Comparative analysis between machine learning reinforcement methods applied in aeronautics furniture production

Abstract

This study seeks new forms of automation with robots applied to aeronautics. The hypothesis to be tested suggests that the insertion of reinforcement learning methods, the robots will behave more versatile, efficient and reconfigurable. Using sensors, it is possible to reproduce industrial environments in a quite plausible way, providing the virtual characteristics similar to real-world system, and the interaction between robots and the environment may be modeled, simulated controlled and the resulting behaviors may be screened prior to performed on the actual situation by the robot. A mechanical impact not planned in the manufacturing environment is an undesirable occurrence, it may damage the robot, other equipment or even injure an operator. The system to be developed in this project, works with applications that involve inherently contact between machines, in particular tasks which robots come in contact with the users and the work environment. We will use this project as a case study, the assembly of aeronautical furniture by a manipulator robot Kuka KR16. Alternatives will be analyzed to explore the ability of conventional robotic systems and favor mainly the means of production, seeking to increase productivity and competitiveness. The proposal is relevant in the current economic scenario, since the robotic need to monitor developments in the global industry, meeting the demands of challenging projects presented by companies based in the country.