Empirical and comparative validation of an original model to simulate the thermal behaviour of outdoor test cells

Calorimetric methods for the performance assessment (e.g. for the determination of the solar factor) of transparent building components have been largely applied in indoor laboratories under steady-state conditions and in outdoor test cells under dynamic boundary conditions provided by real weather. In the latter case the accuracy of the measurements depends significantly on the temporary storage of energy in the test cell envelope. An analysis by Pagliano et al. (2017) developed a dedicated lumped thermal model in Matlab environment in order to improve the design of calorimeters for the measurement of the solar factor by minimizing the energy storage effects in the envelope of the calorimeter and estimating precisely their entity. The developed model was based on literature studies on buildings’ dynamic energy simulations and adopted some common hypotheses used by existing building energy simulation software tools. However, when modelling light-mass and highly insulated buildings, such as test cell facilities, small variations in the power inputs can generate significant variations of the internal temperatures, challenging for the model to follow accurately. In order to verify the accuracy of the developed model in predicting the thermal behaviour of an outdoor test cell, an extensive validation work has been carried out. In particular, this paper summarises (i) an experimental validation carried out using a data set from the BESTLab facility, located at the research centre Électricité de France R&D Les Renardières (FR) and (ii) an intermodel comparison between the code developed in the Matlab environment and TRNSYS, a well-established building energy simulation tool. Concerning the validation at the BESTLab, the results show that the model is able to predict the temperature evolution of the internal air and of the internal surfaces of the envelope with good accuracy, with residuals lying within a range of ± 1 °C; reasons for discrepancies between measurements and predictions are discussed in the paper. As regards the intermodel comparison, the correspondence between the two software tools is generally good, with residuals lying most of the time within a range of ± 0.5 °C. The residuals are lower for the intermodel comparison, partly because input values are in this case not affected by uncertainty. Although TRNSYS and the developed Matlab code adopt some similar assumptions and simplifications, they also present some modelling differences that are highlighted in the paper. © 2017 Elsevier B.V