Detecting and Classifying Parametric Faults in Analog Circuits Using an Optimized Attention Neural Networks

In analog circuits, the essential variability in component parameters and the various distribution of fault component parameters present challenges like unpredicted faults, inconsistent data and high computational complexity for effective classification diagnosis. To address these issues, this research introduces an Optimized Attention Neural Networks, integrating three machine learning classifiers: Pyramidal convolution split Attention Neural Networks, Graph Visual Attention Neural Networks and Capsule Shuffle Attention Neural Networks. Initially, the approach incorporates fuzzy rough mutual information, spatial distribution principal component analysis and enhanced minimum redundancy maximum relevance techniques to select crucial features for distinguishing and categorizing parametric faults in analog circuits. Subsequently, the proposed classifiers leverage with the enhanced termite alate optimization algorithm for the recognition and categorization of parametric burdens. The experiments are conducted using MATLAB, demonstrate notable outcomes across three filters. The achieved results indicate an average accuracy of 99.92% for the sallen-key band-pass filter, 99.88% for the four op-amp biquad high-pass filter and 99.86% for t*he leap frog filter. Furthermore, these filters exhibit higher precision and recall values when compared to existing approaches. Additionally, the proposed approach demonstrates reduced computational time (4.18 s, 3.89 s and 5.59 s) of three filters by employing a minimal number of features, while still yielding excellent results compared to other existing approaches. These findings underscore the superior performance of the introduced approach in the domain of analog circuits, positioning it as a promising solution for the complex task of detecting and classifying specified faults in analog circuits.