Desenvolvimento de rede neural convolucional para classificação de telhas cerâmicas defeituosas

Abstract

This study presents the development of a Convolutional Neural Network (CNN) aimed at classifying ceramic roof tiles into two categories: normal and defective. Given the growing demand for high-quality ceramic products, automating quality control processes has become essential for reducing waste and increasing production efficiency. The pro- posed methodology encompassed manual image acquisition, data preprocessing, CNN modeling, and performance evaluation. An initial dataset of 1,600 images of Roman- type ceramic tiles was collected and expanded through data augmentation techniques to approximately 7,000 images for training purposes. The results demonstrated high performance, achieving an accuracy of 99.7% and greater effectiveness in identifying normal tiles. The findings suggest that CNN-based approaches are promising tools for the automated classification of Roman ceramic roof tiles. Future work includes the development of an Internet of Things (IoT) application and further testing in real industrial environments.