Physics-Enhanced machine learning refers to the fusion of sensing data, physical constraints and engineering knowledge within a common learning environment - also known as physics-informed ML, hybrid modelling, grey-box modelling, or scientific ML. More concretely, physics-enhanced schemes strive to integrate first-principles knowledge and physical biases with real-world observations and machine learning pipelines, thereby improving predictive accuracy, quantifying uncertainty, and enhancing computational efficiency and real-time feasibility. Such capabilities are critical for Structural Health Monitoring (SHM), Digital Twinning, and Engineering Decision Support.

This special session welcomes contributions on both fundamental research and industrial applications, including but not limited to:

• Physics-informed forecasting and anomaly detection in SHM,
• Methods for system identification, uncertainty quantification, and state estimation under incomplete or noisy sensing,
• Hybrid models for real-time inference and predictions in structural systems,
• Model adaptability under environmental and operational variability,
• Scalable and transferable architectures for digital twinning.

The session aims to foster dialogue between communities in mechanics, data science, and machine learning, showcasing advances that push structural engineering towards intelligent, self-adaptive, and resilient systems.

Keywords: Physics-enhanced, Physics-informed, Scientific Machine Learning

Track chairs

Konstantinos Vlachas, ETH Zürich

Marcus Haywood-Alexander, ETH Zürich

Eleni Chatzi, ETH Zürich