Neural Network Recognition And Credibility Evaluation Of Sealed Relay Components Signals

As an important basic component in the field of aerospace,hermetically sealed relay is closely associated with the reliability of the entire aerospace system.The particle impact noise detection method is currently the main method for detecting remainder particles in aerospace electronic components in China.Aiming at the problem of low accuracy of component signal recognition by traditional detection methods,the thesis applies machine learning method to the detection of signal of sealed relay components,uses neural network model and combines multiple features to classify and identify remainder particles signals and component signals.The thesis also intends to optimize the hyper-parameters of the neural network so as to find the optimal combination of hyper-parameters to build the model.It is proved through experiments that this method can improve the accuracy of classification.Sealed relay is mostly used in aerospace field.If an error occurs in terms of the classification results,it may cause huge losses.In this thesis,the reliability evaluation method for soft and hard classifiers is designed to solve the problem of lack of reliability of evaluation for single detection.Considering the common problem of uneven sample distribution in engineering practice,this thesis proposes a single-pulse credibility evaluation method combining experimental sample distribution and sample contribution rate based on this method.The experimental results show that the single-pulse credibility evaluation method can provide a reliability evaluation value for a single detection result,which is of great practical value in practical applications.