guangzhou Railway Group, we can found that:entering of the fault data lacks of standardization, resulting in a low using value of the fault data; Category of the fault in the table is unreasonable, which brings some difficulties in the statistics and processing of the fault data. Railway signaling equipment maintenance managers hope to deal with these problems by making some rules and regulations, but the effect is not good as they expected.For the above problem, this thesis conducted the following research:Based on the structure composition and work principle of the turnout equipment, the turnout fault is classified by using the turnout fault recording data; By using text mining, a fault classifier for the turnouts was designed to realize the automatic classification of turnout faults; Using the reset turnout fault equipment category and the proposed turnout fault classifier, the form of turnout fault data entering was designed, in which the real-time data is recorded with the designed form such that an accurate and effective entering of the data is ensured.The specific tasks and achievements of this thesis are as follows:(1) The structure and working principle of turnout is analyzed and the related equipment is found with the recorded fault data. Then, the faults are classified according to the fault related equipment;(2) The jargon is extracted from turnout fault data. Then, the jargon is put into the user dictionary of NLPIR Chinese word segmentation system (ICTCLA) such that segmentation processing on the turnout fault text is fulfilled;(3) For the diversification of turnout fault data description, PLSA topic model feature extraction method is applied to extract the fault text feature;(4) The support vector machine (SVM) method is applied to design turnout fault classifier, which implements automatic classification of the turnout fault equipment category. The proposed classification methods are compared with others. Comparison results show that the proposed method by combining the PLSA features extraction and SVM classification is better;(5) According to the classification results and the practical requirement, the form for entering the turnout fault data is designed, which normalizes the entering of the turnout fault.This thesis studies the classification of the turnouts and the from for entering the turnout fault, which enhances the standardization of turnout fault data entering, ensures the effectiveness of the fault data entering, and then improves the efficiency of the turnouts maintenance and management.