The research demonstrates how Physics-Guided Neural Networks (PGNNs) can combine first-principles engineering with machine learning to achieve highly accurate feedforward control while maintaining the stability and predictability required for demanding industrial applications.
IBS Precision Engineering has been involved from the early stages of the research, with a strong focus on its practical application. The aim was not only to develop the control theory, but to ensure that the approach could ultimately be implemented in real machines to improve motion performance.
To bridge that gap, different experimental set-ups were developed and extensively tested, including validation on real machine hardware. This practical work helped evaluate and refine the control approach under realistic operating conditions.
For ultra-precision motion systems, this is particularly relevant. Machine learning can capture complex behaviour that is difficult to represent accurately in conventional physical models, but its application in industrial control requires confidence in how the system will behave. By integrating physical knowledge into the learning approach and providing stability guarantees, the research takes an important step towards applying machine learning with confidence in ultra-precision motion control.
The research brought together expertise from IBS Precision Engineering, ASML, Canon Production Printing and Eindhoven University of Technology.
The authors of the award-winning paper are:
• Max Bolderman, IBS Precision Engineering
• Hans Butler, ASML
• Sjirk Koekebakker, Canon Production Printing
• Eelco van Horssen, ASML
• Ramidin Kamidi, ASML
• Theresa Burke, IBS Precision Engineering
• Nard Strijbosch, IBS Precision Engineering
• Mircea Lazar, Eindhoven University of Technology
We thank the International Federation of Automatic Control (IFAC), Elsevier and the Control Engineering Practice editorial team for this recognition. The award encourages us to continue translating advanced control research into practical technologies for the next generation of ultra-precision motion systems.
Read the award-winning research paper (online)
Physics-guided neural networks for feedforward control with input-to-state-stability guarantees
Read more on our PGNN research (PDF): Physics–Guided Neural Networks for Feedforward Control: with application to an industrial linear motor