Research On Electric Gate Valve Fault Diagnosis Based On Convolutional Neural Network And Meta-Learning

As a class of important equipment in nuclear power plants,valves mainly play the role of opening and closing pipelines,controlling flow direction,regulating and controlling the parameters of the transported medium,etc.Due to the harsh working environment and being used frequently in nuclear power plants,valves have a higher probability of failure compared with other equipment.According to statistics,valve failure occupies a large proportion of nuclear power plant shutdown factors,which not only limits the economic benefits of nuclear power plants,but also greatly increases the threat of radioactive leakage in nuclear power plants.Among the types of fault valves,gate valves and solenoid valves account for the highest proportion of failures,up to 48%.Futhermore,in the actual industrial production,the equipment is in normal operation in most cases,and there are few fault samples.This puts forward higher requirements for the generalization ability and robustness of the model.Therefore,the electric gate valve is taken as the research object in this paper.The experimental data of electric valve is collected,and the key technologies of signal processing and fault diagnosis of electric valve are studied.On the basis of the above work,the method for electric gate valve fault diagnosis based on improved Hilbert-Huang transform-convolutional neural network and the few-shot electric gate valve fault diagnosis method based on metalearning with discriminant space optimization are proposed in this research.The following research work is carried out.(1)Three types of electric gate valve faults are collected,electric gate valve three-phase unbalance fault,electric gate valve sealing packing damage fault and electric gate valve internal leakage fault.According to the occurrence mechanism and fault characteristics,the sensor is arranged to complete the three types of fault data and normal state experimental data collection.(2)The improved Hilbert-Huang transform method is used for feature extraction to obtain the IHHT time-frequency diagram.The feature extraction of vibration signal and acoustic emission signal is completed.(3)The lightweight convolutional neural network model is established.Input the extracted IHHT time-frequency diagram into the convolutional neural network for training.And the trained model is applied to the test set to verify the accuracy of the model fault diagnosis.(4)The few-shot electric gate valve fault diagnosis model based on meta-learning with discriminative space optimization is established.And the construction and optimization of the few-shot fault diagnosis model are completed to verify the accuracy of the model in fault diagnosis under the few-shot condition.(5)Two fault diagnosis methods,the convolutional neural network and model-agnostic meta-learning are set as comparison methods to complete the few-shot fault diagnosis comparison experiment of electric gate valve to verify the superiority of the proposed few-shot electric gate valve fault diagnosis model based on meta-learning with discriminative space optimization.The experimental results show that the method based on IHHT-CNN has a high accuracy rate in the classification of electric gate valve faults;the few-shot electric gate valve fault diagnosis method based on meta-learning with discriminative space optimization achieves good fault diagnosis results for electric gate valve.Compared with other methods,this method has better diagnostic effect under the few-shot condition and has better anti-noise ability.