In the realm of energy efficiency and building renovation, accurately determining the thermal conductivity of materials is crucial for assessing and improving energy performance. Researchers Ali Waseem and Malcolm Mielle from the University of Luxembourg have developed a novel approach using Physics-Informed Neural Networks (PINNs) to tackle this challenge. Their work, published in the journal “Neural Networks,” offers a promising solution for estimating material properties in situ, under real-world conditions.
Traditional methods for measuring thermal conductivity often fall short due to their invasive nature, lengthy observation periods, or sensitivity to environmental factors. Waseem and Mielle’s research introduces a PINN-based iterative framework designed to estimate the thermal conductivity (k) of a wall using thermographic data. This approach alternates between solving the forward heat problem with a PINN for a fixed k and optimizing k by comparing the predicted and observed thermographs and surface temperatures. This process repeats until the estimated k converges to a stable value.
The researchers validated their framework using both real-world environmental data captured by a weather station and simulated data generated from Finite-Volume-Method software. Their findings demonstrate accurate predictions of k across various environmental conditions and data collection sampling times, provided the temperature profile of the wall at dawn is close to steady state. Even when the steady-state assumption is violated, the framework maintains a relatively low maximum mean absolute error (MAE) of 4.0851.
The practical applications of this research for the energy sector are significant. By enabling reliable, non-invasive, and efficient estimation of thermal conductivity in situ, this method can streamline the assessment of building energy efficiency. This is particularly valuable for evaluating the impact of facade renovations on thermal transmittance, a key factor in reducing energy consumption and improving sustainability. The researchers hope their work will inspire further exploration of machine learning, particularly PINNs, for solving in-situ inverse problems in the energy industry.
The research was published in the journal “Neural Networks,” highlighting the innovative intersection of physics, machine learning, and energy efficiency.
This article is based on research available at arXiv.