Field test method for thermal performance of prefabricated external wall of near zero energy building

Abstract

This invention belongs to the technical field of evaluating thermal defects in building envelopes using heat conduction measurement, specifically involving a field testing method for the thermal insulation performance of prefabricated near-zero energy building exterior walls. The method includes: acquiring environmental data sequences from a micro-weather station and verifying constraints related to precipitation, wind speed, and sunshine duration; acquiring the target seam detection area and determining a seamless benchmark comparison area; acquiring multiple frames of infrared thermal images and performing inter-frame affine transformation alignment to obtain aligned temperature sequences for the two areas; obtaining a relative temperature difference sequence based on the difference between the two temperature sequences; identifying the heat storage and dissipation characteristics of the relative temperature difference sequence and removing areas that meet these characteristics; calculating the variance and mean of the relative temperature difference sequence for the unremoved areas, and determining them as true through-type thermal bridge defects based on preset variance and mean thresholds; acquiring the defect pixel coordinates and converting them to building physical coordinates, matching prefabricated node attributes, and generating repair instructions. This invention continuously tracks the evolution of the relative temperature difference between the seam and the intact insulation area, separating natural heat dissipation from continuous heat leakage via thermal bridges, and inversely mapping the digital detection coordinates to the physical wall surface to guide repair.

Description

Technical Field

[0001] This invention relates to the field of assessing thermal defects in building envelopes using heat conduction measurements. More specifically, this invention relates to a field testing method for the thermal insulation performance of exterior walls in prefabricated near-zero energy buildings. Background Technology

[0002] Prefabricated near-zero energy buildings reduce heating energy consumption to extremely low levels through highly insulated building envelopes and airtight designs. Their exterior walls typically utilize precast concrete sandwich insulated wall panel systems, with insulation layers exceeding 200mm in thickness. The wall panels are connected by vertical and horizontal joints, filled with polymer sealant and sealed with airtight tape to prevent air exchange between indoors and outdoors. In frigid regions, where the temperature difference between indoors and outdoors often exceeds 35°C in winter, even minor through-hole defects at the joints can create continuous heat leakage channels, leading to localized condensation, decreased insulation performance, and increased energy consumption. Therefore, accurate on-site detection of thermal defects at the joints is crucial for ensuring the actual operational performance of near-zero energy buildings.

[0003] Infrared thermography is a common non-destructive testing method for thermal defects in building envelopes. Its principle involves recording the surface temperature distribution of the envelope using an infrared thermal imager and identifying thermal bridge locations by utilizing the temperature difference between defective and intact areas. However, in the practical testing of near-zero energy buildings in frigid regions, this method faces fundamental challenges. The specific heat capacity of the polymer sealant filling the prefabricated joints is 1.5 kJ / (kg·K), while the specific heat capacity of the precast concrete on both sides is 0.88 kJ / (kg·K), a difference of approximately 70%. During the daytime solar radiation heating phase, the sealant absorbs and stores significantly more heat than the concrete; after sunset, as the temperature cools, the sealant maintains a higher surface temperature due to its greater initial heat storage, forming a clear bright line along the joint on the infrared thermal image. This bright line is essentially a surface heat dissipation driven by the difference in material thermal properties, not a thermal bridge caused by indoor air leakage. However, in the existing infrared thermal imaging detection process, operators mainly rely on the temperature difference amplitude of single-frame or short-time infrared thermal images for interpretation, which makes it difficult to distinguish between surface heat storage artifacts and real through-type thermal bridge defects, resulting in both high false alarm rates and high false alarm rates.

[0004] Furthermore, existing infrared detection methods rely on operator visual judgment when selecting benchmark comparison areas. Inconsistencies in geometric dimensions and unequal amounts of local solar radiation often exist between the selected intact insulation area and the inspected seam area, introducing systematic bias into temperature difference analysis from the outset. During long-term monitoring lasting several hours, micro-vibrations of the thermal imager support platform and environmental wind disturbances can cause spatial drift between image frames, leading to incorrect physical locations for pixel-level temperature curves. In the extreme cold conditions at the end of a cooling period, the temperature difference between intact and defective seams is typically in the range of 0.3K to 1.0K, approaching the noise-equivalent temperature difference level of conventional infrared thermal imagers, placing high demands on the reliability and reproducibility of the judgment logic.

[0005] Chinese patent application CN121213562A discloses a method and system for detecting hollow areas in building exterior walls based on single-frame infrared thermal imaging and heat flow assessment. This method generates a virtual heat flow image sequence by combining a single-frame infrared thermal image with the heat conduction control equation, and then identifies hollow areas through spatiotemporal frequency domain transformation and cylindrical harmonic function inversion. It focuses on extracting heat flow information from single-frame images through mathematical and physical inversion, and is suitable for detecting internal interlayer defects such as hollow areas. However, it does not address the temporal separation of surface heat storage and through-thermal bridges during cooling. Chinese patent application CN116087270A discloses a method and electronic device for identifying defects in building exterior walls. It determines the exterior wall area through affine transformation matching of visible light and infrared images, and then identifies the defect area and defect type based on temperature data. This achieves registration of visible light and infrared images and temperature difference determination. However, it does not establish a reference comparison area selection mechanism with equivalent size and radiation, nor does it distinguish between heat dissipation and steady-state heat flow leakage in long-term monitoring.

[0006] In the on-site infrared detection of thermal defects in the exterior wall joints of prefabricated near-zero energy buildings in frigid regions, the following technical challenges currently exist: how to establish a detection benchmark with equivalent dimensions and radiation; how to maintain spatial alignment accuracy during long-term monitoring; how to distinguish between the heat storage and dissipation process caused by differences in material specific heat capacity and the steady-state thermal bridge temperature difference caused by indoor heat flow leakage at the physical level; and how to convert the defect pixel coordinates in the infrared image into physical construction coordinates that can guide on-site repair. Summary of the Invention

[0007] To address the technical challenges of on-site detection of thermal defects in the exterior wall joints of prefabricated near-zero energy buildings in frigid regions, including establishing a detection benchmark equivalent to dimensional and radiation conditions, maintaining spatial alignment during long-term monitoring, distinguishing between heat dissipation caused by differences in material specific heat capacity and steady-state thermal bridge temperature differences caused by indoor heat flow leakage, and converting infrared detection coordinates into physical construction coordinates that can guide repair, this invention provides an on-site testing method for the thermal insulation performance of the exterior walls of prefabricated near-zero energy buildings. The method includes: acquiring environmental data sequences from a micro-weather station and verifying that the environmental data sequences meet preset constraints on precipitation, wind speed, and sunshine duration; acquiring the target joint detection area and determining a seamless benchmark comparison area around the target joint detection area based on the principle of a solid mask sliding window and equivalent radiation; and acquiring multiple frames of infrared thermal images after sunset, using the first frame as the basis for comparison. Based on the benchmark, inter-frame affine transformation alignment is performed using physical component feature point matching to obtain the aligned temperature sequence of the target seam detection area and the temperature sequence of the seamless benchmark comparison area. The difference between the temperature sequence of the target seam detection area and the temperature sequence of the seamless benchmark comparison area is calculated to obtain the relative temperature difference sequence. In the initial cooling window, the heat storage and dissipation characteristics of the relative temperature difference sequence are identified, and areas that meet the heat storage and dissipation characteristics are marked and removed from the thermal bridge alarm candidate list. In the quasi-steady-state window at the end of the observation period, the variance and mean of the relative temperature difference sequence are calculated for the areas that have not been removed. When the variance is less than the preset variance threshold and the mean is greater than the preset mean threshold, it is determined to be a real through-type thermal bridge defect. The pixel coordinates of the defect in the infrared thermogram are obtained and converted into building physical coordinates through homography mapping. Based on the building physical coordinates, the prefabricated node attributes are matched to generate repair instructions.

[0008] This invention presents a field testing method for the thermal insulation performance of prefabricated near-zero energy building exterior walls. By acquiring environmental data sequences from a micro-weather station and verifying constraints related to precipitation, wind speed, and sunshine duration, interference sources such as latent heat evaporation, convective disturbances, and insufficient heat storage energy are eliminated before data acquisition begins, reducing misjudgments caused by environmental factors in infrared thermal images. A seamless reference comparison area is determined based on a sliding window using a physical mask and the principle of equivalent radiation, ensuring strict equivalence between the reference area and the target seam detection area in terms of physical dimensions and total accumulated solar radiation throughout the day, reducing temperature difference analysis errors introduced by reference selection deviations. Furthermore, by acquiring multiple frames of infrared thermal images after sunset and performing inter-frame affine transformation alignment using physical component feature point matching, the spatial positions of pixels in each frame are unified to the reference frame coordinate system, reducing errors caused by micro-vibrations of the imaging platform. This reduces the risk of temperature curves corresponding to incorrect physical locations; by identifying the heat storage and dissipation characteristics of the relative temperature difference sequence in the initial cooling window and removing areas that meet the conditions from the thermal bridge alarm candidate list, the initial temperature difference bright lines caused by material specific heat capacity differences are separated from the actual thermal bridge, reducing the possibility of misjudging sealant heat storage as a through-type thermal bridge defect; by calculating the variance and mean of the relative temperature difference sequence for the unremoved area in the quasi-steady-state window at the end of the observation period and comparing it with a preset threshold, the temperature difference is confirmed to come from the indoor continuous heating channel after the physics enters the quasi-steady-state stage, improving the reliability of the through-type thermal bridge defect judgment; by converting the defect pixel coordinates into building physical coordinates through homography mapping and matching the prefabricated node attributes to generate repair instructions, a complete link from thermal image detection to accurate on-site repair is established.

[0009] The verification of environmental data sequences in this invention to satisfy preset precipitation constraints, wind speed constraints, and sunshine duration constraints includes: determining that the precipitation constraint is not satisfied when the cumulative precipitation in the past preset time period is greater than zero; determining that the wind speed constraint is not satisfied when the average wind speed in the consecutive preset time period before the current moment is greater than a preset wind speed threshold; and determining that the sunshine duration constraint is not satisfied when the cumulative duration of the solar irradiance on the current day being greater than a preset irradiance threshold is less than a preset sunshine duration threshold.

[0010] This invention sets specific evaluation and judgment conditions for precipitation constraints, wind speed constraints, and sunshine duration constraints, transforming environmental suitability judgment from manual experience into repeatable automatic verification, reducing invalid data and misjudgments caused by starting detection under unfavorable weather conditions.

[0011] The present invention describes the determination of a seamless reference comparison area around a target seam detection area based on a sliding window using a solid mask and the principle of equivalent radiation. This includes: defining an annular search domain with a preset inner and outer diameter centered on the center point of the target seam detection area; extracting the geometric contours of the seams, embedded parts, and door / window openings and expanding them by a preset size to form an inaccessible area mask; sliding a window template with the same size as the target seam detection area within the annular search domain at a preset step size, recording window positions that do not intersect with the inaccessible area mask as candidate areas; calculating the total daily cumulative solar radiation between each candidate area and the target seam detection area, and selecting candidate areas with a relative deviation rate of radiation less than a preset deviation threshold as seamless reference comparison areas.

[0012] This invention automatically selects seamless benchmark comparison areas that are the same size as the target seam detection area and have a cumulative solar radiation deviation rate of less than a threshold by defining a ring search domain, constructing an inaccessible area mask, and traversing a sliding window. This reduces the comparison deviation caused by inconsistent size and radiation conditions when manually selecting benchmark areas.

[0013] The inter-frame affine transformation alignment based on physical component feature point matching described in this invention includes: ORB feature points of the reference frame and each frame to be registered are extracted and brute-force matching is performed. After removing mismatched points using the RANSAC algorithm, the affine transformation matrix is ​​calculated. Based on the affine transformation matrix, reverse resampling interpolation is performed on the frames to be registered to align each frame to the coordinate system of the reference frame.

[0014] This invention uses a combination of ORB feature point extraction and RANSAC mismatch removal to calculate the inter-frame affine transformation matrix and performs reverse resampling interpolation on each frame. This unifies the spatial position of each frame of infrared thermal image to the reference frame coordinate system within several consecutive hours, reducing the risk of spatial misalignment of pixel-level temperature sequences due to slight displacement of the shooting platform.

[0015] The present invention describes identifying the heat storage and dissipation characteristics of a relative temperature difference sequence during the initial cooling window, marking regions that meet the heat storage and dissipation characteristics, and removing them from the thermal bridge alarm candidate list. This includes: determining a reference peak frame for the relative temperature difference sequence; extracting relative temperature difference values ​​from a predetermined number of consecutive frames starting from the reference peak frame; calculating the overall linear regression slope of the relative temperature difference sequence within the predetermined number of frames using the least squares method, and calculating the attenuation magnitude of the mean of the last predetermined number of frames and the sequence peak value; when the overall linear regression slope is negative and the attenuation magnitude exceeds a predetermined attenuation ratio, it is determined that the heat storage and dissipation characteristics are met.

[0016] This invention determines the reference peak frame of the relative temperature difference sequence and calculates the overall linear regression slope and the attenuation magnitude of the mean and peak values ​​within a preset number of frames. It uses the overall convergence trend of the temperature difference in the heat storage and dissipation process as an identification feature, reducing the situation where the actual thermal bridge is mistakenly stripped away during the outdoor cooling phase when the slope sign is determined solely by a single moment.

[0017] The duration of the initial cooling window described in this invention is determined based on the thermal diffusion time constant of the wall material, and is not less than a preset multiple of the thermal diffusion time constant; the value of the preset number of consecutive frames ensures that the covered observation time is not less than a preset multiple of the thermal diffusion time constant.

[0018] This invention correlates the initial cooling window duration and the value of a preset number of frames with the thermal diffusion time constant of the wall material, enabling the time window for identifying heat storage and dissipation characteristics to be adjusted according to the actual thermal response characteristics of different wall materials and insulation layer thicknesses, thereby improving the adaptability of the method to different building projects.

[0019] The starting time of the quasi-steady-state window at the end of the observation period described in this invention is determined based on the time point when the rate of change of the wall surface temperature decays to below a preset rate of change threshold.

[0020] This invention associates the start time of the quasi-steady-state window at the end of the observation period with the time point when the rate of change of the wall surface temperature decays to a preset threshold, rather than fixing it to a certain absolute time. This allows the start judgment of the quasi-steady-state analysis window to be automatically adjusted according to the actual cooling process of the exterior wall under different seasons and weather conditions.

[0021] The preset variance threshold of the present invention is determined based on the noise equivalent temperature difference of the infrared thermal imager, and the preset mean threshold is determined based on the noise equivalent temperature difference of the infrared thermal imager.

[0022] This invention correlates the variance threshold and mean threshold with the noise equivalent temperature difference (NETD) of the infrared thermal imager, enabling the judgment threshold to be automatically adjusted to follow the background noise level of thermal imagers of different performance levels, reducing the risk of judgment standard mismatch due to the replacement of detection equipment.

[0023] The present invention describes a method for converting building physical coordinates using homography mapping, and generating repair instructions based on matching prefabricated node attributes using these building physical coordinates. This includes: calculating a homography mapping matrix from the thermal image pixel plane to the CAD base map plane using at least four non-collinear rigid feature point pairs extracted from infrared thermal images and CAD base maps; converting the defect pixel coordinates into physical coordinates on the CAD base map using the homography mapping matrix; performing spatial inclusion judgment between the physical coordinates and the bounding boxes of each prefabricated node in the CAD base map to determine the node classification label to which the defect belongs; and retrieving the corresponding repair instruction template based on the node classification label to generate repair instructions containing physical coordinates and construction operation steps.

[0024] This invention converts defect pixel coordinates into physical coordinates of a CAD base map by calculating a homography mapping matrix using at least four non-collinear rigid feature point pairs. Then, it determines node classification labels by spatial inclusion judgment and retrieves maintenance instruction templates, thereby realizing the automatic conversion from infrared thermal image detection coordinates to on-site construction instructions with structural node attributes.

[0025] The node classification labels of this invention include vertical splicing cross joints of precast concrete wall panels, horizontal splicing joints of precast concrete wall panels, metal connectors for triple-glazed double-cavity exterior windows, joints between exterior window frames and wall panels, and pre-embedded sleeve joints for through-wall pipes; the maintenance instruction template includes the operation content of cutting the finish, adding filling material, applying airtight tape, or installing heat insulation pads for the corresponding node type.

[0026] This invention covers common joint construction types and targeted repair operations for the exterior walls of prefabricated near-zero energy buildings by listing specific node classification labels and corresponding maintenance instruction templates, so that the test results can be directly converted into specific construction steps for the corresponding structural nodes.

[0027] Compared with the prior art, the present invention has at least the following beneficial effects: This invention provides a field testing method for the thermal insulation performance of prefabricated near-zero energy building exterior walls. It establishes meteorological access conditions for testing by implementing triple environmental constraint verification, reducing invalid testing due to adverse environments. Automatic optimization of the benchmark area, which is equivalent to both size and radiation, establishes a physically comparable reference basis for temperature difference analysis. Furthermore, by acquiring long-term thermal images after sunset and rigidly aligning component feature points frame by frame, it locks the temperature pixels corresponding to the physical materials in the spatial dimension, ensuring that the temperature curve maintains a continuous and consistent physical position over several hours.

[0028] This invention distinguishes between heat dissipation from sealant and temperature differences that maintain true thermal bridges by identifying the overall convergence trend of the temperature difference sequence during the initial cooling phase. This removes heat storage artifacts from the thermal bridge alarm candidate list, rather than relying on a single temperature difference threshold for absolute classification in the final judgment stage. This ensures that the judgment logic corresponds to the physical mechanism of the difference in the thermal properties of the wall material. At the end of the observation period, it confirms through-type thermal bridge defects by using the dual thresholds of variance and mean within the quasi-steady-state window. The judgment threshold is set using the equivalent temperature difference of the infrared thermal imager's own noise as a benchmark, making the judgment conclusion reproducible and independent of the device.

[0029] This invention projects the coordinates detected by infrared thermal imaging onto the CAD base map through homography mapping, automatically matches the attributes of assembled nodes and generates repair instructions with construction labels, enabling on-site workers to accurately locate defects and perform corresponding construction node repair operations under the uniform surface layer, reducing the amount of ineffective work from blindly disassembling the surface. Attached Figure Description

[0030] Figure 1 This schematically illustrates a flowchart of the on-site testing method for the thermal insulation performance of prefabricated near-zero energy building exterior walls according to the present invention; Figure 2 This diagram illustrates a comparison of the evolution characteristics of relative temperature difference sequences. Figure 3 The diagram illustrates the determination of relative temperature difference during the quasi-steady-state stage. Detailed Implementation

[0031] This embodiment uses a prefabricated near-zero energy building project in a frigid region as a specific application scenario. The project's exterior walls adopt a precast concrete sandwich insulated wall panel system. The insulation layer material is graphite polystyrene board, 200mm thick. The wall panels are connected by vertical and horizontal joints, which are filled with polymer sealant and covered with EPDM airtight tape. The project has a total building area of ​​approximately 12,000 square meters, with 12 floors above ground, and a total length of approximately 4,800 meters for the exterior wall joints. In frigid regions, the indoor-outdoor temperature difference in winter is consistently above 35°C, placing extremely stringent requirements on the airtightness and insulation continuity of the building envelope. Any hidden defects at any joint could cause indoor heat to continuously escape along the through-channels, forming a weak abnormal temperature rise area that is difficult to detect in conventional infrared sampling.

[0032] The method is implemented on a portable on-site testing device for thermal defects in exterior walls. The main control processing unit of the testing device uses an embedded processor based on a quad-core ARM Cortex-A76 architecture with a clock speed of 2.0 GHz, equipped with 2 GB of LPDDR4 RAM and 32 GB of non-volatile flash memory. The main control processing unit establishes data connections with sensors and terminal devices through the following four types of communication modules.

[0033] The first type is the infrared thermal imaging acquisition module, configured with a cooled long-wave infrared thermal imager. The infrared thermal imager's detector resolution is 640×512 pixels, the noise equivalent temperature difference (NETD) is no greater than 0.03K, the temperature measurement accuracy is ±0.5℃, and the lens field of view is 24 degrees × 18 degrees. It is directly connected to the main control processing unit via a USB 3.0 interface. The infrared thermal imager is fixed to a motorized two-axis gimbal, with an angle repeatability of 0.01 degrees, used to maintain field-of-view stability during automatic scanning of exterior facades.

[0034] The second type is the visible light-assisted imaging module, which is equipped with a 12-megapixel visible light camera. The focal length of the visible light camera's lens is matched with the field of view of the infrared thermal imager, and it is connected to the main control processing unit via a MIPI interface to synchronously acquire visible light texture images for feature point extraction.

[0035] The third category is wireless environmental sensing and communication modules, which establish a data link with miniature weather stations deployed on-site via the LoRa spread spectrum communication protocol. The miniature weather stations are equipped with rain sensors, ultrasonic anemometers, and total radiation meters, with a sampling period of 1 minute. The LoRa communication range can reach 2 kilometers under line-of-sight conditions, and data packets are transmitted using AES-128 encryption.

[0036] The fourth type is the industrial tablet terminal communication module, which establishes a connection with the operator's handheld terminal via the WiFi 6 protocol to receive detection commands, transmit processing results, and push repair work orders.

[0037] The non-volatile flash memory pre-stores the CAD base drawing of the prefabricated construction panels for the building facade. The CAD base drawing includes the outer contour geometric boundaries of each wall panel, the center lines of the joints, the positioning coordinates of the embedded parts, and the precise demarcation of the door and window openings. When the test device is powered on, the main control processing unit loads the CAD base drawing from the non-volatile flash memory into the running memory and completes the coordinate system initialization.

[0038] The following describes the specific execution process of the test method in detail with reference to the five steps of this embodiment.

[0039] Step S1: Perform a preliminary assessment and verification of the on-site environmental constraints, and delineate the target seam detection area and the seamless benchmark comparison area based on the principle of solid mask sliding window and equivalent radiation.

[0040] It should be noted that the effectiveness of on-site infrared thermal imaging inspection is highly dependent on the controllability of environmental conditions. In frigid regions, strong solar radiation during winter and spring makes the surface temperature distribution of wall panels easily dominated by uneven heating caused by sunlight. If inspection is initiated during periods when residual heat has not yet dissipated, the slight temperature difference caused by surface heat accumulation and actual through-type thermal bridge defects will be completely mixed. At the same time, rain or strong winds can introduce latent heat of vaporization and strong convective disturbances, rendering the infrared thermal images uninterpretable. On the other hand, traditional inspection methods rely heavily on the operator's visual judgment when selecting a reference area. The so-called intact insulation area selected often has inconsistent geometric dimensions with the inspected joint area, and the cumulative solar radiation is not equivalent, resulting in a systematic deviation between the two cooling curves at the source. This invention aims to establish a reference group that is fully equivalent in physical dimensions and energy input through triple verification under rigid environmental constraints and automated reference area optimization based on CAD base map masking and equivalent radiation criteria.

[0041] Specifically, before initiating the on-site infrared thermal image acquisition process, the main control processing unit first retrieves the real-time environmental data stream from the micro weather station via the LoRa communication module. The data stream continuously pushes the recorded sequence of the past 24 hours at one-minute intervals, and the main control processing unit performs three pre-emptive hard evaluation constraint verifications on the sequence.

[0042] The first step is precipitation constraint verification. The main control processing unit retrieves minute-level precipitation records from the past 12 hours and sums all recorded values ​​to obtain the cumulative precipitation. When the cumulative precipitation is greater than 0 mm, it is determined that there is residual liquid water on the exterior wall surface or inside the joints. During the nighttime cooling process, the evaporation of water will take away heat and produce local low-temperature patches on the surface. These patches are easily confused with the abnormal temperature rise areas caused by real through-type thermal bridge defects on infrared thermographs. At this time, the main control processing unit sends a forced interlock signal to the operation terminal with the content of "environmental conditions exceeding limits" and the reason description of "interference from latent heat evaporation of precipitation", suspending all subsequent detection processes.

[0043] The second step is wind speed constraint verification. The main control processing unit extracts the average wind speed and corresponding wind direction for the 30 consecutive minutes preceding the current moment. In frigid regions, the average wind speed is high in winter. When there is lateral shear airflow within the angle range between the infrared thermal imager lens axis and the wall normal, the convective heat transfer coefficient on the wall surface will exhibit spatial non-uniformity, disrupting the symmetry of the thermal boundary layer between the target seam detection area and the seamless reference comparison area. The convective heat transfer safety threshold can be configured based on the surface roughness of the wall and the temperature resolution of the thermal imager. For the precast concrete wall panel surface and the infrared thermal imager with a NETD of 0.03K used in this project, based on the empirical relationship between the convective heat transfer coefficient and wind speed, it is calculated that when the wind speed exceeds 5m / s, the surface temperature difference introduced by the unevenness of the convection space is close to or exceeds 0.1K. This magnitude constitutes an interference source for the subsequent steady-state temperature difference determination. Therefore, in this embodiment, the threshold is set to 5m / s. When the determination exceeds the configured threshold, the main control processing unit sends a forced interlock signal to the operation terminal, with the reason described as strong wind disturbance causing surface heat flow to be covered, requiring the operator to wait for the wind speed to stabilize and decrease before starting the detection.

[0044] The third step is the verification of daytime sunshine duration constraints. The main control processing unit retrieves the total radiation meter data from sunrise to sunset of the day and counts the cumulative duration of solar irradiance greater than 120W / m². This irradiance threshold corresponds to the direct sunlight level when the solar altitude angle exceeds approximately 10 degrees under clear winter weather at 40 degrees North latitude; the set condition is that the cumulative effective sunshine duration must be greater than 4 hours. If the weather is cloudy or overcast all day, the solar radiation energy reaching the wall is insufficient, the surface heat storage effect is weak, and the absolute value of the temperature difference caused by the difference in specific heat capacity between different materials will be drowned out by the background noise. The temperature difference evolution trend used to identify the heat storage and dissipation characteristics in the subsequent step S3 will not have a sufficient signal-to-noise ratio. When the constraint is determined to be unsuccessful, the main control processing unit sends a forced lockout signal to the operation terminal, with the reason described as insufficient daytime heat storage energy, requiring a new detection window to be selected after consecutive clear days.

[0045] The verification logic for the three environmental constraints is executed in a serial hard-determination manner, that is, if any one of them fails, the process will be terminated, and the process will proceed to the next sub-step only after all three are passed.

[0046] After all environmental constraints were verified, on-site operators used the interactive interface of the operating terminal to select a rectangular area on the preview image provided by the visible light camera. This area was required to completely cover the section of the exterior wall joint to be inspected. The main control processing unit received the selected coordinates and performed two processes: First, the main control processing unit called the visible light camera to capture a high-resolution image frame and used the Canny edge detection operator to extract the linear direction of the joint within the selected area. A strip-shaped area, traversed by the center line of the joint and extending 50mm outwards on both sides, was officially registered as the target joint detection area. Second, the main control processing unit extracted the minimum bounding rectangle of the target joint detection area, recorded the two-dimensional coordinates of the center point in the building's physical coordinate system, and recorded the physical length and width of the rectangle. The origin of the building's physical coordinate system was defined at the lower left corner of the first-floor exterior wall of the building. The X-axis extended eastward horizontally, and the Y-axis extended upward vertically, with units of meters.

[0047] After the target seam detection area is registered, the main control processing unit automatically starts the equivalent radiation optimization algorithm. The purpose is to find a seamless benchmark comparison area in the intact insulation area around the target seam detection area that has the same physical size as the target seam detection area and whose total daily solar radiation is statistically equivalent.

[0048] The search space of the optimization algorithm is a ring-shaped physical domain. The main control processing unit sets a ring-shaped area with an inner diameter of 0.5m and an outer diameter of 1.0m, centered on the center point of the target seam detection area. The inner diameter is set to avoid the search results falling into the seam itself or its adjacent heat-affected zone. The outer diameter is set to ensure that the seamless reference comparison area and the target seam detection area are within the same height and orientation range, sharing the same macroscopic solar motion trajectory and building self-shading conditions.

[0049] The main control processing unit then retrieves the CAD base map from the running memory and extracts the precise geometric contours of all embedded parts boundary lines, joint center lines, and door and window openings within the annular search domain. Since the embedded parts are made of metal, their surface temperature response differs significantly from that of the insulation board; the joints themselves are the objects to be inspected and cannot be used as a reference. The main control processing unit performs a uniform morphological dilation operation on the geometric contours, expanding them outwards by 10cm on each side; the dilated contours are merged into a single binary image, forming an inaccessible area mask; any pixels located within the mask are considered invalid candidate positions in subsequent sliding window traversals.

[0050] The main control processing unit constructs a rectangular sliding window template with physical dimensions strictly equal to the length and width of the target seam detection area. The physical sliding step size is set to 5cm. The window template is slid sequentially along the horizontal and vertical directions within the annular search domain. For each sliding step increment, a Boolean intersection operation is performed between the currently occupied area of ​​the window template and the unreachable area mask. If the intersection result is true, it indicates that the window has touched a seam, embedded part, or door / window edge, and the position is discarded. If the intersection result is false, the current center coordinates of the window are recorded as a valid candidate area, and it is added to the candidate area set.

[0051] After traversing the entire annular search domain, the main control processing unit calculates the total solar radiation for each candidate region in the candidate region set. The wall normal vector required for radiation calculation is uniformly assigned by the facade orientation angle determined by the CAD base map. Each candidate region and the target seam detection area are on the same facade, thus sharing the same solar incidence angle cosine time series. The solar position is calculated using an astronomical algorithm based on the local latitude and detection date, calculating the solar altitude angle and azimuth angle minute by minute with a one-minute time step. The judgment of building self-shading and shading by adjacent buildings is achieved by converting the outer wall outline and surrounding building outline in the CAD base map into a three-dimensional shading body. With the sunlight as the ray direction, if the ray intersects with any shading body before reaching the wall, the candidate region is determined to be in a shading state for that minute, and the direct radiation contribution is counted as zero. The calculation of the scattered radiation component uses an isotropic sky model, multiplying the scattered radiation value recorded by the total radiation table by the visibility angle coefficient of the candidate region relative to the sky. This angle coefficient is obtained by integrating the wall orientation and surrounding shading outline through a hemispherical field of view. The main control processing unit performs an irradiance accumulation every minute, accumulating the direct radiation component by multiplying it by the cosine of the incident angle and adding the scattered radiation component. The effective accumulation period is from sunrise to sunset on the same day when the solar irradiance is greater than 120 W / m². Following the procedure of this invention, the total daily cumulative radiation of each candidate region is calculated, as is the total daily cumulative radiation of the target seam detection area. The main control processing unit iterates through the candidate region set one by one, calculating the relative percentage deviation between the radiation of each candidate region and the radiation of the target seam detection area, i.e., the absolute value of the difference divided by the radiation of the target seam detection area and then multiplied by 100%. The main control processing unit selects candidate regions with a relative deviation rate strictly less than 2%. If multiple regions meet the condition, the one with the smallest absolute deviation rate is selected, officially locked, and registered as the seamless reference comparison area.

[0052] When no candidate region within the entire annular search domain meets the 2% relative radiation deviation constraint, the main control processing unit sorts the regions by deviation rate from smallest to largest, sends the deviation rate values ​​of the top three regions back to the operation terminal, and simultaneously pushes a prompt message, requiring the operator to confirm whether to accept the relaxation to 3% or manually reselect the seamless reference comparison area.

[0053] After registration is completed, the main control processing unit stores the four corner pixel coordinates of the target seam detection area and the seamless reference comparison area into the spatial anchor point table in the running memory.

[0054] Step S2: After sunset, start acquiring long-term dynamic thermal imaging data streams and perform rigid alignment between frames based on physical component feature point matching.

[0055] It should be noted that step S1 establishes a pair of equivalent target seam detection areas and seamless reference comparison areas in the spatial dimension. However, in this embodiment, the exterior wall insulation performance of near-zero energy buildings is extremely high. The surface cooling rate of the area with intact seams is almost synchronous with that of the surrounding concrete area. Only when there is a through-type thermal bridge defect will a slight temperature difference of about 0.3K to 1.0K appear in the quasi-steady-state convergence stage. The temperature difference is less than the temperature fluctuation of several degrees or even more than ten degrees caused by daytime solar radiation. Therefore, any spatial pixel misalignment will mix the temperature values ​​of adjacent materials into the time domain sequence of the target area, creating a false temperature jump in the time domain. This invention controls the pixel spatial position of each frame of infrared thermal image at a sub-pixel accuracy alignment level through fixed arcsecond-level locking of an electric two-axis gimbal and dual-track correction based on affine transformation of rigid feature points of physical components.

[0056] Specifically, after sunset when the sun's direct heat source is lost, the main control processing unit begins monitoring the rate of change in air temperature at the micro weather station. When it detects a monotonous downward trend in air temperature for 15 consecutive minutes, it determines that the exterior wall has entered the natural cooling phase. At this point, the main control processing unit automatically times the timer and sends continuous trigger commands to the infrared thermal imager one hour after sunset. The purpose of the one-hour delay is to ensure that the initial rapid release phase of residual heat disturbance within the wall panel has largely passed, and the temperature change enters a relatively gradual cooling phase.

[0057] The infrared thermal imager captures a panoramic infrared thermal image of the exterior wall facade every 3 minutes at a fixed time step. Before each capture, a motorized two-axis pan-tilt unit performs a position reset check, returning to the preset angle position and locking it to ensure that the lens optical axis remains constant for several consecutive hours. The total continuous capture time is set to 4 hours, covering the entire process from the middle of the nighttime cooling phase to the end of the quasi-steady-state convergence phase; a total of 80 infrared thermal images are acquired, forming the original dynamic thermal image data stream matrix. Each infrared thermal image is accompanied by a timestamp accurate to milliseconds, provided by the system clock of the main control processing unit and periodically calibrated with the micro weather station data stream using a network time protocol.

[0058] The moment each frame of infrared thermal image is written to the non-volatile flash memory, the main control processing unit immediately initiates the spatial rigid alignment preprocessing module for that frame. The processing flow uses the first frame of the acquired sequence as the absolute spatial reference frame; the main control processing unit combines the grayscale image of the reference frame with each subsequent frame of infrared thermal image to be registered to form an image pair. For each image pair, the main control processing unit calls the ORB feature point detection algorithm to extract approximately 500 to 800 rotation- and scale-invariant feature point descriptors across the entire area of ​​the reference frame; for each frame of infrared thermal image to be registered, feature point descriptors are similarly extracted across the entire area, and brute-force matching is performed to generate a coarse matching point pair set.

[0059] On actual construction sites, although the repeatability of the electric two-axis gimbal has reached the arcsecond level, the construction lifting platform or tripod may still experience slight displacement of several millimeters during intermittent gusts of wind over a 4-hour period. This displacement is reflected in a 640×512 pixel resolution infrared thermal image as a shift of several pixels. Simultaneously, the signal-to-noise ratio and contrast of infrared thermal images are inherently lower than those of visible light images, and feature points may be sparse in dark, low-temperature regions. Therefore, the main control processing unit further performs two screening processes on the coarsely matched point pair set.

[0060] First, the main control processing unit performs distance threshold filtering, calculates the Euclidean distance of all matching point pairs, removes outlier point pairs whose distance is greater than twice the median, and retains the matching subset with high spatial consistency.

[0061] Then, the main control processing unit calls the RANSAC (Random Sample Consensus) algorithm on the subset to estimate the homography matrix. The RANSAC algorithm randomly selects four pairs of points from the matching subset, calculates the candidate homography matrix, projects the remaining point pairs according to this matrix, and counts the number of interior points with a projection error of less than 3 pixels. After 200 iterations, the candidate matrix corresponding to the round with the most interior points is selected as the final output two-dimensional affine transformation matrix. This matrix is ​​a 3x3 floating-point matrix that can describe the composite geometric transformation relationship of translation, rotation, and scaling between two frames of infrared thermal images.

[0062] If the number of interior points extracted from a frame of infrared thermal image to be registered is less than the preset minimum threshold of 20, the main control processing unit determines that the image is blurred or vibrated due to the instantaneous excessive strength of the gusts, marks it as an invalid frame and discards it, and does not participate in subsequent time series analysis.

[0063] For each frame that passes the quality inspection, the main control processing unit uses the obtained affine transformation matrix as the transformation parameter to perform global pixel reverse resampling interpolation. For the coordinates of each target pixel in the infrared thermal image to be registered, the inverse matrix of the transformation matrix is ​​used for backprojection to the coordinate space of the reference frame. Bilinear interpolation sampling is then performed at the transformed coordinates, and the interpolation result is assigned to the target pixel. After traversing all pixels, the infrared thermal image to be registered is spatially aligned to a coordinate system strictly consistent with the reference frame.

[0064] After rigid alignment is completed, the main control processing unit retrieves the spatial anchor point table registered in step S1 from the running memory. On the aligned infrared thermal image frame, pixel arrays of the target seam detection area and the seamless reference contrast area are cropped according to the anchor point coordinates. The cropping operation is repeated on each aligned infrared thermal image frame, ultimately outputting two sets of pure temporal data matrices that are absolutely static in the spatial dimension and change only with physical cooling in the temporal dimension. Each data matrix is ​​a three-dimensional array, with dimensions equal to the number of frames multiplied by the number of pixel rows in the region multiplied by the number of pixel columns in the region.

[0065] Step S3: Extract the characteristics of relative temperature difference changes in the initial stage of cooling, and identify and remove the surface heat storage illusion caused by the difference in specific heat capacity.

[0066] It should be noted that after completing the standardized acquisition of spatial and temporal data, the core analytical challenge faced by the main control processing unit is the false thermal bridge signal caused by the difference in material thermal properties. In the prefabricated exterior wall system involved in this embodiment, the specific heat capacity of the polymer sealant filling the joint is 1.5 kJ / (kg·K), while the specific heat capacity of the precast concrete on both sides is 0.88 kJ / (kg·K), with the polymer sealant having a specific heat capacity approximately 70% higher than that of the concrete. Under the same daytime solar irradiance input, the polymer sealant absorbs and stores far more heat per unit mass than the concrete. After sunset, both release heat to the cold air below -10 degrees Celsius, but the polymer sealant has a greater initial heat storage capacity, and its surface temperature remains higher than that of the concrete surfaces on both sides during the initial cooling period, forming a clear bright line along the joint on the infrared thermogram. Bright lines are, in essence, a natural manifestation of the thermal inertia of materials and do not necessarily indicate the presence of a through-type thermal bridge defect in the seam. However, in conventional infrared defect screening, such initial temperature differences caused by differences in specific heat capacity are the source that on-site interpreters are most likely to misjudge as through-type thermal bridge defects.

[0067] The physical basis of this invention lies in the fact that the temperature difference driven by the difference in heat storage is a passive dissipation process of a finite heat source. Its thermodynamic behavior is characterized by a monotonically decreasing temperature difference amplitude over a timescale on the order of a constant surface heat diffusion time. In contrast, a through-type thermal bridge defect driven by continuous indoor heat leakage often exhibits a long-term sustained surface temperature difference. That is, although the absolute value of the temperature difference may fluctuate or even decrease when the external ambient temperature changes, a residual difference greater than the instrument's background noise is always maintained between the thermal response and the well-insulated area. The heat storage and dissipation process results in a convergent temperature difference change tending towards zero, while a true thermal bridge results in a temperature difference change tending towards a non-zero stable value. Physically, the sign of the temperature difference slope alone cannot uniquely distinguish between these two mechanisms, because when the outdoor temperature continues to drop at night, the relative temperature difference in the true thermal bridge region may also exhibit a slowly decreasing negative slope phase. Therefore, the design goal of this invention is not to make a one-time absolute judgment on heat storage and thermal bridge, but to temporarily isolate the region that meets the typical dynamic characteristics of heat storage dissipation from the thermal bridge alarm logic of the current detection round, and concentrate the analysis resources on the reserved region that does not show the typical heat storage dissipation trend, and hand it over to the quasi-steady-state convergence analysis in the subsequent step S4 to complete the final confirmation.

[0068] The main control processing unit extracts the initial cooling analysis window from the two sets of time-series data matrices output in step S2, representing the beginning of the observation period. The window duration must cover the characteristic thermal diffusion time constant of the heat storage body on the wall panel surface. Taking the 200mm thick graphite polystyrene board insulation layer and precast concrete panel used in this embodiment as an example, according to the thermal diffusion theory of a one-dimensional semi-infinite object, the response time constant of the surface temperature to a step thermal boundary condition is a function of the material's thermal diffusivity. The thermal diffusivity of the graphite polystyrene board is... m² / s, the surface thermal diffusivity of the concrete panel with a thickness of approximately 50mm is... In a composite wall structure with two materials, the thermal response of the surface heat storage body transitioning from solar heating to nighttime radiative cooling after sunset mainly occurs within the first 2 to 3 hours. Therefore, in this embodiment, the initial cooling analysis window is set from the first frame to the 40th frame, corresponding to a time interval of 1 to 3 hours after sunset, spanning 2 hours. The window can be adjusted according to the actual insulation layer material and thickness of the wall being inspected, with the adjustment principle being that the window duration is not less than 3 times the thermal diffusion time constant of the wall surface.

[0069] At each frame within the initial cooling analysis window, the main control processing unit first performs spatial averaging on all pixels in the target seam detection area to obtain a scalar temperature value, which is recorded as the target area temperature. Simultaneously, the main control processing unit performs spatial averaging on all pixels in the seamless reference comparison area at the same time to obtain the reference area temperature. The averaging operation is performed independently at each time step, thereby constructing two sets of one-dimensional temperature curves that change over time in the 80-frame sequence.

[0070] The main control processing unit performs frame-by-frame subtraction on the two sets of temperature curves to generate a time series representing the relative thermal evolution of the target seam detection area relative to the seamless reference comparison area, which is the dynamic change sequence of the target area temperature minus the reference area temperature over time, or simply the relative temperature difference sequence.

[0071] The main control processing unit then executes heat storage and dissipation feature identification logic on the relative temperature difference sequence. First, the main control processing unit traverses all data points in the sequence, identifies the location of the global maximum value, and marks it as the reference peak frame.

[0072] Starting from the reference peak frame, the main control processing unit extracts the temperature difference change characteristics of N consecutive frames. The value of N is based on the fact that, under the condition of a 3-minute sampling interval of the infrared thermal imager, the continuous observation time covered by N frames should not be less than 1 times the thermal diffusion time constant of the wall panel surface. For the wall material and structure used in this embodiment, the thermal diffusion time constant is calculated to be 40 minutes, therefore N is taken as 15 frames, corresponding to a time span of 45 minutes. This value of N can be adjusted according to the actual wall material after the thermal diffusion time constant is calibrated through pre-experimentation. The adjustment range is 10 to 20 frames. When the wall type or insulation material is changed, it should be recalibrated.

[0073] The main control processing unit examines the relative temperature difference trend of these N frames. The key feature that distinguishes heat storage and dissipation from real thermal bridges is not the sign of the slope at a certain moment, but whether the temperature difference sequence shows an overall convergence trend of continuously declining from the peak throughout the entire window. Therefore, the main control processing unit performs feature discrimination: it uses the least squares method to calculate the overall linear regression slope of the relative temperature difference sequence within N frames, and at the same time calculates the attenuation magnitude between the mean of the sequence in the last 3 frames of N frames and the peak value of the sequence; if the overall linear regression slope is negative and the attenuation magnitude exceeds 50% of the peak value, then the target seam detection area is determined to exhibit typical heat storage and dissipation characteristics.

[0074] When the conditions are met, the main control processing unit marks the target seam detection area as having heat dissipation characteristics in the data table of its running memory. The purpose of this marking is to lower the defect analysis priority of this area in the current detection round, removing it from the thermal bridge alarm candidate list for this round, and preventing it from undergoing quasi-steady-state diagnostic judgment in step S4. In the actual test of this embodiment, approximately 80% to 90% of the polymer sealant seams were marked as having heat dissipation characteristics in this step, significantly reducing the remaining area to be confirmed. It should be clarified that the marking means that the current temperature difference performance of the area does not meet the sufficient conditions for confirming a thermal bridge defect, rather than a final conclusion on whether a thermal bridge defect actually exists in the area; if the same area has been peeled off in multiple consecutive detection days, but other abnormal signs still exist on-site, the operator can manually re-include it in the analysis scope.

[0075] If the overall linear regression slope within N frames does not show a significant downward trend, or the attenuation is much less than 50% of the peak value, it indicates that the temperature difference in this region is maintained at a high level and is not consistent with typical heat storage and dissipation characteristics. The seam area is not peeled off and is retained to enter step S4 for further quasi-steady-state diagnosis.

[0076] For example, Figure 2 This is a schematic diagram comparing the evolution characteristics of relative temperature difference sequences. The diagram shows the change in the relative temperature difference between the surface heat storage illusion detection area and the real through-type thermal bridge detection area as the acquisition frame number increases. The relative temperature difference in the surface heat storage illusion detection area shows a significant decreasing trend in the early stage of cooling; while the relative temperature difference in the real through-type thermal bridge detection area decreases less and remains at a higher level, without showing typical heat storage dissipation convergence characteristics. This indicates that the present invention can distinguish between heat storage illusion caused by material specific heat capacity differences and real continuous heat flow leakage by the temperature difference evolution trend in the early stage of cooling.

[0077] Step S4: Extract the relative temperature difference residual at the end of the quasi-steady-state stage of the observation period to determine the real through-type thermal bridge defect formed by indoor steady heat flow leakage.

[0078] It should be noted that the seam area that remains after the heat storage and dissipation feature annotation process in step S3 indicates that the temperature difference did not exhibit typical convergence characteristics of heat storage and dissipation during the mid-cooling phase. The design objective of this invention is to extract the temperature difference signal that has approached the residual stable value and whose amplitude is consistently higher than the background noise level of the infrared thermal imager after the cooling process enters the quasi-steady-state convergence stage, through dual verification of residual stability and amplitude, as a confirmatory criterion for determining a true through-type thermal bridge defect.

[0079] The physical basis of this invention is as follows. When observations enter the final stage of nighttime cooling, the limited daytime heat stored in the exterior wall surface material has been largely released. At this point, the time-varying rate of temperature change of the exterior wall surface has entered a slow-moving stage controlled by both the slowly decreasing outdoor air temperature and constant indoor heating. During this stage, the exterior surface temperature of the well-insulated area closely follows the slow decrease in outdoor air temperature, showing a slight downward trend. Therefore, the strict steady-state condition requiring zero time derivative of the temperature difference is difficult to meet in practice. However, it can be determined that the rate of change of the relative temperature difference between the well-insulated area and the joint area with through-type thermal bridge defects has slowed down in this stage, entering a quasi-steady-state convergence state. For the through-type thermal bridge defect area, because the constant indoor heating provides a continuous low-thermal-resistance heating channel, the exterior surface temperature is always higher than that of the well-insulated area. The residual temperature difference between the two tends to a stable positive value in the quasi-steady-state stage, while the residual temperature difference in the well-insulated joint area tends to zero.

[0080] Specifically, the main control processing unit extracts the quasi-steady-state analysis window from the 80 frames of time-series data output in step S2, representing the end of the observation period. In this embodiment, the quasi-steady-state analysis window is taken as the time interval from frame 60 to frame 80, corresponding to the 3rd to 4th hour after sunset, spanning one hour. The basis for selecting this window is that after the cooling of the first 3 hours, the time change rate of the wall surface temperature has decayed to less than 1°C per hour, entering the quasi-steady-state interval jointly controlled by the slow decrease of outdoor temperature and constant indoor heating.

[0081] The main control processing unit extracts the net residual relative temperature difference sequence corresponding to all frames within the quasi-steady-state analysis window. The sequence contains 20 independent sample points. The method for obtaining these 20 temperature difference values ​​is the same as in step S3, that is, performing a subtraction operation on the spatial average value of the target seam detection area and the seamless reference comparison area for each frame.

[0082] The main control processing unit introduces quasi-steady-state residual determination logic into the sequence, which includes two serial determination thresholds.

[0083] The first criterion is the discrete fluctuation variance test of the sequence. The main control processing unit calculates the variance of the sequence using the standard unbiased variance formula, which is the sum of the squares of the differences between each sample value and the sequence mean divided by the number of samples minus 1. If there is no steady-state heat flow leakage at the seam, the temperature difference between the target seam detection area and the seamless reference comparison area should have decayed to near zero and fluctuate randomly around zero due to the influence of detector readout noise. At this time, the variance is mainly determined by the noise equivalent temperature difference NETD of the infrared thermal imager. In this embodiment, the NETD of the infrared thermal imager is 0.03K. The main control processing unit sets the upper limit threshold of the convergence variance to the square of the product of 0.5 and NETD, which is calculated to be 0.000225K². When the actual calculated variance is strictly less than this threshold, it indicates that the high-frequency fluctuation amplitude of the temperature difference sequence has contracted below the detector's background noise level, and the sequence has moved away from the stage of severe transient fluctuations. The value of this variance threshold is squared to the nominal NETD value of the infrared thermal imager under test. For devices with different NETD parameters, the same formula can be used to recalculate the threshold to ensure device independence.

[0084] The second judgment threshold is a magnitude test of the mean temperature difference. The main control processing unit calculates the mathematical mean of the sequence; the main control processing unit sets a safe threshold for the mean at 3 times NETD, i.e., 0.09K. When the mean is strictly greater than this threshold, it indicates that the observed quasi-steady-state temperature difference amplitude has exceeded the range that the instrument noise can explain. This signal originates solely from a continuous heat source, i.e., the steady-state heat flow from indoor heat to the outside along the through-type thermal bridge defect. The setting of the 3 times NETD threshold is based on a general engineering criterion in signal detection theory corresponding to approximately 99.7% confidence level. Verified by measured data from winter nights in multiple projects in frigid regions, this threshold can effectively distinguish between detector noise and the actual thermal bridge temperature difference; the threshold can also be adjusted between 2.5 and 4 times NETD according to the on-site trade-off between false alarm rate and missed detection rate.

[0085] Only when both judgment thresholds are met simultaneously—that is, the variance is less than the convergence threshold and the mean is greater than 3 times NETD—will the main control processing unit confirm the target seam detection area as a true through-type thermal bridge defect and immediately activate the true through-type thermal bridge defect alarm command. The main control processing unit will also record the mean value as the heat leakage temperature difference assessment value of the defect, in degrees Celsius, accurate to two decimal places.

[0086] If the variance condition is met but the mean is no greater than 3 times NETD, the main control processing unit will determine that the target seam detection area has qualified thermal insulation performance and will not issue an alarm. If the variance condition is not met, it means that the area has not yet entered the thermal response stability stage within the quasi-steady-state window and there is external time-varying boundary condition interference. The main control processing unit will mark it as a state to be retested and prompt the operator on the operation terminal that there may be intermittent environmental disturbance interference, and suggest retesting at night.

[0087] For example, Figure 3 This is a schematic diagram of relative temperature difference determination during the quasi-steady-state stage. The diagram shows the relative temperature difference residual sequences of two detection zones within the quasi-steady-state analysis window, and the preset mean threshold is marked. At the end of the observation period, the variance of the relative temperature difference sequence in the real through-type thermal bridge detection zone is extremely small, and its mean is significantly higher than the preset mean threshold. In contrast, the mean of the relative temperature difference in the surface heat storage illusion detection zone has decayed to near zero and is lower than the preset mean threshold. This indicates that the present invention, through dual-threshold determination logic, diagnoses real through-type thermal bridge defects formed by indoor stable heat flow leakage during the quasi-steady-state stage.

[0088] Step S5: Map the infrared detected thermal defect coordinates to the building physical coordinates, match the node construction attributes, and generate on-site repair instructions.

[0089] It should be noted that, after the diagnosis in step S4, the main control processing unit has locked one or more pixel clusters in the digital coordinate system of the infrared thermal image. These pixel clusters correspond to the seam segments in the real physical world where steady-state heat flow leakage exists. However, there is still an inconsistency in coordinate space between the pixel positions in the digital image and the actual wall positions that workers can operate on. Factors such as the shooting angle of the infrared thermal imager, wall flatness deviations, and lens distortion can cause a non-linear mapping relationship between the pixel coordinates on the infrared thermal image and the building's physical coordinate system. In addition, prefabricated exterior walls are usually covered with a layer of decorative mortar or paint of a uniform color, completely obscuring the seam positions, making it impossible for on-site workers to find the accurate cutting and repair positions by visual inspection. The goal of this invention is to use a homography perspective mapping algorithm to reverse-project the coordinates of the thermal image defects in the digital dimension to the CAD base map coordinate system with millimeter-level precision, and automatically match the node attributes of the coordinates to generate component-level repair work orders that can directly guide construction operations.

[0090] Specifically, while issuing the alarm command for a true through-type thermal bridge defect in step S4, the main control processing unit automatically extracts the two-dimensional pixel center coordinates of the defect in the reference infrared thermal image. These coordinates are described in the first reference frame coordinate system established in step S2, representing the geometric center of the defect area in the infrared thermal image.

[0091] The main control processing unit then retrieves the CAD base map that has been loaded into the running memory in step S1. The CAD base map is a two-dimensional vector drawing, with the origin of the building's physical coordinate system consistent with the setting of the testing device, that is, the lower left corner of the first-floor exterior wall is the origin, the horizontal X-axis is to the right, and the vertical Y-axis is upward, with the unit being millimeters and the accuracy reaching 1mm.

[0092] The main control processing unit uses the rigid geometric feature points extracted and stored in step S2 as a bridge across coordinate systems. These feature points correspond to geometric entities that do not physically deform, such as the intersection of the right-angle frames of a triple-glazed, double-cavity exterior window or the chamfered turning point of a precast wall panel. The main control processing unit selects at least four non-collinear point pairs from the stored feature point database that are clearly visible in the infrared thermal image and have precise corresponding coordinates in the CAD base map.

[0093] The main control processing unit uses a point-pair-based direct linear transformation algorithm to analyze a 3x3 homography mapping matrix from the thermal image pixel plane to the CAD base map plane. After the calculation is completed, the main control processing unit represents the pixel coordinates in homogeneous coordinate form, multiplies them by the mapping matrix on the left, and outputs the absolute physical location coordinates of the defect on the CAD base map after normalization, in millimeters.

[0094] After obtaining the physical coordinates, the main control processing unit immediately and automatically starts the node attribute classification and matching program. The program determines the spatial inclusion relationship between the physical coordinates of the defect and the bounding boxes of each prefabricated node in the CAD base drawing. When the defect coordinates fall within the two-dimensional bounding box of a node, the program reads the node's construction classification label. In this embodiment, the node classification labels mainly include five categories: vertical splicing cross joints of precast concrete wall panels, horizontal splicing joints of precast concrete wall panels, metal connectors for triple-glazed double-cavity exterior window frames, joints between exterior window frames and wall panels, and pre-embedded sleeve joints for through-wall pipes.

[0095] The main control processing unit retrieves the corresponding construction plan text from the pre-compiled maintenance instruction template library based on the matched classification label. When the classification label is a cross joint of a vertical splicing of a precast concrete wall panel, the corresponding maintenance instruction is: The coordinate confirms a steady-state through-type thermal bridge defect, and the measured heat leakage temperature difference assessment value is 0.85℃; the instruction is to cut a 200mm diameter circular finishing layer at the coordinate, remove the existing failed polymer sealant in the joint, inject two-component polyurethane foam filler into the depth of the joint until it is completely filled, reapply EPDM airtight tape after the foam has cured, and finally restore the finishing layer. When the category label is "metal connector of triple-glazed double-cavity exterior window frame", the instruction is as follows: The coordinates confirm a through-type thermal bridge defect in the metal component, and the measured heat leakage temperature difference assessment value is 0.62℃; the instruction is to remove the decorative cover plate of the frame at this location, install a 5mm thick rigid polyurethane thermal break pad between the metal connector and the window frame, and cover the outside of the connector with a self-adhesive airtight waterproof and breathable membrane, ensuring that the overlap width between the membrane and the surrounding concrete is not less than 50mm.

[0096] The repair instruction file is sent to the operation terminal via a WiFi 6 communication module. The work order includes the precise physical coordinates of the defect, the measured temperature difference assessment value for heat leakage, the corresponding repair operation steps, and the specifications of the required consumables. After opening the work order, the on-site worker uses a laser rangefinder and a measuring tape to locate the coordinates on the wall, accurately finding the location of the defect hidden behind the finish layer and performing precise repairs.

[0097] After the repair work is completed, the operator submits a completion confirmation through the operating terminal. The main control processing unit marks the defect item as repaired and retains the coordinates, heat leakage temperature difference assessment value, and repair date in non-volatile flash memory as part of the building's full life cycle thermal performance traceability archive. Thus, the method in this embodiment completes a closed-loop process from rigorously defining the detection benchmark, dynamically stripping away the impact of heat storage and dissipation, accurately diagnosing through-type thermal bridge defects during the quasi-steady-state convergence stage, to issuing corrective instructions for actual construction.

Claims

1. A field testing method for the thermal insulation performance of prefabricated near-zero energy building exterior walls, characterized in that, include: Obtain environmental data sequences from micro weather stations and verify that the environmental data sequences meet preset constraints on precipitation, wind speed, and sunshine duration. Obtain the target seam detection area, and determine the seamless reference comparison area around the target seam detection area based on the principle of solid mask sliding window and equivalent radiation; After sunset, multiple frames of infrared thermal images are acquired. Using the first frame as a reference, inter-frame affine transformation alignment is performed based on physical component feature point matching to obtain the aligned target seam detection area temperature sequence and the seamless reference comparison area temperature sequence. The temperature sequence of the target seam detection area is calculated to be the difference between the temperature sequence of the seamless reference comparison area, and the relative temperature difference sequence is obtained. In the initial cooling window, the heat storage and dissipation characteristics of the relative temperature difference sequence are identified, and the areas that meet the heat storage and dissipation characteristics are marked and removed from the thermal bridge alarm candidate list. During the quasi-steady-state window at the end of the observation period, for the regions that have not been removed, the variance and mean of the relative temperature difference sequence are calculated. When the variance is less than the preset variance threshold and the mean is greater than the preset mean threshold, it is determined to be a real through-type thermal bridge defect. Obtain the pixel coordinates of the defect in the infrared thermal image, convert them into building physical coordinates through homography mapping, match the prefabricated node attributes based on the building physical coordinates, and generate repair instructions.

2. The on-site testing method for the thermal insulation performance of prefabricated near-zero energy building exterior walls according to claim 1, characterized in that, The verification environment data sequence satisfies preset precipitation constraints, wind speed constraints, and sunshine duration constraints, including: If the cumulative precipitation over a preset time period is greater than zero, the precipitation constraint is not met; if the average wind speed over a preset time period before the current moment is greater than a preset wind speed threshold, the wind speed constraint is not met; if the cumulative duration of the solar irradiance on the current day being greater than a preset irradiance threshold is less than a preset sunshine duration threshold, the sunshine duration constraint is not met.

3. The on-site testing method for the thermal insulation performance of prefabricated near-zero energy building exterior walls according to claim 1, characterized in that, The determination of a seamless reference comparison area around the target seam detection area based on the principle of solid mask sliding window and equivalent radiation includes: A pre-defined annular search domain with inner and outer diameters is defined using the center point of the target seam detection area as the center. The geometric contours of the seams, embedded parts, and door and window openings are extracted and expanded by a pre-defined size to form an inaccessible area mask. Within the annular search domain, a window template of the same size as the target seam detection area is slid at a pre-defined step size. The positions of windows that do not intersect with the inaccessible area mask are recorded as candidate areas. The total cumulative solar radiation of each candidate area and the target seam detection area throughout the day is calculated. Candidate areas with a relative deviation rate of radiation less than a pre-defined deviation threshold are selected as seamless reference comparison areas.

4. The on-site testing method for the thermal insulation performance of prefabricated near-zero energy building exterior walls according to claim 1, characterized in that, The inter-frame affine transformation alignment based on physical component feature point matching includes: ORB feature points of the reference frame and each frame to be registered are extracted and brute-force matching is performed. After removing mismatched points using the RANSAC algorithm, the affine transformation matrix is ​​calculated. Based on the affine transformation matrix, reverse resampling interpolation is performed on the frames to be registered to align each frame to the coordinate system of the reference frame.

5. The on-site testing method for the thermal insulation performance of prefabricated near-zero energy building exterior walls according to claim 1, characterized in that, The step of identifying the heat storage and dissipation characteristics of the relative temperature difference sequence during the initial cooling window, marking regions that meet the heat storage and dissipation characteristics, and removing them from the thermal bridge alarm candidate list includes: A reference peak frame for the relative temperature difference sequence is determined, and the relative temperature difference values ​​of a consecutive preset number of frames are extracted starting from the reference peak frame. The overall linear regression slope of the relative temperature difference sequence within the preset number of frames is calculated using the least squares method, and the attenuation magnitude of the mean of the last preset number of frames and the peak value of the sequence is calculated. When the overall linear regression slope is negative and the attenuation magnitude exceeds the preset attenuation ratio, it is determined that the heat storage and dissipation characteristics are met.

6. The on-site testing method for the thermal insulation performance of prefabricated near-zero energy building exterior walls according to claim 5, characterized in that, The duration of the initial cooling window is determined based on the thermal diffusion time constant of the wall material, and is not less than a preset multiple of the thermal diffusion time constant; the value of the preset number of consecutive frames ensures that the covered observation time is not less than a preset multiple of the thermal diffusion time constant.

7. The on-site testing method for the thermal insulation performance of prefabricated near-zero energy building exterior walls according to claim 1, characterized in that, The starting time of the quasi-steady-state window at the end of the observation period is determined based on the time point when the rate of change of the wall surface temperature decays to below a preset rate of change threshold.

8. The on-site testing method for the thermal insulation performance of prefabricated near-zero energy building exterior walls according to claim 1, characterized in that, The preset variance threshold is determined based on the noise equivalent temperature difference of the infrared thermal imager, and the preset mean threshold is determined based on the noise equivalent temperature difference of the infrared thermal imager.

9. The on-site testing method for the thermal insulation performance of prefabricated near-zero energy building exterior walls according to claim 1, characterized in that, The process involves converting the coordinates to building physical coordinates using homography mapping, matching the prefabricated node attributes based on these coordinates, and generating repair instructions, including: Using at least four non-collinear rigid feature point pairs extracted from infrared thermal images and CAD base maps, a homography mapping matrix from the thermal image pixel plane to the CAD base map plane is calculated; the defect pixel coordinates are converted into physical coordinates on the CAD base map through the homography mapping matrix; the physical coordinates are spatially contained with the bounding boxes of each assembly node in the CAD base map to determine the node classification label to which the defect belongs; the corresponding maintenance instruction template is retrieved according to the node classification label to generate a repair instruction containing physical coordinates and construction operation steps.

10. The on-site testing method for the thermal insulation performance of prefabricated near-zero energy building exterior walls according to claim 9, characterized in that, The node classification labels include vertical splicing cross joints of precast concrete wall panels, horizontal splicing joints of precast concrete wall panels, metal connectors for triple-glazed double-cavity exterior window frames, joints between exterior window frames and wall panels, and pre-embedded sleeve joints for through-wall pipes; the maintenance instruction template includes the operation content of cutting the finish, adding filling material, applying airtight tape, or installing heat insulation pads for the corresponding node type.

Images