Using neural networks to better predict behavior of mechatronic applications

(12-01-2022) In his PhD, Wannes De Groote looked for methods in which physics-inspired models can be combined with neural networks to better predict the behavior of mechatronic systems.

Mechatronic applications (mechatronics = combination of electrical, mechanical and control engineering) such as robots, production machines and electric cars have become indispensable in our modern society. More and more use is being made of accurate system models that can predict the complex behaviour of these mechatronic applications.

The construction of these system models traditionally relied on physical knowledge, using basic laws to create simplified representations of the considered application.

"A frequent problem is that there is often insufficient knowledge to describe all the interactions of the system by known physical laws," says Wannes.

In recent years, machine learning (artificial intelligence), which requires no prior knowledge of the system and by fitting - non-physically interpretable - model parameters to the measurement data, is able to learn relationships from the data itself.

"Unfortunately, these models are often barely interpretable and reliability can often not be guaranteed," explains Wannes.

"Therefore, in my PhD I have been looking for methods where physics-inspired models can be combined with neural networks, in order to obtain interpretable, accurate and robust architectures for predicting the nonlinear behaviour in mechatronic systems," Wannes concludes.

Read the entire PhD


PhD Title: Hybrid Physics-Based Neural Network Models for Predicting Nonlinear Dynamics in Mechatronic Applications


ContactWannes De Groote, Guillaume Crevecoeur, Sofie Van Hoecke

Wannes De Groote

Wannes De Groote was born in 1993 in Knokke-Heist, Belgium. In 2017 he obtained his degree Master of Science in Electromechanical Engineering with specialization in control engineering and automation.

In 2017, Wannes had the opportunity to start a PhD within the Department of Electromechanics, Systems and Metal Engineering (ESME) under the supervision of Guillaume Crevecoeur and Sofie Van Hoecke.

His research was focused on combining physics-inspired and data-driven models for predicting the behavior of mechatronic systems.

This research was conducted within the (Flanders Make) MODA ICON project and the Flanders AI Research Program. This led to the collaboration with multiple industrial partners. Subsequently, he obtained a personal grant (Strategic Basic Research) at the FWO in 2018.

Wannes eventually became lead author of 4 peer-reviewed A1 articles and 1 peer-reviewed P1 conference contribution.


Editor: Jeroen Ongenae - Final editing: Ilse Vercruysse - Illustrator: Roger Van Hecke