Fabricated wall non-uniform heat transfer analysis method based on infrared image segmentation

Abstract

The invention discloses a fabricated wall non-uniform heat transfer analysis method based on infrared image segmentation. The method comprises the following steps: S1, taking a splicing seam as a center, and obtaining a group of fabricated wall body infrared images through an infrared thermal imager; s2, carrying out preprocessing operation on the obtained infrared image, and segmenting the preprocessed image by utilizing an improved Gaussian mixture model algorithm; s3, training an improved Res-UNet network in combination with an improved joint loss function based on segmented image data, identifying and segmenting a non-uniform heat transfer area of the fabricated wall infrared image, and obtaining feature parameters such as infrared temperature and pixel points; s4, for the infrared temperature, calculating a non-uniform heat transfer evaluation index by using the identified characteristic parameters; and S5, representing the calculated non-uniform indexes and analyzing the non-uniform heat transfer condition of the fabricated outer wall. According to the method, accurate identification and more accurate evaluation analysis of the heat transfer non-uniform area of the fabricated building wall are realized.

Description

Technical Field

[0001] The invention relates to the technical field of building thermal engineering and building energy conservation, and in particular to a method for analyzing non-uniform heat transfer of assembled walls based on infrared image segmentation. Background Art

[0002] Due to the different construction methods, the structure, materials and construction of prefabricated building walls have undergone significant changes, resulting in significant differences in their heat transfer laws from traditional buildings. Among them, non-uniformity is a significant heat transfer feature of prefabricated walls. For dry-connected walls, when prefabricated components are connected, a large number of joints will be generated between the components, and the width of the joints is about 15 to 35 mm. At present, the treatment of the joints is usually filled with waterproof glue and thermal insulation mortar, but its heat transfer characteristics are quite different from those of the main parts, which can produce a thermal effect similar to a thermal bridge, which has a serious impact on the uniformity and one-dimensionality of the heat transfer of the wall, and the energy consumption of the building. After the prefabricated building is built, the joints between its components will be covered by the surface layer, which seriously affects the detection and analysis of the heat transfer uniformity of the wall.

[0003] At present, on-site wall heat transfer analysis technology mostly uses contact measurement, such as Du Lin. Research on on-site detection method of heat transfer coefficient of exterior wall of building envelope structure [D]. Nanchang University, 2017. However, the measurement process will encounter many difficulties, such as being greatly affected by the environment, the measurement process is cumbersome, especially for high-rise buildings, there are problems such as difficulty in arranging measurement points, difficulty in operation, long measurement time, low accuracy, etc., making it difficult to obtain high-precision measurement data simply and quickly. Summary of the invention

[0004] In order to overcome the above technical problems, the purpose of the present invention is to provide a method for analyzing non-uniform heat transfer of prefabricated walls based on infrared image segmentation, which can realize accurate identification and more precise evaluation and analysis of non-uniform heat transfer areas of prefabricated building walls.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is:

[0006] A method for analyzing non-uniform heat transfer of assembled walls based on infrared image segmentation comprises the following steps;

[0007] S1: With the joint as the center, a set of infrared images of the assembled wall are obtained by infrared thermal imager;

[0008] S2: performing preprocessing operations on the obtained infrared image, including grayscale conversion and image denoising, and segmenting the preprocessed image using an improved Gaussian mixture model algorithm;

[0009] S3: Based on the segmented image data and the improved joint loss function, the improved Res-UNet network is trained to identify and segment the non-uniform heat transfer area of ​​the prefabricated wall infrared image, and obtain the characteristic parameters of the infrared temperature and pixel points;

[0010] S4: For infrared temperature, the non-uniform heat transfer evaluation index is calculated using the identified characteristic parameters;

[0011] S5: Characterize the calculated non-uniform heat transfer evaluation index and analyze the non-uniform heat transfer of the prefabricated exterior wall.

[0012] In S1, after the seam is found by the infrared thermal imager, contact measurement is performed on the seam area and infrared data is acquired;

[0013] The acquisition of infrared images is non-contact. First, before acquiring the image, the wall is photographed as a whole with an infrared instrument to determine the specific location of the wall joints.

[0014] Secondly, to ensure that the infrared image of the joint is obvious, the acquisition position is 1.5m away from the wall, and the processing time of each image is 2s. When taking pictures, it is necessary to use a tape measure to mark the wall size to facilitate the calculation of the area affected by the thermal bridge;

[0015] The contact detection method uses the heat flow meter method, which regards the heat transfer process of the wall as a one-dimensional process. The thermal performance of the wall is calculated by detecting the temperature on both sides of the wall and the heat flux density of the heat flow meter. In the actual detection process, the thermal resistance is calculated using the average value of the detection data within a period of detection time. The calculation formula is as follows:

[0016]

[0017] Where, R is the thermal resistance of the enclosure structure, W / (m2·K); θ ij is the jth measured temperature of the inner surface of the enclosure structure, °C; is the jth measured temperature of the outer surface of the enclosure structure, °C; q j is the jth measured value of heat flow, W / m 2 ;

[0018] For the building under test, after finding the joints of the wall with an infrared thermal imager, points are arranged at the joints and at positions with horizontal distances of 10cm and 20cm respectively, and points are arranged at the same positions on the outer wall of the wall. Thermocouples are arranged at the six measuring points; heat flux sheets are arranged at the joints of the inner and outer walls of the wall and at a horizontal distance of 20cm.

[0019] Furthermore, in S2, the obtained infrared image is subjected to grayscale processing based on a weighted average method, and image denoising is performed using a non-local mean filter;

[0020] Firstly, the three-channel components of the color image are weighted averaged according to the importance of each channel, and the obtained value is used as the grayscale value of the grayscale image; secondly, the obtained grayscale image is used to obtain a new grayscale value by weighted averaging the pixels in the image with a similar neighborhood structure to the target pixel, that is, non-local mean filtering is used to perform image denoising to achieve image enhancement;

[0021] The image is segmented using an improved Gaussian mixture model algorithm. The Gaussian mixture model refers to a probability distribution model as shown in formula (2):

[0022]

[0023] In formula (2), P(y|θ) means that the observed data y follows the probability distribution determined by the model parameter θ, α k is the kth model parameter representing the weight of each Gaussian distribution in the mixture model, and φ(y|θ k ) represents the probability density function of the kth Gaussian distribution, u k is the mean of the density function, is the variance of the density function;

[0024] Then the kth Gaussian distribution probability density function is:

[0025]

[0026] By analyzing the infrared image segmentation results of the prefabricated wall, the infrared image needs to be segmented into three regions. Therefore, the K value of the Gaussian mixture model is selected as 3, and the Gaussian mixture model parameter estimation based on the EM algorithm is used to obtain the mean value u of the density function. k , the variance of the density function The mixing coefficient α of the Gaussian distribution k ;

[0027] In S3, ResNet18 is used as the basic encoding part of the UNet network, and multiple cross-layer connections are added. The image dimension is unified through the maximum pooling layer. The decoding part also adopts the UNet network structure to form a denser Res-UNet network structure. An improved joint loss function is designed based on the weighted cross entropy and weighted Dice loss functions to improve the performance of the Res-UNet network. After segmentation, accurate feature parameters are obtained, including the total pixel points N in the thermal bridge affected area and the temperature of a single pixel point. The total number of pixels M on the entire wall and the temperature of a single pixel

[0028] By increasing the loss ratio of the foreground pixel value, the total loss of the improved joint loss function is made the weighted sum of the weighted cross entropy and weighted Dice loss functions, as shown in formula (4):

[0029] L CD =ω t ×L CEL +(1-ω t )×L DICE (4)

[0030] Where L CEL Expressed as weighted cross entropy, L DICE It is expressed as the weighted Dice loss function, which is specifically expressed as formula (5); ω t Represented as weights between loss functions;

[0031]

[0032] Among them, ω pos is the proportion of foreground pixel value, P pos is the predicted label of the foreground pixel, T pos is the true label of the foreground pixel, P neg is the predicted label of the background pixel, T neg is the true label of the background pixel, L cel represents the cross entropy loss, L dice Represents Dice loss.

[0033] In S4, since the measured temperature (thermocouple temperature) is different from the infrared temperature, the infrared temperature before correction T is used. IR , Indoor air temperature T IA and indoor air quality R IA As the input variable of the temperature correction model, the calibrated temperature (i.e. the temperature measurement data T of the thermocouple) TC ) is used as the output to construct the BP neural network, and its relationship can be expressed as T TC =f(T IR ,T IA ,R IA ) indicates that a correction model is established based on genetic algorithm to optimize BP neural network to correct infrared temperature;

[0034] In addition, the characteristic parameters are used to calculate the cumulative temperature difference ratio T' and the equivalent heat transfer coefficient K' of the wall non-uniform heat transfer evaluation index. T ;

[0035] The original calculation formula of the cumulative temperature difference ratio T' is shown in formula (6);

[0036]

[0037] In formula (6), t(x) is the wall temperature distribution function, °C; t0 is the temperature value of the main part, °C; l is the coordinate of the area affected by non-uniform heat transfer, m; L is the coordinate of the main part, m; T is the cumulative temperature difference of the thermal bridge affected area, T 0为 The accumulated temperature difference of the whole wall;

[0038] The calculation formula (7) converted into the cumulative temperature difference ratio based on thermal imaging is as follows:

[0039]

[0040] In formula (7), N is the total number of pixels in the seam-affected area; is the temperature of a single pixel in the unblocking affected area; M is the total number of pixels on the entire wall; The temperature of a single pixel point on the entire wall;

[0041] Equivalent heat transfer coefficient K T The calculation formula (8) is as follows:

[0042]

[0043] h is the height of the wall; q(x) is the function of the heat flux density at different distances from the joint normal, which can be obtained through FLUENT simulation; x0 and x1 are the starting point and end point coordinates of the thermal bridge influence distance; l0 is the actual structural width of the thermal bridge part; is the physical area of ​​the thermal bridge, m2; ΔT is the temperature difference between indoor and outdoor;

[0044] The equivalent heat transfer coefficient calculation formula (9) based on thermal imaging is as follows:

[0045]

[0046] In formula (9), T a (x) is the temperature difference function of the inner and outer walls at different distances from the joint normal, T i and M are the temperature of a single pixel and the total number of pixels within the thermal bridge influence range, respectively. T is the thermal bridge affected area, ΔT is the temperature difference between the inner and outer surfaces of the wall, is the total thermal resistance per unit area of ​​the flat wall, x0 and x1 are the starting and ending coordinates of the thermal bridge influence distance, and h is the wall height. The non-uniform heat transfer evaluation index is used to quantitatively evaluate the non-uniformity of heat transfer of prefabricated exterior walls to analyze the impact of non-uniformity on heat transfer performance and building energy consumption.

[0047] The non-uniform heat transfer evaluation index is proposed based on the non-uniform heat transfer of prefabricated exterior walls. It is used to accurately reflect the degree of non-uniform heat transfer of the wall. It is considered that the heat transfer is non-uniform when the cumulative temperature difference ratio T′>0.8, otherwise the heat transfer is uniform. The equivalent heat transfer coefficient K′T The influence of thermal bridge effect on the overall heat transfer of prefabricated building walls can be judged. The larger the value of the equivalent heat transfer coefficient, the more significant the influence of thermal bridge effect on the heat transfer of walls near the joints. By comparing the equivalent heat transfer coefficients of different areas, the non-uniform heat transfer of the wall can be judged.

[0048] Beneficial effects of the present invention:

[0049] In view of the uneven heat transfer caused by the joints in the assembled exterior walls, the infrared image segmentation technology is used to identify the non-uniform heat transfer area of ​​the assembled wall, which can directly observe the parts of the wall that are most prone to heat loss, such as the thermal bridge effect. Through infrared image analysis and based on infrared image processing technology, the image accuracy can be enhanced to obtain more accurate feature parameters to accurately calculate the non-uniform heat transfer evaluation index to evaluate the thermal performance of the wall.

[0050] On the other hand, the image is segmented by an improved Gaussian mixture model algorithm, which fully retains the image features. The improved Res18-UNet network and joint loss function greatly improve the recognition accuracy and speed. This method is not affected by the type and structure of prefabricated walls. Only the infrared image data of the wall is needed to know the uniformity of heat transfer of the wall and understand the influence of the joints on the uniformity of heat transfer of the wall.

[0051] This invention fills the gap in the research on the identification of non-uniform heat transfer of prefabricated building exterior walls, conducts a more accurate quantitative analysis of the non-uniform heat transfer of prefabricated exterior walls, and provides an infrared image-based analysis method for the non-uniform heat transfer effect of prefabricated walls. It effectively solves the problems of existing wall heat transfer identification technology, such as low identification accuracy and difficulty in obtaining high-precision calculation parameters, and provides valuable technical support for quality control and energy consumption optimization of prefabricated buildings. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 This is the improved Res-Unet network structure diagram.

[0053] Figure 2 This is the image segmentation result based on the improved Gaussian mixture model algorithm.

[0054] Figure 3 To predict the segmentation results and the pixel value ratio of the infrared image segmentation area after using the improved Res-Unet training.

[0055] Figure 4 It is the internal and external surface temperature of the prefabricated exterior wall and the air temperature.

[0056] Figure 5The temperatures of the wall joint area, joint influence area and heat transfer uniform area before and after processing by the temperature correction model, as well as the contact measurement temperature.

[0057] Figure 6 The evaluation indexes obtained by calculation are compared with the measured data. DETAILED DESCRIPTION

[0058] The present invention will be further described in detail below in conjunction with the accompanying drawings.

[0059] This example selected an assembled building in Nanjing, which has an exterior wall structure with metal carved panels and rock wool insulation. The total thickness of the wall panels is 160mm, of which the inner and outer wall panels are both 30mm, and the inner layer of rock wool is 100mm. After finding the joints with an infrared thermal imager, contact measurement and infrared data acquisition were performed on the joint area. In order to prove that this method is more accurate, a comparative analysis was conducted using the measured (contact measurement) data and the data processed by infrared imaging technology.

[0060] S1. Acquisition of infrared image data. The acquisition of infrared images is non-contact. The infrared imaging instrument used for acquisition is FLIR ONE Pro, with a thermal resolution of 160×120 pixels and a visual resolution of 1440×1080 pixels. Before acquiring images, the wall needs to be photographed as a whole with an infrared instrument to determine the specific location of the wall joints. Secondly, to ensure that the infrared image of the joints is obvious, the acquisition position is 1.5m away from the wall. The processing time for each image is 2s. When taking pictures, the wall size needs to be marked with a tape measure to facilitate the calculation of the area affected by the thermal bridge. According to different time periods, a total of 400 infrared images of wall joints were collected. The contact detection method uses the heat flow meter method, the principle of which is to determine the heat transfer coefficient through the corresponding relationship between the internal and external temperature difference and the heat flow. It regards the heat transfer process of the wall as a one-dimensional process, and calculates the thermal performance of the wall by detecting the temperature on both sides of the wall and the heat flux density of the heat flow meter. In the actual detection process, the thermal resistance can be calculated by the average value of the detection data within a period of detection time. The calculation formula is as follows:

[0061]

[0062] Where R is the thermal resistance of the enclosure structure, W / (m2·K); θ ij is the jth measured temperature of the inner surface of the enclosure structure, °C; is the jth measured temperature of the outer surface of the enclosure structure, °C; q j is the jth measured value of heat flow, W / m 2 .

[0063] For the above-mentioned building under test, after finding the joints of the wall with an infrared thermal imager, points were arranged at the joints and at positions with horizontal distances of 10cm and 20cm respectively, and points were arranged at the same positions on the outer wall of the wall. Thermocouples were arranged at the six measuring points; heat flux sheets were arranged at the joints of the inner and outer walls of the wall and at a horizontal distance of 20cm.

[0064] S2. Preprocessing the obtained infrared image and segmenting the infrared image;

[0065] The preprocessing operations of the obtained infrared image include grayscale processing based on weighted average and image denoising based on non-local mean filtering. First, the weighted average grayscale method weights and averages the three-channel components of the color image according to the importance of each channel, and the obtained value is used as the grayscale value of the grayscale image. Secondly, the obtained grayscale image is used to obtain a new grayscale value by weighted averaging the pixels in the image with a neighborhood structure similar to the target pixel, that is, non-local mean filtering is used to perform image denoising to achieve image enhancement. The pre-processed image is then subjected to image segmentation based on the improved Gaussian mixture model, such as Figure 2 The figure shows two infrared images of prefabricated walls and their grayscale images after segmentation by the improved Gaussian mixture model. It can be seen that the infrared images are divided into three areas after segmentation, namely, the heat transfer uniform area, the joint area, and the joint influence area, which provides a basis for the subsequent more detailed segmentation.

[0066] The Gaussian mixture model refers to a probability distribution model as shown in formula (2):

[0067]

[0068] In formula (4), P(y|θ) means that the observed data y follows the probability distribution determined by the model parameter θ, α k is the kth model parameter representing the weight of each Gaussian distribution in the mixture model, and φ(y|θ k ) represents the probability density function of the kth Gaussian distribution, u k is the mean of the density function, is the variance of the density function;

[0069] Then the kth Gaussian distribution probability density function is:

[0070]

[0071] By analyzing the infrared image segmentation results of the prefabricated wall, the infrared image needs to be segmented into three regions. Therefore, the K value of the Gaussian mixture model is selected as 3, and the Gaussian mixture model parameter estimation based on the EM algorithm is used to obtain the mean value u of the density function. k , the variance of the density function The mixing coefficient α of the Gaussian distribution k ;

[0072] The segmented images are infrared images that have been preprocessed by grayscale and denoising. These images have undergone preliminary image segmentation using clustering algorithms and Gaussian mixture models. The segmented images can clearly show key information such as the wall joint area, the joint influence area, and the heat transfer uniformity area, providing a basis for subsequent more detailed segmentation and heat transfer performance analysis.

[0073] The K value of the Gaussian mixture model is selected as 3. Although the three-classification Gaussian mixture model is more practical, the fitting result of the traditional Gaussian mixture model is not accurate enough, which leads to low image segmentation accuracy and errors in the determination of the non-uniformity of heat transfer of building walls. Therefore, the k-means clustering algorithm is used to optimize the initial value of the iteration of the EM algorithm based on the Gaussian mixture model, thereby accelerating the algorithm's iteration efficiency and segmentation accuracy.

[0074] S3. Considering that the semantics of the wall joint area image after segmentation is relatively simple, and the detailed information at the edges of the joint area and the joint affected area is more important, that is, both the high-level semantics and low-level features of the infrared image need to be considered, a Res-UNet network structure that combines ResNet18 and Unet networks and uses ResNet18 as the basic encoding network of the UNet network is proposed. It can extract the deep features of the prefabricated wall infrared image and reuse the shallow features, thereby improving the segmentation performance and completing related recognition and segmentation tasks.

[0075] The UNet network structure can fully extract and utilize image features, but too deep a network layer will cause gradient disappearance and network degradation, while ResNet can effectively solve this problem. Therefore, ResNet18 is used as the encoding part of the UNet network, and multiple cross-layer connections are added. At the end, like the UNet network, the image dimension is unified through the maximum pooling layer; the decoding part also uses the UNet network structure to form a denser network structure. The improved Res-UNet network structure is shown below: Figure 1 .

[0076] The performance of Res-UNet network is improved by designing an improvement of the joint loss function based on weighted cross entropy and weighted Dice loss functions. After segmentation, accurate feature parameters including the total number of pixels N in the thermal bridge affected area and the temperature of a single pixel are obtained. The total number of pixels M on the entire wall and the temperature of a single pixel

[0077] In order to enhance the network model’s learning of the wall joint area and the joint-affected area, a joint loss function is designed based on the weighted cross entropy and weighted Dice loss functions. By increasing the loss ratio of the foreground pixel value, the total loss function is made the weighted sum of the two, as shown in formula (4):

[0078] L CD =ω t ×L CEL +(1-ω t )×L DICE (4)

[0079] Where L CEL Expressed as weighted cross entropy, L DICE It is expressed as the weighted Dice loss function, which is specifically expressed as formula (5); ω t Expressed as weights between loss functions.

[0080]

[0081] Among them, ω pos is the proportion of foreground pixel value, P pos is the predicted label of the foreground pixel, T pos is the true label of the foreground pixel, P neg is the predicted label of the background pixel, T neg is the true label of the background pixel, L cel represents the cross entropy loss, L dice Represents Dice loss.

[0082] The ruu is trained to identify and segment the heat transfer of the prefabricated wall joint area to obtain characteristic parameters.

[0083] The collected data set has a total of 400 images. After 90-degree rotation, 270-degree rotation and horizontal mirroring, the data set increases to 800 images, and the ratio of the training set to the validation set is set to 8:2, that is, 640 images for the training set and 160 images for the validation set. The improved Res-UNet network is trained, and its structure is as follows Figure 1 ResNet18 is used as the encoding part of the UNet network, and multiple cross-layer connections are added. At the end, like the UNet network, the image dimension is unified through the maximum pooling layer; the decoding part also uses the UNet network structure to form a denser network structure.

[0084] The heat transfer recognition results of the joint area of ​​the test set of wall infrared images are shown in Table 1. By screening and analyzing the test infrared images, a total of 60 images at different time periods every day were selected for analysis and recognition, and the partial results of the segmentation results and the proportion of pixel point values ​​in each area of ​​the infrared image segmentation were predicted. Figure 3 , infrared temperature wipe parameters such as Figure 5 .

[0085] Table 1 Training results

[0086]

[0087] S4: Correct the infrared temperature and calculate the non-uniform heat transfer index using characteristic parameters.

[0088] The test results show that the temperature of the inner and outer surfaces of the prefabricated building and the air temperature are as follows: Figure 4 , Figure 4 (a) is a comparison between the wall temperature at the joints of the exterior walls of prefabricated buildings in summer and the indoor air temperature, showing the impact of air conditioning on the non-uniform heat transfer of the wall in summer. Figure 4 (b) shows the wall temperature at different locations of the indoor and outdoor joints of the exterior wall of the prefabricated building in summer, indicating that environmental factors such as solar radiation have a great influence on the non-uniformity of wall heat transfer. It also provides data support for the calculation of evaluation indicators.

[0089] In S4, since the measured temperature (thermocouple temperature) is different from the infrared temperature, the infrared temperature before correction T is used. IR , Indoor air temperature T IA and indoor air quality R IA As the input variable of the temperature correction model, the calibrated temperature (i.e. the temperature measurement data T of the thermocouple) TC ) is used as the output to construct the BP neural network, and its relationship can be expressed as T TC =f(T IR ,T IA ,R IA ) represents, and based on the genetic algorithm to optimize the BP neural network to establish a correction model to correct the infrared temperature. ;

[0090] In addition, the characteristic parameters are used to calculate the cumulative temperature difference ratio T' and the equivalent heat transfer coefficient K' of the wall non-uniform heat transfer evaluation index. T

[0091] The original calculation formula of the cumulative temperature difference ratio T' is shown in formula (6).

[0092]

[0093] In formula (6), t(x) is the wall temperature distribution function, °C; t0 is the temperature value of the main part, °C; l is the coordinate of the area affected by non-uniform heat transfer, m; L is the coordinate of the main part, m; T is the cumulative temperature difference of the thermal bridge affected area, T 0为 The accumulated temperature difference of the whole wall;

[0094] The calculation formula (7) converted into the cumulative temperature difference ratio based on thermal imaging is as follows:

[0095]

[0096] In formula (7), N is the total number of pixels in the seam-affected area; is the temperature of a single pixel in the unblocking affected area; M is the total number of pixels on the entire wall; It is the temperature of a single pixel point on the entire wall.

[0097] Equivalent heat transfer coefficient K T The calculation formula (8) is as follows:

[0098]

[0099] h is the height of the wall; q(x) is the function of the heat flux density at different distances from the joint normal, which can be obtained through FLUENT simulation; x0 and x1 are the starting point and end point coordinates of the thermal bridge influence distance; l0 is the actual structural width of the thermal bridge part; is the physical area of ​​the thermal bridge, m2; ΔT is the temperature difference between indoor and outdoor.

[0100] The equivalent heat transfer coefficient calculation formula (9) based on thermal imaging is as follows:

[0101]

[0102] In formula (9), T a (x) is the temperature difference function of the inner and outer walls at different distances from the joint normal, T i and M are the temperature of a single pixel and the total number of pixels within the thermal bridge influence range, respectively. T is the thermal bridge affected area, ΔT is the temperature difference between the inner and outer surfaces of the wall, is the total thermal resistance per unit area of ​​the flat wall, x0 and x1 are the coordinates of the starting and ending points of the thermal bridge influence distance, and h is the height of the wall.

[0103] The temperature-related changes in the wall joint area, joint influence area and heat transfer uniform area after being processed by the temperature correction model are shown in Figure 2. Figure 5 shown.

[0104] Figure 5 (a) is the temperature change related to the wall joint area, (b) is the temperature change related to the joint influence area, and (c) is the temperature change related to the uniform heat transfer area. By analyzing the temperature changes in these three areas, it can be seen that the temperatures of the three areas obtained by the infrared camera are slightly higher than the temperature measured by the thermocouple. There is almost no error between the corrected infrared temperature and the actual measured temperature, indicating that the infrared temperature data correction model mentioned above is very accurate, and the average temperature of each area after segmentation is significantly different. The temperature at the joint is the highest, followed by the joint influence area, indicating that the temperature at the joint fluctuates greatly, resulting in more obvious heat transfer non-uniformity.

[0105] The cumulative temperature difference ratio and equivalent heat transfer coefficient of the prefabricated wall measured by contact are calculated by the formula. The wall temperature distribution function T(x) and the wall heat flux density function q(x) are obtained by FLUENT simulation software. By building the same structural wall as the tested prefabricated building and setting the same internal and external surface temperatures as during the test, the temperature distribution of the inner wall surface of the wall is simulated. For the cumulative temperature difference ratio and equivalent heat transfer coefficient of the prefabricated wall in the thermal imaging image.

[0106] S5. Characterize the calculated non-uniform index and analyze the non-uniform heat transfer of the prefabricated exterior wall.

[0107] like Figure 6 The figure shows the characterization of the results of the non-uniform heat transfer evaluation index. The heat transfer uniformity of the prefabricated wall is evaluated by the non-uniform heat transfer index calculated by the measured value and the infrared image data. Figure 6 It includes the characterization of the heat transfer non-uniformity evaluation results of prefabricated walls based on both contact measurement and infrared image data.

[0108] from Figure 6 It can be obtained that the measured cumulative temperature difference ratio of the prefabricated building wall is around 0.8, indicating that this building has significant heat transfer unevenness, and the joints have a significant impact on the overall heat transfer, which has a negative impact of about 20% on the overall thermal performance. The error between the cumulative temperature difference ratio calculated based on the thermal imaging image and the actual measurement is within 5%. Based on the equivalent heat transfer coefficient, it can be seen that the joints of the walls of this prefabricated building have a significant impact on the overall heat transfer. At the same time, the error between the actual measured value and the thermal imaging image is small, both within 10%, indicating that the equivalent heat transfer coefficient calculated based on the thermal imaging image can accurately realize the thermal performance evaluation of the area affected by non-uniform heat transfer. This shows that the non-uniform heat transfer evaluation index calculated based on the thermal imaging image has a high accuracy. Therefore, this prefabricated building has significant heat transfer unevenness, and the joints of the walls of this prefabricated building have a significant impact on the overall heat transfer.

[0109] In summary, the method for analyzing the non-uniform heat transfer of prefabricated walls based on infrared image segmentation proposed in the present invention is simple and effective to operate. The method is not affected by the type and structure of the prefabricated wall. It only needs to collect the infrared image data of the measured wall to know the uniform heat transfer of the wall. In addition, the method uses an improved UNet network to identify the non-uniform heat transfer area of ​​the wall, which can effectively extract and fuse the deep and shallow features of the image, thereby improving the performance of the identification of non-uniform heat transfer of prefabricated walls. The BP neural network optimized by genetic algorithm is used to correct the infrared temperature, which improves the accuracy of the recognition result and obtains more accurate feature parameters. It provides a more accurate analysis method for the thermal performance analysis of prefabricated buildings.

[0110] The above embodiments illustrate the technical concept and features of the present invention, and are intended to enable people familiar with the technology to understand the content of the present invention and implement it accordingly, and cannot limit the scope of protection of the present invention. All equivalent changes and modifications made to the scope of the claims of the present invention should fall within the scope of the claims of the present invention.

Claims

1. A method for analyzing non-uniform heat transfer of assembled walls based on infrared image segmentation, characterized in that: The steps include: S1: With the joint as the center, a set of infrared images of the assembled wall is obtained by infrared thermal imager; S2: performing preprocessing operations on the obtained infrared image, including grayscale conversion and image denoising, and segmenting the preprocessed image using an improved Gaussian mixture model algorithm; S3: Based on the segmented image data and the improved joint loss function, the improved Res-UNet network is trained to identify and segment the non-uniform heat transfer area of ​​the prefabricated wall infrared image, and obtain the characteristic parameters of the infrared temperature and pixel points; S4: For infrared temperature, the non-uniform heat transfer evaluation index is calculated using the identified characteristic parameters; S5: Characterize the calculated non-uniform heat transfer evaluation index and analyze the non-uniform heat transfer of the prefabricated exterior wall.

2. The method for analyzing non-uniform heat transfer of assembled walls based on infrared image segmentation according to claim 1 is characterized in that: In S1, after the seam is found by the infrared thermal imager, contact measurement and infrared data acquisition are performed on the seam area; The acquisition of infrared images is non-contact. First, before acquiring the image, the wall is photographed as a whole with an infrared instrument to determine the specific location of the wall joints. Secondly, the acquisition position is 1.5m away from the wall, and the processing time of each image is 2s. When taking pictures, it is necessary to use a tape measure to mark the wall size; The contact detection method uses the heat flow meter method, which regards the heat transfer process of the wall as a one-dimensional process. The thermal performance of the wall is calculated by detecting the temperature on both sides of the wall and the heat flux density of the heat flow meter. In the actual detection process, the thermal resistance is calculated using the average value of the detection data within a period of detection time. The calculation formula is as follows: Where, R is the thermal resistance of the enclosure structure, W / (m2·K); θ ij is the jth measured temperature of the inner surface of the enclosure structure, °C; is the jth measured temperature of the outer surface of the enclosure structure, °C; q j is the jth measured value of heat flow, W / m 2 ; For the building under test, after finding the joints of the wall with an infrared thermal imager, points are arranged at the joints and at positions with horizontal distances of 10cm and 20cm respectively, and points are arranged at the same positions on the outer wall of the wall. Thermocouples are arranged at the six measuring points; heat flux sheets are arranged at the joints of the inner and outer walls of the wall and at a horizontal distance of 20cm.

3. The method for analyzing non-uniform heat transfer of assembled walls based on infrared image segmentation according to claim 1 is characterized in that: In S2, grayscale processing is performed on the obtained infrared image based on a weighted average method, and image denoising is performed using a non-local mean filter; First, the three-channel components of the color image are weighted averaged according to the importance of each channel, and the obtained value is used as the grayscale value of the grayscale image; secondly, the obtained grayscale image is used to obtain a new grayscale value by weighted averaging the pixels in the image with a similar neighborhood structure to the target pixel, that is, non-local mean filtering is used to perform image denoising and achieve image enhancement; The image is segmented using an improved Gaussian mixture model algorithm. The Gaussian mixture model refers to a probability distribution model as shown in formula (2): In formula (2), P(y|θ) means that the observed data y follows the probability distribution determined by the model parameter θ, α k is the kth model parameter representing the weight of each Gaussian distribution in the mixture model, and φ(y|θ k ) represents the probability density function of the kth Gaussian distribution, u k is the mean of the density function, is the variance of the density function; Then the kth Gaussian distribution probability density function is: By analyzing the infrared image segmentation results of the prefabricated wall, the infrared image needs to be segmented into three regions. Therefore, the K value of the Gaussian mixture model is selected as 3, and the Gaussian mixture model parameter estimation based on the EM algorithm is used to obtain the mean value u of the density function. k , the variance of the density function The mixing coefficient α of the Gaussian distribution k .

4. The method for analyzing non-uniform heat transfer of assembled walls based on infrared image segmentation according to claim 1, characterized in that: In S3, ResNet18 is used as the basic encoding part of the UNet network, and multiple cross-layer connections are added. The image dimension is unified through the maximum pooling layer. The decoding part also adopts the UNet network structure to form a denser Res-UNet network structure. An improved joint loss function is designed based on the weighted cross entropy and weighted Dice loss functions to improve the performance of the Res-UNet network. After segmentation, accurate feature parameters are obtained, including the total pixel points N in the thermal bridge affected area and the temperature of a single pixel point. The total number of pixels M on the entire wall and the temperature of a single pixel By increasing the loss ratio of the foreground pixel value, the total loss of the improved joint loss function is made the weighted sum of the weighted cross entropy and weighted Dice loss functions, as shown in formula (4): L CD =ω t ×L CEL +(1-ω t )×L DICE (4) Where L CEL Expressed as weighted cross entropy, L DICE It is expressed as the weighted Dice loss function, which is specifically expressed as formula (5); ω t Represented as weights between loss functions; Among them, ω pos is the proportion of foreground pixel value, P pos is the predicted label of the foreground pixel, T pos is the true label of the foreground pixel, P neg is the predicted label of the background pixel, T neg is the true label of the background pixel, L cel represents the cross entropy loss, L dice Represents Dice loss.

5. The method for analyzing non-uniform heat transfer of assembled walls based on infrared image segmentation according to claim 1, characterized in that: In S4, since the measured temperature is different from the infrared temperature, the infrared temperature before correction T is used. IR , Indoor air temperature T IA and indoor air quality R IA As the input variable of the temperature correction model, the calibrated temperature is used as the output to construct the BP neural network. The relationship can be expressed as TC =f(T IR ,T IA ,R IA ) indicates that a correction model is established based on genetic algorithm to optimize BP neural network to correct infrared temperature; In addition, the characteristic parameters are used to calculate the cumulative temperature difference ratio T' and the equivalent heat transfer coefficient K' of the wall non-uniform heat transfer evaluation index. T ; The original calculation formula of the cumulative temperature difference ratio T' is shown in formula (6); In formula (6), t(x) is the wall temperature distribution function, °C; t0 is the temperature value of the main part, °C; l is the coordinate of the area affected by non-uniform heat transfer, m; L is the coordinate of the main part, m; T is the cumulative temperature difference of the thermal bridge affected area, T 0为 The accumulated temperature difference of the whole wall; The calculation formula (7) converted into the cumulative temperature difference ratio based on thermal imaging is as follows: In formula (7), N is the total number of pixels in the seam-affected area; is the temperature of a single pixel in the unblocking affected area; M is the total number of pixels on the entire wall; The temperature of a single pixel point on the entire wall; Equivalent heat transfer coefficient K T The calculation formula (8) is as follows: h is the height of the wall; q(x) is the function of the heat flux density at different distances from the joint normal, which can be obtained through FLUENT simulation; x0 and x1 are the starting point and end point coordinates of the thermal bridge influence distance; l0 is the actual structural width of the thermal bridge part; is the physical area of ​​the thermal bridge, m2; ΔT is the temperature difference between indoor and outdoor; The equivalent heat transfer coefficient calculation formula (9) based on thermal imaging is as follows: In formula (9), T a (x) is the temperature difference function of the inner and outer walls at different distances from the joint normal, T i and M are the temperature of a single pixel and the total number of pixels within the thermal bridge influence range, respectively. T is the thermal bridge influence area, ΔT is the temperature difference between the inner and outer surfaces of the wall, is the total thermal resistance per unit area of ​​the flat wall, x0 and x1 are the coordinates of the starting and ending points of the thermal bridge influence distance, and h is the height of the wall.

Images