Surrogate-Assisted Inverse Thermal Characterization of Façade Elements in Testing Facilities

Current methods for the thermal characterization of façade elements typically rely on costly heat-flux sensors, controlled indoor conditions, and long monitoring campaigns. This study presents a surrogate-assisted inverse framework for digital thermal characterization in façade testing facilities. To this end, two main contributions are introduced. First, long short-term memory (LSTM) metamodels were developed, trained on ten years of high-resolution EnergyPlus simulations, to reproduce the dynamic thermal response of opaque and transparent façade elements from environmental boundary conditions together with measured surface and air temperatures. Second, these metamodels were integrated with a genetic algorithm to estimate thermal conductivity from short monitoring windows. Unlike conventional quasi-steady or heat-flux-based inverse methods, the proposed framework enables characterization under fully free-running conditions without requiring controlled indoor environments, sustained temperature gradients, or direct heat-flux measurements, thereby reducing instrumentation requirements, energy demand, and testing complexity. The methodology was demonstrated on a modular façade testing facility in Valladolid (Spain) and evaluated across multiple façade configurations. The LSTM metamodels achieved high predictive accuracy relative to EnergyPlus, with worst-case RMSE below 0.175°C for opaque elements and 0.220°C for transparent elements over 1.4 million unseen data points. Forty-eight characterization runs covering four materials and twelve months showed robust conductivity estimates. Additional evaluation using independent free-running experimental data from Salta (Argentina) further supported the applicability of the framework under different climatic conditions, including low-gradient cases where conductivity identifiability is more challenging.