The Research And Implementation Of A Method For Recognizing Switch Fault Current Curve Based On Similarity

With the continuous development of China’s railways, especially the constant growth of the high-speed railway operating mileage, the safe operation of the railway signaling equipment put forward higher requirements. Switch is one of the main railway signal equipments and is important to ensure the safety of railway operation. At present, in most of railway administration, maintenance workers rely on their experience to take periodic maintenance way to maintain switch. But, not all maintenance workers have enough work experience to accurately determine fault and maintenance. At the same time, the periodic maintenance repair mode does not guarantee timely detection switch failures and risks. These will affect the efficiency of the railway transport and even security. In order to solve these problems, on the basis of grasping the operating status and various data of the switch, researching a system which is able to identify and diagnosis switch faults and provide the specific help to maintenance workers.Based on the comparison of the research status at home and abroad, this paper studies on the working principle and failure modes of the switch, especially researches and summarizes the regular change of switch machine operating current. By comparing the difference between the normal operating current and the abnormal operating current, judgments based on similarity, this paper respectively introduces Fuzzy Neural Network and Support Vector Machines to diagnosis switch faults. The results show that both methods can identify the switch faults, but, in comparison, switch fault recognition method based on SVM shows a better performance and a higher recognition rate.This dissertation mainly completes the following work:First, based on the basic structure and working principle of switch device, switch conversion process will be divided to unlock, transforming and locking structure and connecting indication circuit four working process. And according to the four working process of switch, switch action current curve is divided into four segments. Then select typical switch faults in each time zone and analysis their current curves and possible causes of faults.Secondly, A fault diagnosis method based on fuzzy neural network is put forward. Preprocess the data of switch action current and compose the training samples and test samples of switch fault diagnosis model first. Then construct a fuzzy neural network model and train the model by the training samples. Input the test samples and test model diagnosis correct rate. From the simulation results it can be concluded that: this method can achieve the switch fault diagnosis, and the diagnostic accuracy rate is normal.Third, the machine learning is applied to the fault diagnosis of switch. The switch fault diagnosis method based on support vector machine is put forward. First to support vector machine theory is expounded, and analyze the influence of various kernel functions and parameters on the classification results of support vector machines. Finally a switch fault diagnosis model based on Gauss kernel function is determined. Adopt different feature extraction methods, processing the current curves of action switch machine, then get the training samples and test samples. The training samples are used for model training to get the model of fault diagnosis. Input the test samples and test model diagnosis correct rate. The experimental results show that the fault diagnosis method based on support vector machine in small samples can still achieve fault diagnosis.Finally, this paper compares the performance of the two models, summaries their advantages and defects, and draws the conclusion. This paper also predicts the practical application in the future.