Research On Fault Current Recognition For ZD6 Switch Machine

As the scale of train transportation continues to expand,the safe and stable operation of the railway system is significant.It is necessary to conduct more comprehensive real-time monitoring of the equipment in the system to identify and troubleshoot hidden troubles quickly through the obtained data.The switch machine is one of the most important equipment in the turnout system.For a long time,the recognition and positioning of its faults mostly rely on the experience of the staffs.This traditional method has low recognition efficiency and is prone to misjudgment.Focusing on the shortage of analysis on the fault feature contribution degree in the traditional fault recognition for ZD6 switch machine,the multiple feature extraction methods are used to establish the initial feature set based on the actual switch machine operating current data.The best feature subset is established by feature evaluation and selection techniques which contributes the most to the fault recognition classification.Then,the recognition model is designed with neural network combined and is applied to the switch machine fault current diagnosis.Firstly,the basic structure and working principle of ZD6 switch machine are expounded.The current curve under normal working conditions is determined by analyzing the working process of the switch machine and the principle of current acquisition,and several common fault current curves are summarized.Secondly,the features parameters of the switch machine operating current are extracted including time domain features,time-frequency domain features and morphological features.The time-frequency feature is extracted by the Wavelet Transform and Empirical Mode Decomposition with their strong time-frequency analysis ability obtaining the frequency band component of the signal and extracting the features.The morphological features refers to the human visual perception process to calculate accumulated difference value by establishing the set of current shape contour points of different curves.A current recognition model based on feature-evaluation-selection and neural network is established.The model can identify the best feature subset for fault recognition from the original feature set,which overcomes the blindness of feature selection in traditional recognition methods.The results of recognition and verification based on the current data collected by the microcomputer monitoring system shows that the feature-evaluation-selection plays an important role in the fault current recognition and improve the accuracy of the fault current recognition.Finally,the design of the fault current recognition software of the ZD6 switch machine is finished.Based on the recognition model with feature evaluation selection and neural network method,the design of the software is realized by MATLAB and Qt joint programming.This software includes functions such as curve recognition,introduction of new fault types and model updating.The test sample is proved that the system can effectively identify the type of the fault curve and assist the worker in quickly troubleshooting.