AI Researchers Teach Machine Learning Physics

Related Vendors

TU GRAZ

Taiwan Machine Tool & Accessory Builders' Association (TMBA)

Wuxi InfiMotion Technology Co., Ltd.

SIMTOS (Seoul International Machine Tool Show)

By combining physics-based methods with machine learning, a team at the Institute of Thermodynamics and Sustainable Propulsion Systems at TU Graz (Austria) is developing models that aim to deliver better results despite limited training data.

However, for machine learning models to deliver good results, they require a large amount of training data. In industrial applications, this data is often either insufficiently available or can only be obtained through very costly methods. For this reason, laboratory director Stefan Posch and his team at the Institute of Thermodynamics and Sustainable Propulsion Systems at TU Graz are working in the newly inaugurated "Christian Doppler Laboratory (CD Laboratory) for Physics-Based Machine Learning in Industrial Applications" to combine traditional machine learning with physics-based methods.

Less Training Data, More Accuracy

The resulting physics-driven machine learning models achieve their results not only based on the available data but also by adhering to the laws of physics they must follow. According to the researchers, this reduces the required amount of training data while simultaneously increasing accuracy. By using such models, numerical simulations in areas such as structural or fluid mechanics can be accelerated, significantly shortening product development times.

If we directly incorporate physical knowledge into machine learning models, we require significantly less data—a major advantage in fields such as mechanical engineering.

Laboratory manager Stefan Posch

"For digital models to be useful for industrial applications, their results must align with the real world," explains Stefan Posch. "By integrating physical knowledge directly into machine learning models, we need significantly less data—a major advantage in fields such as mechanical engineering, where data is often difficult or very costly to generate. At the same time, the models become more transparent, comprehensible, and, to some extent, capable of making reliable predictions for situations not included in the original data. This opens up new possibilities for the safe and efficient application of artificial intelligence in engineering."

Universal Methods As the Goal

Stefan Posch has been working in the relatively young research field of physics-driven machine learning for about five years. At that time, the COMET module LEC HybTec was launched at the COMET K1 Competence Center Large Engines Competence Center (LEC), laying the foundational groundwork for the current CD Laboratory. The LEC, along with several institutes of TU Graz and the Graz Center for Machine Learning, continues to support the research on the scientific side.

Industry partners of the CD lab are: Andritz Hydro GmbH (drives and generators), BRP-Rotax GmbH & Co KG (small engines), the Engineering Center Steyr GmbH & Co KG as part of Magna International Inc. (chassis), Everllence SE (formerly MAN Energy Solutions SE) (large engines), and Palfinger Europe GmbH (cranes and lifting platforms).

Challenge of Model Complexity

They contribute their in-house data, realistic problem scenarios, and experience with numerical simulations. The goal is to develop fundamental, universally applicable methods through which the combination of machine learning and physics delivers reliable and reproducible results that align with reality. These results can then be tailored by corporate partners to meet their specific needs. The biggest challenge is that incorporating physical rules significantly increases the model's complexity. Therefore, a major part of the research will focus on optimizing the model by reducing the physical parameters without sacrificing accuracy.

About the Christian Doppler Laboratories

Christian Doppler Laboratories are technical-scientific research institutes at Austrian universities. They conduct application-oriented basic research at a high level, with scientists cooperating with companies. The Christian Doppler Research Association is considered an international best-practice example for promoting such collaboration.

Future-Proof Authentication with Universal RFID Readers

Universal RFID Readers: One Solution for All Credentials