SPOT Faults: Parametric Fault Detection in Linear Analog Circuit Via Sooty Egret Optimization Based Iterative Decisioned CNN

Analog circuits present intrinsic variability in component parameters, inconsistent data, and high computational complexity, causing unpredicted faults and problems with classification diagnosis. Mainly, 5G and beyond communication systems suffer from high-frequency noise disturbance. To overcome these challenges, Sooty Egret Optimization based Iterative Decisioned Convolutional Neural Network (SEO-IDCNN) is proposed for the Linear analog circuit parametric fault diagnosis. The proposed SEO-IDCNN techniques improve feature localization and decrease redundancy by extracting the edges, transitions, and directional pattern features using a fast discrete curvelet transform based on wrapping. Moreover, the Iterative Decisioned Convolution Neural Network (IDCNN) classification model is developed for categorizing parametric faults in Linear analog circuits and optimizing the hyperparameters of the IDCNN using Sooty Egret Optimization (SEO). This proposed approach is verified with two filter circuits: the four-op-amp Recursive Biquad high pass filter (four-op-amp-RBqHF) and the Derivative Sallen key bypass filter (DSK-BpF). These circuits are appropriate benchmarks for assessing the efficacy of fault diagnostic systems due to their distinctive designs and sensitivity to changes in component parameters. According to experimental analysis, the average accuracy rates of the DSK-BpF and the four op-amp RBq-HF were 99.92% and 99.88%, respectively. Furthermore, by using minimum features, the suggested method shows a shorter computing time (3.15 s, 2.75 s) for two filters while still producing superior results when compared to other methods.