Research On Algorithms Of Fault Diagnosis And Health Evaluation For Turnout Switch Machine And Its System Implementation

As an important device for subway rail transport,turnout is used to control the direction of trains.,and The normal operation of that device is important for driving safety.Even for now,the inspection and maintenance of turnout is still depending on manual,which is inefficient and low accuracy.Therefore,this paper completes the fault diagnosis and health status assessment of turnout Based on S700 K turnout machine.The main contents of this thesis are as follows:(1)Research and implementation of fault diagnosis algorithm for turnout.Firstly,According to the actual situation of the scene and combined with the experience of experts,this thesis summarizes the faults of the switch machine to determine the type of fault to be diagnosed.Secondly,in order to diagnose the faults effectively,the thesis designs the DDAE＿BILSTM fault diagnosis model,and compared the model effect from the accuracy,false positive rate,false negative rate and other indicators through experiments,the results show that the model is higher than other models.Finally,In order to further improve the fault recognition rate,the DDAE network is improved in this thesis.The encoding and decoding layer of DDAE network is changed to one-dimensional convolutional neural network,and the DCDAE＿BiLSTM model is constructed finally.The experimental results show that the fault recognition rate is improved to 98.82%.(2)Research and implementation of health assessment algorithm.Firstly,in order to further reduce the failure rate,by analyzing the normal power data on the eve of the failure in the switch conversion process,the time domain,the parameter of ARMA model,sample entropy and other features are extracted to establish the original feature space.secondly,in order to find the appropriate sensitive features,the combined evaluation of monotonic and tendency is used to select the feature of samples,the selected features are used as the input for subsequent degradation state mining and recognition;then a variety of clustering algorithms are used for unsupervised learning,combined with contour coefficient indicators to achieve the automatic identification of the degradation state number,and on this basis,a degraded state sample library is built.Finally,the XGBOOST algorithm is used to automatically identify the degradation state,accuracy reaches 93.23%.(3)Design and realization of turnout health maintenance system.According to demand analysis,the system architecture of the turnout health maintenance system is designed,and the front and rear ends separation technology is used to realize the turnout health maintenance system.