TRIBRID-Optimized MobileNetV3 Framework for Explainable Printed Circuit Board Defect Classification

<title>Abstract</title> <p>Printed Circuit Board (PCB) defect inspection is crucial for successful electronic manufacturing, but creating a highly accurate, computationally efficient solution with model interpretability is challenging. This paper introduces an explainable framework for PCB defect classification based on TRIBRID-optimized MobileNetV3. The TRIBRID optimization algorithm automatically tunes the training hyperparameters, and transfer learning is also used to fine-tune the MobileNetV3 model. To offer visual explanations of classification actions, gradient-weighted Class Activation Mapping is also incorporated. Experiments conducted on a publicly available PCB defect set show that the proposed framework outperforms the baseline models, achieving 100.00% accuracy, precision, recall, and F1 score while retaining the lightweight nature of MobileNetV3, which is suitable for intelligent Automated Optical Inspection systems.</p>