Can thermodynamic quantities estimated from in-situ measurements be extensively used to characterise building elements during energy simulations? Errors and uncertainties in in-situ measurements and parameter estimation processes
Reliable energy simulations require an accurate knowledge of the thermal characteristics (i.e. building materials and their thickness, thermodynamic properties, etc.) of the building element being investigated. Tracing exhaustive information can be very problematic for existing or even new buildings. Consequently, approximate values (e.g., tabulated inthe literature) are usually used during simulations, leading to inaccurate results. The prediction of thermodynamic parameters (e.g., thermal resistance and thermal capacity) from the analysis of in-situ measurements (i.e. temperature and heat flux) is of fundamental importance to characterise the actual thermal behaviour of the building element. The possibility of estimating thermal properties from in-situ measurements is not extensively applicable at present due to a number of limitations in the methods usually adopted. Therefore, new methods to estimate the thermodynamic parameters of building elements from monitoring campaigns should be investigated together with an analysis of the uncertainties they are affected by.
Aim of this dissertation was to apply a novel dynamic Bayesian-based building simulation method (Biddulph et al., 2014) to primary data collected by the author on a solid wall. Six weeks of data were collected during late spring and summer 2013 using two heat flux and four surface temperature sensors, and analysed to compute weekly U-values. An error analysis was performed on the estimated outputs (e.g., R-value and U-value) to investigate: a) systematic errors in the U-value prediction; b) the variation of U-values over the monitoring period; c) how surveys length impact on the accuracy of final results.
Results showed that: a) the method returned sensible U-values (comparable with the U-value calculated inferring the thermal properties of the different layers of the wall from the literature) from time series collected in summer; b) the error analysis showed that accurate U-values were achieved also analysing shorter time series (i.e. weekly U-values felt within a range of -3% and +7% of the U-value obtained analysing the whole six weeks of data); c) although the sensors were placed quite close on the wall, the U-values obtained were affected by systematic errors. This last result highlights that the characterisation of the thermal performance of a whole building element from a single spot measurement may be affected by uncertainties.