Supata is the result of one of the latest collaborations
between EPF and Siemens. It is about a
robotic island designed to set up plants in different industrial
product sectors.
The strength of Supata is the ability to
manipulate a multitude of components, and simplicity in changing size thanks to
a vision system that automatically recognizes the parts inserted in the
database. However, the real change has been AI integration. A
software solution has been developped: integrating
with a hardware ad hoc – it enables the deployment of AI models into an automation
machine with minimum inference times to adapt to machine cycle.
- A Siemens hardware has been used, particularly an that integrates re-engineered video
cards to cover an industrial load and to give the opportunity to Supata to turn
at cycle time it needs.
- Dealing with software, a vision code has been written, that starting from the knowledge of EPF, aimed at
recognizing the piece. It integrates neural network architectures, which are
the state of the art for the "Object Localization" areas, in order to
be able to abstract the task of finding the object even when the environmental
conditions change (lights, position, shape of the object).
Moreover, some apps are used to make it happen:
- the “Performance Insight” app: it allows checking that the vision algorithm always works with the same accuracy, by relating field data at the same time (for example, how many parts move the machine, with how many movements of the robot, etc.). The monitoring system is very useful, because it allows programming the machine’s maintenance, avoiding problems and production stops.
- the “Energy Manager” app: it provides the level of energy consumption per machined part. Thanks to the recent ISO5001 certification, the app is able to give energy transparency directly to the companies that use Supata.
Today,
AI leads robot to take the object through an optimized process in which the
algorithms plan the least amount of vibrations required, using probability
calculations based on the position of the objects to collect. Therefore, a simulation in a photorealistic
environment, inside an
industrial metaverse, has taken place to achieve summary data and so
training AI, before it even existed a new part. All of this is thanks to:
- Omniverse, the simulation
platform by NVIDIA
- the Isaac Sim
app, which allows simulating robotic and artificial intelligence.
In the case
of Supata, we had to understand if the machine was able to meet certain
performances, so if the algorithm is able to detect the objects position which
have not yet created materially, by starting to the model CAD 3D of the object.
Then synthetic images (not real) of the highest quality have been generated in
the industrial metaverse, and we checked that the robot was able to meet the
needs, everything in a simulated environment. Siemens and NVIDIA – pioneer company of graphic accelerated and AI-
have recently announced the extension of their partnership to
enable the
industrial metaverse and to
increase the use of digital twin technology leads
by AI.
