Validation of the Heat3D System for Mapping U-Values in Homes Through a Two-Phase Winter Field Trial

Abstract

Effective evaluation of building fabric is essential to understanding building performance. Traditional methods, such as ISO 9869-1, provide point U-values but do not always capture the complete picture, and new measurement-led approaches are needed to support building retrofit. Heat3D is a novel iOS application that performs rapid U-value measurements of building elements by mapping thermographic images from a mobile infrared camera onto an augmented reality (AR) model of a room, from which heat flux across the element is calculated. A field trial of 22 UK properties during the 2019–2020 winter heating season assessed Heat3D’s heat flux measurements against traditional heat flux plates, with 90% of 295 Heat3D surveys falling within the combined uncertainty of the reference method. A second field trial, conducted over the 2020–2021 winter heating period, evaluated a new timelapse iteration of the method capable of measuring wall U-value within approximately 60 min; 90% of 42 Heat3D surveys fell within the combined confidence interval of the ISO 9869-1-measured U-value. These results indicate that Heat3D offers a viable, rapid alternative to traditional methods for both heat flux and U-value measurement.

1. Introduction

Building fabric performance measurement is central to both the diagnosis and verification of energy retrofit in the existing housing stock. This Introduction establishes the policy context driving demand for improved in situ measurement methods, reviews the current state of U-value measurement and quantitative infrared thermography (QIRT), and identifies the specific limitations of existing approaches that the Heat3D methodology is designed to address.

1.1. Policy

In 2024, the domestic sector was responsible for 25.8% of the total energy use of the UK [1]. On an international scale, buildings account for approximately 40% of the European Union’s final energy consumption, with residential buildings representing around two-thirds of this figure [2]. The European Parliament states that 75% of the European Union’s building stock has poor energy performance [3], and notes that renovating these buildings could reduce the EU’s total energy consumption by 5–6% [4].

In response, the European Parliament adopted a major recast of the Energy Performance of Buildings Directive (EPBD, Directive 2024/1275) in April 2024, which entered into force in May 2024 and must be transposed into national law by May 2026 [5]. The recast sets binding targets to reduce the average primary energy use of EU residential buildings by at least 16% by 2030 and 20–22% by 2035, relative to 2020 levels, and mandates that all new buildings achieve zero-emission status by 2030 [5]. A key instrument introduced under Article 3 is the requirement for each Member State to develop a National Building Renovation Plan (NBRP)—a detailed roadmap setting out policies, targets, and financing to decarbonise the national building stock by 2050 [6].

The UK Government legislated a target of net zero greenhouse gas emissions by 2050 through an amendment to the Climate Change Act 2008, coming into force in June 2019 [7]. However, to achieve such targets, decisive action must be taken to reduce the energy use of both the current and future housing stock [8].

This has been outlined in the Heat and Buildings Strategy, in which the UK Government presents its roadmap to decarbonising not only how buildings are heated, but importantly improving the energy performance of them [9]. The report references the 2019 English Housing Survey [10], in which it was found that only 40% of dwellings achieved an SAP energy efficiency rating (EER) of bands A to C. Although this has significantly increased from 12% in 2009, it highlights the work still required to improve the current housing stock. With regard to retrofit, the Heat and Building Strategy highlights the need for a “fabric first” approach [11], in which the thermal performance of the building’s fabric (walls, ceilings, floors, glazing, etc.) must be improved first before changing the heating system.

Consumer awareness is a further recommendation made by the UK Government in the 2022 “Decarbonising heat in homes” document [12], highlighting the need for a public awareness campaign. This would not only make homeowners aware of changes required to achieve low-carbon housing, but also allow homeowners to make informed decisions on how they use their energy, which can reduce energy consumption by up to 15% [13].

The UK Government’s Warm Homes Plan signals a deliberate policy shift away from a fabric-first approach to domestic retrofit, deprioritising insulation in favour of electric technologies such as heat pumps and solar panels [14]. This carries inherent risk: EPC ratings, the primary metric used to identify homes for upgrade, are known to substantially over-predict energy use in lower-rated properties, by up to 48% in band F and G dwellings, highlighting a fundamental divergence between asset-rated and in-use performance that is characteristic of the energy performance gap [15].

The consequences of inadequate quality assurance in retrofit delivery have been demonstrated starkly by the October 2025 National Audit Office (NAO) investigation into the Energy Company Obligation (ECO4) and Great British Insulation Scheme (GBIS), which found that 98% of homes receiving external wall insulation under the schemes had major installation defects requiring remediation [16]. This finding underscores a critical need for robust, independent measurement of building fabric performance post-retrofit: without verified in situ evidence of whether an intervention has been installed correctly and is performing as intended, neither homeowners, policymakers, nor scheme administrators can confirm that funded works have achieved their intended outcome.

The challenge of verifying as-built fabric performance is not unique to the UK. Internationally, a substantial body of peer-reviewed research has developed and validated in situ U-value measurement methods as alternatives to tabulated assumptions, including work in Italy [17,18,19], Spain [20,21], Portugal [22], Cyprus [23], Croatia [24,25], the Netherlands [26,27] and Canada [28]. Recent international reviews [22,24,29] identify a common set of drivers—ageing housing stock, energy-labelling schemes reliant on assumed U-values, and large-scale retrofit programmes—and reach a common conclusion: measured performance frequently diverges from calculated values, and faster, lower-cost measurement methods are required if in situ verification is to be deployed at scale. The UK policy context described above is therefore one instance of an international problem, and the method presented in this paper is intended to be applicable beyond the UK housing stock on which it was validated.

1.2. Literature Review

1.2.1. Energy Performance Gap

A key step in achieving a fabric-led retrofit is to accurately assess the heat loss of a property. This is usually done using predetermined U-values in energy modelling calculations, typically using the Energy Performance Certificate, which relies on the Reduced Data Standard Assessment Procedure (RdSAP) [30,31]. However, such calculations have been shown to often underestimate the building’s total energy use when compared to the measured performance of a dwelling [32,33,34].

This “energy performance gap” (EPG) applies to both the new-build and retrofit markets, with occupants not always getting the improved performance they expect from their homes. This not only costs the occupant more money due to the property’s energy use but also increases the CO₂ emissions. Fitton defines the EPG using the following equation [35]:

$$EPG={\displaystyle \frac{Actual Consumption-Theoretical Consumption}{Theoretical Consumption}}\cdot 100$$

(1)

Equation (1)—Energy Performance Gap [35].

Both the actual (measured) and theoretical (modelled) consumption figures can have their own “gaps” associated with them, further increasing the EPG of a dwelling. The modelling gap is derived from how different energy modelling systems handle varying physical parameters and also how different users of the systems interpret the results [36].

The measurement gap stems from the fact that for the many physical parameters of a building (heat transfer coefficient (HTC), U-values, air permeability), there are multiple methods describing different ways in which they should not only be measured but also analysed. This was highlighted in a report by Butler and Dengel, in which six teams were tasked with independently measuring the HTC of a standard test house, in which there was variation of 30% across all their results [37].

To tackle this issue, reliable measurements of the buildings’ as-built performance need to replace the current norm of using predetermined U-value estimations. A report by the Zero Carbon Hub focused on the EPG highlights that there is a clear need to accurately and cost effectively measure the performance of buildings [38].

1.2.2. U-Value

The thermal transmittance (U-value) of a building element is defined as “heat flow rate in the steady state divided by area and by the temperature difference between the surrounding on both sides of a flat uniform system” [39]. It can be thought of as an efficiency rating, with lower U-values reducing the amount of heat transfer and, as such, are more energy efficient.

In the UK, it is typically measured through the heat flow meter method as described in ISO 9869-1 [40]. A heat flux plate (HFP) is used to measure the spot-measured heat flux (Q) through the building element. This is divided by the internal–external air temperature difference (T_i and T_e respectively).

$$U\left[\mathrm{W}{\mathrm{m}}^{-2}{\mathrm{K}}^{-1}\right]={\displaystyle \frac{Q\left[\mathrm{W}{\mathrm{m}}^{-2}\right]}{T_i\left[\mathrm{K}\right]-T_e\left[\mathrm{K}\right]}}\equiv {\displaystyle \frac{Q}{\mathsf{\Delta }T}}$$

(2)

Equation (2)—Thermal transmittance (U-value) equation as defined in ISO 9869-1 [40], units defined in square brackets.

The method takes a minimum of 72 h and typically takes around 1 week. Due to the long measurement period, the expertise required, and the cost of the equipment, it is not feasible for an energy assessor to measure every element of a building, which gives rise to the use of the previously identified U-value estimations.

Another issue with the HFP method is that it relies on a point measurement of heat flux. It is known that the U-value can vary across a building element, due to construction details such as thermal bridging [41,42], thermal bypass [43] and construction defects (i.e., gaps in the insulation) [32]. To combat this, multiple HFP measurements are taken across the elements, and an average is produced. This method can be costly if new equipment is required (as shown in

Table 1. Typical equipment costs for a basic HFP U-value measurement using industry standard equipment (2026).

Item	Qty	Cost	Total
Novus Data Logger	1	£1072.00	£1072.00
Heat Flux Plate (HFP01)	3	£418.00	£1254.00
Thermocouple Type T (25 m)	1	£15.11	£15.11
Thermal Paste	1	£6.42	£6.42
TOTAL			£2347.53

) and requires a skilled person to configure the data logger, correctly place sensors for measurement, monitor data throughout the test period (repeatedly re-attending site), remove equipment, and finally analyse the data. Additionally, if the building is occupied, this measurement can be quite disruptive with sensor wires running throughout the building for the duration of the measurement.

A method to overcome the localised heat flux measurement issue presented by ISO 9869-1 is to use the infrared thermography method defined in ISO 9869-2 [44]. This method allows for the heat flux to be measured over a large area of the building element; however, this approach still requires the minimum of 72 h to conduct the measurement, and a high-quality thermal camera, generally costing more than the equipment required for the heat flow meter method. A rapid and inexpensive solution is required if measurement is going to replace the use of the predetermined U-values.

Quantitative Infrared Thermography (QIRT) extends conventional thermographic surveying by extracting physically meaningful values, such as heat flux and U-value, from thermal imagery, rather than providing purely qualitative visualisation of surface temperature distributions [19,45]. The approach was first demonstrated for opaque building elements by Albatici and Tonelli [18], using an external infrared thermovision technique, and by Fokaides and Kalogirou [23], who determined envelope U-values from internal thermographic measurements of surface temperature, ambient temperature, and reflected temperature. The internal-surface formulation, in which the convective and radiative heat exchange at the internal wall surface is reconstructed from thermal imagery, was subsequently developed and validated by Tejedor et al. [46], whose physics model underpins the heat flux calculation used in Heat3D. The standardisation of the infrared method in ISO 9869-2 [44] reflects the growing maturity of the technique; however, the standard’s scope is limited to frame-structure dwellings and retains the multi-day measurement period of the heat flow meter method.

The accuracy of QIRT measurement is governed by a well-documented set of uncertainty sources: the assumed emissivity of the target surface, the estimation of reflected radiation, the accuracy and stability of the infrared detector, the air temperature measurement, and the convective heat transfer model applied at the surface [19,45,47,48]. Operating conditions impose further constraints. Tejedor et al. [20] found that variance in internally measured U-values was predicted principally by changes in outdoor air temperature, and identified an optimal internal–external temperature difference window of approximately 7–16 K; subsequent work on heavy multi-leaf walls demonstrated the additional complications introduced by thermal mass under dynamic boundary conditions [49]. For external QIRT approaches, wind velocity materially affects the surface heat balance and must be accounted for [18,50]. These sensitivities explain why published QIRT studies impose quasi-steady-state requirements on the internal environment, and they motivate the environmental controls and effective external temperature weighting adopted in the present work.

Published validation studies of QIRT-based U-value measurement have generally been conducted on small numbers of walls under laboratory or semi-controlled conditions: Albatici et al. [17] validated the external technique on an experimental building over three years; Tejedor et al. [20,46] validated the internal technique on instrumented case-study walls; Gaši et al. [25] compared infrared and heat flux methods under dynamic conditions; and Mahmoodzadeh et al. [28] applied external thermography to wood-framed assemblies across buildings in Canada. In parallel, automated processing has been developed to produce two-dimensional U-value maps from internal thermography [21], and alternative rapid non-thermographic methods have emerged, including the transient Excitation Pulse Method, which measures wall thermal resistance within approximately two hours [26,27], and the QUB method, which measures whole-building heat loss and U-values overnight [51,52]. What remains largely absent from the literature is large-scale field validation of a QIRT method in occupied dwellings against the ISO 9869-1 reference method, the contribution of the present study.

Taken together, the literature reveals four practical limitations shared by existing QIRT implementations. First, they depend on high-specification research-grade thermal cameras and skilled thermographers, placing them beyond routine surveying budgets [19,47]. Second, they yield measurements of a limited region of the element from single images, rather than a spatially complete, geometrically referenced model of the whole wall. Third, image capture, geometric referencing, and heat flux computation are performed manually and offline, requiring specialist post-processing for every survey. Fourth, published validation datasets are small, typically single buildings or laboratory walls, leaving open the question of performance across the diversity of real, occupied housing [19,24]. Alternative rapid methods address the time constraint but not the rest: the Excitation Pulse Method requires bespoke heating apparatus and remains a point measurement [26,27], while QUB measures both whole-building heat loss but still point-measured element-level U-values [51,52]. Heat3D was developed to address these limitations in combination: it implements the internal QIRT formulation [46] on consumer hardware (a smartphone-attached FLIR One Pro), automates geometric referencing through augmented reality room capture, computes heat flux in-app across the full wall area, and in the present study is evaluated against ISO 9869-1 at a scale not previously reported for a QIRT method: 295 heat flux surveys across 22 properties, and 42 U-value surveys across 14 walls, in UK dwellings.

1.3. Heat3D Overview

Heat3D is an iOS application which allows the user to rapidly measure the U-value of an external wall through QIRT. The user first creates a model of the room within the application using Apple’s augmented reality (AR) software (ARKit 3 & 4) (tolerance of ±2 cm over 4 m (0.5%) at 2 m from the wall), and then scans the IR image of the entire wall onto the AR model using the FLIR One Pro attachment [53]. The steps of conducting a survey is demonstrated in Figure 1.

An air temperature target and reflective temperature target are placed at a fixed distance (100 mm) from the wall. These temperatures, along with the surface temperature of the heat loss element, can be used to calculate the heat flux for each pixel using the physics model based upon Tejedor’s paper [46].

Heat3D reports heat flux averaged over 50 cm × 50 cm regions across the wall, with an underlying IR spatial resolution of approximately ±5–10 cm at 2.5 m. This gives a better measure of the heat flux across the full wall surface, including areas of thermal bridging, which may be missed with the point measurement of HFPs. The user can exclude areas from the measurement, such as windows, radiators or service voids, by drawing a rectangle over the area to omit, which will not be used in the heat flux calculation.

In the first iteration of the application’s development, the application would use a single scan of the wall and calculate its thermal characteristics. This yielded a good agreement with the traditional HFP method in terms of measured heat flux; however, it was found that an accurate U-value measurement could not be achieved from a single Heat3D thermographic survey.

Further development of the application identified that the cause of the poor U-value agreement was typically attributed to periods of greater and lower flux due to a heating system cycling, and thermal lag effects when measuring the external temperature. To rectify these issues, the following changes to the application were made:

Instead of an instantaneous survey, the user would scan the full wall as before but then define an area on the wall to record a timelapse and leave the iOS device set up on a tripod for an hour, recording a new IR image every minute.

An effective external temperature is now calculated and used within the application. This takes local historic weather data from an API and weights this against the current temperature.

The room in which the survey is being conducted must now be isolated and heated using a portable convective heater attached to a proportional–integral–derivative (PID) thermostatic controller for approximately 12 h prior to the Heat3D timelapse survey. The PID thermostatic controller vastly reduces the level of hysteresis, resulting in a more stable internal temperature. This creates a quasi-steady-state environment, reducing cycling effects caused by typical domestic heating systems.

The above changes were used throughout the 2020–2021 field trial (FT2) and can be seen to have yielded good U-value agreement between Heat3D and the ISO 9869-1 HFP method.

2. FT1: 2019–20 Winter Field Trial: Heat Flux Measurement Validation

2.1. Methodology

2.1.1. Properties/Test Cases

Following on from a 3-week testing campaign conducted at the University of Salford Energy House in September 2019 [54], the application was tested in a field trial during the 2019–20 winter heating period. This first field trial consisted of 295 Heat3D surveys across 22 different properties throughout England and Wales, with ISO 9869-1-measured U-values ranging from 0.13 Wm⁻²K⁻¹ (new-build Passivhaus flat) to over 2 Wm⁻²K⁻¹ (18th century Welsh farmhouse with 2 m thick solid stone walls).

The aim of this field trial was to validate that Heat3D can accurately measure the heat flux through a variety of different wall constructions when compared to a traditional heat flux plate.

Table 2. An overview of the different properties surveyed in the Heat3D 2019–2020 winter field trials.

Ref	Type	Detachment	Wall Age	Orientation	Wall Type	Insulation
FT1-1	House	Detached	2012 onwards	NW	Cavity	Filled
FT1-2a	House	End-Terrace	1983–1990	NE	Cavity	Filled
FT1-2b	House	End-Terrace	1983–1990	NE	Cavity	Filled
FT1-2c	House	End-Terrace	1983–1990	NE	Cavity	Filled
FT1-3	House	Detached	1991–1995	E	Cavity	Filled
FT1-4	House	Detached	1983–1990	NW	Cavity	Unfilled
FT1-5	House	Detached	2007–2011	NW	Cavity	Filled
FT1-6	Other	Enclosed End-Terrace	before 1900	N	Solid Brick	None
FT1-7	House	Mid-Terrace	before 1900	E	Solid Brick	External
FT1-8	House	Mid-Terrace	before 1900	N	Solid Brick	None
FT1-9	House	Detached	2012 onwards	NE	Cavity	Filled
FT1-10	House	Detached	before 1900	W	Stone (sandstone)	None
FT1-11a	House	Detached	2012 onwards	N	System Built	Filled
FT1-11b	House	Detached	2012 onwards	N	System Built	Filled
FT1-12	Other	Mid-Terrace	1950–1966	NW	Cavity	None
FT1-13	Bungalow	Detached	1967–1975	E	Cavity	Filled
FT1-14	House	Semi-Detached	2012 onwards	SW	Cavity	Filled
FT1-15	House	Detached	1983–1990	NW	Cavity	Filled
FT1-16	Bungalow	Detached	1976–1982	E	Cavity	Filled
FT1-17	Other	Enclosed End-Terrace	1996–2002	S	Cavity	Filled
FT1-18	Flat	Mid-Terrace	1930–1949	NE	Solid Brick	None
FT1-19	House	End-Terrace	2012 onwards	NE	Cavity	Filled
FT1-20	Flat	Mid-Terrace	2012 onwards	N	Timber Frame	Filled
FT1-21	Flat	Mid-Terrace	2012 onwards	E	Timber Frame	Filled
FT1-22	Other	Detached	2012 onwards	N	Timber Frame	Internal

details the different wall constructions measured throughout the field trial.

The 22 properties surveyed span a range of wall constructions broadly representative of the UK owner-occupied housing stock, with cavity (filled) walls accounting for the largest share (55%), consistent with national stock data from the English Housing Survey [55]. Solid wall and stone properties (18% combined) reflect the pre-1919 hard-to-treat stock that is a primary target for retrofit intervention.

2.1.2. Test Procedure

The equipment used at each field trial location consisted of three heat flux plates, an internal air temperature type T thermocouple and an external air temperature type T thermocouple with solar shield. Prior to deployment in the field, the type T thermocouple was calibrated to an accuracy of ±0.1 K in a silicon oil micro-bath against a calibrated (ITS-90) SPRT temperature probe at three set points. The heat flux plates were shipped calibrated, and this was validated following an in-house process using a Fox 314 heat flux meter and a material sample of known thermal resistance.

A notable development in the equipment used in conducting a Heat3D survey was the incorporation of a BlueMaestro Bluetooth temperature sensor [56] into the surface of the air temperature target (Figure 2), with an accuracy of ±0.2 K [57]. This allowed the application to not only have a more accurate air temperature measurement, but it meant that all other IR temperature measurements could be calibrated against it by calculating an offset between the IR air temperature value and the Bluetooth temperature value. This negated the need for the constant temperature box used in a previous three-week testing campaign, consisting of 200+ surveys, conducted at the University of Salford Energy House [54].

Throughout the field trials, the general layout of the equipment remained the same (as shown in Figure 3). A 1 m² area was left exposed on the wall for Heat3D imaging, with the heat flux plates and required temperature targets placed around it. In both heat flux surveys and timelapse U-value measurements, the measurement area was positioned in a central, visually representative section of the wall, following the placement guidance of ISO 9869-1: at least 0.5 m from wall edges, corners, floor, ceiling, and any window or door reveal, and away from areas of visible surface anomaly. This is consistent with the requirement to avoid localised flux concentrations from thermal bridging or construction defects, which would not be representative of the bulk wall performance. For the heat flux surveys (FT1), the Heat3D measurement was taken from a full-wall scan of the exposed 1 m² area; for the timelapse U-value surveys (FT2), the timelapse area was defined on the same representative central zone. The field of view of the FLIR One Pro at typical survey distances means the timelapse area does not capture the full wall, and is not intended to: it captures a representative sub-area (1 m²) from which the mean U-value is derived. Radiative heat sources within the field of view, including visible radiators, lamps, or secondary glazing, were excluded either automatically by the application’s temperature-threshold filter or by manual definition of exclusion zones within the AR model prior to survey completion. Where the wall height placed areas of the survey outside the effective working range of the FLIR One Pro, those areas were excluded from the timelapse region.

Each property would have a minimum of three Heat3D surveys at the start of the seven-day test period and three at the end, with more surveys throughout the week if access to the building was available.

The Heat3D survey procedure was largely the same as that used for the Energy House testing; however, two equipment modifications were made:

• Foil-lined air temperature targets were found to improve the air temperature measurement through reducing the effect of radiative cooling.
• Thermal imaging calibration by a Bluetooth thermocouple located within the air temperature target meant that a temperature correction offset could be applied to the full thermal image. This offset was calculated by taking the difference between the Bluetooth thermocouple and FLIR One Pro air target measurement. This offset in temperature measured by the IR camera can stem from camera settings, internal camera mechanics, or environmental factors.

Mid-way through the heating period, a new development in the application required the user to conduct full room surveys, whereby after imaging the target wall, the user would stand with their back to said wall, and image the rest of the room. Doing so allowed the application to calculate a $T_{Ref}$ for each point on the target wall through raytracing, instead of depending on a single reflective target.

To safeguard the objectivity of the validation, the reference ISO 9869-1 measurements, data curation, and the statistical agreement analyses reported in Section 2.2 and Section 3.2 were led by the University of Salford research team, whose members hold no financial interest in the Heat3D product. Heat3D survey outputs were generated by the application without post hoc adjustment, and the two datasets were processed independently before comparison. In addition, the measurement chain underlying Heat3D, including sensor traceability, target design, camera stability, and the heat flux model, was subject to an independent external metrology review by the National Physical Laboratory (NPL) under the Measurement for Recovery programme, prior to the second field trial.

2.2. Results

It can be seen from Figure 4 that there was a strong agreement between the measured heat flux using the heat flux plates and the heat flux measured using Heat3D, with a slope of 0.87. Despite the strong agreement, the measured Heat3D heat flux was on average 1.9 Wm⁻² lower than the heat flux plate measurement across all the field trial properties, with a standard deviation of 3.3 Wm⁻².

To supplement the correlation analysis, the Bland–Altman method [58] was applied to formally assess agreement between Heat3D Q and HFP measurements. Unlike correlation, which quantifies association, the Bland–Altman approach evaluates whether two methods agree sufficiently to be used interchangeably [59]. Consequently, for each survey pair, the differences between the two methods were plotted against their mean value. The Shapiro–Wilk test was also used to indicate whether the differences’ distribution deviate from normality [60]. The test indicated that they do deviate, which was attributed to the methodological variability inherent in FT1, in which measurement protocols were still being refined, and no room preheating was applied, resulting in a greater spread of systematic error across surveys. Due to the abovementioned deviation, the limits of agreement in the Bland–Altman plot (Figure 5) were determined using the 2.5th and 97.5th percentiles of differences [61,62] rather than the mean difference ± 1.96 × the Standard Deviation (SD), which would be used to specify the upper and lower limits for the case where the Shapiro–Wilk test would not indicate deviation from normality. Note that the 1.96 is the z-critical value for the two-tailed 90% confidence interval.

Figure 5 indicates that the Heat3D (heat flux (Q)) method underestimates the measured values relative to the HFP method; the Difference (y-axis) is defined as the HFP subtracted from the Heat3D-measured values. This observed bias could be attributed to the previously described measurement conditions, where the minimum measurement protocols were still under refinement, as well as the absence of room preheating in FT1. Note that the upper and lower limits of agreement are relatively narrow, indicating a constrained level of variability between the two methods. Although the results are limited, they indicate that Heat3D has the potential to replace the existing method, while retaining the benefits associated with the Heat3D approach as described in the Introduction.

3. FT2: 2020–2021 Winter Field Trial: U-Value Measurement Validation

3.1. Methodology

3.1.1. Properties/Test Cases

A second field trial was conducted the following year, consisting of 42 surveys across 14 walls spread throughout England and Wales (

Table 3. An overview of the different properties surveyed in the 2020–2021 Heat3D Field Trial.

Ref	Type	Detachment	Wall Age	Orientation	Wall Type	Insulation
FT2-1	Maisonette	Enclosed Mid-Terrace	1950–1966		Cavity	Unfilled
FT2-2	House	End-Terrace	1983–1990	NE	Cavity	Filled
FT2-3	House	Detached	1991–1995	E	Cavity	Filled
FT2-4	House	Detached	1983–1990	NW	Cavity	Unfilled
FT2-4b	House	Detached	1983–1990	NW	Cavity	Filled
FT2-5	Other	Mid-Terrace	1950–1966	NW	Cavity	Unfilled
FT2-6	House	Detached	1983–1990	NW	System Built	EWI
FT2-7	House	End-Terrace	2007–2011	W	Cavity	Filled
FT2-8	Bungalow	Semi-Detached	1950–1966	S	Cavity	Filled
FT2-9	House	Detached	2012 onwards	N	Cavity	Filled
FT2-10	House	Detached	2012 onwards	NE	Cavity	Filled
FT2-11	Flat	Mid-Terrace	1930–1949	SE	Solid Brick	None
FT2-12	House	Semi-Detached	1930–1949	E	Solid Brick	EWI
FT2-13	Maisonette	End-Terrace	1983–1990	NW	Cavity	Filled

). Again, there was a diverse range of wall constructions, with ISO 9869-1-measured U-values ranging from 0.15 Wm⁻²K⁻¹ to 1.44 Wm⁻²K⁻¹.

The 13 properties surveyed span a range of wall constructions broadly representative of the UK owner-occupied housing stock [55]. Cavity walls account for the largest share (85%), with solid wall properties (15%) reflecting the pre-1919 hard-to-treat stock that is a primary target for retrofit intervention. Notably, three properties (FT2-1, FT2-4, FT2-5) have unfilled cavity walls, offering a direct comparison with their filled counterparts and providing an opportunity to assess the sensitivity of Heat3D measurements to cavity fill status. One system-built property with external wall insulation (FT2-6) extends coverage to non-traditional construction types increasingly common in the post-1980 stock.

3.1.2. Application Developments

As with the previous field trial, the focus of the application’s development evolved from accurately measuring the heat flux specifically, to accurately measuring the U-value of a wall. Although Heat3D was found to measure the instantaneous heat flux of a wall with good agreement, it was not possible to obtain the U-value with just one measurement. To enable Heat3D to make a U-value measurement, the ”timelapse” and ”Effective External Temperature” (TExtEff) features were developed.

In timelapse mode, the user conducts the initial survey in the same manner as before; however, once complete, a timelapse area is defined on the wall, and the iOS device is set on a tripod (Figure 6). The application will then take an IR image every minute for a defined length of time (typically set to 1 h). This allows the application to create an average of the heat flux, and therefore, reduces how sensitive the application is to internal environmental change.

TExtEff considers the thermal lag of the wall to external temperature variations and the historic values for the external temperature. By inputting a time constant (τ) based on the wall’s construction and thermal capacity, the application creates a weighted average of the historic external temperature, using a normal distribution curve with sigma equal to half the time constant (τ) (Figure 7).

Such a weighting reflects the fact that the most recent external temperature fluctuations have little impact if the wall has a long external time constant. This was found to have significant improvement for solid and empty brick cavity walls but less so on well insulated walls.

3.1.3. Test Procedure

Unlike the previous field trial, the room was now required to be preheated to a steady temperature for a minimum of 12 h prior to the survey, consistent with the thermal stabilisation requirements of ISO 9869-2 [44]. This was achieved either through the use of PID temperature controllers and convective panel heaters (set to achieve an air temperature of 21 °C) or through the room’s existing heating system. If the latter was used to preheat the room, the PID controller setpoint would be set to equal that of the current measured temperature of the room when the surveyor arrived (typically 18–21 °C) and the heating source switched from the existing heating system to the convective panel heater for the duration of the survey. This reduced dynamic effects by creating a quasi-steady-state environment to avoid continued charging/discharging of the wall. The same measurement area was applied, as described in Section 2.1.2.

After completing the initial Heat3D survey, the iOS device was set up on a tripod, with a timelapse area defined upon the wall/AR model. The user must then tap the record button and exit the room. A minimum of three timelapse surveys were conducted at each field trial property.

3.1.4. Measurement Uncertainty

The combined uncertainty intervals used throughout this paper comprise the uncertainty of the reference ISO 9869-1 measurement and that of the Heat3D measurement. ISO 9869-1 itself attributes a total measurement uncertainty of 14–28% to the heat flow meter method under typical operating conditions [28]. For Heat3D, the principal uncertainty components were quantified through laboratory characterisation of the measurement chain, subsequently examined in an independent metrology review by the National Physical Laboratory, and are summarised in

Table 4. Uncertainties associated with the Heat3D U-value measurement.

Component	Value	Basis/Traceability
Bluetooth air temperature sensor (Disc Mini Temp, Blue Maestro)	±0.2 K typical, ±0.4 K max; 0.1 K resolution; drift < 0.01 K/yr	Manufacturer specification [57]; calibration service available
FLIR One Pro absolute radiometric accuracy	±3 K or ±5%	Manufacturer specification [53]; largely eliminated by in-frame differential use
Camera temporal stability (within FFC cycle)	±0.1 K	Laboratory characterisation; survey protocol requires completion within one flat-field-correction calibration cycle to prevent drift
Camera spatial field uniformity	±0.2 K	Laboratory characterisation
Wall/air-target emissivity	0.9 ± 0.05; no correction below 0.8	Measured by comparison with known-emissivity reference
Reflective target temperature	<±0.2 K	Laboratory testing freezer differential method; independent metrology review
Internal–external air temperature difference	±1.5 K	Estimated from combined sensor uncertainties; dominated by the effective external temperature term
Propagated U-value uncertainty at ΔT ≥ 10 K	≈±15%	Propagated from the dominant ΔT term

. Because heat flux is derived from temperature differences within a single thermal frame, the absolute radiometric accuracy of the FLIR One Pro (±3 K or ±5%) is largely eliminated by the in-frame calibration against the Bluetooth air temperature reference; the governing camera contributions to uncertainty are instead its temporal stability (±0.1 K within a flat-field-correction cycle) and spatial field uniformity (±0.2 K), with each survey captured within a single calibration cycle to prevent drift. Emissivity of the wall surface and air temperature target is taken as 0.9 ± 0.05, with a first-order correction applied where the wall emissivity is measured to differ; no correction is applied below an emissivity of 0.8, which bounds the method’s applicability. The dominant term in the U-value uncertainty is the internal–external air temperature difference, estimated at ±1.5 K, reflecting the greater uncertainty of the effective external temperature relative to the internally measured air temperature. For this reason, the survey procedure specifies a minimum internal–external temperature difference of 10 K, limiting the resulting U-value uncertainty to approximately ±15%—within the published 7–16 K optimal operating window for internal QIRT [20] and comparable to the 14–28% uncertainty of the ISO 9869-1 reference method itself. Timelapse averaging over 60 one-minute frames further suppresses random error in the heat flux by approximately the square root of the number of independent frames; however, environmental fluctuations within the hour are correlated, and the effective reduction is smaller; the residual scatter is captured empirically in the standard deviation of repeat surveys reported in Section 3.2.

3.2. Results

Figure 8 shows a strong agreement between the ISO 9869-1-measured U-value and the Heat3D timelapse method U-value, with a slope of 0.91. The mean difference between Heat3D U-value and ISO 9869-1 U-value was 0.00 Wm⁻²K⁻¹, with a standard deviation of 0.09 Wm⁻²K⁻¹.

The Bland–Altman method was also applied to the U-value comparison. In contrast to FT1, the Shapiro–Wilk test did not provide evidence against normality in the U-value differences, consistent with the more controlled and repeatable FT2 protocol, in which preheating and timelapse averaging substantially reduced systematic variability, yielding a mean difference of 0.00 Wm⁻²K⁻¹. The limits of agreement in Figure 9 were therefore calculated as the mean difference ± 1.96 × SD.

Figure 9 indicates good agreement between the Heat3D (U-value (U)) and the measurements obtained using the HFP method. In addition, the upper and lower limits of agreement are narrow, suggesting that the differences between the two methods are confined within a limited range.

Taken together, the Bland–Altman analyses for both field trials indicate that Heat3D offers a viable alternative to the HFP method for both heat flux and U-value measurements.

Figure 10 shows this agreement broken down by the different wall constructions. Heat3D showed a very strong agreement across all wall types. However, the statistical limitations of the FT2 dataset warrant explicit acknowledgement. The 42 surveys span 14 walls, of which filled cavity constructions account for the large majority; sample sizes for unfilled cavity, solid brick, system-built and externally insulated walls are small, and no timber frame or stone walls were included in FT2. The reported 90% agreement should therefore be read as strong evidence for the cavity-walled dwellings that dominate the UK owner-occupied stock, and as preliminary evidence for other constructions, where per-type confidence intervals are necessarily wide. Regarding highly insulated constructions, the divergence observed below 0.2 Wm⁻²K⁻¹ (Section 4.2) already bounds the method’s applicability for Passivhaus-standard and Future Homes Standard walls at typical UK winter temperature differences. Regarding transferability beyond the UK, three factors require validation in other contexts: construction types and surface finishes with emissivities outside the 0.8–1.0 range for which the current correction applies; climates in which the minimum 10 K internal–external temperature difference is infrequently available; and external temperature dynamics different from those against which the effective external temperature weighting was developed. Extension of the validation dataset across these dimensions is identified as a priority for further work, aligned with the international validation objectives of IEA EBC Annex 94 [63].

Table 5. FT2 (2020–2021) U-value measurement—breakdown by wall type.

Wall Type	n (Surveys)	Mean Diff. (Wm−2K−1)	SD (Wm−2K−1)	Min/Max (Wm−2K−1)	Within Combined Uncertainty (n)	% Within Combined CI
Cavity (filled)	23	−0.01	0.10	−0.25/0.22	19	83%
Cavity (unfilled)	8	0.03	0.08	−0.12/0.11	8	100%
System built (EWI)	5	−0.05	0.05	−0.09/0.02	5	100%
Solid brick (none)	3	−0.03	0.02	−0.04/−0.01	3	100%
Solid brick (EWI)	3	0.13	0.004	0.12/0.13	3	100%
TOTAL	42	0.00	0.09	−0.25/0.22	38	90%

presents the FT2 U-value agreement broken down by wall construction type, supplementing Figure 10. Per-type sample sizes are small for all types other than cavity (filled) walls, and results for individual construction categories should be interpreted accordingly; the overall 90% agreement figure is driven by the cavity-walled majority. These limitations are discussed further in Section 4.2.

4. Discussion

4.1. FT1—Heat Flux Validation

The above results provide strong evidence that Heat3D can accurately measure the heat flux across a variety of wall constructions, with 90% of the 295 measurements being within the combined uncertainty interval.

It was found that in both methods, the measured heat flux was highly susceptible to climatic changes in the internal environment, such as someone entering a room to conduct a survey. This meant that the instantaneous measured heat flux during a survey was usually elevated above the average than that would be observed throughout a typical ISO 9869-1 measurement.

4.2. FT2—U-Value Validation

The 2020–2021 field trial yielded good results for Heat3D U-value measurement through the new timelapse methodology, with 90% of the 42 surveys being within the combined level of uncertainty of the traditional ISO 9869-1 U-value measurement across all sampled wall constructions.

It can be seen in Figure 8 that there is very strong agreement for U-values above 0.2 Wm⁻²K⁻¹; however, results begin to diverge below this. This may indicate that Heat3D has a minimum heat flux resolution of ±2 Wm⁻² as a result of the FLIR One Pro hardware. Figure 11 shows the calculated heat flux for varying U-values for a range of temperature differences, with any heat flux below the minimum resolution of Heat3D highlighted in red.

For a typical UK winter ∆T of 15 K, Heat3D is able to measure U-values greater than 0.2 Wm⁻²K⁻¹, compared to 2.2 Wm⁻²K⁻¹ for a basic data logger and HFP measurement kit. This therefore makes the application suitable to measure the U-value of walls found in most new-build and retrofit dwellings. However, it would likely struggle to measure U-values found in Passivhaus (≤0.15 Wm⁻²K⁻¹) [64] or Future Homes Standard-compliant construction (U-value ≤ 0.18 Wm⁻²K⁻¹) [65], requiring a ∆T of at least 20 K.

Taken together, these constraints define the practical applicability range of the current Heat3D implementation: walls of U-value ≥ 0.2 Wm⁻²K⁻¹ with a minimum internal–external temperature difference of 10 K (rising to ≥0.3 Wm⁻²K⁻¹ at a shoulder-season ΔT of 9 K) and surfaces of emissivity ≥ 0.8 for which the current correction applies.

Measurements are theoretically possible in warmer climates during periods when the external temperature is significantly higher than the internal temperature, provided the same absolute internal–external temperature difference (∆T > 10 K) is maintained. However, this specific operational configuration was not tested within this work. This limitation is likely to affect a small but growing proportion of the stock, primarily dwellings built to post-2012 Part L standards or Passivhaus specification, estimated at fewer than 5% of the current English housing stock [65]. This is expected to become more significant as retrofit and new-build standards tighten toward the Future Homes Standard.

The characterisation of Heat3D as rapid warrants clarification. The full measurement protocol comprises approximately 12 h of unattended room preheating followed by roughly 1–1.5 h of attended survey time, including the 60 min timelapse acquisition. The appropriate comparison with ISO 9869-1 is therefore not elapsed time alone but operator time, equipment commitment, and scheduling flexibility. The heat flow meter method requires equipment to be installed, left in situ, and recovered across a minimum 72 h (typically 7-day) monitoring period, occupying a set of instrumentation per wall for the duration of the process and requiring repeated site visits; the preheating phase of a Heat3D survey, by contrast, is passive, requires no equipment on site, and can run overnight ahead of a scheduled visit. A single operator with one equipment set can consequently survey multiple properties per day, which the HFP method cannot match at any equipment budget. The workflow is therefore most advantageous where visits can be scheduled and heating controlled in advance: vacant properties, retrofit quality-assurance inspections, new-build commissioning, and social-housing stock surveys. It is least advantageous in ad hoc surveys of occupied homes, where the surveyor depends on the occupant to have operated the heating correctly, and where a failed preheat, through heating system failure or occupant error, necessitates a revisit, as acknowledged in the Conclusion. Where measurement periods of a week and sensor setup in the property are acceptable, the ISO 9869-1 method remains appropriate; Heat3D’s contribution is to make element-level measurement viable in the set of circumstances where they are not.

5. Conclusions

The two field trials provide strong evidence that Heat3D can rapidly measure both the heat flux and U-value of an external wall, with agreement comparable to the traditional ISO 9869-1 heat flux plate method.

The first field trial showed good agreement between Heat3D and calibrated heat flux plate measurements, with 90% of the 295 measurements being within the combined uncertainty. However, it highlighted the fact that instantaneous measured heat flux from either method was too susceptible to climatic changes in the internal environment, and as such was unsuitable for U-value measurement.

The second field trial tested the new timelapse and TExtEff features, both developed to allow the application to accurately measure the U-value of an external wall through a rapid new method. This, when compared to the traditional ISO 9869-1 heat flux plate method, yielded positive results, with 90% of the 42 surveys being within the combined uncertainty. However, it was shown that results begin to diverge for walls with U-values less than 0.2 Wm⁻²K⁻¹, due to limitations in the FLIR One Pro hardware (minimum resolution ±0.1 K). It is worth noting, however, that many IR cameras that are commercially available have a similar resolution, and as such would limit the application in the same way.

Although the application yields good results, its dependence on the room having been heated constantly for up to 12 h prior to the survey leaves the test process vulnerable to potential failure outside of the application’s control. Heating system failure, or even the surveyor’s dependence on either a homeowner or site manager to have remembered to heat the room, could result in a failed survey and require additional tests to take place during a subsequent visit.

A further limitation is the requirement of historic external temperature data via a weather API for the TExtEff calculation, particularly in sparsely monitored areas. This can be resolved using a separate external temperature sensor, and the user can enter this value at the start of a survey. The separate sensor would need to be installed at least 6 h in advance of the survey, and the TExtEff calculation applied, which places a higher dependence on user competence. Sensor placement may also skew the results.

However, the application does offer many potential benefits. The rapid U-value evaluation would allow surveyors to measure multiple properties in a single day, making quality assurance easier to schedule around retrofit and site works. It also has the potential to reduce cost to the surveyor, as expensive equipment is not required to be left in situ for prolonged periods for a single measurement, as is the case for the ISO 9869-1 method. In addition, the Heat3D system provides a measurement over the whole area of the building element (wall) compared to HFP point measurements and, therefore, is more likely to catch inhomogeneous U-value discrepancies. Further, the 3D thermographic model of a building can highlight snagging issues in both new-build and the retrofit markets, but also increase the homeowners’ awareness of where their home loses heat, which is highlighted in the UK Government’s “decarbonising heat in homes” [12]. Such increased awareness and understanding has been shown to reduce homeowner energy consumption by up to 15% [13].

Further research and development of this application should relate to the following:

Development of the application so that other building elements (floors, roofs, etc.) can be measured. This could potentially lead to a whole room heat loss calculation, which could be used to accurately size heating systems for a property. Additionally, improvements to the heat flux resolution so that building elements of lower U-values can be measured. This will likely be through hardware changes or through allowing thermal images from better-resolution IR cameras to be used within the application’s algorithm.

The growing demand for scalable building performance diagnostics, driven by large-scale retrofit programmes such as the Warm Homes Plan, has accelerated interest in rapid in situ measurement methods. Whilst such methods offer clear practical advantages, ensuring they are validated and appropriately applied across their intended use cases will be critical to closing the performance gap at scale. Further, the ability to conduct such measurements outside of the heating season, and in cooling-dominated climates, is critical to ensuring our building stock performs as expected and reduces our reliance on assumptions. These are objectives being pursued internationally through IEA EBC Annex 94 Sub-Task 4 [63].

6. Patents

S.B. and co-inventors hold US Patent No. 12,135,242 B2 (“Thermal Quality Mappings”), assigned to Electric Pocket Limited, filed 6 September 2021 and granted 5 November 2024, covering the Heat3D method described in this work. The reference measurements, data curation, and statistical analyses were led by the university-based authors independently of the commercial partners, as described in Section 2.1.2.