Published: Jul 22, 2022 by Lizzy Cross

Keynote presentation for the 10th European Workshop on Structural Health Monitoring, 2022

I was thrilled to give a keynote talk at EWHSM2022 on physics-informed machine learning for Structural Health Monitoring. It was a great conference and so good to see so many people again.

In the talk I outlined the motivation for wanting to try and put some physics back into our much loved data-driven algorithms and gave a bit of an overview of the group’s work - which you can read more about in these webpages!

I showed a spectrum of white to black-box models with examples across it (the book chapter on the publications page is a good place to read about these examples) before talking about ongoing and future challenges. For me two of these are model validation and getting that balance right between the physics and data-based components of your model. For the former I showed off some work going on at our amazing lab (check out lvv.ac.uk) and for the latter I started to try and formulate this discussion by thinking about where any given problem sits in terms of balancing evidence from physical insight and evidence from data.

Where do you think your modelling problem lies in this space? Clearly there is no absolute answer or measure but I think that trying to think in this way might help to point to which model architecture might be the most suitable - the spectrum of models available maps really nicely into this space!

If you want to see more, check out the presentation slides and watch this space for a more formal exploration of whether this makes sense…