A Novel Hybrid Model for Printed Circuit Board Defect Classification: CNN Meets Random Forest

The study's goal is to accurately categorize printed circuit board (PCB) defects by employing a hybrid strategy that combines Random Forest and Convolutional Neural Networks (CNNs). The model was rigorously tested both during training and evaluation. Four convolution layers, four max-pooling layers, & a fully linked layer made up its intricate architecture. When it comes to identifying positive cases for all defect types—such as breakouts, pinholes, missing holes, missing conductors, and incorrectly sized holes—the model is incredibly precise. This fact is emphasized by the precision measurements, which vary from 96.77% to 98.55%. The model, which used a dataset of 4818 photos, classified printed circuit board flaws with an exceptional 97.55% overall accuracy. The model is effective in capturing the majority of true positive cases, as confirmed by recall values, which vary from 97.14% to 98.15%. F1-Score values, on the other hand, provide a reasonable assessment of both accuracy and recall, ranging from 97.12% to 97.95%. The Macro, Weighted, and Micro Average metrics all support the model's ability to function in equilibrium across various defect types and class sizes, with a mean score of 97.56%. The model's 99% overall accuracy attests to its robustness and validates its reputation as an extremely dependable automated quality control tool in PCB manufacturing. The study concludes with a noteworthy observation: the model's practical utility is highlighted, and it is envisioned that its integration will serve as a foundation for enhancing the efficiency and reliability of electronic device manufacturing through advanced defect identification and classification processes. The model's usefulness can be enhanced by using it to improve quality assurance procedures in PCB manufacturing, as seen by the precision, recall, and F1-Score values.