Sensor de identificação de solda por fusão de dados para robô inspetor de vasos de pressão da indústria de petróleo e gás

Abstract

Large metallic structures, such as tanks in the oil and gas sector, require recurring inspections to ensure their integrity and prevent catastrophic accidents. Ultrasonic Non-Destructive Testing (NDT) of weld seams is a commonly used method for such inspections, as it allows for the practical acquisition of detailed information regarding the structure’s weld profile. Given the harsh environment and the size of the structures involved, robotic inspection significantly enhances safety and allows the process to be carried out while the structure remains in service while being inspected. An inspection robot must precisely follow the weld seam while applying the NDT instrument. To achieve this, it relies sensors such as cameras and 3D profile sensors to recognize the seam and assist the operator in conducting the inspection correctly. This thesis presents the construction of a climbing robot designed for ultrasonic weld inspection, and introduces an innovative composite sensing system that uses point cloud data from a 3D camera and a profile sensor to identify the weld seam. Inferences regarding the weld seam’s position, orientation, and width are derived from the 3D camera data using techniques such as RANSAC and weighted regression. These same characteristics are also inferred via the profile sensor, and both sets of inferences are used as inputs for a fuzzy fusion system that robustly identifies the weld seam; the result is displayed in a 3D grid representation of the space ahead of the robot, highlighting the seam for the operator. The evaluation of these systems is detailed through qualitative analysis across several scenarios, achieving an error of 2 mm for position and 2 degrees for orientation on a real oil tank plate.