FAULT IDENTIFICATION USING OF NN BASED GA PARAMETERS IN ANALOG CIRCUITS BY SIMULATION IMPLEMENTATION

Abstract

The research titled as “Fault identification using of NN based GA parameters in analog circuits by simulation implementation” proposes a fault identification approach for analog circuit using hybrid evolutionary techniques and neural network in GA parameters. With the increasing demand for analog integrated circuit technology and the complexities and shrinking equipment dimensions, integrated circuit performance is becoming sensitive to inherent parameter deviations. Therefore fault detection and location in analog circuits have received more attention in recent decades. Fault detection in analog circuits is challenging due to poor circuit model, limited test nodes, component tolerance, non-linearity of output, limited access to internal nodes. The genetic algorithm is used as an evolutionary technique for the optimization and learning of neural networks. The proposed method has been validated through a state-variable filter circuit and all possible parametric variations have been derived for faulty and non-faulty conditions. A large area of research has been done on fault detection in analog circuits, but neural-network-based detection methods have proven to be more efficient as it has good robustness, adaptability, and learning capability. But neural networks have poor generalization capability and require a large number of iterations. So in this work, the boundaries of neural networks are overcome by designing a hybrid scheme of neural networks and evolutionary algorithms. All experiments are conducted on MATLAB R2015a. Experimental results are presented to show that the hybrid scheme is more efficient in terms of fault detection rate and time constraints than the neural network method.