Mapeamento semântico de subestações de energia elétrica

Abstract

Considering the increasing use of robotic systems in substation environments, this work proposes a semantic mapping algorithm capable of estimating the positions and dimensions of equipment in such settings. The study was conducted using a simulated substation model developed in the CoppeliaSim software, from Coppelia Robotics. This virtual environment enabled the implementation of a wheeled robot for movement and data acquisition within the simulated substation. Through the ROS platform, control commands were sent to the robot, and data referring to RGB images and the point cloud were exported for further processing in a Python environment. A YOLO-based model was trained to recognize six different types of objects present in the virtual substation. During the object detection and localization stage, the model achieved accuracy rates above 90% for the mAP metric, considering the intersection over union (IoU) threshold of 𝛼 = 0.5. Subsequently, by correlating regions of the point cloud with the corresponding detection areas in the RGB images, it was possible to estimate the positions of the equipment relative to the robot. Finally, the DBSCAN clustering algorithm, with optimized parameters, was applied to group the detected position samples in order to estimate both the location and dimension of the equipment in the substation. The results enabled an evaluation of the strengths and limitations of the proposed approach, demonstrating that it is a promising strategy for semantic mapping in electrical substations.