Flexible framework to model industry 4.0 tasks for process-oriented virtual simulators involving automation and smart robots

Resumo

The advent of Industry 4.0 (I4.0) has made the industry to redefine its processes to include new technologies with the purpose of improving its production and therefore become more efficient and economically competitive. This inclusion has the drawback of making the processes more complex for the workers and the industry. One strategy to manage this growing complexity is to create simulation models to help with the decision-making a priori, i.e. before the physical system is available. In particular, virtual simulations can help multidisciplinary teams to share their expertise regarding the production processes, which is beneficial for increasing productivity and identifying issues beforehand, thus, preventing unexpected costs. However, the development of immersive simulators oriented to the industry can be difficult as it must consider many different situations, actors, and workflows as close to the physical systems as possible. As the project evolves, the development can even become unmanageable without proper engineering tools. For this reason, a new framework to model industrial processes that involve I4.0 features was developed. This framework is flexible enough to be adapted to different industrial domains, such as energy, manufacturing, or aerospace, for several purposes that include prototyping, design, process engineering, or decision-making. It is prepared for multiple I4.0 technologies including Virtual Reality (VR), Augmented Reality (AR), Human-Robot Collaboration (HRC), Motion Capture (MoCap), Digital Twin (DT), and Reinforcement Learning (RL), without losing generality. The framework supports the interaction among multiple actors, such as humans and automated devices. It also considers different types of tasks to model processes, including assembly, disassembly, and logistics. It comprises two modules, the process definition and the simulator with an embedded process controller, which communicate through an interface. The proposed framework has been applied to the development of four industrial scenarios: an aircraft Final Assembly Line (FAL) simulator, a guidance tool for high-voltage cell security, an application for machine-tool usage training, and a DT of a robotic Non-Destructive Testing (NDT) system. For the former, a comprehensive study of the productivity and ergonomics of several strategies with different automation levels was made. This study includes the VR simulation of 13 fully automated and 10 semi-automated basic scenarios for cabin and cargo assembly of sidewall panels, hatracks, and linings. The data collected during these simulations served to create 81 whole aircraft new assembly combinations of these parts and evaluate them in terms of 5 Key Performance Indicators (KPIs): assembly time, worker cost, investment, Return of Investment (ROI), and ergonomics. The results show that most of the new proposed scenarios improve the assembly time, worker cost, or ergonomics of the process, with an investment varying between 100K and 200K euros and ROI of 1-2 years. As many I4.0 processes include smart robotics, a workflow for integrating RL technologies to the framework was created. With this workflow, a robotic task can be formalized as a RL problem by leveraging the Markov Decision Process (MDP) theoretical background. Then, a RL method can be chosen to train an agent in the virtual environment. Finally, the model obtained by this training is used to perform the autonomous robotic tasks inside the simulator. Two use examples of this workflow are presented: an agent for robotic reaching task and an agent for the assembly planning of an aircraft part.