Digital Theses Archive


Tesi etd-11252021-211007

Type of thesis
Vision-guided human assistance systems for industrial environment
Scientific disciplinary sector
Istituto di Biorobotica - BIOROBOTICS
relatore Prof. STEFANINI, CESARE
  • assistance
  • Imaging
  • industry
Exam session start date
Welding is one the most efficient and widely used joining technique in the industrial environment. Nevertheless, the formation of defects is inevitable due to a number of causes, e.g., wrong welding techniques, adverse environmental conditions (i.e., temperature, humidity), human factors. Since welding processes are highly dependent on manual interventions for both the actual procedure and the following quality monitoring and inspection, the operator is inevitably subjected to a protracted exposure to radiations, smoke, heat, and noise. Minimizing hazard occurrences during industrial processes to improve safety in the welding scenario is one of the most important aim of this thesis. This goal also influences the optimization of product quality, and increases task flexibility, adaptability, and rate of production. This thesis describes the development of robotic prototypes adaptable and flexible to harsh environment through the application of machine vision information placed on-board of customized platforms for both manual and automated welding processes. As for the manual process, we designed a wearable real-time tracking system employing a vision sensing unit and an inertial unit placed on the welder mask. It was able to detect, track and measure the speed of the welding pool despite the noises introduced by smoke, sparks and high-density illumination, by fusing the information provided the two typologies of sensors. In this way, we provided a real-time and online standalone monitoring and quality inspection system for manual procedure completely, adaptable and flexible to different typologies of welding techniques and welder positions, reducing the exposure of the welder to the hazards. Concerning the automated welding process, we designed a robotic platform able to autonomously localize, implement welding process, and inspect the quality of the product, increasing welder safety during the inspection phase. The tasks of localization and inspection were performed by applying vision sensors to provide adaptability and intelligence to custom products characterized by high dimensional variability, high reflectivity on surfaces, and very narrow spaces to weld and inspect. This system is, thus, a robotic platform able to manage customized products, variable dimensions, and shapes, improving the performance of current robotic platforms, which have been designed for repetitive and well-known products.