Fault Detection Using ReliefF Algorithm with Canonical Correlation Analysis Based on Improved Support Vector Machine

Recently, the increasing time-complexity of electronic circuit has made fault detection and diagnosis a challenging task. The existing Deep Belief Networks (DBN) has been used for fault detection and circuit diagnosis however they possess a significant challenge due to their complex behavior and sensitivity to component variations. Hence, this research proposes ReliefF algorithm to optimize Canonical Correlation Analysis combined with Improved Support Vector Machine (RCCA-ISVM) for accurate nonlinear circuit diagnosis and fault detection by learning patterns as well as anomalies in circuit behavior. Initially, the input data is collected from the Sensor Fault Detection dataset. Further, fault characteristics are extracted using the Wavelet Packet Transform (WPT) from the time-domain, statistical and frequency-domain features. Canonical Correlation Analysis (CCA) algorithm is applied to enhance the correlation between the features based on their respective weights, improving the relationship between the selected features as informative for fault diagnosis. Then the fused features undergo dimensionality reduction using Principal Component Analysis (PCA). Finally, an ISVM classifier is employed to perform the fault diagnosis based on the reduced set of features. From the results, the proposed achieved outstanding results in terms of accuracy (99.30%) when compared with the existing RCCA-SVM.