Inspeção de falhas em conectores PTH de placas de circuito impresso utilizando Machine Learning

Abstract

This dissertation addresses the development of a machine learning model to predict failure modes in PTH connector terminals and screws on printed circuit boards. Using image processing and algorithms such as KNN, Decision Tree, and SVM, the model segments and classifies defects from a custom database of 3,800 images representative of real production conditions. Integration of this system into a production line demonstrated that automated inspection outperforms manual inspection, with accuracy of up to 97%, in addition to reducing inspection time and costs related to rework and downtime, improving the final quality of the products.