Loft insulation is one of the most common household energy efficiency upgrades in the UK, yet there is a scarcity of in-situ data evidencing the thermal performance of insulated roofs in real houses. Such data is vital in characterising the thermal performance of the wider housing stock and improving the accuracy of energy-use estimates included in modelling tools which
are used to inform government policy aimed at reducing carbon emissions
in this sector.
Motivated by this gap in knowledge, this research aims to gather insitu data in a sample of case study roofs in order to quantify the potential ‘building fabric thermal performance gap’ associated with residential
cold-pitched roofs. The research also aims to investigate the range of issues
which may influence a performance gap; including variability in weather
conditions, installation issues and the functional complexity of lofts which
often house building services and may be accessed by occupants for storage.
In-situ data from four case study houses with cold-pitched roofs is therefore used to estimate localised and area weighted in-situ ceiling and roof
U-values. In addition to quantitative measurements, visual inspections and
thermal imaging surveys were conducted to investigate the installation and
condition of the loft insulation in the case studies. The results suggest that
the in-situ thermal performance of the case study roofs is better than expected and this positive building fabric thermal performance gap is expected to be a consequence of convective and radiative heat flows in the
loft void. If such a performance gap exists in the wider stock, the expected energy and cost savings associated with the installation of additional loft
insulation may not be realised, making it harder for the UK to achieve its
carbon emissions targets.
The second key aim of this thesis is to trial a dynamic probabilistic
method combining a lumped parameter model and Bayesian optimisation
for estimating in-situ U-values. The dynamic approach which quantifies
the thermal mass of the element can incorporate the effects of solar gains
and has been shown to reduce the amount of data required to estimate Uvalues for wall elements. For a typical ceiling or roof, the thermal mass of
the element is low and there is a functional complexity associated with the
presence of a large void in the roof.
A collection of lumped parameter models are trialled, including models
which incorporate an additional heat input to account for solar gains on the
surface of the roof. The results suggest that a simple model with a single
effective thermal mass can sufficiently characterise the heat transfer through
a roof, producing physically reasonable parameter estimates and sensible
in-situ U-value estimates; in some cases, where the conventional steadystate average method failed. The models incorporating solar gains were of
limited value in this research; however, this is expected to be due to the low
solar gains observed during the data collection period. Further testing of
the solar models with additional datasets is therefore recommended as well
as refinement of the 1C2R model so that the model is able to identify each of
the parameters independently, thereby improving the statistical significance
of the results.