A view on Physics Enhanced ML
Check out this latest publication in the Data-Centric Engineering journal, led by Marcus Haywood-Alexander in collaboration with Wei Liu, Kiran Bacsa, Zhilu (Oscar) Lai and Eleni Chatzi.
📄 “Discussing the Spectrum of Physics-Enhanced Machine Learning: A Survey on Structural Mechanics Applications”
📖 Published in Data-Centric Engineering
Open access Paper: external page https://doi.org/10.1017/dce.2024.33
Code: external page https://github.com/ETH-IBK-SMECH/PIDyNN
This work essentially organizes our thoughts and experiences with various methods and flavors across the spectrum of so-called physics-enhanced machines learning for digital twinning and monitoring applications. The work delves into the exciting paradigm of Physics-Enhanced Machine Learning (PEML), which combines the strengths of physics-based modeling and machine learning. By surveying recent applications in structural mechanics, we showcase the two-dimensional spectrum of PEML techniques and discuss their characteristics, motivations, and potential.
To bring these methods to life, we apply different PEML approaches to the Duffing oscillator example, showcasing the unique advantages of different techniques. To foster transparency and collaboration, we’ve also shared the code used to generate these examples.
⚙️ Key contributions include:
• A categorisation of PEML techniques.
• Insights into how PEML bridges the gap between physics-only and data-driven methods.
• Practical guidance for selecting PEML approaches tailored to specific engineering challenges.
The paper emphasises the transformative role of PEML in pushing the boundaries of science and engineering by harmonising physical insights with machine learning.
The work is conducted with our team members at the Chair of Structural Mechanics and Monitoring at ETH Zurich and the Future Resilient Systems (FRS) of the Singapore-ETH Centre. We also wish to acknowledge the ReCharged MSCA Staff Exchanges program for valuable interactions.