Controle servo visual para replanejamento da tarefa de um robô manipulador para inspeção de peças automotivas

• Ano de defesa: 2023
• Autor(a) principal: Lucca Garcia Leão
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; Brasil; ENG - DEPARTAMENTO DE ENGENHARIA ELÉTRICA; Programa de Pós-Graduação em Engenharia Elétrica; UFMG
• Palavras-chave em Português: Robótica; Visão de máquina; Controle de qualidade; Replanejamento de tarefa; Controle servo visual; Engenharia elétrica; Robôs; Automóveis - Peças; Automação; Robôs - Sistemas de controle; Visão por computador; Controle automático; Processos de fabricação
• Link de acesso: http://hdl.handle.net/1843/64392

The fourth Industrial Revolution, also called Industry 4.0, brought many changes to manufacturing systems and quality control processes through new technologies such as robotics and machine vision. Machine vision is one of the most commonly employed techniques for quality control of industrial processes, as it allows the capture, processing and inspection of products with the intent of identifying failures in the production process, such as absence of certain components, displacement, smudges and cracks. However, the inspection of objects with increasingly complex geometry demands machine vision systems, which are commonly composed of one single static camera, to be able to capture images in various different perspectives. Thus, the combination of robotic manipulators and machine vision systems can be a solution to move the camera around to desired positions, increasing the flexibility and effectiveness of said systems. This dissertation is based on the case study of the Invent Vision automated inspection cell. This cell performs automatic inspection using computer vision algorithms for quality control of the produced parts. The cell is equipped with a robotic manipulator with a camera mounted in its end-effector to perform visual inspection of complex vehicle parts. Upon operating the machine, the object is placed in its interior and the robot performs an inspection, defined by a sequence of desired poses where images are captured. However, the robot expects the inspected part to always be in the same position in order to obtain the correct image perspectives and perform a succesfull inspection, and therefore, problems could occur due to mispositioning of the object inside the cell. In this sense, this dissertation proposes a task replanning strategy for the robot movement during an inspection. The replanning is calculated with relation to the pose of the inspected component, which is estimated by identifying a fiducial marker. The proposed system is based on Visual Servo Control techniques and aims to improve the original workflow of the inspection cell by avoiding the imposition of part position, while performing the task replanning based on its estimated pose. The system was initially validated in a simulation environment. Then, an experimental setup was constructed with a real Kuka KR4 R600 and an industrial Invent Vision camera identical to the ones in real production environments. The inspections were performed using a real part, and a template matching algorithm to detect the presence of certain components. Experimental results show that the proposed system was capable of performing the inspection of a part with success, even when subject to great pose variation in the workspace. The experiments also show that the choice of movement result in different trajectories between the inspection points, but without altering the inspection result. The proposed methodology is versatile and can be also used in other applications.