VISION-BASED EMI/EMC FAULT DETECTION AND MITIGATION USING RANDOM FOREST CLASSIFIERS

Abstract

High-frequency electronic circuits, switched-mode power supplies (SMPS), and Class-D amplifiers frequently encounter electromagnetic interference (EMI) and electromagnetic compatibility (EMC) compliance failures. This paper presents an automated computer vision and machine learning-powered framework designed to accept raw circuit schematic images as direct input to automatically detect EMI/EMC problems and generate immediate engineering mitigation guidance. The end-to-end pipeline integrates OpenCV and Tesseract OCR to extract passive filter component parameters (inductance L, capacitance C, damping resistance R_damp) and switching frequencies (f_sw) directly from schematic diagrams. A numerical simulation module constructs transfer function models to synthesize 300-point magnitude frequency profiles across high-frequency spectra. An ensemble Random Forest classifier evaluates these feature vectors to identify specific failure modes, notably switching frequency noise leakage caused by filter notch misalignment. Upon fault detection, an integrated expert mitigation engine calculates exact component adjustments and structural filtering rules to rectify the failure. Tested across Class-D audio amplifiers, SMPS flyback filters, and DC-DC converters, the proposed vision system provides robust automated fault detection and prevention without manual netlist creation.