Super-large fireproof door copper veneer fitting quality detection method based on thermal diffusion modeling

Abstract

The invention discloses an oversized fireproof door copper veneer fitting quality detection method based on thermal diffusion modeling, and belongs to the technical field of building construction, and the technical scheme is as follows: exciting a copper veneer fitting area through an array type pulse heat source, and setting excitation power and duration; collecting a thermal diffusion image sequence by a thermal imager; modeling is carried out on the temperature change of each pixel point on a time axis in each obtained thermogram sequence, and a pixel-level time-temperature fitting model is constructed; identifying a thermal diffusion abnormal region in the atlas based on a convolutional neural network and a time sequence model, and generating a defect probability graph; and extracting defect area position and boundary information according to the defect probability graph, and outputting a labeled image and structured defect data. The method has the beneficial effect that the method for detecting the lamination quality of the copper veneer of the oversized fireproof door based on thermal diffusion modeling is provided.

Description

Technical Field

[0001] The present invention belongs to the technical field of building construction, and in particular relates to a method for detecting the lamination quality of copper facings of super-large fire doors based on thermal diffusion modeling. Background Art

[0002] Oversized fire doors are widely used in industry and construction, often coated with copper cladding for both decorative purposes and enhanced fire protection. However, uneven adhesive bonding and insufficient pressure-bonding between the copper cladding and the base layer often result in hollows or unbonded areas, seriously compromising safety and fire resistance. Traditional inspection methods such as manual tapping, ultrasonic, and visual inspections are inefficient, subjective, and unsuitable for large structures.

[0003] To this end, the present invention proposes a method for detecting the lamination quality of copper facings of super-large fire doors based on thermal diffusion modeling. Summary of the Invention

[0004] The object of the present invention is to provide a method for detecting the lamination quality of copper facings of super-large fire doors based on thermal diffusion modeling.

[0005] The present invention is achieved through the following measures: The present invention is achieved through the following measures: A method for detecting the fitting quality of copper veneers of super-large fire doors based on thermal diffusion modeling, characterized in that it includes: exciting the copper veneer fitting area through an array pulse heat source, and setting the excitation power and duration; acquiring a thermal diffusion image sequence with a thermal imager, and performing synchronous control of thermal excitation and image acquisition; modeling the temperature change of each pixel point on the time axis in each acquired thermal image sequence, and constructing a pixel-level time-temperature fitting model; identifying abnormal thermal diffusion areas in the map based on a convolutional neural network and a time series model, and generating a defect probability map; extracting the position and boundary information of the defect area according to the defect probability map, and outputting the labeled image and structured defect data Furthermore, an array-type pulse heat source is arranged directly above the detection surface through a scanning platform, and the array-type pulse heat source excites the copper finish bonding area and is arranged directly above the copper finish to be tested.

[0006] Furthermore, the excitation power and duration are set, including: Get copper finish density Specific heat capacity ,thickness , adhesive density Specific heat capacity ,thickness , (density in kilograms per cubic meter, specific heat in joules per kilogram Kelvin, thickness in meters), detection area (unit: square meters), target temperature rise (equal to the difference between the excitation target temperature and the initial ambient temperature, in degrees Celsius), is the excitation energy transfer efficiency coefficient (indicating the energy coupling efficiency between the heat source and the material interface, with a value range of 0.5 to 0.9), The heat loss adjustment factor is equal to (in is the interfacial heat transfer coefficient in watts per square meter Kelvin, is the total thermal resistance of the interface, expressed in Kelvin per watt). The target excitation temperature should be able to raise the temperature of the copper surface and the adhesive layer to a reasonable range without affecting the performance of the adhesive itself.

[0007] The excitation power is calculated based on the above physical parameters , the calculation formula is: ; According to the excitation power Get incentive duration , then the excitation duration and excitation power satisfy the following relationship: ; Where, is the excitation power; is the target maximum temperature; is the initial ambient temperature, is the heat transfer leakage coefficient, The value range is 0 to 1, which is used to indicate the degree of heat loss in the adhesive-substrate interface; is the interface heat transfer coefficient; is the total thermal resistance of the interface.

[0008] The interface heat transfer coefficient can be estimated through experimental measurement, simulation experiment or reference standard data based on the interface material combination. For the copper finish on the fire door, epoxy thermal conductive adhesive is generally used for hot pressing or vacuum pressing. Therefore, the interface heat transfer coefficient between copper and adhesive is Generally, it is taken as 1200–2000 W / m²·K. The total thermal resistance of the interface is obtained based on the interface heat transfer coefficient, specifically:

[0009] Where, The thermal conductivity of the adhesive is generally 1.5-3W / (m·K) for epoxy thermal conductive adhesives.

[0010] Furthermore, thermal imaging of the copper veneer bonding area is captured using a high-frame-rate infrared thermal imager. The acquisition frequency should be no less than 100 Hz, preferably in the range of 120–200 Hz, to ensure coverage of the critical time window for thermal diffusion. The acquisition cycle should be set based on the excitation power and the thermal diffusion characteristics of the material, generally ranging from 0.5–5 seconds, with a total of 50–1000 frames captured.

[0011] A thermal diffusion image sequence refers to a set of time-sequential thermal image frames continuously acquired using an infrared thermal imager during the thermal excitation process. Each frame reflects the temperature distribution of the bonding area at a specific point in time. The entire image sequence records the temperature diffusion process of the material after heating. Furthermore, the temperature variation of each pixel in each acquired thermal image sequence along the time axis is modeled to construct a pixel-level time-temperature fitting model; this includes: Data construction: Image sequences acquired during thermal excitation , each pixel Temperature changes over time Organized as a one-dimensional time series vector ; Fitting function model: Use the following composite function model to fit the series:

[0012] Where, is the predicted temperature value of the pixel at time t in degrees Celsius, and t is the frame timestamp of the thermal image acquisition in seconds; is the rapid thermal diffusion amplitude factor in degrees Celsius; The unit of the rapid diffusion attenuation coefficient is per second; The unit of the rise for slow diffusion is degrees Celsius; The unit of response rate for slow diffusion time is per second; is the local temperature rise peak amplitude in degrees Celsius; The unit of time for the temperature peak to occur is seconds; is the peak diffusion width in seconds.

[0013] Rapid diffusion stage, simulating the rapid heat transfer and surface response of the copper surface; Slow diffusion response, reaction heat is transferred to the substrate with hysteresis through the adhesive layer; Gaussian peak function simulates local thermal anomaly peaks caused by lamination defects or non-uniform bonding.

[0014] Fitting method: Use the least squares method to perform nonlinear parameter optimization fitting, perform parameter fitting independently for each pixel point, and output the optimal parameter vector: .

[0015] Spectrum construction: All fitting parameters are constructed into seven-channel thermal diffusion parameter maps according to their spatial positions.

[0016] Atlas processing and output: Each channel map is uniformly normalized to standardize the pixel value range to between 0 and 1. The pseudo-color mapping function can be used to visualize the map as a pseudo-color image such as Jet, Viridis or Plasma for manual assisted recognition. At the same time, the normalized map data is organized into a seven-channel tensor structure in floating-point matrix format and output in standard formats such as .npy, .mat or HDF5 for direct call by subsequent convolutional neural network models. The atlas construction process also includes evaluating pixel-level fitting residuals, marking invalid values ​​at pixel positions where the fitting error exceeds a preset threshold, and marking the area with zero or null values ​​in the generated atlas. In the subsequent neural network recognition process, the system implements feature shielding or input weighted attenuation strategies for these invalid areas to improve the model's robustness in identifying defective areas and data fault tolerance.

[0017] Furthermore, based on the convolutional neural network and time series model, the abnormal heat diffusion areas in the map are identified and the defect probability map is generated, which includes: Defect identification is performed using a deep learning model that combines spatial distribution features with temporal response features; The seven-channel thermal diffusion parameter map generated by fitting is used as the input tensor, and the spatial feature sub-network (CNN) is input to extract the spatial features of local thermal anomalies to obtain the spatial feature map. ,This figure reflects the changing pattern of heat diffusion pattern in the fitting area; Select several key frames from the thermal image sequence to form a time series input, and each pixel forms a temperature-time vector , input long short-term memory network (LSTM or TCN) for dynamic evolution modeling, output time response feature map , represents the relative activity and offset characteristics of the heat diffusion process of the pixel over time; Fusion decision module: and For channel fusion, you can use channel-by-channel splicing followed by convolution fusion or attention mechanism superposition, and finally enter the fully connected classification layer to output the defect probability map , each pixel value reflects the probability that the area is a thermal diffusion abnormal area. The spatial feature is calculated according to the following formula and time characteristics Perform fusion judgment and output defect probability map. The formula is:

[0018] Where, is the image coordinate position The defect probability prediction value at the point, whose value range is between 0 and 1, indicates the possibility that the point belongs to the fitting abnormal area; The Sigmoid function is used to normalize the fusion output value to the probability range; The index of the atlas channel ranges from 1 to L; L is the total number of atlas channels, which is 7; For the The fusion weight coefficient of each channel feature is learned during the training process to represent the contribution strength of the channel to the final discrimination; The time response feature fusion coefficient is used to adjust the influence of time series features on the discrimination results; The value range of the time series frame index is 1 to K; K is the number of key frames used for dynamic analysis in the thermal image sequence; For the Frame time point in pixels The temperature change rate at is in degrees Celsius per second, which represents the instantaneous change in the heat diffusion rate; For the The dynamic response weight coefficient of a frame represents the sensitivity or importance of abnormal judgment at that time point; It is a very small number stabilization term used to avoid calculation anomalies when the denominator is zero. The value is usually 1e-6; For the CNN spatial channels at pixel locations The characteristic map value of Represents the pixel The dynamic thermal diffusion response characteristic value output after modeling the temperature change sequence.

[0019] Furthermore, the defect area location and boundary information are extracted based on the defect probability map, and the annotated image and structured defect data are output, including: Setting probability thresholds ; Defect Probability Plot Perform binarization and convert all > The pixels are marked as suspected defect areas; After filtering out small noise areas through image morphological operations, a contour extraction algorithm is used to obtain the boundary coordinates, area, minimum bounding rectangle, and center centroid coordinates of each defect area. A unique number is then generated for each area to construct a defect structured data record. Each defect record includes the defect number, spatial location, area, average defect probability value, and boundary coordinate set. At the same time, the defect area is superimposed on the original thermal diffusion image to form a two-dimensional pseudo-color defect annotation map with numbers and boundary marks. The defect map and data record can be output as image files and structured data files in parallel, including but not limited to TIFF, PNG, JSON, or CSV formats, for manual review, defect location, quality traceability, or docking with industrial quality inspection systems for automatic removal.

[0020] The beneficial effects of the technical solution provided by the embodiment of the present invention are: by combining infrared thermal imaging with thermal excitation, non-destructive detection of ultra-large copper finish areas can be achieved. The excitation power and duration can be adaptively set according to the copper finish and adhesive layer parameters, and are highly versatile. By constructing a composite time-temperature fitting function that includes fast diffusion, slow diffusion, and local thermal anomaly response, the actual thermal diffusion process of the bonded structure under thermal excitation is accurately restored, effectively reflecting the difference in interface bonding quality. The seven-channel thermal diffusion parameter map is generated by fitting the time series of each pixel, which enhances the map's ability to express small defects and non-uniform distribution characteristics of the bonding structure, and improves the system's recognition accuracy of complex bonding states. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solution of the present invention, the following is a brief introduction to the drawings used in the embodiments. Obviously, the drawings listed below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0022] Figure 1 The present invention is a flowchart of a method for detecting the lamination quality of copper facings of super-large fire doors based on thermal diffusion modeling according to an embodiment of the present invention. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below in conjunction with the embodiments. Of course, the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0024] See also Figure 1 , the present invention is achieved by the following measures: a method for detecting the lamination quality of copper veneer of an oversized fire door based on thermal diffusion modeling, characterized by comprising: S1, exciting the lamination area of ​​the copper veneer by an array pulse heat source, and setting the excitation power and duration; The array-type pulse heat source is set directly above the detection surface through the scanning platform. The array-type pulse heat source is used to stimulate the copper finish bonding area. Generally, four groups of short-wave infrared heat source arrays are used, which are arranged directly above the copper finish to be tested. The distance between the infrared heat source unit and the copper surface is controlled between 5 and 10 cm. The heat sources are activated in sequence to achieve regional coverage pulse heating excitation and effectively control the uniformity of heat energy distribution.

[0025] Set the excitation power and duration, including: S101. Obtain copper surface density Specific heat capacity ,thickness , adhesive density Specific heat capacity ,thickness , (density in kilograms per cubic meter, specific heat in joules per kilogram Kelvin, thickness in meters), detection area (unit: square meters), target temperature rise (equal to the difference between the excitation target temperature and the initial ambient temperature, in degrees Celsius), is the excitation energy transfer efficiency coefficient (indicating the energy coupling efficiency between the heat source and the material interface, with a value range of 0.5 to 0.9), The heat loss adjustment factor is equal to (in is the interfacial heat transfer coefficient in watts per square meter Kelvin, is the total thermal resistance of the interface in Kelvin per watt).

[0026] The target excitation temperature should be able to raise the temperature of the copper surface and the adhesive layer to a reasonable range without affecting the performance of the adhesive itself.

[0027] S102: Calculate the excitation power based on the above physical parameters , the calculation formula is: ; S103, based on the excitation power Get incentive duration , then the excitation duration and excitation power satisfy the following relationship: ; Where, is the excitation power; is the target maximum temperature; is the initial ambient temperature, is the heat transfer leakage coefficient, The value range is 0 to 1, which is used to indicate the degree of heat loss in the adhesive-substrate interface; is the interface heat transfer coefficient; is the total thermal resistance of the interface.

[0028] The interface heat transfer coefficient can be estimated through experimental measurement, simulation experiment or reference standard data based on the interface material combination. For the copper finish on the fire door, epoxy thermal conductive adhesive is generally used for hot pressing or vacuum pressing. Therefore, the interface heat transfer coefficient between copper and adhesive is Generally, it is taken as 1200–2000 W / m²·K. The total thermal resistance of the interface is obtained based on the interface heat transfer coefficient, specifically:

[0029] Where, The thermal conductivity of the adhesive is generally 1.5-3W / (m·K) for epoxy thermal conductive adhesives.

[0030] S2, collecting a thermal diffusion image sequence with a thermal imager, and performing synchronous control of thermal excitation and image acquisition; Use a high-frame-rate infrared thermal imager to capture thermal images of the copper veneer bonding area. The acquisition frequency should be no less than 100 Hz, preferably in the range of 120–200 Hz, to ensure coverage of the critical time window for thermal diffusion. The acquisition cycle should be set based on the excitation power and the thermal diffusion characteristics of the material, generally 0.5–5 seconds, with a total of 50–1000 frames collected.

[0031] A thermal diffusion image sequence refers to a set of time-sequential thermal image frames continuously acquired using an infrared thermal imager during the thermal excitation process. Each frame reflects the temperature distribution state of the bonding area at a certain point in time. The image sequence as a whole records the temperature diffusion process of the material after heating. S3. Model the temperature change of each pixel point on the time axis in each acquired thermal image sequence to construct a pixel-level time-temperature fitting model; including: S301. Data construction: Image sequence collected during thermal excitation , each pixel Temperature changes over time Organized as a one-dimensional time series vector ; S302, fitting function model: fitting the sequence using the following composite function model:

[0032] Where, is the predicted temperature value of the pixel at time t in degrees Celsius, and t is the frame timestamp of the thermal image acquisition in seconds; is the rapid thermal diffusion amplitude factor in degrees Celsius; The unit of the rapid diffusion attenuation coefficient is per second; The unit of the rise for slow diffusion is degrees Celsius; The unit of response rate for slow diffusion time is per second; is the local temperature rise peak amplitude in degrees Celsius; The unit of time for the temperature peak to occur is seconds; is the peak diffusion width in seconds.

[0033] Rapid diffusion stage, simulating the rapid heat transfer and surface response of the copper surface; Slow diffusion response, reaction heat is transferred to the substrate with hysteresis through the adhesive layer; Gaussian peak function simulates local thermal anomaly peaks caused by lamination defects or non-uniform bonding.

[0034] S303, fitting method: Use the least squares method to perform nonlinear parameter optimization fitting, perform parameter fitting independently for each pixel point, and output the optimal parameter vector: .

[0035] S304, map construction: construct all fitting parameters into seven-channel thermal diffusion parameter maps according to spatial positions.

[0036] S305. Graph processing and output: Each channel graph is uniformly normalized to standardize the pixel value range to between 0 and 1. A pseudo-color mapping function can be used to visualize the graph as a pseudo-color image such as Jet, Viridis or Plasma for manual assisted recognition. At the same time, the normalized graph data is organized into a seven-channel tensor structure in a floating-point matrix format and output in a standard format such as .npy, .mat or HDF5 for direct call by the subsequent convolutional neural network model. The graph construction process also includes evaluating the pixel-level fitting residuals, marking invalid values ​​at the pixel positions where the fitting error exceeds a preset threshold, and marking the area with zero or null values ​​in the generated graph. In the subsequent neural network recognition process, the system implements feature shielding or input weighted attenuation strategies for these invalid areas to improve the model's robustness in identifying defective areas and data fault tolerance.

[0037] S4. Based on convolutional neural networks and time series models, abnormal heat diffusion areas in the map are identified and defect probability maps are generated.

[0038] S401. Defect identification using a deep learning model that combines spatial distribution features and temporal response features; S402, using the fitted seven-channel thermal diffusion parameter map as the input tensor, input the spatial feature sub-network (CNN) to extract the spatial features of local thermal anomalies, and obtain the spatial feature map ,This figure reflects the changing pattern of heat diffusion pattern in the fitting area; S403, select several key frames (such as 0.1s, 0.3s, 0.5s, etc.) from the thermal image sequence to form a time series input, and each pixel point forms a temperature-time vector , input long short-term memory network (LSTM or TCN) for dynamic evolution modeling, output time response feature map , represents the relative activity and offset characteristics of the heat diffusion process of the pixel over time; S404, fusion decision module: and For channel fusion, you can use channel-by-channel splicing followed by convolution fusion or attention mechanism superposition, and finally enter the fully connected classification layer to output the defect probability map , each pixel value reflects the probability that the area is a thermal diffusion abnormal area. The spatial feature is calculated according to the following formula and time characteristics Perform fusion judgment and output defect probability map. The formula is:

[0039] Where, is the image coordinate position The defect probability prediction value at the point, whose value range is between 0 and 1, indicates the possibility that the point belongs to the fitting abnormal area; The Sigmoid function is used to normalize the fusion output value to the probability range; The index of the atlas channel ranges from 1 to L; L is the total number of atlas channels, which is 7; For the The fusion weight coefficient of each channel feature is learned during the training process to represent the contribution strength of the channel to the final discrimination; The time response feature fusion coefficient is used to adjust the influence of time series features on the discrimination results; The value range of the time series frame index is 1 to K; K is the number of key frames used for dynamic analysis in the thermal image sequence; For the Frame time point in pixels The temperature change rate at is in degrees Celsius per second, which represents the instantaneous change in the heat diffusion rate; For the The dynamic response weight coefficient of a frame represents the sensitivity or importance of abnormal judgment at that time point; It is a very small number stabilization term used to avoid calculation anomalies when the denominator is zero. The value is usually 1e-6; For the CNN spatial channels at pixel locations The characteristic map value of Represents the pixel The dynamic thermal diffusion response characteristic value output after modeling the temperature change sequence.

[0040] S5. Extract the defect area location and boundary information based on the defect probability map, and output the annotated image and structured defect data, including: Setting probability thresholds , generally greater than 0.5 and less than 1; Defect Probability Plot Perform binarization and convert all > The pixels are marked as suspected defect areas; After filtering out small noise areas through image morphological operations, a contour extraction algorithm is used to obtain the boundary coordinates, area, minimum bounding rectangle, and center centroid coordinates of each defect area. A unique number is then generated for each area to construct a defect structured data record. Each defect record includes the defect number, spatial location, area, average defect probability value, and boundary coordinate set. At the same time, the defect area is superimposed on the original thermal diffusion image to form a two-dimensional pseudo-color defect annotation map with numbers and boundary marks. The defect map and data record can be output as image files and structured data files in parallel, including but not limited to TIFF, PNG, JSON, or CSV formats, for manual review, defect location, quality traceability, or docking with industrial quality inspection systems for automatic removal.

[0041] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for testing the lamination quality of copper finishes on oversized fire doors based on thermal diffusion modeling, characterized in that: include: The copper veneer bonding area is stimulated by an array pulse heat source, and the stimulation power and duration are set; Collecting heat diffusion image sequences with a thermal imager; The temperature changes of each pixel point on the time axis in each acquired thermal image sequence are modeled to construct a pixel-level time-temperature fitting model; based on the convolutional neural network and time series model, abnormal heat diffusion areas in the image are identified and a defect probability map is generated; the location and boundary information of the defect area are extracted based on the defect probability map, and the annotated image and structured defect data are output.

2. The method for detecting the lamination quality of copper facings of super-large fire doors based on thermal diffusion modeling according to claim 1 is characterized in that: Set the excitation power and duration, including: Obtain copper finish density, specific heat capacity, thickness, adhesive density, specific heat capacity, thickness, test area, and target temperature rise; The excitation power is calculated based on the above physical parameters; The excitation duration is obtained according to the excitation power.

3. The method for detecting the lamination quality of copper facings of super-large fire doors based on thermal diffusion modeling according to claim 1 is characterized in that: Build a pixel-level time-temperature fitting model, including: For the image sequence collected during the thermal excitation process, the temperature value of each pixel changing with time is organized into a one-dimensional time series vector; The following composite function model is used to fit the series; The least squares method is used for nonlinear parameter optimization fitting. Parameter fitting is performed independently for each pixel point, and the optimal parameter vector is output: All fitting parameters are constructed into seven-channel thermal diffusion parameter maps according to spatial positions; Process the graph and output it.

4. The method for detecting the lamination quality of copper finishes on super-large fire doors based on thermal diffusion modeling according to claim 3 is characterized in that: The graph is processed and output as follows: Each channel map is normalized uniformly, and the normalized map data is organized into a seven-channel tensor structure in floating-point matrix format; It also includes evaluating the pixel-level fitting residuals, marking invalid values ​​at the pixel positions where the fitting error exceeds the preset threshold, marking the area with zero or empty values ​​in the generated map, and implementing feature masking or input weighted attenuation strategies for invalid areas.

5. The method for detecting the lamination quality of copper finishes on super-large fire doors based on thermal diffusion modeling according to claim 3 is characterized in that: Based on convolutional neural networks and time series models, we identify abnormal heat diffusion areas in the map and generate defect probability maps, including: Defect identification is performed using a deep learning model that combines spatial distribution features with temporal response features; The fitted seven-channel thermal diffusion parameter map is used as the input tensor and input into the spatial feature sub-network to extract the spatial features of local thermal anomalies and obtain the spatial feature map. Select several key frames from the thermal image sequence to form a time series input. Each pixel forms a temperature-time vector, which is input into the long short-term memory network for dynamic evolution modeling and outputs a time response feature map. The spatial features and temporal features are fused to perform channel fusion, and after channel-by-channel splicing, they enter the fully connected classification layer and output the defect probability map.

6. The method for detecting the lamination quality of copper finishes on super-large fire doors based on thermal diffusion modeling according to claim 1 is characterized in that: Extract the defect area location and boundary information based on the defect probability map, and output the annotated image and structured defect data, including: Setting probability thresholds; Binarize the defect probability map and mark all pixels that meet the requirements as suspected defect areas; After filtering out small noise areas through image morphological operations, the contour extraction algorithm is used to obtain the boundary coordinates, area, minimum enclosing rectangle and center centroid coordinates of each defect area. Then, a unique number is generated for each area to construct a defect structured data record.

Images