Fuzzy Inference System And Neuro - Fuzzy System for Analog Circuits Fault Diagnosis

Abstract

A new fault diagnosis procedure for analog circuits is presented. The imprecision I and uncertainty associated with faulty circuits are handled utilizing fuzzy logic system. On the other hand, the remarkable abilities of the neural networks (NNs) to model nonlinear systems and in isolating analog circuits' faults are also utilized. Combining neural networks and. fuzzy logic systems, using neuro-fuzzy systems, automates the process of tuning the rules of a Fuzzy Inference System (FIS) for better diagnosis performance.

This thesis introduces the use of the FIS in modeling faults of the analog circuits by proposing a new FIS. It is constructed based on a fuzzy clustering algorithm. In addition, hybrid neuro-fuzzy systems, both Neural-Fuzzy Systems (NFS) and Fuzzy• Neural Networks (FNN), are exploited in modeling faults in analog circuits. tt'his work marks the first application of such systems in analog fault diagnosis. Three different well-known neuro-fuzzy models are investigated: the Adaptive Neural Fuzzy Inference System (ANFIS), A NEuro-Fuzzy approach; for the CLASSification of data (NEFCLASS) and Fuzzy Neural Network (FuNN). The first two models are considered as NFS while the last one is a FNN. Some modifications in the use of the considered models are introduced to suit the nature of the diagnosis problem. I Comparisons between different models are also conducted. i

Three benchmark circuit examples are considered to demonstrate the potential of the proposed algorithm. The salient measurements that characterize the Circuit Under Test (CUT) behavior are selected by feature extraction and dimensionality �eduction

techniques like, wavelet transform (WT), Principle Component Analysis (PCA} which in

turn reduce the complexity of the model and the training time.