Research On Simulation And Diagnosis Method Of Internal Leakage Failure Of The Three Gorges Ship Lift Anti-collision Buffer Hydraulic System

The Three Gorges Ship Lift Anti-collision Buffer Hydraulic System plays a role in buffering the ship’s impact force during the operation of the Three Gorges Ship Lift Anti-collision Device.Due to its more complex working environment,leakage faults can occur in actual work.Internal leakage fault,a form of leakage fault is concealed.When they occur,it is difficult for staff to detect them,and it is difficult to diagnose them in real time.When noise is mixed in the diagnostic signal,the diagnostic accuracy of the diagnostic method will decrease.Aiming at the internal leakage fault of the hydraulic system,this paper establishes an internal leakage fault simulation model of the system,and studies its internal leakage fault simulation,obtains fault samples for subsequent fault diagnosis method research.Based on convolutional neural network,the internal leakage fault diagnosis method of the buffer hydraulic system is designed,and it can realize the internal leakage fault diagnosis.Finally,the diagnostic method was improved to improve its anti-noise performance.The main research work of the paper is as follows:(1)The causes and mechanisms of leakage faults in the hydraulic system were analyzed.The normal working conditions of the Three Gorges Ship Lifter’s anticollision buffer hydraulic system and the internal leakage fault simulation model of the hydraulic cylinder and electro-hydraulic directional valve were established.The influences of piston displacement,pressure and flow of rod cavity and pressure and flow of rodless cavity is researched and it provides data support for subsequent research on fault diagnosis methods.(2)Using the structure of the convolutional neural network and Adam’s optimization algorithm,a convolutional neural network model for leakage fault diagnosis in hydraulic systems is established,a simulation automation program is written,so a large number of fault data samples are obtained.The data sets completed the training and testing of the network model,visualized its classification process,and compared the diagnosis accuracy rate with two machine learning algorithm-based diagnostic methods.Finally,it analyzed the training hyperparameters how to affect the diagnosis accuracy rate of the network model.(3)The BN and dropout algorithms and the corresponding formulas of the minibatch training method are discussed.Based on the convolutional neural network established in Chapter 3,it is improved,and the anti-noise performance of the diagnostic method is improved.There are: deepened network structure,added BN layer and dropout layer,and adopted mini-batch training method.Different levels of noise were added to the sample set,and the improved network model was trained,tested,and visualized in the classification process.Finally,in terms of anti-noise performance,it was compared with diagnostic methods based on machine learning algorithms and convolutional neural networks.