Detection and Classification of Defects on Printed Circuit Board Assembly through Deep Learning

Contemporary production of electronic devices and modules requires the final product to be of high quality and at the lowest possible price. To ensure this, the production process includes various testing phases, thus resulting defects can be detected in time. The paper analyzes and evaluates a process for classifying defected and non-defected Printed Circuit Board Assemblies (PCBAs) by applying a deep learning algorithm. Our own datasets of PCBAs images with and without defects are created, experiments are performed utilizing TensorFlow, and classification models are evaluated. The results show that the discussed approach is characterized with high accuracy in detecting defective PCBAs at the conductance of two-class classification tasks.