Combining Physics and Deep Learning

When it comes to combining physics and deep learning, you may be surprised at the outcome.

PhysiNet helps combine learning with physics models, such as digital twins, which are replicas but virtually of an object. One of the major issues faced with digital twins is their ability to combine empirical data with the physics aspect.

However, as digital twins require a tremendous amount of developmental resources, they can be a very effective alternative when physical test are outrageously expensive or entirely to dangerous.

Key Takeaways:

• Digital twins rely on black box models and rule based models but since both models can fail, the best way is to combine them into one linear weighted combination.
• Digital twins usually rely on rules or empirical data but the best models combine those rules and empirical data in a variety of different ways.
• Digital twins are an alternative to physical tests which are typically much more expensive, and the framework of the models are actually pretty simple to understand.

“In 2021, researchers at the University of Sheffield developed a very simple digital twin framework called PhysiNet to solve this problem. PhysiNet combines deep learning with physics models to develop robust forecasts of device performance.”

More details: here

References:

• Towards Data Science (Website)
• MITCBMM (YouTube Channel)