Periodic Reporting for period 1 - Edge Twins HPC (Bringing Digital Twins to the Edge for mass Industry 4.0 applications)
Berichtszeitraum: 2020-06-01 bis 2021-11-30
Digital Twins, along with the Internet Of Things and Edge computing, are expected to play a crucial role in the next decade’s industrial markets (Industry 4.0) enabling dramatic improvements in design and operation processes. A Digital Twin should be understood as a digital representation of a physical product, production process, or product’s utilization, in operation, with the required accuracy and fidelity to predict actual, physical performance. The twin replicates the real-world system throughout its lifecycle, being continuously updated, through sensors embedded in the physical system, to replicate its improvements, performances and issues. Digital Twins are usually run and maintained in the cloud while fed with data from field devices.
The added value of a Digital Twin is that it can provide a link between the physical and digital worlds. This link, or physical-digital loop, allows to understand past and present operations and make predictions based on real-time data by leveraging machine-learning approaches to condition monitoring, anomaly detection and failure forecasting.
Despite their potential, Digital Twins have not yet been widely used in engineering, other than some applications related very complex and demanding systems like in automotive and aerospace. The reason for this is that the development of digital twins is still very challenging, since it requires the collaboration of experts in multiple fields, as well as the use of robust and affordable computational tools, able to integrate multiple physics as well as diverse solving technologies. The kind of resources that are only available to OEMs or Tier-1 manufacturers.
The EdgeTwins project aims to develop a software toolchain to allow the generation of very compact Digital Twin apps able to run on the Edge; that is, twins that are installed and operate on the physical asset they represent, enabling a new breed of novel real-time applications from autonomous vehicles to small devices.
The new software tools will be open source, eventually deploying and freely sharing source code as needed, and will also allow leveraging HPC hardware for the training phase.
The added value of a Digital Twin is that it can provide a link between the physical and digital worlds. This link, or physical-digital loop, allows to understand past and present operations and make predictions based on real-time data by leveraging machine-learning approaches to condition monitoring, anomaly detection and failure forecasting.
Despite their potential, Digital Twins have not yet been widely used in engineering, other than some applications related very complex and demanding systems like in automotive and aerospace. The reason for this is that the development of digital twins is still very challenging, since it requires the collaboration of experts in multiple fields, as well as the use of robust and affordable computational tools, able to integrate multiple physics as well as diverse solving technologies. The kind of resources that are only available to OEMs or Tier-1 manufacturers.
The EdgeTwins project aims to develop a software toolchain to allow the generation of very compact Digital Twin apps able to run on the Edge; that is, twins that are installed and operate on the physical asset they represent, enabling a new breed of novel real-time applications from autonomous vehicles to small devices.
The new software tools will be open source, eventually deploying and freely sharing source code as needed, and will also allow leveraging HPC hardware for the training phase.
The EdgeTwins project explored the potentiality of computer-based reduced order models in the development of predictive twins able to run on small form factor compute device.
The work done includes a prototype implementation of SVD able to leverage HPC hardware as well as some initial scripts to demonstrate the possibility of running parallel training in a simple setting. In this sense, work is performed to produce and publish to the open repository scripts that ease the construction of new models and allow leveraging supercomputers by the combination of the PyCOMPSs library with the Kratos software capabilities.
The work also includes the selection of a testcase for which a tentative business plan is drafted. A simple demo, running on a Raspberry PI 4 as testbed is produced and shown for feedback to potential customers.
The work done includes a prototype implementation of SVD able to leverage HPC hardware as well as some initial scripts to demonstrate the possibility of running parallel training in a simple setting. In this sense, work is performed to produce and publish to the open repository scripts that ease the construction of new models and allow leveraging supercomputers by the combination of the PyCOMPSs library with the Kratos software capabilities.
The work also includes the selection of a testcase for which a tentative business plan is drafted. A simple demo, running on a Raspberry PI 4 as testbed is produced and shown for feedback to potential customers.
The construction of Digital Twins requires normally a large investment both in terms of human resources as well as of simulation software expenditure. EdgeTwins contributes to the minimization of such costs by the production of open source code and of know-how on the subject.
The newly defined twins also allow advances towards obtaining:
- Lower analytics latency. Applications that require sub-second latencies are possible, i.e. system protection functions, including shutoff, can be instantly activated if a threat is identified or forecasted by the digital twin analytics.
- Closed-loop integration of analytics and device control. Analytics produced by the digital twin can guide the local controller. This leads to proactive control applications resulting in autonomous operation. For example, a forecasted critical anomaly could be mitigated without human intervention.
- Faster evolution of the digital twin. By opening the door to using approaches like online machine learning on streaming data and real-time reinforcement learning, the twins developed with the new techonolgy could potentially continuously self-learn and evolve. This results in optimized system operation and self-tuning devices.
Executing the Digital Twins at the Edge instead of in the cloud also has business advantages:
• Democratization of real-time simulation extending it to new applications including small devices.
• Reduced cloud hosting costs. Sending all data to the cloud for storage and analysis can be costly.
• Reduced data transmission costs. Data pre-processing reduces the data volume transmitted to the cloud.
• Cybersecurity and privacy. Sensitive data need not be sent to the cloud.
• Higher Resilience. Analytics can be performed even when the digital twin is disconnected from the cloud.
These advantages can trigger novel applications related to assets that require faster and localized decision making at the edge: Autonomous vehicles (or their subsystems), Utilities (i.e. smart solar converters for grid regulation), Industrial, logistics and smart cities (i.e. image recognition in edge devices, infrastructures structural integrity monitoring), etc.
The newly defined twins also allow advances towards obtaining:
- Lower analytics latency. Applications that require sub-second latencies are possible, i.e. system protection functions, including shutoff, can be instantly activated if a threat is identified or forecasted by the digital twin analytics.
- Closed-loop integration of analytics and device control. Analytics produced by the digital twin can guide the local controller. This leads to proactive control applications resulting in autonomous operation. For example, a forecasted critical anomaly could be mitigated without human intervention.
- Faster evolution of the digital twin. By opening the door to using approaches like online machine learning on streaming data and real-time reinforcement learning, the twins developed with the new techonolgy could potentially continuously self-learn and evolve. This results in optimized system operation and self-tuning devices.
Executing the Digital Twins at the Edge instead of in the cloud also has business advantages:
• Democratization of real-time simulation extending it to new applications including small devices.
• Reduced cloud hosting costs. Sending all data to the cloud for storage and analysis can be costly.
• Reduced data transmission costs. Data pre-processing reduces the data volume transmitted to the cloud.
• Cybersecurity and privacy. Sensitive data need not be sent to the cloud.
• Higher Resilience. Analytics can be performed even when the digital twin is disconnected from the cloud.
These advantages can trigger novel applications related to assets that require faster and localized decision making at the edge: Autonomous vehicles (or their subsystems), Utilities (i.e. smart solar converters for grid regulation), Industrial, logistics and smart cities (i.e. image recognition in edge devices, infrastructures structural integrity monitoring), etc.