Optimal Search And Fault Prediction Based On High-power Millimeter Wave Test Systems

Since the testing of high-power millimeter wave devices is a multi-input and multioutput parameter testing process,there are many test parameters obtained in the whole process.The analysis process of the data collected from the tests is still mainly carried out manually,mainly by the research-related personnel based on their actual calculations and experience,which makes it difficult to find the best working parameters.And the existing high-power millimeter-wave test platform's basic protection focuses on the hardware level of protection,mainly in response to the occurrence of failure,which can cause certain device damage and economic losses,for the software level of protection measures are lacking.The existing intelligent test system for the industry is mainly focused on the data collection,automatic data cleaning and storage,basic control protection and other needs,however,the analysis capability of these data is weak,and based on these data analysis of the software level functions such as the best working parameters search and fault prediction can not be carried out.This is also not a good use of the test data that is continuously collected,which also leads to the inability of these data to provide more effective feedback to researchers,especially device designers.In this paper,in order to solve the problems in the above,we study the application of predictive models for high-power millimeter wave test data and expand the functions of intelligent test systems by introducing intelligent learning algorithms in related fields such as neural networks,as described below:1.The expansion scheme of optimal search and fault prediction based on intelligent test system is proposed.The existing test system has made a lot of work for in automatic data processing,however,the function of prediction analysis based on data is weak,and the data modeling of test data is expanded by introducing neural networks and other related intelligent learning algorithms for more functions.By analyzing the existing system framework,the expanded functions are analyzed from the software level.2.The impact and significance of the prediction study on this topic is introduced through the test data modeling prediction as an entry point,and the core prediction model of the extended module is analyzed in detail by using different neural networks and different ways to model the data algorithmically,and single-output prediction of the "efficiency" attribute is performed based on the available data,and finally The results are analyzed.3.By using principal component analysis to reduce the dimensionality of the test data,a prediction model was developed based on the optimized data and compared with the prediction model in the previous section.Based on the existing model,the most searchable module is developed,and the optimal working "efficiency" is the goal to find the working parameters in this state.Finally,the failure prediction modeling is analyzed and feasible solutions are given.