Research On Fault Diagnosis Of Analog Circuit Based On Deep Extreme Learning Machine

With the rapid development of electronic industry,circuit fault diagnosis plays a very important role in reliable operation and good industrial system maintenance,which can ensure the production of higher quality products,reduce product scrap rate and meet increasingly stringent safety and environmental requirements.At present,although fully automatic fault diagnosis technology is widely used for digital circuits,but for analog circuits,due to the complex fault model,component tolerance and non-linear problems,it is difficult to detect and diagnose faults,which causes the high cost of circuit chips and the technical bottleneck of the development of integrated circuits.Therefore,in the analog circuit,the effective diagnosis technology is still an open research topic.In recent years,with the emergence of deep learning and the development of intelligent algorithm,it provides a new research way for analog circuit fault diagnosis.Based on the theory of deep learning,GA and PSO algorithm,this paper proposes a fault diagnosis model of analog circuit based on deep limit learning machine,and studies the fault diagnosis of analog circuit by analyzing the output signal of the circuit.The main work and achievements are as follows:1.Aiming at the difficulty of feature extraction and slow model training time in analog circuit fault diagnosis,this paper adopts an analog circuit fault diagnosis algorithm based on Deep Extreme Learning Machine(DELM).This algorithm introduces the idea of Auto Encoder(AE)into Extreme Learning Machine(ELM),builds an ELM-AE model with double random hidden layers,and then uses the original ELM as a classifier.ELM-AE is a neural network that reproduces the input as much as possible.It can be used as a feature extractor in a multi-layer learning framework.It has feature representation capabilities and can achieve data dimension compression and sparse expression.In order to further enhance the robustness of the DELM network,a regularized extreme learning machine(RELM)is used instead of the original ELM.ELM-AE is used as the basic component of training DELM,and ELM is used as a diagnostic tool to form the underlying fault features into more abstract advanced features,which can learn data features independently,avoid the problem of feature extraction and selection for data alone,and improve the diagnosis efficiency.Training DELM first extracts the original fault data through simulation,and then directly uses it as the input of the DELM.Without separate feature extraction and selection,you can quickly locate the fault and realize fault diagnosis.Finally,Sallen-Key,four op amp high-pass filtering and two-stage four op amp bi-second-order low-pass filter circuits are tested.The analysis results show the feasibility of this method and can realize fault identification quickly and accurately.2.Based on GA and PSO algorithm to optimize the fault diagnosis of analog circuit of DELM.It is difficult to select the number of hidden layer nodes in deep network and artificially selected hidden layer nodes will miss the effective feature information of the data itself,which will cause a large training error problem,selecting the appropriate hidden node can effectively reduce the time complexity and make the network very good generalization ability,so genetic algorithm(GA)is used to find the optimal number of nodes in each hidden layer of DELM network,and it is further compared with the deep extreme learning machine optimized by particle swarm optimization(PSO)and proved that the GA-DELM algorithm can adaptively search the global best,and can avoid falling into the local optimal.Finally,it is verified by a nonlinear rectifier circuit,and the diagnosis result shows the feasibility of the algorithm.