DVU Group - Physics-enhancing Machine Learning

Physics-Enhancing Machine Learning

(AKA: Scientific Machine Learning, Physics-informed ML)

This page is under development and covers Physics-enhanced Machine Learning topics - last updated: 02/2024.

Useful online resources

Many excellent seminar series on this topic:
Blog
- Data-Centric Engineering blog
Workshops:
- Physics Enhancing Machine Learning in Applied Solid Mechanics, Institute of Physics - already held in 12/2022
- SciMLWorkshop: Workshop Materials for Training in Scientific Computing and Scientific Machine Learning (SciML)
Books/Notes
- Physics-based Deep Learning Book (v0.2)

Some must read papers:

- Karniadakis et al., Physics-informed machine learning, Nature Reviews Physics, 3, 422-440 (2021)

- Chuang and Barba, Experience report of physics-informed neural networks in fluid simulations: pitfalls and frustration, Proceedings of the 21st Python in Science Conference (SciPy 2022).

DVU group resources

If you are looking for an intro lecture tutorial on Physics-enhanced machine learning (see below), for a brief review of Machine Learning Concepts, have a look at this page of this website.

A 12 pages position paper (pre-print on arxiv, open access) will introduce you to:

- Why Machine Learning is not enough?

- What is Physics-Enhanced Machine Learning (PEML)?

- How to setup PEML architectures?

- Which PEML strategy should be used?

- What are open challenges and opportunities in PEML?

Moreover, recent applications of PEML to real-world problems are summarised.

Physics-Enhanced Machine Learning: a position paper for dynamical systems investigationsThis position paper takes a broad look at Physics-Enhanced Machine Learning (PEML) -- also known as Scientific Machine Learning -- with particular focus to those PEML strategies developed to tackle dynamical systems' challenges. The need to go beyond Machine Learning (ML) strategies is driven by: (i) limited volume of informative data, (ii) avoiding accurate-but-wrong predictions; (iii) dealing with uncertainties; (iv) providing Explainable and Interpretable inferences. A general definition of PEML is provided by considering four physics and domain knowledge biases, and three broad groups of PEML approaches are discussed: physics-guided, physics-encoded and physics-informed. The advantages and challenges in developing PEML strategies for guiding high-consequence decision making in engineering applications involving complex dynamical systems, are presented.

DVU presenting at the TU Delft Summer School "The era of AI and digitalization for structural applications" - 19/06/2024

2024_06_summer_School_share.pdf

TUD_AI_summer_school_dpivae.pdf

Google Colab

My attempt to have a wider general intro to the topic of PEML, and showcasing the latest work within the DVU group - CSAR talk, 27/05/2024

2024_05_Physics_enhancing_ML_CSAR.pdf

A concise summary of what is Scientific ML, Physics-Enhanced ML and the use of PINNs for ODE and PDE investigations... with application in mechanics and materials in less than 42 slides!

This is a biased view (shared on 31/01/2024), this is how I would teach this topic.
I would be happy to receive your feedback for improving this slides deck (and yes, normally the slides are presented with transitions : )). If anything is missing, or is not clear, please, let me know.

Note:

the material builds up on the content of the slides deck share before this on Intro to ML, FCNN and CNN

SCIML_PEML_PINNs_ac685.pdf

A very brief intro to Physics-informed Machine Learning
This is a biased view. I gave this talk as opening of the workshop Physics Enhancing Machine Learning in Applied Solid Mechanics (2023)

Programme & slides from all the speakers of the workshop are available here

2023_11_20_workshop_Physics_informed_lecture.pdf

A physics-enhanced machine learning perspective to SHM

This is a biased view. I gave this talk as part of the seminars hosted by the Chair of Structural Mechanics and Monitoring at ETH Zurich (27/02/24)
The recording of the talk is available here - note that 1 slide has been slightly changed wrt video version

2024_02_27_ETH_visit_to_share.pdf

Special issue collections

Special issue collection on DCE: deadline: still accepting papers!

Data-Centric Engineering - an open access journal published by Cambridge University Press at the interface of data science and all areas of engineering - is pleased to be partnering with the Institute of Physics (IOP) workshop on Physics Enhancing Machine Learning in Applied Solid Mechanics (London, December 12th 2022). Articles developed through the workshop will be published in a dedicated, specially curated collection in DCE after peer review. We encourage workshop participants but also those who did not attend the workshop to contribute to the DCE special collection.

We welcome contributions on advanced techniques and industrial applications showcasing recent progress, strengths and limitations of using physics knowledge to enhance Machine Learning strategies in applied solid mechanics.

Particular interest is given to contributions focusing on how physics domain knowledge and the availability of a causal physics-based model enable one to move from accurate-but-wrong predictions, to explainable and interpretable inferences fully exploiting machine learning techniques in applied solid mechanics.

Relevant topics include, but are not limited to:

Probabilistic Model updating,
Virtual Sensing,
Structural Health Monitoring,
Identification of system parameters and non-linear relationships,
Uncertainty Quantification,
Reduced Order Modelling of Nonlinear problems,
Physics-informed Neural Networks,
Reinforcement Learning,
Transfer Learning.

Editors:

Alice Cicirello (TU Delft)
Zack Xuereb Conti (The Alan Turing Institute)

Call for Papers on Physics Enhancing Machine Learning in Applied Solid MechanicsCall for Papers on Physics Enhancing Machine Learning in Applied Solid Mechanics (deadline April 17th 2023)

PAGE UNDER DEVELOPMENT