Prior LDA and SVM Based Fault Diagnosis of Vehicle On-board Equipment for High Speed Railway

With the accumulation of maintenance data from the operation of vehicle on-board equipment (VOBE), it plays an important role in fault diagnosis and prognosis. However, natural language in maintenance data is a big challenge for fault diagnosis due to its irregular feature and uncertainty semantics. Some researchers have introduced text mining methods to deal with this problem, but they lose sight of the real meaning of the topics and some prior knowledge related to these topics which are important to efficient feature extraction. In this paper, we put forward prior Latent Dirichlet Allocation (prior LDA) and Support Vector Machine (SVM) based fault diagnosis. Firstly, Term Frequency & Inverse Topic Frequency (TFITF) method is proposed to extract prior knowledge around fault symptom, which then is integrated into basic Latent Dirichlet Allocation (LDA) to build prior LDA model. Next, we extract feature for classifiers with the prior LDA model from maintenance records. Thirdly, we give hierarchical classification model based on SVM and feature fusion method which are used for fault diagnosis. Finally, F-measure method is introduced to evaluate the performance of the proposed model with real data from high speed railway system in Guangzhou Railway Corporation. Experiments show that the proposed method outperforms text mining method which reckons without prior knowledge and other common methods of fault diagnosis.