Research And Implementation Of Attention Based Prognostics And Radar Health Management System

Today,manufacturing craft and digital circuits are highly developed,various kinds of electronic equipment have been continuously upgraded,showing increasingly advanced but more complex features.The development of electronic equipment has brought unprecedented development opportunities to all walks of life,but it is also more prone to failure and more difficult to maintain.The problems of "under-repair" and "overrepair" are common in the later maintenance.Based on the research of PHM technology and prognostics technology,this thesis proposes and implements a radar health management system based on prognostics algorithm,aiming at that traditional maintenance methods such as "periodic maintenance" and "manual decision" of radar equipment are hard to meet the needs of modern radar maintenance.The main research contents are including:(1)To solve the defect of traditional prognostics algorithms that cannot extract the order dependencies among long sequence data,a new prognostics method based on the combination of self-attention and long short-term memory networks is proposed.The core idea of this method is to use the LSTM layer to capture the time-dependent relationship of the monitoring data,and then to filter the feature information in the intermediate output results of the LSTM layer by the self-attention mechanism,which realized the selection and preservation of long-term memory and solved the problem that the traditional prognostics algorithms are inferior in long sequence scenes.(2)To solve the problem that the traditional prognostics algorithms have a significant decrease in prediction performance and modeling difficulties under multiple operating conditions,the use of the multi-head attention mechanism enhances the ability of the prediction model to identify different working environments and improves the prediction accuracy of the algorithm in complex environments.To solve the problem of loss of time information in the monitoring data during training,the positional relationship information is embedded by an automatic encoding method.(3)Designed and developed a radar health management system based on the Django framework,and used phased array radar as the application object,and realized the condition monitoring of PAR,the assessment of the health status of the array,the prediction of the RUL of the channel,and the maintenance arrangement,model training and other important functions.Finally,comprehensively test the system's performance and function.The Attention-LSTM based prognostics method proposed in this thesis solves the problem that traditional prognostics algorithms cannot extract the deep characteristic information in the data and the problem of long-term memory to a certain extent;the health management system designed and implemented in this thesis also improves the quality of radar equipment maintenance and reduces the difficulty of radar equipment maintenance to a certain extent.