Analog Circuit Fault Diagnosis Based On EMD Algoirthm

Since the1960s the development of analog c ircuit fault diagnosis has beenconstrained by comple x fa ilure scenarios and the growing size of scale integratedcircuit. With the growing numbers of faults and more and more complex o f integratedcircuit the diffic ulty o f ana log c ircuit fault d iagnosis has become harder and harder.The failure number of a nalogue inte grated circuit ho lds more than e ighty percent ofthe total number of electrica l fault. So conducting research on ana log circuit faultdiagnosis has theoretical and practical significance.Firstly this paper reviews the current research progress according to the anvancedworks of worldwide and ho mela nd and presents the fu ndame ntal theory anddiagnostic methods of analo g circuit. At the basis of the existing research find ings thispapers puts forward a new method by using EMD algorithm to extract fault features.This method solves the diffic ult extraction of fault features. At the aim of solving thedifficulty of fault d iagnosis and recognition this paper puts forward two new differentways of fault dia gnos is.One of the two new methods is connecting EMD a lgorithmand BP neura l network whic h is optimized by genetic algorithms. The other method isconnecting EMD algorithm and BP neural network whic h is optimized by partic leswarm.The method of fault feature extraction on the basis of EMD algorithm dosealternating simulation and monte carlo analysis and collects response message fromaccessib le node. The collected response message is normalized and divided intomultiple unit IMF and Hilbert margina l sepctrum by us ing EMD algorithm. Fina llythis paper calc ulates energy value of multiple IMF units and Hilbert marginalspectrum and extracts the energy value by using the fault feature vector.The method of analog circ uit fault dia gnos is basing on the EMD algorithm firstlyneeds to extract the characteristic value of the simulating circ uit and using multip leIMF units and Hilbert marginal spectrum as fault characteristic value. Then this paperputs the extracting characteristic value into BP neura l network whic h is optimized bygenetic algorithms and BP neural network which is optimized by partic le swarm. BPneural network which is optimized by genetic algorithms is mainly us ing geneticalgorithms to optimize the initial we ights and basis of the BP neura l network. Theoptimized BP neural network can predict the output of functio n better. The optimized BP neural network by partic le swarm uses the strong ability of particle swarm’s globalsearching and quick local searching ability.This paper uses the above method of ana log c ircuit fault dia gnos is to dosimulation experiment on Sallen-Key filter circuit and Elliptic Filter circuit at thebasis of EMD algorithm and analyzes the added10db white noise and30db whitenoise by Sallen-Key filter circuit and Elliptic Filter circuit’s simulation of faultdiagnos is on analo g circuit. The simulation results of e xamp les give n in thisdissertation show that the fault dia gnos is methods proposed above have gooddiagnos is effect and feasib ility in analyzing the fault response of analog c ircuits andcan locate the faults in analog circuits correctly.