Physics AI

See also our blog about machine learning.

See also our whitepaper on physics AI.

What are the use cases of physics AI?

Fast surrogate models

The advantage of machine learning models is that they provide answers very quickly. This makes them ideal for model-in-the-loop control, scenario studies where numerous scenarios must be calculated, and crisis situations where rapid forecasts are needed.

A fast machine learning model for a scientific process can be obtained by training it with the results of a model based on scientific knowledge. See also this case study.

Note that machine learning is just one way to make a model faster. See also our whitepaper on speeding up computational models.

Machine learning models for sensor data

Sensor data often contains errors and is typically limited in size or scope. This makes it virtually impossible to train a reliable machine learning model with sensor data. The solution is to train the model using scientific knowledge of the process that generates the data in addition to the data itself.

Note that there are other methods to use sensor data in combination with scientific knowledge. See our page on models and sensor data.

Adding Missing Physics

Some simulations or forecasting calculations also contain components for which no good scientific model exists. These parts can be filled in with machine learning based on the difference between computing results and observations.

How Does Physics AI Work?

Simulations and predictive computations that are based on the laws of nature are often relatively slow. Moreover, they can only compute processes for which an explicit scientific description exists.

Machine learning has the advantage of producing answers very quickly once the model is trained. Moreover, it learns purely from data and does not require a scientific formulation.

The disadvantage of machine learning is that it requires a lot of data, which is often not available for processes in industry and environmental management. And even with a lot of data, a machine learning model can still produce results that are scientifically incorrect. This makes such a model unreliable.

Physics AI explicitly uses scientific knowledge when training a machine learning model. This knowledge complements the data, providing enough information to train the model. Moreover, it ensures that the model complies with the laws of science. This makes the model reliable enough for use in a critical environment.

How can VORtech help you?

VORtech offers the end-to-end services related to physics AI:

• Assessment of the potential of physics AI for your application.
• Assistance in training a machine learning model for an industrial process or environmental management.
• Assistance in training a machine learning model based on sensor data.
• Audit of machine learning models for industrial processes or environmental management.

If you are interested in these services, please contact us for a free, no-obligation consultation.

Why choosing VORtech as your partner for physics AI?

VORtech offers the following advantages as a partner:

• We have longtime experience with applications in environmental management and industry. We work for both government institutes and large corporations.
• Our colleagues combine deep mathematical knowledge and excellent software development skills with a background in physics or engineering. This makes them uniquely qualified to collaborate with your experts.
• With a wide range of expertise, we can select the method that is best suited to your challenge, whether it is machine learning or an entirely different approach.
• VORtech fully shares all the knowledge and experience that goes into the solution of your challenge. We train your experts so that they can work with it and extend it whenever needed.
• You will work with the same VORtech employees across multiple assignments, making any subsequent assignments ever more efficient.