| Accompanying with the rapid development of the rail transit industry,the metro has gradually become the primary option of green travel.The metro platform screen door system is an important link to ensure the safe operation of metro rail transit,and its safety and reliability are of vital importance.The metro door control unit is the most frequently used equipment in the metro platform screen door system,so the working status of its main components directly affects the stable operation of the screen door system.The metro door control unit is driven by brushless DC motor,and the rectifier,filter,inverter and other power conversion devices contained in the driving system are the components with the highest failure rate in the entire door control unit system,so it is of crucial significance to conduct early fault diagnosis research on the power conversion devices of the metro door control unit driving system.The traditional diagnostic method on the basis of model and signal analysis is difficult to adapt to the fault diagnosis research of complex systems composed of multiple signals,while knowledge-based intelligent fault diagnosis method without analyzing the precise mathematical model principle of the system can achieve feature acquisition and classification,which has high recognition accuracy.Therefore,this thesis adopts the intelligent fault diagnosis method of wavelet time-frequency analysis combined with deformable convolutional neural network to conduct the diagnosis research on open-circuit fault of power conversion devices.Firstly,analyzing the fault mechanism and characteristic changes of the devices according to the principle and characteristics of the research object.Taking the metro door control unit driving system as the research object,this thesis analyzed the fault principle of power devices with frequent faults,observed the changes of the system working state under various fault types,and clarified the correspondence between the fault types and characteristic changes.Secondly,in order to collect fault signals,this thesis made a simulation model of the power conversion part of the brushless DC motor drive to simulate various fault states.Based on the mathematical model and principle of the motor driving system,this thesis simulated the dual closed-loop control system in Simulink by means of the control switches to simulate the normal and fault states,collecting the current and voltage signals of various fault types.Then,using the wavelet time-frequency analysis method to obtain the time-frequency images of the fault signals.The principle of wavelet time-frequency analysis and the characteristics of the generating function are introduced,and this thesis adopts the continuous wavelet transform to perform time-frequency conversion on the collected fault signals under different fault types,to obtain time-frequency images that can reflect the fault characteristics,and to construct the time-frequency image data set.Finally,constructing a deformable convolutional neural network to identify and diagnose power conversion devices faults of brushless DC motor driving systems in metro door control units.The algorithm principle,training process and optimization strategy of convolutional neural network are introduced.This thesis adopts the deformable convolutional algorithm to extract features and classify the data set of time-frequency images.And in order to reduce and update the parameters of the network model,made an introduction of Dropout regularization and Adam gradient descent algorithm,continuously optimizing and improving the accuracy of the model.In addition,this thesis obtained the real fault data on the experimental platform to verify the accuracy of the algorithm model. |
