Soldering Defect Identification Using Deep Learning and Machine Learning Approach

Electronic device functionality and dependability can be severely compromised by soldering flaws in printed circuit boards (PCBs). Effective quality control in manufacturing processes depends on automated flaw identification. Using a bespoke dataset of 500 photos of defective PCBs, this research proposes a novel approach for soldering defect recognition and summary creation on printed circuit boards (PCBs) using a combination of Deep Learning (DL) and Machine Learning (ML) techniques. A framework for creating computer vision models is used to carefully classify these pictures according to six different defect categories: solder-bridge, solder-skip, overheated, offtrack-solder, overwetted, and underwetted. Total 300 images are classified into mentioned categories. To improve the model's performance and expand the dataset, the authors implemented augmentation methods, which led to the creation of a comprehensive 30% of total image dataset. The proposed system uses the Google Collab platform to train our unique dataset using the YOLOv5s model architecture, a deep learning model that helped us optimize the model to identify and categorize soldering faults properly. To ensure smooth implementation and reproducibility, we carefully recorded and downloaded the necessary dependencies for YOLOv5s. Preprocessing techniques for image enhancement were carried out on SVM algorithm to deploy the machine learning approach. To enable effective quality control measures in PCB manufacturing processes, our suggested methodology shows promising results in precisely identifying soldering faults.