Analog Circuit Fault Prognostic Approach Based On Machine Learning

With the continuous improvement of the safety and reliability requirements of electronic systems,analog circuits,as an important part of electronic systems,fault prediction methods have become the focus of current research.Therefore,aiming at the key problem of fault prognostic of analog circuits,it is proposed that the incipient fault diagnosis of analog circuits should be carried out to diagnose the components with fault tendency,and then the failure time of analog circuit can be predicted by analyzing the Remaining Useful Performance(RUP)of the component.Firstly,Deep Belief Network(DBN)method is used to extract the incipient fault features in fault data,in which the key parameters in DBN were optimized by Quantum-behaved Artificial Bee Colony(QABC)algorithm.Based on the fault features data,the fault diagnosis model is established by using multi-kernel and multi-classification Relevance Vector Machine(RVM)to accurately identify the incipient faults of analog circuits.Secondly,for the incipient fault components diagnosed,the output response of the analog circuit is extracted as the fault features,and the Pearson Product-moment Correlation Coefficient(PPMCC)is used to preprocess the fault features.Combining the discrete gray model with the multi-kernel RVM,using the fault features data to establish the fault prognostic model.The RUP of the analog circuit is predicted by the fault prognostic model.Finally,Sallen-key band-pass filter circuit and Biquad low-pass filter circuit are used as verification circuits.The above methods are verified and analyzed by simulation experiments and comparative experiments.In the incipient fault diagnosis experiment of analog circuits,the incipient fault feature extraction using DBN has achieved good effect.The multi-kernel multiclassification RVM incipient fault diagnosis model has a certain improvement in the recognition accuracy of the fault type compared with other methods.In the experiment of fault prognostic,the discrete gray multi-kernel RVM fault prognostic method is used to predict the fault of analog circuit,and the prediction results is more accurate than the traditional methods and the related literature methods.