Research On Comprehensive Fault Evaluation Of Maglev Train Based On Machine Learning

Comprehensive fault evaluation is one of the most important technologies of operational safety. However, the traditional evaluation methods commonly have difficulty in system modeling and effective adaptation, etc. Since the self-developing maglev is turning to commercial use, developing a new kind of effective evaluation method is of great urgency. The development of machine learning attracts great attention in recent years. It is efficient not only in obtaining knowledge automatically but also in adapting the model, which is especially suitable for maglev fault evaluation. Aiming at the improvement of performance of evaluation, a technology of comprehensive fault evaluation based on machine learning is studied in this dissertation, the major work include:Firstly, a fault sorting method based on minimal cut sets of the fault tree is proposed. By this method, the component faults of maglev train are sorted, which greatly decreased the number of objects to be evaluated and the complexity of comprehensive evaluation problem;Secondly, the key issues of applying machine learning to comprehensive fault evaluation of maglev are studied and a model of it is built. Representative machine learning algorithms are analyzed and tested in the model. Experimental results show that machine leaning method is faster in getting the relations of fault influences automatically than traditional ones and has good accuracy;Thirdly, a comprehensive fault evaluation method based on ensemble learning is proposed, such that the performance of evaluation can be improved through the ensemble and complementation of different classifiers. Experimental results show that the performance of this method is commonly better than the machine learning algorithms it combined;Finally, the maglev supervision systems on train and off train are designed, on which the comprehensive fault evaluation system can work practically.