Research On Key Technologies Of Turnouts Fault Diagnosis Based On Monitoring Current Curves

The railway turnout system controls the direction of the train,and its health directly affects the safety of the train.When a turnout fails,being able to find and identify the type of fault in a timely manner is of great significance to ensure train operation.At present,improving the safety of turnouts mainly depends on relevant technical personnel observing the turnout current or power curves collected by the computer monitoring system to judge the health status of the turnouts.This manual-led diagnosis method is inefficient,and the accuracy of diagnosis mainly depends on the work experience and professional knowledge of the technical personnel,which leads to frequent false alarms and missed reports,and serious waste of manpower and resources.In response to the above problems,this thesis proposes a rapid detection and diagnosis method for turnout faults based on monitoring current curves.The main work is as follows:(1)Whatever the cause of the turnout fault,the fault current curve will inevitably deviate from the normal current curve.In view of this,a quick detection method for turnout faults based on the similarity of curves is proposed by analyzing the similarity between normal and fault current curves.Firstly,using the historical current curve of the same turnout,a normal template current curve generation method is proposed based on the mode and median,so that the normal template current curve can be updated according to the stage difference of the turnout itself,thereby improving the adaptive ability of the method.Secondly,according to the change of the slope of the curve in the three stages of unlocking,switching and locking of the turnout,the current curve is divided into three sections using the slope energy characteristics.Then,FastDTW is introduced to align the template curve of each section with the curve to be inspected in the time domain,the regular path distance(curve similarity)is calculated.Finally,the best similarity threshold is obtained through experimental search,and the similarity threshold is used to determine whether the turnout has failed.Experimental results show that the proposed method has higher accuracy and fast detection speed.(2)Aiming at 6 typical turnout fault modes,a turnout typical fault diagnosis method based on FS-WKNN is proposed.Firstly,the discrete wavelet transform and common statistical parameters are used to extract the two-phase fusion features of the current curve.Then,the Fisher feature selection method is used to calculate the Fisher Score of each fault mode feature and the normal mode,and features with great contribution to subsequent classification are selected.Finally,a K-Nearest Neighbor algorithm based on class features weighting(FS-WKNN)is proposed by using Fisher Score calculated from each fault mode and normal mode as features weight,FS-WKNN is used as a classifier of different fault modes and normal modes to realize fault diagnosis.The experimental results show that FS-WKNN can effectively improve the diagnostic accuracy,and the average diagnostic accuracy rate reaches 90.47%.