This post has been republished via RSS; it originally appeared at: Microsoft Tech Community - Latest Blogs - .
Microsoft Project Bonsai has partnered with Ansys to integrate Ansys Digital Twin with the Bonsai platform. The integration brings new opportunities for Ansys customers to unlock new levels of optimization with Bonsai, Machine Teaching, and Deep Reinforcement Learning.
Performance and accuracy are the most critical aspects when training AI agents in digital environments. One of the challenges with training AI is that it requires a lot of data for the agent to learn anything meaningful. Ansys Digital Twin builder provides all the tools you need to create a reduced order model of your original simulation model, wrapped, and exported as a digital twin. Using the Digital Twin instead of the original Computational Fluid Dynamic (CFD) model enables easy scaling in the cloud and improved runtime performance, allowing you to get to the needed data samples in a reasonable amount of time. Ansys Digital Twin builder provides all the tools you need to create a twin model of your original simulation model.
Why are reduced order models (ROMs) so important for Training AI?
Traditionally mathematically complex simulations, such as CFD simulations, are too slow for the effective training of AI agents in control scenarios. This means that optimization of processes that contain CFD models are blocked from the advantages a platform like Bonsai offers, unless the AI agents learn from an environment that can be executed at much faster speeds. This is where reduced order models (ROM) come into play. ROMs are a representation of the original model using a neural net that has learned to match input and output values without solving complex equations.
ROMs, and the speed they provide, enable businesses to develop intelligent control systems using the Bonsai toolchain and existing Ansys simulation models.
As a specific example, consider a model of an aircraft environmental control system (ECS) that we used to train an intelligent controller. The original model was converted to a ROM using Ansys Twin Builder.
Bonsai provides an easy-to-use workflow for Ansys Digital Twins that allows quick integration and scaling simulation instances as needed. Once the twin is connected, we define our optimization goals. Bonsai will interact with the digital twin to learn a control policy for temperature and pressure management as the plane climbs and descends. Pressure and temperature are crucial for safety and passenger comfort and thus it is important that the brain learns to control them optimally. During training, users can monitor data flowing between the twin and the AI learning service.
Once the brain has successfully trained in Bonsai, it can be easily exported and independently assessed against the ROM and full simulation to validate the AI capabilities. Independent assessment is an important step in the workflow as it also ensures that they brain is providing decisions within the safety parameters of the model.
An Ansys Digital Twin is a wrapped and exported ROM, ideal for AI training scenarios. Find out more here: Ansys 2021 R2: Enhancements in Ansys Twin Builder Deployment Workflow | Ansys
Where can ANSYS Digital Twins be applied?
Ansys Digital Twins are complete virtual prototypes of real-world systems that can be deployed to manage the entire lifecycle of products and assets across multiple industries. For example:
|
Use Case |
Applications |
Target Industries |
|
Industrial Flow Networks |
Management of fluid networks, machines, and mixing/blending |
O&G Process Industry, Healthcare |
|
Drive Systems and Electrification |
Optimize performance characteristics of electric motors, batteries, and power |
Automation, Automotive and Off Highway, Health care electronics |
|
Heating and Cooling |
HVAC/Refrigeration, thermal processing, and thermal stresses on component during operations |
Discrete Manufacturing, Utilities, and Infrastructure |
These use case areas present opportunities for using the Bonsai toolchain to develop and deploy intelligent AI-based control system for further system or process optimization. It’s good to remember that the Bonsai toolchain can potentially be applied anywhere where traditional control methods are struggling, and human intelligence is needed to operate the process efficiently, or wherever rapidly changing conditions require a robust intelligent controller that has learned from edge cases in simulation.
Useful AI is an ensemble of techniques
The emergence of the next wave of useful AI combines subject matter expertise and human collaboration through Machine Teaching and models of many different kinds. Observing this evolution of AI, we see that it is an ensemble of techniques which leads to success.
At Bonsai we think of AI brains as combining best in class solutions in supervise learning, mathematical modelling, programming, and deep reinforcement learning. These brains are taught skills by subject matter experts and learn in best in class simulated environments, taking their place besides human operators to drive the next level of business value.
Conclusion
Microsoft Project Bonsai is always seeking new simulation use cases to enable the best environments for brains to learn. The Ansys Digital Twin integration with Bonsai is a step on all our journey to unlocking new levels of business value with AI.
- Example code and documentation: https://github.com/microsoft/bonsai-twin-builder
- More on Ansys Digital Twins: https://www.ansys.com/products/digital-twin
- Contact Microsoft Autonomous Systems: https://aka.ms/asgetstarted