Inspecting and maintaining industrial plants is an important and emerging field in robotics. A particular case is represented by the inspection of oil and gas refinery facilities consisting of different long pipe racks to be inspected repeatedly. This task is costly in terms of human safety and operation costs due to the high altitude location in which the pipes are placed. In this domain, we propose a visual inspection system for unmanned aerial vehicles (UAVs), allowing the autonomous tracking and navigation of the center line of the industrial pipe. The proposed approach exploits a depth sensor to generate the control data for the aerial platform and, at the same time, highlight possible pipe defects. A set of simulated and real experiments in a GPS-denied environment have been carried out to validate the visual inspection system.

High-precision surface defect detection in manufactured products is still a key and open problem for industry; many manufacturers still require the use of human inspectors for ensuring defect-free production. 2D machine vision is the option commonly considered for defects detection automation but cannot reach the high sensitivity required when defects are small distance variations on the surface without a clear light contrast; for these cases, 3D inspection becomes the only option. One of the most accurate 3D inspection techniques is the use of profilometric sensors. Using these techniques, it is possible to register a point cloud with the 3D geometry of a part, where surface defects can be detected.To develop and compare different inspection algorithms, it is crucial to have a wide database of all possible defects. This is more and more important for the use of Artificial Intelligence algorithms. In most real cases, scanned data for defects is not easily available, especially in the development phase. Actual acquisition systems are expensive and must be installed on production lines to scan massive amounts of data to obtain a good number of defect samples, that must be manually classified. An additional complication comes from the fact that, in many industrial processes, defects occur very occasionally.To solve this problem of paucity of data, in this paper we propose a method to simulate realistic surface defects in 3D models of any product using a free-form deformation (FFD) approach. Also, we propose a mathematical model for 3 of the most common defects in sheet metal formed products: bumps, peaks, and cracks. The depth, size and orientation of all defects can be parameterized by the user.Also, we propose a realistic simulation of data capture: we developed an inspection system simulator including a profilometric triangulation sensor and its relative movement of the sensor over any CAD model representing a piece, generating a realistic 3D point cloud very close to the real object scan. We introduce the use of Perlin noise to simulate sensor measurement noise and material roughness. We validate our proposal by comparing the computer-generated data with real measurements.

Localization of mobile robots in industrial environments is key in an increasingly automated industry. Nowadays, the inspection and repair of heavy steel plates is performed by human workers. Repair work often requires long hours in uncomfortable postures that can cause problems for the worker. We propose a mobile robot placed on top of a steel plate that must move along the plate to inspect and repair it, without leaving the sheet. Robot localization on the plate is key to generate the inspection and repair trajectories.There are different methods of localization, the most widely used require the use of expensive laser sensors to create a map using information from the environment and localize from it. This paper proposes a less expensive localization system for a mobile robot based on the installation of ArUco markers in the environment and the use of a ring of 8 calibrated cameras mounted on the robot that allow a 360° vision. This ensures a correct localization regardless of the working area. It is necessary to map the markers with respect to a common coordinate system.We propose a method to create the map using the ring. We validate the proposal through experiments comparing the localization obtained with the proposed system and a localization using a state-of-the-art SLAM method employing laser sensors.

Visual inspection of manufactured products is a field in constant expansion. In this work we present a method to create a high resolution panorama (1mm/pixel) of a large rectangular plate using a mobile robot with two RGB-D cameras. The panorama is intended to analyze the surface in search of possible defects and identify areas of interest that have been encircled using a high contrast mark. Identifying which points belong to the surface plane and estimating the amount of distortion caused by the perspective correction we are able to form a panorama of a 4400 by 2500 mm plate with errors lower than 2% and a resolution of 1 mm/pixel.

espanolSe pretende desarrollar un robot movil capaz de inspeccionar y reparar chapas de acero en un entorno industrial de grandes dimensiones. Uno de los principales problemas a resolver es la localizacion del mismo mientras navega sobre la chapa. El presente documento propone un sistema de localizacion de un robot movil basado en la instalacion de marcadores ArUco y el uso de un anillo de 8 camaras calibradas dispuesto sobre el robot que permiten una vision de 360o. Es preciso conocer la posicion de los marcadores respecto a un sistema de coordenadas comun. Por ello, se propone una forma de localizar los marcadores y crear un mapa de los mismos con una sola camara, de tal manera que, posteriormente, pueda ser utilizado para desplazarse sobre la chapa con la mayor precision posible. La estrategia escogida se desarrollo en forma de paquetes de ROS capaces de proporcionar el estado del robot a otros algoritmos encargados de tareas como la navegacion. EnglishThe goal of this project is to develop a mobile robot capable of inspecting and repairing steel sheets in an industrial environment. This document proposes a localization system of a mobile robot base on ArUco markers and a ring formed by 8 cameras. The ring is mounted on top of the robot so that provides a 360o vision. Previously to the localization process, it is necessary to map the pose of ArUco markers. In this way, a mapping system has been developed that estimates the poses of the markers with respect to a common reference system, attached to the first marker found. We have developed ROS Kinetic packages, to solve the marker mapping problem and are able to provide the localization of the robot to other modules, such as navigation.

espanolEl presente documento describe un algoritmo de generacion de trayectorias para un robot movil encargado de sanear chapas de acero. Las trayectorias generadas deben asegurar la cobertura total de las superficies poligonales que delimitan los defectos con la herramienta acoplada al robot. Ademas, el robot debera resolver la tarea sin abandonar el interior de la chapa. Para resolver el problema, la chapa se divide en distintas zonas de trabajo derivadas de las orientaciones seguras que permiten al robot reparar sin abandonar la superficie de la chapa. Los defectos se recibiran en forma de poligonos que seran, en primer lugar, divididos de acuerdo con las zonas de trabajo y, a continuacion, descompuestos en formas mas simples teniendo en cuenta la orientacion de trabajo de cada zona. El recorrido completo del defecto podra realizarse calculando trayectorias paralelas para cada uno de los poligonos simples que lo componen. La trayectoria asi calculada corresponde a la seguida por la herramienta, por lo que para calcular la trayectoria del robot bastara con aplicar una traslacion. EnglishThis paper describes a path planning algorithm to allow a mobile robot to repair surface defects of a steel sheet. The resulting trajectories ensure that the polygonal surface that encloses the defect is completely covered by the repairing tool carried by the robot. In addition, the robot should work within the surface of the metal sheet. Defects will be described as polygonal shapes. First, they will be divided according to diffierent working areas. Then, each resulting shape will be divided again into simpler triangular and trapezoidal shapes. To guarantee that the defect is completely covered, the tool trajectory is defined as a set of rectinilear paths parallel to the axis of the sheet is generated. This trajectory is then shifted to obtain the robot trajectory.