The advance in machine learning drives activities in data collection also in the manufacturing sector, where there is increasing data collected from physical assets (industrial robots, machines, etc). This data is used to create a virtual representation of physical assets or the so-called digital twins. Ideally, the digital twin should resemble the state of the physical asset as close as possible (internet of reality). However, with the fact that the digital twin might be executed in a remote location, there is a delay between the data collected from the physical asset (and stored to form the digital twin) regarding the physical asset due to communication delay. Consequently, any user who observes or interacts with the digital twin at any point in time always sees the digital twin as the “past” state of the physical asset. In addition, the condition of the underlying communication network may affect the data being transmitted to the digital twin, the collected data might be missing in transport, or arrives scrambled, demanding the use of timestamps and intensive checking. This affects the overall user experience of the user who interacts with the digital twin.
At Aalto Factory of the Future, we are experimenting with an approach based on visualizing in real-time dynamic objects in a virtual model of the shop floor. Here, a more intuitive interface based on a 3D representation of physical assets plays an important role to make the interface intuitive to operate and easy to understand. However, connecting the virtual environment to a highly dynamic and adaptive physical one is not that simple. Our approach corresponds to a theory-practical approach to support adaptability in manufacturing through the virtual and physical assets in our facility. We develop algorithms and demonstrators applicable to common tools for cloud-based 3D visual interfacing towards the conception of an integrated Factory of the Future demonstrator.
Currently, we work on a web-based remote operation of a MIR-100 Automated Guided Vehicle, applying the best of distributed computing. The web application is based on an enhanced visualization system deployed as an Azure cloud service and the navigation system is deployed as an edge AI platform. The real robot in the lab can be visualized and controlled remotely by users connected securely through a VPN connection. But also they can simulate it locally for testing. The AGV is equipped with a height-adjustable cobot arm, able to reach objects up to 2m high.
EIT Manufacturing - Mirrorlabs is an excellent example of how this approach to digital twin thorough internet of reality makes it possible to provide a platform for remote experimentation and simulation. The tools in development on the AFoF support operational data analytics and IIoT, enabling physical and virtual assets to students anywhere in the world. A set of open source tools as ROS and Gazebo are an important part of the framework, as well as a professional networking platform for VPN and networking.
In the future, we expect to expand our approach to include other possible scenarios, for example, plan or allow the operator to see the robot movement plan in simulation, before the physical.