Research On Fault Prediction Method Of Analog Circuit Based On Deep Learning

Satellite communication system plays an important role in sea,land and air fields,because of its wide coverage,long communication distance and high quality.Once it breaks down,it will not only affect the safety and stability of the system,but also may lead to the failure and paralysis of the whole communication system,which will cause huge economic losses.Therefore,it is urgent to change from post-maintenance to premaintenance.As an important part of the communication system,analog circuits are important causes of system failures.Therefore,the research on fault prediction method becomes a hot spot.The essence of analog circuit fault prediction is to accurately predict the future operation trend according to the historical operation data of the circuit,so as to provide basis for advance maintenance and guarantee the reliability of the circuit.Due to the high complexity of the system,the occurrence of its failure is generally based on the equipment,but the cost of replacing the equipment is high,and there is maintenance surplus for some minor failures.Therefore,this paper studies the fault prediction methods of components and devices respectively,and the main work and innovation points are as follows:1.Research on component fault prediction method.Research on component fault prediction methods.Component fault prediction mainly includes two key technologies:fault feature extraction and fault prediction.Aiming at the problem that the early fault signals of analog circuits are weak,and traditional feature extraction methods rely heavily on manual labor,it is difficult to extract deep and essential features.This paper uses PSO-DBN network for feature extraction.Dealing with the problems that current prediction methods cannot fully mine the time sequence of features as well as effectively combine historical data,this paper introduces the relative entropy-LSTM network for fault prediction.First,we use PSpice simulation software to build a SallenKey band-pass filter circuit,perform Monte Carlo analysis,and use the collected circuit output voltage frequency response as the input of the PSO-DBN network for feature extraction.Second,the extracted fault features are calculated by relative entropy to obtain the health degree that characterizes the health state of the circuit,and finally the health degree is used as the input of the LSTM network for fault prediction.Through KPCA to visualize the extracted features,it can be seen that the separation of features extracted by PSO-DBN network is higher than that of traditional wavelet packet feature extraction.2.Research on equipment fault prediction method.Aiming at the problems of complex equipment-level fault factors,low reliability of single variable characterization of equipment failure,and high complexity of multi-variable characterization of equipment failure,this paper proposes a fault prediction method based on Spearman correlation coefficient and BP neural network.First,we establish a prediction model combining Spearman and BP neural network,calculate the monotonicity of 21 features in the C-MAPSS data set through Spearman correlation coefficient,and then select the monotonically increasing feature as the fault feature according to the irreversible characteristics of equipment degradation,effectively reducing the redundancy between features as well as the size of the data set.We divide the retained features into training set and test set and input to BP neural network for fault prediction.Experimental results show that the prediction accuracy of the combination of Spearman and BP neural network is nearly 3 times higher than that of BP neural network,which verifies the effectiveness of the method.3.Design and implement the fault prediction system.3.Design and implementation of the system.In order to verify the theoretical model and simulation results presented in this paper,a corresponding fault prediction system is designed and developed.It includes two functional modules: condition monitoring and fault prediction.The system realizes the visual visualization of circuit diagram,monitoring result and prediction result,and demonstrates the reliability of the system by taking component R4 fault as an example.