New pre-print: PINN-it — Physics-Informed Neural Networks for Thermophysical Property Retrieval.
Measuring the thermal conductivity of building walls in situ is hard: traditional methods are invasive, require prolonged observation periods, or are highly sensitive to environmental variability.
In this work, Ali Waseem and I developed an iterative physics-informed neural network framework that estimates the thermal conductivity of facades using thermographic data and environmental measurements — no invasive procedures, reduced measurement time.
- Accurate prediction when the wall’s initial temperature profile is at steady state
- Robust across seasons and sampling strategies
- A path toward practical, scalable in-situ material property estimation
This work is part of the INSULATED project.