Implementación de sistema de detección de objetos mediante ESP32-CAM para el control de calidad de la estación FMS-210

Abstract

The interest in artificial intelligence and machine learning has grown significantly in recent years, with object detection becoming a critical application of these technologies. In industrial settings, automated product development is increasingly adopted by micro and small enterprises to enhance production efficiency. However, quality control, a vital aspect of production lines, is often overlooked due to the high cost and complexity of implementing sophisticated technologies. This oversight impacts overall production efficiency, as effective quality control enables necessary corrections for cost-effective manufacturing. Modern tools such as Google Cloud Vision AI, Labelbox, Simplismart, TensorFlow, and Edge Impulse have simplified the development of machine learning models, making it possible to address these challenges. These advancements now allow micro and small enterprises to implement accessible quality control systems. This project focuses on designing a machine learning model for the ESP32 microcontroller to improve quality control in production lines. The proposed system evaluates detection speed and accuracy compared to the standard vision-based quality control system of the FMS-210 station. The solution is structured into four stages: capture, analysis, segmentation, and recognition. A custom model is developed for quality control, exported as an Arduino-compatible library. The Arduino platform facilitates real-time image capture and processing, utilizing the model to determine whether a piece is complete. Additional metrics, such as detection time, connection status, and approximate dimensions, are also recorded. Data is transmitted via Bluetooth from the ESP32 to a computer or mobile device and displayed on a user interface developed using MIT App Inventor.