Classificação de defeitos na pintura de peças metálicas no processo industrial utilizando algoritmos de inteligência artificial.

Abstract

Surface coatings have been used for thousands of years, with their consumption gradually increasing. Painting aims to deposit a film of paint on a metallic or non-metallic surface, serving a wide range of purposes, including corrosion protection, aesthetics, safety, and more. Industrial painting processes involve several steps, including surface preparation, paint application, and drying. In recent years, there has been an exponential increase in the use of artificial intelligence algorithms in industrial processes. This work suggests, analyzes, and classifies the application of these algorithms to classify defects in the painting of metal parts in industries using artificial intelligence algorithms. In this work, an artificial intelligence model based on convolutional neural networks (CNN) was developed to automatically identify parts that meet and do not meet the standards stipulated by the industry where the research was conducted. The model was trained using an image database provided by the industry itself, which contains a collection of defective parts. For parts that meet the standards, a database of images captured by the author himself was created. These images underwent preprocessing to highlight the main defects in the parts and reduce the computational load by applying a grayscale filter. After training for 50 epochs and using metrics such as accuracy, precision, recall, and F1-score, the model achieved 88.86% accuracy, 88.18% precision, 88.86% recall, and an F1-score of 91.39%, demonstrating excellent performance in identifying defective parts. After validating a predictive model, a mobile application called AÇO_CHECK was designed to collect field data, enabling empirical validation of the model. This application uses a machine learning model developed with the Keras library (integrated with TensorFlow), which analyzes user-submitted images and draws inferences from the visual data.