Research On The Technologies Of Fault Diagnosis Of Multi-machine Tracting Switches Based On Kernel Fisher

With the opening of the large-scale passenger dedicated lines and the increasing demand value of train speed,transportation operations have put forward higher requirements for the safety of switching equipment in railway lines.As the speed of passing lateral switch machines in main lines for high-speed railway is mostly more than 80km/h,switch machines with number 18 or above are usually installed,which are multi-machine tracting switches.Currently,the domestic research mainly focuses on single machine tracting switches.Since there are some technical problems of synchronous conversion in multi-machine tracting switches,the types of fault are various and difficult to detect.Meanwhile,current technologies of fault diagnosis mainly rely on the experience of signaling maintenance and the data collected by centralized signaling monitoring system(short for CSM),which takes a long time and has poor reliability in fault diagnosis.Thus,the research on intelligent fault diagnosis of multi-machine tracting switches is one of the most pressing problems.In the thesis,the current curve of switches collected by CSM is taken as the research object.For further research on the fault diagnosis of multi-machine tracting switches,a method of fault diagnosis based on kernel Fisher was proposed.Combined with a large number of practical cases and the principle of switch control circuit,according to the section characteristics of current curves of switch and the influence degree of traffic safety,four kinds of typical fault modes of switch are summarized and analyzed.Firstly,according to the process of fault diagnosis,the fault features are represented intelligently.The fault feature representation of intelligent split switch current curve based on time domain analysis is designed.This method not only adapts to the switch under different working conditions,but also can collect the characteristic value of switch current curve intelligently,which is able to construct switch fault feature representation sets with high efficiency and low error.Then,the kernel Fisher analysis method after kernel function optimization is applied to extract fault features from the fault feature set,and the data is preprocessed to achieve the dimension reduction.Finally,a fault diagnosis system based on support vector machine model is established to classify the feature vectors after preprocessing.At the same time,the kernel function is further optimized and complex kernel function with high nonlinear classification ability is used.A fusion algorithm parameter optimization strategy based on intelligent algorithm and grid search method is proposed.The optimized support vector machine model is simulated by MATLAB software,and the test set which is extracted by feature is inputted into the training fault diagnosis model,and the fault classification result is output.The experimental results show that the kernel Fisher analysis method with great robustness is more effective than the linear Fisher analysis and kernel principal component analysis,and it can better represent the characteristics of the original data.Furthermore,the fault diagnosis model after parameter optimization has high convergence speed,classification accuracy rate and test accuracy rate.In a word,the method proposed in this thesis has high classification accuracy for fault diagnosis of switches.The fault diagnosis system has excellent reliability,and can be applied to assignments of field maintenance for railway.