Research On Key Technologies Of Analog Circuit Prognostics And Health Management

Complex electrical and electronic systems play an increasingly important role in aviation,aerospace,navigation,industrial manufacturing,nuclear,new energy,high-speed rail,etc.Analog circuit is an important part of complex electrical and electronic system,and its prognostics and health management(PHM)technology has gradually become a research hotspot.The PHM of analog circuit is important and significant for the complex electrical and electronic system's running state evaluation,anomaly detection,rapid diagnosis,accurate maintenance,and improving system operation reliability and security.In this paper,deep belief network(DBN)technology of deep learning is used to extract analog circuit incipient fault features,where the restricted Boltzmann machine(RBM)'s learning rates of DBN are generated by using a quantum-behaved particle swarm optimization(QPSO)algorithm.An incipient fault diagnosis model based on support vector machine(SVM)is established to identify incipient faults.After the incipient fault components are identified,the output voltages of circuit are extracted as features,and then the Pearson product-moment correlation coefficient(PPMCC)is calculated as the health degree in the analog circuit fault prognostic work.Multiple kernel relevance vector machine(MKRVM)method is applied to predict the remaining useful performance(RUP).The main work of this paper is as follows:(1)In view of the current weakness of fault diagnosis features extraction methods,DBN technology of deep learning is researched to extract analog circuit incipient fault features.The RBM's learning rates of DBN are the key parameters and affecting the extraction results deeply,and they are generated by using QPSO algorithm.The analog circuit incipient fault features are extracted as sample data by using the proposed DBN method,and the incipient faults are identified by using the SVM basd incipient fault diagnosis model.Meanwhile,SVM's regularization parameter and width factor are optimized by using QPSO algorithm,which can effectively improve the accuracy of incipient fault diagnosis.(2)For the analog circuit fault prognostic researches are fewer currently,an analog circuit fault prognostic framework is presented in the paper.Firstly,output voltages are extracted as analog circuit fault prognostic features,which is simple and effective.The concept of health degree of the analog circuit component is presented,and the component's healthy degrees are produced by calculating PPMCC between the analog circuit's output voltages of component's degradation condition and no fault condition.Through measuring the output voltages and calculating the component's health degree for a long time,the sample data are generated.The sample data are learnt by using MKRVM algorithm,and a fault prognostic model is set up to predict failure time of the component and calculate its RUP,where the kernels' sparse weights of MKRVM are optimized by the QPSO algorithm.(3)Sallen-Key band-pass filter circuit,four-op-amp biquad high-pass filter circuit and nonlinear rectifier circuit are used as experiment circuits,and the components in the circuits are selected to perform analog circuit incipient fault diagnosis and fault prognostic.In the incipient fault diagnosis experiment,the features are extracted by the DBN based feature extraction method proposed in the paper.Features of the same incipient fault class are high degree of aggregation,and features of different incipient fault classes are low degree of overlapping.By using the SVM based incipient fault diagnosis model optimized by the QPSO algorithm,the incipient diagnosis accuracies of the three circuits are respectively 100%,96.41% and 100%.In the fault prognostic experiment,the MKRVM method proposed in the paper is used to perform fault prognostic on the eight component identified in the incipient fault diagnosis.The prediction results are all within the set confidence interval,which reflects that the MKRVM has high precision on fault prognostic.Finally,the MKRVM method is better than the traditional RVM based on single kernel learning method on the problem of fault prognostic through comparative experiment.