A transient simulation-based approach for in-situ estimation of U-values using artificial neural networks

The U-value is the most important criterion for evaluating the quality of thermal envelopes. However, reliable estimations of in-situ U-values are still subject to current research activities. This paper presents the application of artificial neural networks (ANNs) for estimating U-values using synthetic input data derived from transient building element simulations. Building elements, represented in established building typologies, are modeled using finite volume methods. The boundary conditions as well as the thermal impacts induced by internal and external climate are modeled with a transient behavior. The results of the aforementioned simulations, in particular internal and external surface temperatures as well as heat flux density, serve as inputs for the ANNs. The authors study different ANNs, including the hyperparameters and training strategies, focusing on the performance of predicting U-values. Finally, the proposed method is applied to in-situ measurements.