Automatização e desenvolvimento de um Sistema de Inspeção de PCB utilizando Robótica Colaborativa e Processamento de Imagem

Abstract

The advancement of enabling technologies under the concept of Industry 4.0, such as collaborative robotics and artificial intelligence (AI), has contributed to achieving high levels of automation in the industrial sector. One of the major challenges of AI applied to industries is the lack of specific datasets for diverse products and models, which are essential for training intelligent systems. The electronics segment is among the most impacted by these transformations. This work proposes an innovative solution to automate the inspection of printed circuit boards (PCBs), including an intelligent registration system, using computer vision techniques based on deep learning, such as the pre-trained FastSAM model combined with Template Matching, to build an efficient automatic registration system. The automation proposed in this work was implemented through a collaborative robot capable of inspecting items in any position and orientation without posing risks to operators performing other tasks during the process, while also ensuring flexibility to adapt to changes in the factory layout. The proposed method eliminates the need for labeled datasets or defective samples, enabling more efficient and safer quality control. The system achieved remarkable results in evaluation metrics, such as a precision of 83.77%, recall of 90.43%, IoU of 76.95%, F1-Score of 86.97%, and pixel accuracy of 98.69%, demonstrating high efficiency in identifying and segmenting electronic components on PCBs. As a distinctive feature, the solution shows potential to significantly improve the efficiency of processes in Pólo Industrial de Manaus (PIM) and other industrial scenarios.