| The switch device is a instrument for railway to change the position of the turnout,which can change the running path of train or trainset and ensure the safe operation of rolling stock on the turnout.The switch machine is an important component to complete the switch transition.Its working status detection mainly uses the current or power collected by the microcomputer monitoring system to set the threshold to determine whether a fault occurs.This method has low flexibility and high missed judgment rate.Therefore,manual screening is added to perform a second judgment,resulting in poor real-time performance and low efficiency,and can not form a systematic and intelligent maintenance plan.Therefore,this thesis takes the ZDJ9 switch machine,which is currently used for speed-increasing turnouts as the research object.Based on the current action curve collected by the microcomputer monitoring system,time domain analysis and wavelet packet transform are used to extract the fault feature set.The extracted feature set is used as the input of the genetic algorithm and the BP neural network optimized by the improved genetic algorithm to realize the fault diagnosis of the switch machine.The main research work of the thesis is as follows:(1)By analyzing the basic structure and working principle of the switch machine,combined with the literature and field investigation,summarized 8 common fault types of the switch machine,and analyzed the current curve of each fault type and the cause of the fault.Then,two methods of time domain analysis and wavelet packet transformation are used to extract the features of the action current.Aiming at the problem that the time domain analysis feature parameters lack the local features of the signal and the wavelet packet transformation feature parameters lack the overall features of the signal,a method of combining the time domain and the wavelet packet transform is proposed.This method is more comprehensive and detailed in signal characterization.(2)Using the extracted three feature parameters as the input of the neural network,a basic model of fault diagnosis for the switch machine based on the BP neural network is constructed.Through simulation and comparison,the fault diagnosis rate is up to 90%,and the corresponding training steps are 205.Aiming at the frequent occurrence of over-fitting,unstable training,and local optimization of the neural network during the training process,the genetic algorithm is used to optimize the input feature value,ownership value,and threshold value of the BP neural network.The results show that the diagnostic accuracy of the optimized model is increased to 95.6%,and the training steps is also reduced to 94.(3)In view of the premature phenomenon of genetic algorithm when optimizing the weights and thresholds of BP neural network,which leads to the poor ability of distinguishing similar states of switch machine,an improved genetic operation method in genetic algorithm is proposed to optimize the weights and thresholds.The results show that the improved algorithm has only one set of recognition errors in 90 sets of test data,the diagnosis accuracy rate reaches 98.9%,and the number of training steps is reduced to 69.It is better than the original in terms of distinguishing similar features and optimal fitness functions.The algorithm has also been greatly improved.The research method in this thesis has solved the problems of redundant input features,poor ability to distinguish similar states,and low diagnostic accuracy during the diagnosis process.It can efficiently and accurately diagnose the fault type of the switch machine,and provide a theoretical basis for the maintenance and repair of the railway field equipment. |
