Research On The Key Technology Of Meteorological Radar Fault Diagnosis And Health Management

With the diverse climate changes and the frequent occurrence of extreme weather,the agricultural sector is becoming increasingly dependent on meteorological radars.Meteorological radars are generally established in sparsely-occupied open-air places.The damage to various weathers and the natural aging of parts and components all affect the performance of the meteorological radars and the normal business.At present,such methods as periodic "timed maintenance" and "after-the-fact maintenance" after a failure are generally adopted.Such strategies waste human and financial resources and have poor reliability.Therefore,it is necessary to study the fault diagnosis and health management of meteorological radar,and realize the transformation of maintenance strategy from "after-the-fact maintenance" and "scheduled maintenance" to preventive maintenance in order to improve the maintenance efficiency and ensure the operational capability of weather radar.The main work of this paper includes:First,according to the composition and working principle of the meteorological radar system,combined with the historical data of meteorological radar faults,the fault types,time of failure,and the proportion of various faults are analyzed to provide support for subsequent fault diagnosis.At the same time,the theoretical knowledge of fault diagnosis and health management was sorted out,and the health management system architecture for weather radar was designed.Secondly,in-depth research on the fault diagnosis technology of meteorological radar servo motors.Four methods are used to realize the precise diagnosis of servo motor faults.First,the signal features are manually extracted in the time domain and sent to the neural network.Fault diagnosis;Next,the LSTM long-term and short-term memory network with excellent processing ability for time series is used to complete the fault diagnosis;then,combined with the artificially extracted features are time-dependent,they are sent to the LSTM network for fault diagnosis;Finally,the Conv1 D convolutional neural network is used to automatically extract the features of the signals.In combination with the LSTM network,a Conv1D_LSTM deep neural network method is proposed for fault diagnosis.This method can simultaneously feature automatic feature extraction and time dependence.The experimental results show that the Conv1D_LSTM deep neural network method is significantly better than other methods and can well solve the problem of meteorological radar servo motor fault diagnosis in this project.Third,the design and implementation of a health management system for weather radar.Aiming at various problems in the real-time data collection of meteorological radars across the country,a four-layer meteorological radar health management system software structure was designed for data collection,data storage,data mining and analysis,and data visualization.Characteristics,detailed technical requirements analysis,and finally chose to use Kafka technology in the data collection layer,HDFS and Redis technology in the data storage layer,Spark real-time computing framework for data mining and analysis,and Django web framework for data visualization layer.After completing the software architecture design and technology selection,a test system was set up to implement modules such as simulation data generation,real-time data reception,model training,model diagnosis,and data display.