Autonomous robotic systems for batch-size-one production: an oil and gas case study application

Robotics and Automation are playing a key-role in the new trends of Industry 4.0. According to Industry 4.0 paradigm, industrial robotic applications are shifting from repetitive and fixed tasks to variable and flexible ones, by exploiting Human-Machine Interaction or the increasing development of perceptual features, such as vision systems or tactile sensing. These capabilities allow the employment of robotics also for small batch production: the flexibility and the intelligence of these systems reduce programming effort for different components, making automation for such variable applications feasible from an economic standpoint. In this thesis, principles of Industry 4.0 are applied on a batch-size-one manufacturing process: robotic external welding for centrifugal compressor impellers is a new technology developed and qualified by Baker Hughes Innovation Welding Laboratory, and, since centrifugal compressor are highly customized products and oftentimes each impeller is different from any others, the programming and in-process control phases performed by the operator shall be as short as possible to make the process economically convenient in a production environment. Different self-programming approaches are analyzed for native or neutral 3D models file formats and impeller position strategies: to embrace the most generic and challenging case, a self-programming module for a marker-based vision system is proposed, based on the automatic geometry extraction from the component .stl file and marker recognition and three-dimensional localization through eye-in-hand triangulation with a 2D camera for randomly placed impeller. Then, the industrial application of centrifugal compressor impellers robotic external welding is described, with the industrial robotic cell installed in Baker Hughes Florence manufacturing plant and its perceptual capabilities, such as the use of a laser scanner for path correction, the on-line gas cap movement control and the data acquisition for trouble shooting and process optimization. Finally, on-line methods for autonomous welding imperfection detection for the above mentioned process have been evaluated, developing and testing a laser scanner and relevant profile classification algorithms, and an industrial camera vision system exploiting Artificial Intelligence based Computer Vision.