Research On Process-Oriented Real-time State Monitoring And Fault Diagnosis

A large number of state variables and parameter records will be generated during the development and operation of the aircraft,which can be used to analyze in depth the failure of each part of the aircraft system after the test.Because of the complexity of aircraft system,it is inefficient to interpret test data manually and there is a risk of missed judgment.Aiming at process-oriented real-time condition monitoring and fault diagnosis for complex systems,this paper studies the following technologies with single-axis inertia test and motion turntable as the research object:(1)Based on the research of multi-signal flow graph model theory,the fault analysis of control system of single-axis inertia test and motion turntable is modeled,and monitoring points are set up for data acquisition.(2)Through the research of data visualization technology,the time series characteristics of uniaxial INS test turntable data are mapped to images in the form of graph.(3)Through the research on the theory of image recognition based on neural network and the technology of data set making,data preprocessing is carried out.The training set,test set and verification set are made from the visualized image with normal data and fault data.(4)By studying the theory of convolution neural network,the structure of simple convolution neural network is designed.By studying Inception-v3 model and migration learning,the structure of Inception-v3 model is reformed.The data sets are used to train and optimize the parameters of the two models,and then the models are predicted and validated.The model selection is completed by comparing the data of accuracy,iteration times and training time.The purpose of recognizing the time series characteristics of fault signal data of single-axis inertia test and motion turntable can be achieved through the identification of convolution neural network.(5)Based on the research of rule-based expert system,combined with the result of threshold judgment and the result of sequential feature analysis,the inference engine is used to infer the final conclusion according to the rules and realize fault diagnosis.Based on the above theory and technology,a real-time condition monitoring and fault diagnosis method is proposed,and the key technology is validated through the system construction.This method innovatively uses data visualization technology and convolution neural network to recognize the time series characteristics of monitoring data.It can be applied to condition monitoring and fault diagnosis of complex systems with high time series requirements.It has application value for space equipment,new energy equipment,marine environment monitoring and atmospheric environment monitoring and analysis.