Automatic door and window quality inspection method and system based on machine vision

Abstract

The invention provides an automatic door and window quality inspection method and system based on machine vision. A thermal image sequence and pressure data are synchronously collected in the pressurization process, a heat conduction base line is established in the low-pressure stage, and heat conduction mutation caused by deformation of the sealing structure is captured in the high-pressure stage. Dynamic registration is carried out on thermal image data based on door and window geometric features to generate a displacement vector field, and abnormal heat flow vortexes and discontinuous boundaries are identified through a spatial-temporal feature extraction model. And starting a dynamic data compensation mechanism to generate a tracking atlas in a pressure attenuation stage, finally analyzing the phase correlation of an abnormal region in combination with a three-dimensional seepage model, outputting an air leakage path parameter, comparing the air leakage path parameter with a sealing failure threshold value, and generating a partition sealing performance grading evaluation result. According to the technical scheme provided by the invention, through thermodynamic dynamic modeling and multi-source data fusion, precise positioning of non-contact door and window sealing defects, three-dimensional seepage path reconstruction and quantitative leakage evaluation are realized, and the quality inspection precision and efficiency are remarkably improved.

Description

Technical Field

[0001] This application relates to the field of machine vision technology, and in particular to an automated door and window quality inspection method and system based on machine vision. Background Technology

[0002] With the increasing standards for building energy efficiency, the airtightness testing of doors and windows has become a core part of building quality acceptance. Traditional methods rely on static pressure testing and manual visual inspection, which are difficult to accurately quantify minute leakage defects caused by deformation of the sealing structure during dynamic pressurization. In particular, there is an urgent need to capture local heat conduction anomalies and low-flow-rate leakage paths caused by transient deformation during high-pressure stages. There is a pressing need for a non-contact, high spatiotemporal resolution automated detection technology to synchronously correlate pressure load changes and thermodynamic response characteristics, and to achieve defect location and dynamic modeling of leakage parameters.

[0003] Current mainstream solutions employ static pressure detection technology based on infrared thermal imaging. This involves acquiring the temperature field distribution on the surface of doors and windows under a fixed pressure, and then using image segmentation algorithms to identify areas with abnormal temperature differences. Specifically, the system acquires a single-frame thermal image under a preset pressure value, extracts the contour of the sealing strip using an edge detection algorithm, filters suspected leak hotspots using a thermal gradient threshold, and finally estimates the leakage equivalent by combining the pressure decay curve. Some improved solutions introduce time-series thermal image overlay analysis to enhance sensitivity to minute temperature fluctuations.

[0004] This approach relies on static pressurization conditions and cannot capture the transient deformation characteristics of the sealing structure under dynamic pressure changes, leading to the neglect of local heat conduction abrupt changes caused by elastic deformation during high-pressure stages. Furthermore, single-frame or low-sampling-rate thermal imaging data struggles to resolve heat flow continuity characteristics (such as vortices and boundary fractures), easily resulting in the loss of transient anomaly signals due to insufficient temporal resolution. Simultaneously, the static registration method does not consider frame displacement caused by pressure changes, leading to accumulated spatial alignment errors in the thermal imaging sequence, further affecting the accuracy of leak path reconstruction. In scenarios of sudden pressure changes, the existing approach lacks a dynamic data compensation mechanism, making it difficult to simultaneously track the thermodynamic response and pressure phase relationship, ultimately resulting in errors in leak equivalent calculations and misjudgments of defects. Summary of the Invention

[0005] This application provides an automated door and window quality inspection method and system based on machine vision, which solves the problem of errors caused by transient deformation of the sealing structure under dynamic pressure changes in the prior art.

[0006] In a first aspect, this application provides an automated door and window quality inspection method based on machine vision, including: During the pressurization process of doors and windows, thermal imaging sequence data and pressure data are acquired simultaneously. The initial distribution of the temperature field is collected during the low-pressure stage of the pressurization process and used to establish the heat conduction baseline. During the high-pressure stage of the pressurization process, the heat conduction abrupt change characteristics caused by the deformation of the sealing structure are captured. The thermal imaging sequence data is subjected to thermal flow correction processing. A dynamic registration relationship is established based on the geometric features of the door and window frame, the thermal conduction baseline, and the thermal conduction abrupt change features. A spatially constrained displacement vector field is generated based on the dynamic registration relationship. The displacement vector field is input into a preset spatiotemporal feature extraction model, and the heat flow continuity features of the sealed edge region are extracted through a multi-layer feature fusion structure to identify abnormal heat flow vortices and discontinuous boundaries. A dynamic data compensation mechanism is triggered during the pressure decay process. When a sudden pressure change is detected, the sampling frequency of thermal imaging data is increased and a dynamic tracking map is generated. Based on the topology of the abnormal heat flow vortex, the phase correlation between the abnormal region divided by the discontinuous boundary and the pressure data, and combined with the response events of the dynamic tracking spectrum, a three-dimensional seepage model of the sealing defect is established, and the geometric parameters and leakage equivalent of the leakage path are output according to the three-dimensional seepage model. Based on the leakage equivalent, the geometric parameters, and the preset sealing failure threshold curve, a graded evaluation result associated with the door and window structure partition is generated.

[0007] Optionally, based on the topological structure of the abnormal thermal vortex, the phase correlation between the abnormal region delineated by the discontinuous boundary and the pressure data, and in conjunction with the response events of the dynamic tracking spectrum, a three-dimensional seepage model of the sealing defect is established, including: The topological structure of the abnormal heat flow vortex is decomposed into multiple closed loops. Based on the spatial geometric constraints of the abnormal region divided by the discontinuous boundary, the minimum distance between the center point of each closed loop and the edge of the abnormal region is calculated. Based on the phase correlation of the pressure data, the pressure decay process is divided into a linear decay stage and a nonlinear abrupt change stage, and the time domain interval in the nonlinear abrupt change stage where the pressure change rate exceeds a preset value is extracted. The response events of the dynamic tracking map are matched with the time domain interval, and the set of image frames in the thermal imaging sequence within the time domain interval where the heat flow direction changes in reverse is marked; based on the distribution density of the center point of the circular closed path, the spatial superposition relationship between the minimum spacing and the set of image frames, a three-dimensional grid of porous media permeability distribution is constructed. In the three-dimensional mesh, the difference in the angle between the heat flow vectors of adjacent mesh units is calculated. When the difference in the angle exceeds the structural deformation threshold, a seepage channel connecting the annular closed path and the joint of the door and window frame is generated, and finally a three-dimensional seepage model of the sealing defect is established.

[0008] Optionally, based on the leakage equivalent, the geometric parameters, and a preset sealing failure threshold curve, a graded evaluation result associated with the door and window structural zones is generated, including: Map the extension length of the seepage channel in the geometric parameters to the physical dividing line of the structural partition of the door and window frame, and calculate the number of intersections between the seepage channel and the frame joint in each partition. Based on the projection position of the leakage equivalent onto the sealing failure threshold curve, determine the airtightness attenuation coefficient corresponding to each partition; When the number of intersections within the same partition exceeds the structural strength threshold and the attenuation coefficient simultaneously exceeds the critical value, the partition is marked as a high-risk area for sealing failure. Based on the consistency of the seepage channel extension direction between adjacent zones in the high-risk area, a graded assessment result is generated, which includes priority maintenance path indications and is associated with the window and door structural zones.

[0009] Optionally, a dynamic registration relationship is established based on the geometric features of the door and window frame, the thermal conduction baseline, and the thermal conduction abrupt change characteristics, and a spatially constrained displacement vector field is generated according to the dynamic registration relationship, including: Extract the stable heat flow distribution pattern of the corner points and joint lines in the geometric features of the door and window frame in the heat conduction baseline, and generate a reference heat flow direction constraint template; Identify the frame deformation region in the heat conduction abrupt change characteristics, and calculate the matching error between the heat flow direction deviation at the edge of the deformation region and the reference heat flow direction constraint template; When the matching error exceeds the deformation compensation threshold, the vector weight of the reference heat flow direction constraint template is adjusted according to the gradient distribution of the heat conduction abrupt change characteristics. Based on the adjusted vector weights and the geometric topological relationship between the door and window frame joints, a proportional mapping function between thermal displacement and physical deformation is constructed. The thermal flow vector difference between adjacent frames in the thermal image sequence data is converted into a spatial displacement with millimeter-level precision using the scaling mapping function, thereby generating a spatially constrained displacement vector field.

[0010] Optionally, the displacement vector field is input into a preset spatiotemporal feature extraction model, and the heat flow continuity features of the sealed edge region are extracted through a multi-layer feature fusion structure to identify abnormal heat flow vortices and discontinuous boundaries, including: The displacement vector field is input into a preset spatiotemporal feature extraction model, and the vector direction consistency analysis is performed on the strip-shaped regions on both sides of the sealing strip edge through the multi-layer feature fusion structure in the model. The strip region is divided into equally spaced detection units. The dynamic smoothness of the direction angle of the displacement vectors in three consecutive frames in each detection unit is calculated. When the smoothness is lower than the continuity threshold, it is marked as a heat flow break point. Based on the spatial distribution density of the heat flow breakpoints, a discontinuous boundary is generated along the extension direction of the sealing strip, and the geometric accuracy of the boundary profile is corrected according to the rate of change of the breakpoint spacing. The circular closed path of the displacement vector is detected outside the strip region. When more than two non-concentric circular paths are detected and their rotation direction is opposite to the pressure attenuation direction, they are marked as abnormal heat flow vortices. Based on the consistency between the corrected profile of the discontinuous boundary and the rotation direction of the anomalous heat flow vortex, a topology containing the preferred direction of air leakage diffusion is output.

[0011] Optionally, a dynamic data compensation mechanism is triggered during the pressure decay process. When a pressure change is detected, the sampling frequency of the thermal imaging data is increased and a dynamic tracking map is generated, including: The first derivative of the pressure decay rate is monitored in real time. When the absolute value of the derivative exceeds the pressure change threshold, the rapid sampling mode of the thermal imaging acquisition device is activated. In the fast sampling mode, a dynamic data compensation mechanism is triggered to increase the thermal imaging frame rate to a preset sampling rate and simultaneously record the microsecond-level response delay of the pressure sensor. The thermal imaging frame rate boost rate is dynamically adjusted based on the difference between the duration of the pressure mutation threshold and the response delay. After the fast sampling mode ends, the thermal image data with a sampling rate higher than the standard rate are timestamped and aligned to generate a tracking map containing subframe-level thermal flow dynamics.

[0012] Optionally, the initial temperature field distribution is acquired during the low-pressure phase of the pressurization process and used to establish a heat conduction baseline; during the high-pressure phase of the pressurization process, the characteristics of abrupt changes in heat conduction caused by deformation of the sealing structure are captured, including: During the low-pressure phase of the pressurization process, the pressure is maintained within a preset range below the standard test pressure. The initial temperature field distribution is collected, and a heat conduction baseline including the steady-state heat flow diffusion rate at the sealed joint is established. When the pressure during the pressurization process switches to the high-pressure stage, the sudden change in heat conduction characteristics caused by the deformation of the sealing structure are captured, including significant changes in the heat flow rate at the joint and the synchronous reversal of the heat flow direction. Based on the spatiotemporal correlation between the thermal conduction baseline and the thermal conduction abrupt change feature, a deformation compensation parameter incorporating the deformation hysteresis effect of the frame material is established. During the high-pressure stage, when the phase difference between the thermal conduction abrupt change feature and the pressure change is detected to exceed the dynamic tolerance threshold, the iterative update of the deformation compensation parameter is triggered.

[0013] Secondly, this application provides an automated door and window quality inspection system based on machine vision, comprising: The acquisition module is used to simultaneously acquire thermal imaging sequence data and pressure data during the pressurization process of doors and windows. The initial temperature field distribution acquired during the low-pressure stage of the pressurization process is used to establish a heat conduction baseline, and the heat conduction abrupt change characteristics caused by the deformation of the sealing structure are captured during the high-pressure stage of the pressurization process. The generation module is used to perform heat flow correction processing on the thermal image sequence data, establish a dynamic registration relationship based on the geometric features of the door and window frame, the heat conduction baseline and the heat conduction abrupt change features, and generate a displacement vector field with spatial constraints. The identification module is used to input the displacement vector field into the spatiotemporal feature extraction model, extract the heat flow continuity features of the sealed edge region through a multi-layer feature fusion structure, and identify abnormal heat flow vortices and discontinuous boundaries. The compensation module is used to trigger a dynamic data compensation mechanism during the pressure decay process. When a pressure change is detected, the sampling frequency of thermal imaging data is increased and a dynamic tracking map is generated. The output module is used to establish a three-dimensional seepage model of the sealing defect based on the topology of the abnormal heat flow vortex, the abnormal region divided by the discontinuous boundary and the phase correlation of the pressure data, combined with the response event of the dynamic tracking spectrum, and output the geometric parameters and leakage equivalent of the leakage path based on the three-dimensional seepage model. The generation module is also used to generate a graded evaluation result associated with the door and window structure partition based on the leakage equivalent, the geometric parameters and the preset sealing failure threshold curve.

[0014] Thirdly, embodiments of this application provide a computing device, including a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are to be invoked and executed by the processing component to implement an automated door and window quality inspection method based on machine vision as described in the first aspect above.

[0015] Fourthly, embodiments of this application provide a computer storage medium storing a computer program, which, when executed by a computer, implements an automated door and window quality inspection method based on machine vision as described in the first aspect.

[0016] This application embodiment, by simultaneously acquiring thermal image sequences and pressure data during the pressurization process of doors and windows, and combining the establishment of a thermal conduction baseline in the low-pressure stage and the capture of abrupt changes in thermal conduction characteristics in the high-pressure stage, can dynamically correlate pressure deformation and thermodynamic response, accurately distinguishing between elastic deformation of the sealing structure and leakage defect characteristics; by generating a displacement vector field through thermal flow correction and dynamic registration, the spatial image offset caused by the displacement of the door and window frame can be eliminated, and a dynamic mapping relationship between thermal flow characteristics and geometric structure can be established; by using a spatiotemporal feature extraction model to identify abnormal thermal flow vortices and discontinuous boundaries, it breaks through the limitations of single-frame thermal image analysis and enhances the sensitivity to continuous anomalies such as minute thermal flow breaks and vortices; by triggering high-frequency sampling and generating tracking maps through a dynamic data compensation mechanism, it ensures the complete capture of transient thermal flow characteristics during pressure abrupt changes; based on the analysis of the topological structure of abnormal areas and the pressure phase correlation based on a three-dimensional seepage model, it realizes the three-dimensional reconstruction of the leakage path and the physical quantification of the leakage equivalent; finally, by combining the zonal evaluation of the sealing failure threshold curve, it accurately locates the weak leakage areas of the door and window structure, forms an interpretable defect classification conclusion, and significantly improves the detection efficiency and the engineering applicability of defect modeling.

[0017] Furthermore, the construction of the three-dimensional seepage model specifically includes: decomposing the abnormal heat flow vortex into a closed loop and calculating its distance from the defect edge; dividing the time domain interval by combining the nonlinear pressure change stage; matching image frames of reverse heat flow change in the dynamic tracking spectrum; generating a three-dimensional mesh of porous media permeability based on the center point density, spacing, and image frame superposition relationship; and finally determining the seepage channel by the difference in the angle between the heat flow vectors. By decoupling the loop path topology and matching the pressure-heat flow time domain, the limitations of traditional two-dimensional thermal gradient analysis are overcome, and the three-dimensional expansion path of the seepage channel in the porous medium is accurately quantified. At the same time, combined with the deformation threshold dynamic correction model, the reconstruction accuracy and spatial positioning reliability of small tortuous leakage defects are significantly improved.

[0018] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart of an automated door and window quality inspection method based on machine vision provided in this application is shown; Figure 2 The illustration shows a scenario of an automated door and window quality inspection method based on machine vision provided in this application. Figure 3 A schematic diagram of the structure of an automated door and window quality inspection system based on machine vision provided in this application is shown. Figure 4 A schematic diagram of the structure of a computing device provided in this application is shown. Detailed Implementation

[0021] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0022] In some of the processes described in the specification, claims, and accompanying drawings of this application, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a chronological order, nor do they limit "first" and "second" to different types.

[0023] Current technologies for detecting the airtightness of doors and windows, particularly static pressurization schemes based on infrared thermal imaging, suffer from significant bottlenecks. Relying on single-frame thermal images under fixed pressure, they fail to capture the transient deformation characteristics of the sealing structure during dynamic pressurization, leading to the omission of sudden changes in local heat conduction caused by elastic deformation during high-pressure phases. Furthermore, low-sampling-rate thermal image data struggles to resolve anomalies in heat flow continuity (such as vortices and boundary fractures), and static registration methods neglect frame displacement caused by pressure changes, resulting in accumulated spatial alignment errors in the thermal image sequence, further impacting the accuracy of leak path reconstruction. Moreover, existing schemes lack dynamic response mechanisms to pressure surges, failing to simultaneously track the phase relationship between thermodynamic characteristics and pressure decay, ultimately leading to errors in leak equivalent calculations and misjudgments of defects.

[0024] To address the aforementioned shortcomings, this application proposes a method for detecting sealing defects in doors and windows based on dynamic thermo-mechanical coupling modeling. The core of this method lies in the simultaneous acquisition of thermal image sequences and pressure data during the pressurization process. A thermal conduction baseline is established during the low-pressure phase, and abrupt changes in thermal conduction characteristics are captured during the high-pressure phase, thus decoupling the conduction modes of deformation and leakage. Furthermore, a displacement vector field is generated based on dynamic registration of door and window geometric features to eliminate thermal image offset errors caused by frame displacement. A spatiotemporal feature extraction model identifies thermal flow vortices and discontinuous boundaries, and a dynamic data compensation mechanism enhances the ability to capture transient features. Finally, a three-dimensional seepage model is used to correlate the topology and pressure phase of abnormal areas, quantifying the geometric parameters and equivalents of the leakage path, and generating zonal evaluation results by combining the sealing failure threshold curve. This scheme overcomes the limitations of static analysis by accurately reconstructing seepage channels and eliminating registration errors through dynamic thermo-mechanical data fusion and spatiotemporal modeling. Simultaneously, pressure-triggered high-frequency sampling and phase correlation algorithms address the problem of transient signal loss, significantly improving the detection rate of minute defects and the accuracy of leakage assessment, providing reliable technical support for automated quality inspection of door and window airtightness.

[0025] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] Figure 1 A flowchart of an automated door and window quality inspection method based on machine vision is provided in this application embodiment, as shown below. Figure 1 As shown, the method includes: 101. During the pressurization process of doors and windows, thermal imaging sequence data and pressure data are acquired simultaneously. The initial distribution of the temperature field is collected during the low-pressure stage of the pressurization process and used to establish a heat conduction baseline. During the high-pressure stage of the pressurization process, the heat conduction abrupt change characteristics caused by the deformation of the sealing structure are captured. Optionally, step 101 may specifically include the following steps: 1011. During the low-pressure phase of the pressurization process, the pressure is maintained within a preset range below the standard test pressure, the initial temperature field distribution is collected, and a heat conduction baseline including the steady-state heat flow diffusion rate at the sealing joint is established. 1012. When the pressure during the pressurization process switches to the high-pressure stage, capture the thermal conduction abrupt change characteristics caused by the deformation of the sealing structure, including significant changes in the heat flow rate at the detection joint and the synchronous reversal of the heat flow direction. 1013. Based on the spatiotemporal correlation between the thermal conduction baseline and the thermal conduction abrupt change feature, establish deformation compensation parameters that include the deformation hysteresis effect of the frame material; during the high-pressure stage maintenance process, when the phase difference between the thermal conduction abrupt change feature and the pressure change is detected to exceed the dynamic tolerance threshold, trigger the iterative update of the deformation compensation parameters.

[0027] In the above scheme, the heat conduction baseline is a reference model established by maintaining a preset range below the standard test pressure during the low-pressure stage of door and window pressurization. Specifically, it refers to the heat flow diffusion rate calculated by using infrared thermal imager to collect steady-state temperature field data at the sealing joint and combining it with the Fourier heat conduction equation, which is used to characterize the normal heat conduction law under no structural deformation. The heat conduction abrupt change feature is a general term for the abnormal change and reversal of heat flow rate caused by deformation of the sealing structure during the high-pressure stage. For example, the heat flow direction changes abruptly from outward diffusion to inward convergence after the sealing strip is compressed. The deformation compensation parameter refers to the dynamic correction coefficient generated by fitting the pressure-heat conduction phase delay based on the spatiotemporal correlation between the above baseline data and the abrupt change feature, which is used to offset the influence of the deformation hysteresis effect of the frame material on the detection accuracy.

[0028] In this embodiment, firstly, in step 1011, during the low-pressure stage of the pressurization process, the pressure control module is stabilized within a preset range. An infrared thermal imager is used to collect initial temperature field distribution data of the sealing area at a fixed frequency. After eliminating random noise using a mean filtering algorithm, the steady-state heat flow diffusion rate at the sealing joint is calculated based on the Fourier heat conduction equation to generate a heat conduction baseline model. Secondly, in step 1012, when the pressure rises to the high-pressure stage, pressure sensor data and thermal image sequences are collected simultaneously. A heat flow vector field algorithm is used to calculate the heat flow rate and direction of each pixel in the sealing area in real time. If a region's heat flow rate significantly exceeds the baseline threshold and its direction reverses, it is marked as a heat conduction abrupt change feature and its temporal position is recorded. Finally, in step 1013, the baseline data of the heat conduction baseline and the heat conduction abrupt change feature are analyzed in a spatiotemporal correlation to calculate the phase difference between pressure change and heat conduction response. When the phase difference exceeds a preset tolerance threshold, the pressure-heat conduction delay curve is fitted using the least squares method, and the deformation compensation parameters are iteratively updated to correct the phase deviation of subsequent data processing, ultimately forming a closed-loop dynamic correction mechanism.

[0029] In practical applications, taking the airtightness test of an aluminum alloy sliding window as an example, during the low-pressure stage, the pressure is stably controlled within 50% of the standard test pressure. Temperature field distribution data of the sealing strip area is continuously collected by an infrared thermal imager. After filtering and noise reduction, a baseline model of steady-state heat flow diffusion rate is generated based on the heat conduction equation. When the pressure rises to the high-pressure stage, the pressure sensor data and thermal image sequence are monitored simultaneously. A sudden increase in heat flow rate in the sealing strip area at the lower right corner of the window frame is detected, and the direction changes from outward expansion to inward cohesion, which is marked as a heat conduction abrupt change feature. Subsequently, the temporal difference between the pressure rise and the heat flow abrupt change response is analyzed. It is found that the phase difference exceeds the tolerance threshold. The delay curve is fitted by the least squares method and the deformation compensation parameters are updated, thereby correcting the phase deviation between heat conduction data and pressure changes in real time during the subsequent high-pressure maintenance stage.

[0030] This solution establishes a thermal conduction baseline during the low-pressure stage, effectively distinguishing between environmental interference and actual thermal conduction characteristics, providing a reliable benchmark for defect identification. During the high-pressure stage, it dynamically captures the characteristics of sudden changes in heat flow, accurately locating abnormal areas caused by deformation of the sealing structure, avoiding misjudgments of elastic deformation and leakage defects. Combined with the iterative update mechanism of deformation compensation parameters, it corrects the hysteresis effect of pressure-thermal conduction response in real time, significantly improving the sensitivity and location accuracy of defect detection, while ensuring the reliability of data for subsequent seepage model reconstruction.

[0031] 102. Perform heat flow correction processing on the thermal image sequence data, establish a dynamic registration relationship based on the geometric features of the door and window frame, the heat conduction baseline and the heat conduction abrupt change features, and generate a spatially constrained displacement vector field according to the dynamic registration relationship; Optionally, step 102 may specifically include the following steps: 1021. Extract the stable heat flow distribution pattern of the corner points and seam lines in the geometric features of the door and window frame in the heat conduction baseline, and generate a reference heat flow direction constraint template; 1022. Identify the frame deformation region in the heat conduction abrupt change characteristics, and calculate the matching error between the heat flow direction deviation at the edge of the deformation region and the reference heat flow direction constraint template; 1023. When the matching error exceeds the deformation compensation threshold, the vector weight of the reference heat flow direction constraint template is adjusted according to the gradient distribution of the heat conduction abrupt change characteristics. 1024. Based on the adjusted vector weights and the geometric topological relationship between the door and window frame joints, construct a proportional mapping function between thermal displacement and physical deformation. 1025. The thermal flow vector difference between adjacent frames in the thermal image sequence data is converted into a spatial displacement with millimeter-level precision through the scaling mapping function, thereby generating a spatially constrained displacement vector field.

[0032] In the above scheme, the reference heat flow direction constraint template refers to the direction reference template generated based on the stable heat flow distribution pattern of the corner points and seam lines of the door and window frames in the heat conduction baseline during the low-pressure stage. It is used to constrain the deviation of the heat flow direction in the high-pressure deformation area. The matching error refers to the difference between the heat flow direction in the high-pressure deformation area and the reference template, which is quantified by the cosine similarity of the vector angle. The vector weight is a template correction coefficient that is dynamically adjusted according to the gradient distribution of the heat conduction abrupt change characteristics. It is used to balance the difference between the reference template and the local deformation characteristics. The proportional mapping function is a transformation function established based on the geometric topological relationship and the adjusted weight. It maps the heat flow vector difference into physical deformation to generate a displacement field.

[0033] In this embodiment, firstly, step 1021 extracts the stable heat flow distribution pattern of the corner points and seam lines of the window frame in the heat conduction baseline. For example, the Harris corner detection algorithm is used to locate the four corners of the window frame, and a reference heat flow direction constraint template is generated by combining heat flow vector direction consistency analysis. Specifically, the heat flow direction at the four corners of the window frame diffuses outward. Secondly, step 1022 identifies the frame deformation area in the heat conduction abrupt change characteristics during the high-pressure stage. For example, if the heat flow direction in the upper left corner sealing strip area of ​​the window frame suddenly changes to converge inward, the change in heat flow direction at the edge of the deformation area is tracked using optical flow method. The deviation of the heat flow direction at the edge of the deformation area is calculated and the cosine similarity between the deviation of the heat flow direction at the edge of the deformation area and the reference template is calculated to obtain the matching error. When the matching error is lower than a preset threshold, it is determined to be a significant deformation. Then, step 1023 is performed according to the... The vector weights of the reference template are adjusted based on the thermal conduction gradient distribution in the deformation region. For example, when the thermal flux gradient amplitude is high in the upper left corner, the template weight in that region is increased by 1.5 times to enhance the correction priority. Then, in step 1024, based on the adjusted vector weights and the geometric topological relationship between the door and window frame joints (e.g., the joint angle is 90 degrees), the proportional mapping function between thermal flux displacement and physical deformation is constructed by fitting historical deformation data using the least squares method. The thermal flux vector difference is converted into a physical displacement, for example, a thermal flux difference of 5 W·m⁻² corresponds to a deformation of 4 mm. Finally, in step 1025, the thermal flux vector difference between adjacent frames in the thermal image sequence data is input into the proportional mapping function, and the joint continuity constraint is superimposed to generate a displacement vector field with millimeter-level precision. For example, the output shows that the deformation displacement of the upper left corner frame is 4 mm, and the displacement direction is consistent with the joint extension direction.

[0034] In practical applications, taking the high-pressure airtightness test of thermally broken aluminum windows as an example, in the low-pressure stage, Harris corner detection is used to locate the four corners of the window frame and the joint line of the mullion, and the direction of outward heat flow is analyzed to generate a reference template. In the high-pressure stage, the heat flow direction in the upper left corner sealing strip area is detected to suddenly become cohesive. The optical flow method is used to track the deformation edge and calculate the cosine similarity with the reference template. The matching error is significantly lower than the threshold. The template weight is increased based on the regional heat conduction gradient amplitude, and high gradient areas are corrected first. The proportional mapping function is fitted by combining the right-angle topological relationship of the window frame joint to convert the heat flow difference into physical displacement. Finally, the displacement vector field is generated by the joint continuity constraint, showing the inward deformation of the upper left corner frame. The displacement direction is consistent with the extension of the joint, and the millimeter-level spatial positioning of the deformation area is completed.

[0035] This scheme effectively eliminates the interference of frame geometric features on the heat flow direction by generating a reference heat flow direction constraint template and dynamically adjusting the weights, thereby improving the accuracy of deformation region identification. Based on the matching error threshold determination and gradient-driven weight optimization mechanism, it enhances the sensitivity to local deformation features and the priority of correction. Combined with the proportional mapping function of geometric topological constraints, it achieves high-precision conversion of heat flow data to physical displacement, ensuring the spatial consistency of the displacement vector field and the reliability of deformation modeling. This provides accurate spatial deformation input for subsequent seepage path analysis, and improves the robustness and engineering applicability of defect detection as a whole.

[0036] 103. Input the displacement vector field into a preset spatiotemporal feature extraction model, and extract the heat flow continuity features of the sealed edge region through a multi-layer feature fusion structure to identify abnormal heat flow vortices and discontinuous boundaries; Optionally, step 103 may specifically include the following steps: 1031. Input the displacement vector field into a preset spatiotemporal feature extraction model, and perform vector direction consistency analysis on the strip-shaped regions on both sides of the sealing strip edge through the multi-layer feature fusion structure in the model; 1032. Divide the strip-shaped area into equally spaced detection units, and perform dynamic smoothness calculation on the directional angle of three consecutive frames of displacement vectors in each detection unit. When the smoothness is lower than the continuity threshold, it is marked as a heat flow break point. 1033. Based on the spatial distribution density of the heat flow break points, a discontinuous boundary is generated along the extension direction of the sealing strip, and the geometric accuracy of the boundary profile is corrected according to the rate of change of the distance between break points. 1034. Detect the closed-loop path of the displacement vector around the periphery of the strip region. When more than two non-concentric closed-loop paths are detected and their rotation direction is opposite to the pressure attenuation direction, they are marked as abnormal heat flow vortices. 1035. Based on the consistency between the corrected profile of the discontinuous boundary and the rotation direction of the anomalous heat flow vortex, output a topology containing the preferred direction of air leakage diffusion.

[0037] In the above scheme, the strip-shaped region refers to the strip-shaped detection range on both sides of the edge of the sealing strip, which is used to define the analysis area of ​​the heat flow continuity characteristics; the dynamic smoothness calculation is a quantitative index of the fluctuation of the angle between the directions of displacement vectors in continuous multiple frames, which is used to assess the risk of abrupt change or breakage in the heat flow direction; the heat flow break point refers to the local discontinuity marked when the dynamic smoothness is below the threshold; the discontinuity boundary is the defect boundary line extending along the sealing strip based on the spatial density of the heat flow break point; the annular closed path is the annular trajectory formed by the change of continuous vector direction in the displacement vector field, which characterizes the topological structure of the abnormal heat flow vortex; the vortex rotation consistency refers to the physical correlation between the vortex rotation direction and the pressure attenuation direction, which is used to determine the leakage diffusion trend.

[0038] In this embodiment, firstly, step 1031 inputs the displacement vector field into a preset spatiotemporal feature extraction model. For example, a graph convolutional neural network is used to perform consistency analysis on the vector direction of the left strip region of the sealing strip, detecting that the average vector direction of the strip region is outward diffusion, while the right strip region is inward convergence. Secondly, step 1032 divides the strip region into equally spaced detection units. For example, at 5mm intervals, the dynamic smoothness calculation is performed on the direction angle of three consecutive frames of displacement vectors in a certain unit in the lower right corner. The dynamic smoothness calculation is performed on the direction angle of three consecutive frames of displacement vectors in each detection unit. If the angle sequence is 45°, 120°, and 90°, the smoothness calculation value is lower than the threshold 0. .7 is marked as a heat flow breakpoint; then, in step 1033, the spatial distribution density of the heat flow breakpoints is statistically analyzed to generate an initial discontinuous boundary. For example, when the number of breakpoints per unit length is 3 / cm, the jagged edge of the boundary profile is corrected based on the spacing change rate to improve its geometric accuracy; then, in step 1034, a circular closed path is detected on the periphery of the strip region. For example, if two non-concentric circular paths are found and their rotation direction is opposite to the pressure attenuation direction, they are marked as abnormal heat flow vortices; finally, in step 1035, the consistency between the corrected discontinuous boundary profile and the vortex rotation direction is combined. For example, if the counterclockwise diffusion trend of the vortex matches the boundary break direction, the topology structure in which the air leakage path preferentially diffuses along the lower right boundary is output.

[0039] In practical applications, taking the airtightness testing of thermally broken aluminum windows and doors as an example, the displacement vector field was analyzed using a spatiotemporal feature extraction model. It was found that the vector direction of the left sealing strip strip area was highly consistent, while there were local abrupt changes in direction on the right side. In the detection unit divided in the lower right corner, the angle of the displacement vector direction fluctuated significantly over three consecutive frames, and the dynamic smoothness calculation value was lower than the threshold, which was marked as a thermal flow breakpoint. After generating an initial discontinuous boundary based on the breakpoint distribution density, its sawtooth contour was corrected according to the spacing change rate. Two non-concentric ring paths were detected on the periphery of the strip area, and their rotation direction was opposite to the pressure attenuation direction, which was determined to be abnormal thermal flow vortices. Finally, by combining the consistency between the corrected boundary and the vortex rotation direction, the topology of the leakage path preferentially spreading along the lower right corner boundary was output, clarifying the preferred direction of leakage diffusion.

[0040] This solution accurately locates abnormal heat flow regions at the edge of the sealing strip through multi-layer feature fusion and vector direction consistency analysis; it determines heat flow breakpoints based on dynamic smoothness thresholds, effectively identifying microscopic directional abrupt defects; it improves geometric accuracy by combining breakpoint density and spacing to correct discontinuous boundary contours; and it reconstructs the diffusion trend of abnormal heat flow vortices through ring-shaped closed path detection and vortex consistency matching. Ultimately, it achieves topological modeling of sealing defects and dynamic prediction of leakage paths, significantly enhancing the robustness and reliability of detection results in complex leakage scenarios.

[0041] 104. Trigger a dynamic data compensation mechanism during pressure decay. When a sudden pressure change is detected, increase the sampling frequency of thermal imaging data and generate a dynamic tracking map. Optionally, step 104 may specifically include the following steps: 1041. Monitor the change of the first derivative of the pressure decay rate in real time. When the absolute value of the derivative exceeds the pressure change threshold, start the fast sampling mode of the thermal imaging acquisition device. 1042. In the fast sampling mode, a dynamic data compensation mechanism is triggered to increase the thermal imaging frame rate to a preset sampling rate and simultaneously record the microsecond-level response delay of the pressure sensor. 1043. Based on the difference between the duration of the pressure mutation threshold and the response delay, dynamically adjust the boost rate of the thermal imaging frame rate; 1044. After the fast sampling mode ends, the thermal image data with a sampling rate higher than the standard sampling rate are timestamped and aligned to generate a tracking map containing subframe-level thermal flow dynamics.

[0042] In the above scheme, the first derivative of the pressure decay rate refers to the instantaneous slope of the pressure change over time, which is used to quantify the intensity of pressure mutation; the fast sampling mode is the high-frequency sampling state of the thermal imaging device triggered when the pressure changes, which captures transient heat flow changes by increasing the frame rate; the microsecond-level response delay refers to the signal transmission time deviation between the pressure sensor and the thermal imaging device, which needs to be synchronously calibrated to ensure data synchronization; the timestamp interpolation alignment is a technique that uses algorithms to align high-frequency thermal imaging data with the standard time axis to generate a sub-frame-level accurate dynamic heat flow tracking map.

[0043] In this embodiment, firstly, step 1041 monitors the change of the first derivative of the pressure decay rate in real time. For example, when the slope of the pressure change suddenly increases from -5 kPa / s to -15 kPa / s, if the absolute value exceeds the threshold of 10 kPa / s, the rapid sampling mode of the thermal imaging acquisition device is immediately triggered. Secondly, in step 1042, under the rapid sampling mode, a dynamic data compensation mechanism is triggered to increase the thermal imaging frame rate to a preset sampling rate, for example, increasing the thermal imaging frame rate from 30 Hz to 120 Hz. For example, during a sudden pressure drop in aluminum alloy doors and windows, the response delay of the pressure sensor is simultaneously recorded as 500 μs and stored as time. The calibration parameters are then adjusted. Next, in step 1043, the frame rate boost rate is dynamically adjusted based on the duration of the pressure change and the delay difference. For example, when the pressure change lasts for 80ms and the delay is 500μs, the frame rate is boosted to 150 frames per second to match the dynamic characteristics of pressure decay. Finally, in step 1044, a cubic spline interpolation algorithm is used to align the timestamps of the high-frequency thermal imaging data. For example, 100 frames of data acquired at 120Hz are interpolated and mapped to a standard 30Hz time axis to generate a tracking map showing the dynamic process of the heat flow direction changing from outward expansion to inward cohesion at the window frame seam within 3ms, thus fully restoring the thermodynamic response details of the transient leakage event.

[0044] In practical applications, taking the pressure decay detection of an aluminum alloy sliding window as an example, the system monitors in real time that the first derivative of the pressure decay rate suddenly drops from -8 kPa / s to -18 kPa / s, exceeding the preset threshold of 10 kPa / s, triggering the rapid sampling mode of the thermal imaging device; the frame rate is increased from 30 Hz to 120 Hz, and the pressure sensor response delay is recorded as 550 μs; the frame rate is dynamically adjusted to 140 frames per second based on the difference between the pressure change duration of 70 ms and the delay; after sampling, the high-frequency data is aligned with the standard time axis through cubic spline interpolation, generating a tracking graph that shows the heat flow rate of the sealing strip in the lower left corner of the window frame suddenly increases and reverses direction within 2.5 ms, fully capturing the dynamic details of the heat flow in the transient leakage event.

[0045] This solution accurately captures transient heat flow characteristics during pressure surges through a dynamic data compensation mechanism, eliminating timing deviations caused by equipment delays. Combined with adaptive frame rate adjustment and timestamp interpolation alignment, it achieves sub-frame level dynamic heat flow restoration and improved data synchronization, ensuring strict spatiotemporal correlation between pressure and heat flow responses. Ultimately, it enhances the detection sensitivity and data integrity of transient leakage events, providing high-precision transient process data support for dynamic modeling of sealing defects and leakage path analysis, significantly improving the reliability of defect diagnosis under complex operating conditions.

[0046] 105. Based on the topological structure of the abnormal heat flow vortex, the phase correlation between the abnormal region divided by the discontinuous boundary and the pressure data, and combined with the response events of the dynamic tracking spectrum, a three-dimensional seepage model of the sealing defect is established, and the geometric parameters and leakage equivalent of the leakage path are output according to the three-dimensional seepage model. Optionally, step 105 may specifically include the following steps: 1051. Decompose the topological structure of the abnormal heat flow vortex into multiple closed loops. Based on the spatial geometric constraints of the abnormal region divided by the discontinuous boundary, calculate the minimum distance between the center point of each closed loop and the edge of the abnormal region. 1052. Based on the phase correlation of the pressure data, the pressure decay process is divided into a linear decay stage and a nonlinear abrupt change stage, and the time domain interval in the nonlinear abrupt change stage where the pressure change rate exceeds a preset value is extracted. 1053. Match the response events of the dynamic tracking map with the time domain interval, and mark the set of image frames in the thermal imaging sequence within the time domain interval where the heat flow direction changes in reverse; based on the distribution density of the center point of the circular closed path, the spatial superposition relationship between the minimum spacing and the set of image frames, construct a three-dimensional grid of porous media permeability distribution; 1054. Calculate the difference in the angle between the heat flow vectors of adjacent grid cells in the three-dimensional mesh. When the difference in the angle exceeds the structural deformation threshold, generate a seepage channel connecting the annular closed path and the joint of the door and window frame, and finally establish a three-dimensional seepage model of the sealing defect.

[0047] In the above scheme, the annular closed path refers to the ring-shaped trajectory formed by the continuous vector direction change in the abnormal heat flow vortex; the phase correlation characterizes the matching relationship between the pressure decay time domain characteristics and the heat flow response; the three-dimensional mesh is a three-dimensional unit structure based on spatial topological constraints, used to simulate the permeability distribution; the heat flow vector angle difference is a quantitative index of the directional difference between adjacent units, used to determine the formation of seepage channels.

[0048] In this embodiment, firstly, step 1051 decomposes the topological structure of the abnormal heat flow vortex into multiple closed annular paths. For example, Hough transform is used to detect three annular contours, and the minimum distance from the center point to the boundary of each closed path is calculated based on the spatial geometric constraints of the abnormal region divided by discontinuous boundaries. For example, center point A is 1.5 mm from the boundary, and center point B is 2.0 mm from the boundary. Secondly, step 1052 divides the pressure decay process into linear and nonlinear stages based on the phase correlation of the pressure data. For example, through the second derivative analysis of the pressure-time curve, the time domain interval of 10ms~35ms for the nonlinear abrupt change stage with a pressure change rate exceeding 18kPa / s is extracted. Then, step 1053 matches the response events in the dynamic tracking spectrum with the above time domain interval and marks them. The set of image frames in the thermal imaging sequence within the time domain interval where the heat flow direction reverses abruptly is used. Based on the distribution density of the center point of the closed ring path, the minimum spacing, and the spatial superposition relationship of the image frame set, for example, 5 frames of data with reversed heat flow direction are marked within 10ms~35ms. Combined with the distribution density of the center point of the ring path, such as 3 points / cm², and the minimum spacing, a three-dimensional grid is constructed, for example, the permeability distribution is divided into 1mm³ units. Finally, the difference in the angle between the heat flow vectors of adjacent grid units is calculated in step 1054. For example, when the angle between the vectors of adjacent units is 60° and exceeds the deformation threshold of 45°, an S-shaped seepage channel connecting the ring path and the window frame joint is generated. For example, the path extends from the center point B to the lower right corner of the window joint. The length of the leakage path and the leakage equivalent parameters are output, and finally a three-dimensional seepage model of the sealing defect is established.

[0049] In practical applications, taking the detection of sealing defects in a thermally broken aluminum window as an example, the abnormal heat flow vortex is decomposed into three closed annular paths using Hough transform. The minimum distances from the center point of each path to the discontinuity boundary are calculated to be 1.5mm, 2.0mm, and 1.8mm, respectively. Based on the second derivative analysis of the pressure decay data, the nonlinear abrupt change stage is divided, and the time domain interval where the pressure change rate exceeds the preset value is extracted. Five frames of thermal image data with reversed heat flow direction in the dynamic tracking spectrum during this period are matched, and a three-dimensional mesh is constructed by combining the distribution density of the center point of the annular path with the minimum distance. When the difference in the angle between the heat flow vectors of adjacent units exceeds the deformation threshold, an S-shaped seepage channel connecting the annular path and the window frame joint is generated. The geometric parameters of the leakage path and the leakage equivalent are output, clarifying the leakage diffusion path and the severity of the defect. This solution accurately locates the spatial expansion trend of seepage paths through ring path topology decomposition and boundary constraint analysis; it achieves strict spatiotemporal correlation of pressure-heat flow response by combining pressure phase correlation division and heat flow dynamic matching; and it reconstructs the seepage channel from the sealing defect area to the frame joint based on three-dimensional mesh permeability modeling and heat flow vector difference determination, providing a physical quantitative basis for the geometric parameters of the leakage path and the leakage equivalent. Ultimately, it realizes three-dimensional visualization modeling and dynamic diffusion simulation of leakage defects, significantly improving the engineering interpretability of defect diagnosis and the reliability of maintenance decisions.

[0050] 106. Based on the leakage equivalent, the geometric parameters, and the preset sealing failure threshold curve, generate a graded evaluation result associated with the door and window structure partition.

[0051] Optionally, step 106 may specifically include the following steps: 1061. Map the extension length of the seepage channel in the geometric parameters to the physical dividing line of the structural partition of the door and window frame, and calculate the number of intersections between the seepage channel and the frame joint in each partition. 1062. Based on the projection position of the leakage equivalent on the sealing failure threshold curve, determine the airtightness attenuation coefficient corresponding to each partition; 1063. When the number of intersections within the same partition exceeds the structural strength threshold and the attenuation coefficient simultaneously exceeds the critical value, the partition is marked as a high-risk area for sealing failure. 1064. Based on the consistency of the extension direction of the seepage channels in the adjacent zones of the high-risk area, generate a graded assessment result that includes priority maintenance path indications and is associated with the door and window structure zones.

[0052] In the above scheme, the physical dividing line of the structural partition refers to the boundary line of the area divided by the door and window frame according to mechanical characteristics or function, such as the four corners of the window frame, the mullion joint, etc.; the air tightness level attenuation coefficient is a quantitative index of the decrease in sealing performance calculated based on the projection position of the leakage equivalent on the sealing failure threshold curve; the structural strength threshold is the upper limit of the number of intersections between the allowable seepage channels and joints in the door and window partition, exceeding this value is judged as a risk of structural failure; the priority maintenance path indication is a maintenance priority suggestion generated based on the consistency of the extension direction of the seepage channels in adjacent partitions of high-risk areas, such as prioritizing the repair of areas with concentrated diffusion directions.

[0053] In this embodiment, firstly, step 1061 maps the extension length of the seepage channel in the geometric parameters to the physical dividing line of the window and door structure partition. For example, the thermally broken aluminum window frame is divided into four partitions: upper left, upper right, lower left, and lower right. Based on the geometric topology of the joints, such as the mullion joint line, the number of intersections between the seepage channel and the joint in each partition is calculated. For example, two intersections between the seepage channel and the joint are detected in the lower right partition. Secondly, step 1062 determines the airtightness attenuation coefficient based on the projection position of the leakage equivalent on the sealing failure threshold curve. For example, when the leakage equivalent is 0.75 m³ / h·m², the projection on the threshold curve is located in the medium-high risk transition range, and the attenuation coefficient is calculated to be 0.82 using linear interpolation. Then... Step 1063 determines whether the number of intersections within a zone exceeds the structural strength threshold. When the number of intersections within the same zone exceeds the structural strength threshold and the attenuation coefficient simultaneously exceeds the critical value, the zone is marked as a high-risk area for sealing failure. For example, if the threshold is set to 2 and the attenuation coefficient exceeds the critical value of 0.8, the lower right zone may have 2 intersections equal to the threshold but an attenuation coefficient of 0.82, which is still marked as a high-risk area. Finally, step 1064 analyzes the consistency of the seepage channel extension direction between adjacent zones of the high-risk area. For example, if the cosine similarity between the seepage direction of the lower right zone and the lower left zone reaches 0.9, a priority maintenance path is generated from the lower right to the lower left, and the graded evaluation results are output in conjunction with the window and door structural zones.

[0054] In practical applications, taking the airtightness assessment of an aluminum alloy casement window as an example, if the number of intersections between the seepage channel extending to the lower right corner of the window frame and the joint exceeds the structural strength threshold, the leakage equivalent is projected onto the high-risk area of ​​the sealing failure threshold curve, and the corresponding airtightness attenuation coefficient exceeds the critical value, the area is marked as a high-risk area. The seepage channel of the adjacent upper right corner area extends in the same direction as the high-risk area and has a high similarity in direction cosine, generating a priority maintenance path indicator from the lower right corner to the upper right corner, guiding the engineering team to prioritize the handling of continuous leakage diffusion paths across zones.

[0055] This solution accurately locates high-risk sealing failure areas through spatial mapping of seepage channels and structural zones and dynamic calculation of airtightness attenuation coefficients. By combining seepage direction consistency analysis and maintenance path priority generation, abstract leakage parameters are transformed into actionable engineering decision-making basis. Ultimately, it achieves visualized expression of defect classification and precise allocation of maintenance resources, significantly improving the efficiency of door and window maintenance and the economy of defect treatment, and ensuring the long-term reliability of building airtightness performance.

[0056] Figure 2 This application provides a scenario diagram for an automated door and window quality inspection method based on machine vision, as shown in the embodiment. Figure 2 As shown, a complete embodiment of steps 101-106 includes: In the airtightness test of the thermally broken aluminum casement window, step 101 establishes a thermal conduction baseline for the window frame sealing strip area at a low pressure stage of 300 Pa, with a heat flow diffusion rate of 0.12 W / (m·K). At a high pressure stage of 600 Pa, a sudden increase in the heat flow rate in the upper right corner to 0.38 W / (m·K) with a reversed direction is detected, triggering iterative updates of the deformation compensation parameters. Step 102 calculates the deviation of the upper right corner deformation area to 65% based on the corner heat flow direction constraint template, generating a displacement vector field showing a frame shrinkage deformation of 3.5 mm. Step 103 divides the sealing strip area into detection units, finding that the dynamic smoothness of the upper right corner unit is 0.6, which is below the threshold and is marked as a heat flow breakpoint. At the same time, two non-concentric annular paths are detected with rotation directions opposite to the pressure attenuation direction, which are determined to be abnormal heat. Step 104: During the pressure change phase, at -20 kPa / s, the thermal imaging frame rate is increased to 120 Hz, the sensor delay is recorded as 500 μs, and the interpolation generates a tracking map showing that the heat flow direction reverses within 2 ms in the upper right corner; Step 105: The abnormal vortex is decomposed into two closed annular paths with a distance of 1.8 mm from the center point to the fracture boundary. A three-dimensional mesh is constructed and the difference in the heat flow angle between adjacent units is calculated to be 55°, generating an S-shaped seepage channel and outputting a leakage equivalent of 0.75 m³ / (h·m²); Step 106: The seepage channel is extended to the upper right corner partition, with 4 intersection points exceeding the threshold of 2. The leakage equivalent is projected onto the high-risk interval with a corresponding attenuation coefficient of 0.88. Combined with the cosine similarity of the seepage direction in the adjacent lower right corner (0.91), a maintenance path "upper right corner → lower right corner" is generated. Ultimately, it achieves full-process analysis from dynamic data acquisition to 3D modeling of leakage paths. Through thermo-mechanical coupling response and seepage topology reconstruction, it accurately locates high-risk areas and guides engineering maintenance, significantly improving defect diagnosis efficiency and building airtightness maintenance accuracy.

[0057] This solution employs a multi-step collaborative technology to achieve a closed-loop analysis across the entire chain, from dynamic data acquisition to defect classification and assessment: synchronous acquisition of thermo-mechanical data and baseline modeling eliminate environmental interference and accurately quantify the deformation characteristics of the sealing structure; high-frequency sampling and spatiotemporal feature extraction capture transient thermal flow fractures and abnormal vortex details; a three-dimensional seepage model reconstructs the topology of the leakage path and outputs quantified leakage parameters; and combined with zoning threshold determination and directional consistency analysis, maintenance priority decisions are generated. Ultimately, the abstract thermodynamic response is transformed into an engineering-executable defect location and maintenance strategy, significantly improving the accuracy of airtightness testing, the efficiency of defect diagnosis, and the rationality of maintenance resource allocation, providing reliable technical support for the long-term assurance of building energy conservation and sealing performance.

[0058] Figure 3 This application provides a schematic diagram of the structure of an automated door and window quality inspection system based on machine vision, as shown in the embodiment of the present application. Figure 3 As shown, the system includes: The acquisition module 31 is used to simultaneously acquire thermal imaging sequence data and pressure data during the pressurization process of doors and windows. The initial distribution of the temperature field acquired during the low-pressure stage of the pressurization process is used to establish a heat conduction baseline, and the heat conduction abrupt change characteristics caused by the deformation of the sealing structure are captured during the high-pressure stage of the pressurization process. The generation module 32 is used to perform heat flow correction processing on the thermal image sequence data, establish a dynamic registration relationship based on the geometric features of the door and window frame, the heat conduction baseline and the heat conduction abrupt change features, and generate a displacement vector field with spatial constraints. The identification module 33 is used to input the displacement vector field into the spatiotemporal feature extraction model, extract the heat flow continuity features of the sealed edge region through a multi-layer feature fusion structure, and identify abnormal heat flow vortices and discontinuous boundaries. Compensation module 34 is used to trigger a dynamic data compensation mechanism during pressure decay. When a pressure change is detected, it increases the sampling frequency of thermal imaging data and generates a dynamic tracking map. Output module 35 is used to establish a three-dimensional seepage model of the sealing defect based on the topology of the abnormal heat flow vortex, the abnormal region divided by the discontinuous boundary and the phase correlation of the pressure data, combined with the response event of the dynamic tracking spectrum, and output the geometric parameters and leakage equivalent of the leakage path based on the three-dimensional seepage model. The generation module 32 is also used to generate a graded evaluation result associated with the door and window structure partition based on the leakage equivalent, the geometric parameters and the preset sealing failure threshold curve.

[0059] Figure 3 The aforementioned automated door and window quality inspection system based on machine vision can perform... Figure 1The implementation principle and technical effects of the automated door and window quality inspection method based on machine vision described in the illustrated embodiment will not be repeated here. The specific methods by which each module and unit of the automated door and window quality inspection system based on machine vision in the above embodiments are described in detail in the embodiments related to this method, and will not be elaborated upon here.

[0060] In one possible design, Figure 3 The automated door and window quality inspection system based on machine vision shown in the embodiment can be implemented as a computing device, such as... Figure 3 As shown, the computing device may include a storage component 41 and a processing component 42; The storage component 41 stores one or more computer instructions, wherein the one or more computer instructions are invoked and executed by the processing component 42.

[0061] The processing component 42 is used for the above Figure 1 The embodiment describes an automated door and window quality inspection method based on machine vision.

[0062] The processing component 42 may include one or more processors to execute computer instructions to complete all or part of the steps in the above-described method. Alternatively, the processing component may be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above-described method.

[0063] Storage component 41 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0064] Of course, computing devices may also include other components, such as input / output interfaces, display components, communication components, etc.

[0065] Input / output interfaces provide interfaces between processing components and peripheral interface modules, which can be output devices, input devices, etc.

[0066] The communication components are configured to facilitate wired or wireless communication between computing devices and other devices.

[0067] The computing device can be a physical device or an elastic computing host provided by a cloud computing platform. In this case, the computing device can refer to a cloud server, and the aforementioned processing components, storage components, etc., can be basic server resources rented or purchased from the cloud computing platform.

[0068] This application also provides a computer storage medium storing a computer program, which, when executed by a computer, can perform the above-described functions. Figure 1 The illustrated embodiment presents an automated door and window quality inspection method based on machine vision.

[0069] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0070] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0071] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0072] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. An automated door and window quality inspection method based on machine vision, characterized in that, include: During the pressurization process of doors and windows, thermal imaging sequence data and pressure data are acquired simultaneously. The initial distribution of the temperature field is collected during the low-pressure stage of the pressurization process and used to establish the heat conduction baseline. During the high-pressure stage of the pressurization process, the heat conduction abrupt change characteristics caused by the deformation of the sealing structure are captured. The thermal imaging sequence data is subjected to thermal flow correction processing. A dynamic registration relationship is established based on the geometric features of the door and window frame, the thermal conduction baseline, and the thermal conduction abrupt change features. A spatially constrained displacement vector field is generated based on the dynamic registration relationship. The displacement vector field is input into a preset spatiotemporal feature extraction model, and the heat flow continuity features of the sealed edge region are extracted through a multi-layer feature fusion structure to identify abnormal heat flow vortices and discontinuous boundaries. A dynamic data compensation mechanism is triggered during the pressure decay process. When a sudden pressure change is detected, the sampling frequency of thermal imaging data is increased and a dynamic tracking map is generated. Based on the topology of the abnormal heat flow vortex, the phase correlation between the abnormal region divided by the discontinuous boundary and the pressure data, and combined with the response events of the dynamic tracking spectrum, a three-dimensional seepage model of the sealing defect is established, and the geometric parameters and leakage equivalent of the leakage path are output according to the three-dimensional seepage model. Based on the leakage equivalent, the geometric parameters, and the preset sealing failure threshold curve, a graded evaluation result associated with the door and window structure partition is generated.

2. The method according to claim 1, characterized in that, Based on the topological structure of the abnormal thermal flow vortex, the phase correlation between the abnormal region delineated by the discontinuous boundary and the pressure data, and combined with the response events of the dynamic tracking spectrum, a three-dimensional seepage model of the sealing defect is established, including: The topological structure of the abnormal heat flow vortex is decomposed into multiple closed loops. Based on the spatial geometric constraints of the abnormal region divided by the discontinuous boundary, the minimum distance between the center point of each closed loop and the edge of the abnormal region is calculated. Based on the phase correlation of the pressure data, the pressure decay process is divided into a linear decay stage and a nonlinear abrupt change stage, and the time domain interval in the nonlinear abrupt change stage where the pressure change rate exceeds a preset value is extracted. The response events of the dynamic tracking map are matched with the time domain interval, and the set of image frames in the thermal imaging sequence within the time domain interval where the heat flow direction changes in reverse is marked; based on the distribution density of the center point of the circular closed path, the spatial superposition relationship between the minimum spacing and the set of image frames, a three-dimensional grid of porous media permeability distribution is constructed. In the three-dimensional mesh, the difference in the angle between the heat flow vectors of adjacent mesh units is calculated. When the difference in the angle exceeds the structural deformation threshold, a seepage channel connecting the annular closed path and the joint of the door and window frame is generated, and finally a three-dimensional seepage model of the sealing defect is established.

3. The method according to claim 1, characterized in that, Based on the leakage equivalent, the geometric parameters, and the preset sealing failure threshold curve, a graded evaluation result associated with the door and window structural zones is generated, including: Map the extension length of the seepage channel in the geometric parameters to the physical dividing line of the structural partition of the door and window frame, and calculate the number of intersections between the seepage channel and the frame joint in each partition. Based on the projection position of the leakage equivalent onto the sealing failure threshold curve, determine the airtightness attenuation coefficient corresponding to each partition; When the number of intersections within the same partition exceeds the structural strength threshold and the attenuation coefficient simultaneously exceeds the critical value, the partition is marked as a high-risk area for sealing failure. Based on the consistency of the seepage channel extension direction between adjacent zones in the high-risk area, a graded assessment result is generated, which includes priority maintenance path indications and is associated with the window and door structural zones.

4. The method according to claim 1, characterized in that, A dynamic registration relationship is established based on the geometric features of the door and window frame, the thermal conduction baseline, and the abrupt change characteristics of thermal conduction. A spatially constrained displacement vector field is then generated based on this dynamic registration relationship, including: Extract the stable heat flow distribution pattern of the corner points and joint lines in the geometric features of the door and window frame in the heat conduction baseline, and generate a reference heat flow direction constraint template; Identify the frame deformation region in the heat conduction abrupt change characteristics, and calculate the matching error between the heat flow direction deviation at the edge of the deformation region and the reference heat flow direction constraint template; When the matching error exceeds the deformation compensation threshold, the vector weight of the reference heat flow direction constraint template is adjusted according to the gradient distribution of the heat conduction abrupt change characteristics. Based on the adjusted vector weights and the geometric topological relationship between the door and window frame joints, a proportional mapping function between thermal displacement and physical deformation is constructed. The thermal flow vector difference between adjacent frames in the thermal image sequence data is converted into a spatial displacement with millimeter-level precision using the scaling mapping function, thereby generating a spatially constrained displacement vector field.

5. The method according to claim 1, characterized in that, The displacement vector field is input into a preset spatiotemporal feature extraction model. Through a multi-layer feature fusion structure, the heat flow continuity characteristics of the sealed edge region are extracted, identifying abnormal heat flow vortices and discontinuous boundaries, including: The displacement vector field is input into a preset spatiotemporal feature extraction model, and the vector direction consistency analysis is performed on the strip-shaped regions on both sides of the sealing strip edge through the multi-layer feature fusion structure in the model. The strip region is divided into equally spaced detection units. The dynamic smoothness of the direction angle of the displacement vectors in three consecutive frames in each detection unit is calculated. When the smoothness is lower than the continuity threshold, it is marked as a heat flow break point. Based on the spatial distribution density of the heat flow breakpoints, a discontinuous boundary is generated along the extension direction of the sealing strip, and the geometric accuracy of the boundary profile is corrected according to the rate of change of the breakpoint spacing. The circular closed path of the displacement vector is detected outside the strip region. When more than two non-concentric circular paths are detected and their rotation direction is opposite to the pressure attenuation direction, they are marked as abnormal heat flow vortices. Based on the consistency between the corrected profile of the discontinuous boundary and the rotation direction of the anomalous heat flow vortex, a topology containing the preferred direction of air leakage diffusion is output.

6. The method according to claim 1, characterized in that, A dynamic data compensation mechanism is triggered during pressure decay. When a sudden pressure change is detected, the sampling frequency of thermal imaging data is increased and a dynamic tracking map is generated, including: The first derivative of the pressure decay rate is monitored in real time. When the absolute value of the derivative exceeds the pressure change threshold, the rapid sampling mode of the thermal imaging acquisition device is activated. In the fast sampling mode, a dynamic data compensation mechanism is triggered to increase the thermal imaging frame rate to a preset sampling rate and simultaneously record the microsecond-level response delay of the pressure sensor. The thermal imaging frame rate boost rate is dynamically adjusted based on the difference between the duration of the pressure mutation threshold and the response delay. After the fast sampling mode ends, the thermal image data with a sampling rate higher than the standard rate are timestamped and aligned to generate a tracking map containing subframe-level thermal flow dynamics.

7. The method according to claim 1, characterized in that, The initial temperature field distribution is acquired during the low-pressure phase of the pressurization process and used to establish a heat conduction baseline. During the high-pressure phase of the pressurization process, the abrupt changes in heat conduction caused by deformation of the sealing structure are captured, including: During the low-pressure phase of the pressurization process, the pressure is maintained within a preset range below the standard test pressure. The initial temperature field distribution is collected, and a heat conduction baseline including the steady-state heat flow diffusion rate at the sealed joint is established. When the pressure during the pressurization process switches to the high-pressure stage, the sudden change in heat conduction characteristics caused by the deformation of the sealing structure are captured, including significant changes in the heat flow rate at the joint and the synchronous reversal of the heat flow direction. Based on the spatiotemporal correlation between the thermal conduction baseline and the thermal conduction abrupt change feature, a deformation compensation parameter incorporating the deformation hysteresis effect of the frame material is established. During the high-pressure stage, when the phase difference between the thermal conduction abrupt change feature and the pressure change is detected to exceed the dynamic tolerance threshold, the iterative update of the deformation compensation parameter is triggered.

8. An automated door and window quality inspection system based on machine vision, characterized in that, include: During the pressurization process of doors and windows, thermal imaging sequence data and pressure data are acquired simultaneously. The initial temperature field distribution collected during the low-pressure stage of the pressurization process is used to establish a heat conduction baseline. During the high-pressure stage of the pressurization process, the heat conduction abrupt change characteristics caused by the deformation of the sealing structure are captured. The thermal imaging sequence data is subjected to heat flow correction processing. A dynamic registration relationship is established based on the geometric features of the door and window frame, the heat conduction baseline, and the heat conduction abrupt change features to generate a displacement vector field with spatial constraints. The displacement vector field is input into the spatiotemporal feature extraction model, and the heat flow continuity features of the sealed edge region are extracted through a multi-layer feature fusion structure to identify abnormal heat flow vortices and discontinuous boundaries. A dynamic data compensation mechanism is triggered during the pressure decay process. When a sudden pressure change is detected, the sampling frequency of thermal imaging data is increased and a dynamic tracking map is generated. Based on the topology of the abnormal heat flow vortex, the phase correlation between the abnormal region divided by the discontinuous boundary and the pressure data, and combined with the response events of the dynamic tracking spectrum, a three-dimensional seepage model of the sealing defect is established, and the geometric parameters and leakage equivalent of the leakage path are output according to the three-dimensional seepage model. Based on the leakage equivalent, the geometric parameters, and the preset sealing failure threshold curve, a graded evaluation result associated with the door and window structure partition is generated.

9. A computing device, characterized in that, It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement an automated door and window quality inspection method based on machine vision as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that, The device contains a computer program that, when executed by a computer, implements an automated door and window quality inspection method based on machine vision as described in any one of claims 1 to 7.

Images