Fault Diagnosis Method For Railway Switch Point Based On Fuzzy Neural Network

The core content research on fault diagnosis method of high-speed railway turnout is how to effectively access, analysis, process and make use of historical data and expert experience to predict and recognize the operation state of the turnout accurately in a given environment. The bottleneck lies in the restriction of priori knowledge acquisition, the complex working environment, the uncertain fault mechanism, and the randomness of fault forms, problems of using relationship between historical sample information to deal with fault, over the all above problems, this thesis focuses on feature knowledge representation, preprocessing technology, diagnosis model parameter learning, fault and symptom corresponding rules set. Different faults may exhibit same symptoms, different symptoms may also lead to the same fault, so the relationship between fault and their symptoms is often fuzzy, comparing with human involvement in setting the symptom threshold, fuzzy language variables can describe the symptom characteristics more accurately, therefore, we have Fuzzy Neural Network (FNN) theory in application to fault diagnosis of high speed railway turnout. This paper mainly finishes the following work:First of all, analyze switch fault diagnosis research status at home and abroad, summarizes current situation and problems existing in switch fault diagnosis, proposes the needs of fault diagnosis for switch; through analyzing fault characteristics and the cause of the problem to prepare for the research behind.Secondly, according to failure mechanism, failure mode of the turnout, and turnout operation process, extract characteristics of the switch current curve using three kinds of methods, then select the most significant feature which are high efficiency and most associated with the target concept in distinguish faults through Relief algorithm as diagnostic system input, in order to reduce the dimension of feature, the performance and operation efficiency of diagnosis models.Then, by using fuzzy language to describe the relationship between the symptom and fault, and the established rules based on expert experience, work has been done under learning ability and optimize algorithm of FNN, including automatic updates of the rules, network parameters put forward, and the construction of diagnosis model.Finally, by simulation, validate the efficiency of FNN model, the error rate can meet the requirements and the diagnosis result is satisfying, proving that the method is feasible, then complete fault diagnosis software design under requirements, put forward maintenance advice and fault state.