Sistema de inspeção visual em uma planta didática utilizando aprendizado profundo

Abstract

Inspection during an industrial process is a important factor, as it aims to ensure efficiency and avoid waste. To achieve these objectives, process inspections are conducted after each stage of the industrial process and also at its conclusion. This work proposes a fast and practical system where a previously trained AI model can be saved and, from there, inspections in an industrial process can begin. The system offers result visualization options and allows for the correction of the AI’s hits and misses, generating metrics that enable monitoring the model’s performance. The system was developed using the Streamlit framework, with YOLOV8 for AI training and a MySQL database for information storage. For the development of the system, a didactic plant that simulates an industrial piece capping process was used.