Method for detecting bonding effect of external wall insulation board and AR glasses

Abstract

The invention provides a method for detecting the bonding effect of an external wall insulation board and AR glasses, and belongs to the technical field of image data processing.The method for detecting the bonding effect of the external wall insulation board and the AR glasses comprise the steps that AR glasses image data and ultrasonic detection equipment data are collected and subjected to space-time registration; and generating an optimal observation point vector group through fusion analysis, and guiding an inspector to adjust the position of detection equipment. Collecting data at the optimal position to form a detection data matrix, disassembling the detection data matrix into a basic matrix, a first variation matrix and a second variation matrix based on a stability index, calculating contribution values and feature vectors of the matrixes, inputting the contribution values and the feature vectors into an external wall insulation board bonding effect analysis model, and outputting a bonding effect index value; the technical problem that the bonding effect of the external wall insulation board cannot be accurately and efficiently detected in the prior art is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of image data processing, and in particular relates to a method for detecting the bonding effect of exterior wall insulation boards and AR glasses. Background Art

[0002] Exterior wall insulation systems are a crucial component of modern building energy conservation. The bond quality between insulation panels and walls directly impacts building energy efficiency and safety. Traditional testing methods include tapping, pull-out testing, and infrared thermal imaging. The tapping method relies on the tester's experience and is highly subjective, relying on sound quality. While accurate, the pull-out test is destructive and cannot be applied on a large scale. Infrared thermal imaging testing is significantly affected by ambient temperature and suffers from reduced reliability in complex weather conditions.

[0003] In recent years, ultrasonic testing technology has been increasingly used in exterior wall insulation system inspections due to its non-destructive nature and applicability. However, single-use ultrasonic testing presents challenges such as complex data interpretation and inaccurate spatial positioning. Meanwhile, the application of augmented reality (AR) technology in building inspections offers a visual data presentation method, but lacks effective integration with specialized testing equipment.

[0004] Current technology for testing the adhesion of exterior wall insulation panels faces challenges such as inaccurate data collection, difficulty integrating multi-source data, and inconsistent quantitative standards for test results. In particular, there is a lack of methods to effectively integrate AR visualization technology with ultrasonic testing data, resulting in low testing efficiency and limited reliability. In other words, existing technology hinders accurate and efficient testing of the adhesion of exterior wall insulation panels. Summary of the Invention

[0005] In view of this, the present invention provides a method for detecting the bonding effect of exterior wall insulation boards and AR glasses, which can solve the technical problem in the prior art that the bonding effect of exterior wall insulation boards cannot be accurately and efficiently detected.

[0006] The present invention is implemented as follows: The present invention provides a method for detecting the bonding effect of exterior wall insulation boards and AR glasses, including: collecting AR glasses image data and ultrasonic detection equipment detection data, performing spatiotemporal registration and fusion analysis on the image data and the detection data, and generating an optimal observation point vector group; guiding the inspectors to adjust the AR glasses to the optimal observation point coordinate position of the AR glasses in turn, and at the same time aligning the ultrasonic detection equipment with the optimal detection point coordinate position of the ultrasonic detection equipment; collecting data at the optimal observation point coordinate position of the AR glasses and the optimal detection point coordinate position of the ultrasonic detection equipment to form a detection data matrix; decomposing the detection data matrix into a basic matrix, a first change matrix and a second change matrix based on a stability index and calculating the contribution values and eigenvectors of the three matrices; inputting the eigenvectors and contribution values of the three matrices into a pre-trained exterior wall insulation board bonding effect analysis model; and outputting the detection result matrix through the exterior wall insulation board bonding effect analysis model.

[0007] Among them, spatiotemporal registration specifically aligns the AR glasses image data with the ultrasonic detection equipment detection data in the time dimension and spatial coordinate system to ensure that the two data sources describe information about the same physical location.

[0008] Among them, the optimal observation point vector group is specifically a group of vectors calculated by the data fusion algorithm. Each vector contains two coordinate points, which respectively represent the optimal observation point coordinates of the AR glasses and the optimal detection point coordinates of the ultrasonic detection equipment required to obtain the best detection results.

[0009] Among them, the stability index is a dimensionless parameter that measures the stability of the change trend of elements in the detection data matrix, and is used to distinguish between stable components and components with different degrees of change in the detection data matrix.

[0010] Among them, the basic matrix is specifically a sub-matrix representing the basic structural information in the detection data matrix, reflecting the basic bonding state of the exterior wall insulation board; the first change matrix is specifically a sub-matrix representing the change information in the detection data matrix whose stability index is higher than the preset threshold, reflecting the main change characteristics in the bonding state of the exterior wall insulation board; the second change matrix is specifically a sub-matrix representing the change information in the detection data matrix whose stability index is lower than the preset threshold, reflecting the secondary change characteristics in the bonding state of the exterior wall insulation board.

[0011] The contribution value is specifically a numerical value that quantifies the degree of influence of the basic matrix, the first change matrix, and the second change matrix on the final detection result, which is obtained by normalization calculation and the sum is 1.

[0012] Among them, the bonding effect index is a dimensionless parameter that quantifies the degree of bonding between the exterior wall insulation board and the wall. 1 indicates complete bonding and 0 indicates complete detachment.

[0013] Among them, the contribution values of the three matrices are calculated using the matrix contribution optimization function. The input includes the detection data matrix, the basic matrix, the first change matrix, the second change matrix and the ambient temperature parameter. It is used to dynamically adjust the weight distribution of the basic matrix, the first change matrix and the second change matrix in the final result. The output is the contribution value.

[0014] Among them, the specific structure of the adhesion effect analysis model of external wall insulation boards is a hybrid neural network based on a multi-layer perceptron and a cross-attention mechanism, which includes a feature extraction layer, two cross-attention layers and three fully connected layers. The cross-attention layer uses a sparse attention mechanism to perform weighted analysis on the input basic matrix eigenvector, the first change matrix eigenvector and the second change matrix eigenvector. The number of attention heads in the sparse attention mechanism is dynamically adjusted according to the preset threshold of the stability index. The higher the preset threshold of the stability index, the more attention heads there are. The output layer of the adhesion effect analysis model of external wall insulation boards uses a softmax activation function to map the result to the range of 0 to 1 as the adhesion effect index.

[0015] The present invention also provides an AR glasses for detecting the bonding effect of exterior wall insulation boards, which includes a processor for executing the above method.

[0016] Compared with the existing technology, the present invention provides a method and AR glasses for detecting the bonding effect of exterior wall insulation boards. The present invention proposes a method and AR glasses for detecting the bonding effect of exterior wall insulation boards. Through spatiotemporal registration technology, AR image data and ultrasonic detection data are accurately aligned, and an optimal observation point vector group is generated to guide the detection position, thereby realizing the effective fusion and collaborative detection of multi-source data.

[0017] This method utilizes matrix decomposition and feature extraction techniques to decompose the test data matrix into a basic matrix, a first variation matrix, and a second variation matrix, respectively representing the basic state of insulation board adhesion and varying degrees of variation. A matrix contribution optimization function dynamically adjusts the weights of each matrix, accounting for the impact of external factors such as ambient temperature on the test results, thereby improving the accuracy and adaptability of the test results. Furthermore, a hybrid neural network model based on a multilayer perceptron and a cross-attention mechanism utilizes a sparse attention mechanism to perform weighted analysis on feature vectors, quantifying the adhesion effect as an exponential value ranging from 0 to 1.

[0018] Compared with traditional technologies, the present invention realizes precise guidance of the detection process, intelligent analysis of data and standardized quantification of results, solving the technical problem in the existing technology that the bonding effect of exterior wall insulation boards cannot be accurately and efficiently detected, and provides more efficient and reliable technical support for the quality control of exterior wall insulation systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1A flowchart of the method provided in the first aspect of the present invention. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0021] like Figure 1 FIG. 1 is a flow chart of a method for detecting the bonding effect of exterior wall insulation boards provided by the first aspect of the present invention. The method comprises the following steps: S01. Collect AR glasses image data and obtain ultrasonic detection equipment detection data, and perform spatiotemporal registration of the AR glasses image data with the ultrasonic detection equipment detection data; S02. Perform fusion analysis on the spatiotemporal registration data to generate an optimal observation point vector group, where the optimal observation point vector group includes the coordinates of the optimal observation points of the AR glasses and the coordinates of the optimal detection points of the ultrasonic detection equipment; S03, guiding the inspector to sequentially adjust the AR glasses to the coordinate position of the optimal observation point of the AR glasses, and at the same time align the ultrasonic detection device with the coordinate position of the optimal detection point of the ultrasonic detection device; S04. Collecting data at the coordinate position of the optimal observation point of the AR glasses and the coordinate position of the optimal detection point of the ultrasonic detection device to form a detection data matrix; S05. Decomposing the detection data matrix into a basic matrix, a first variation matrix, and a second variation matrix based on the stability index; S06. Calculate contribution values of the basic matrix, the first variation matrix, and the second variation matrix using a matrix contribution optimization function, where inputs of the matrix contribution optimization function include the detection data matrix, the basic matrix, the first variation matrix, the second variation matrix, and an ambient temperature parameter; S07. Calculate the eigenvectors of the basic matrix, the first change matrix, and the second change matrix respectively to obtain the eigenvector of the basic matrix, the eigenvector of the first change matrix, and the eigenvector of the second change matrix; S08. Inputting the basic matrix eigenvector, the first variable matrix eigenvector, the second variable matrix eigenvector, and the contribution value into a pre-trained exterior wall insulation board bonding effect analysis model; S09. Output a detection result matrix through the external wall insulation board bonding effect analysis model, where the elements in the detection result matrix are the bonding effect indexes of the external wall insulation boards at corresponding coordinate points, and the bonding effect index ranges from 0 to 1.

[0022] Among them, spatiotemporal registration specifically aligns the AR glasses image data with the ultrasonic detection equipment detection data in the time dimension and spatial coordinate system to ensure that the two data sources describe information of the same physical location.

[0023] Among them, the optimal observation point vector group is specifically a group of vectors calculated by the data fusion algorithm, and each vector contains two coordinate points, which respectively represent the optimal observation point coordinates of the AR glasses and the optimal detection point coordinates of the ultrasonic detection equipment required to obtain the best detection results.

[0024] The stability index is a dimensionless parameter that measures the stability of the change trend of elements in the detection data matrix, and is used to distinguish between stable components and components with different degrees of change in the detection data matrix.

[0025] The basic matrix is specifically a sub-matrix representing the basic structural information in the detection data matrix, reflecting the basic bonding state of the exterior wall insulation board.

[0026] The first variation matrix specifically represents a variation information submatrix in which the stability index in the detection data matrix is higher than a preset threshold, and reflects the main variation characteristics in the bonding state of the exterior wall insulation board.

[0027] The second variation matrix specifically represents a variation information submatrix in which the stability index in the detection data matrix is lower than the preset threshold, reflecting minor variation characteristics in the bonding state of the exterior wall insulation board.

[0028] The contribution value is specifically a numerical value that quantifies the degree of influence of the basic matrix, the first change matrix, and the second change matrix on the final detection result, which is obtained by normalization calculation and the sum is 1.

[0029] Among them, the bonding effect index is a dimensionless parameter that quantifies the degree of bonding between the exterior wall insulation board and the wall. 1 indicates complete bonding and 0 indicates complete detachment.

[0030] The ambient temperature parameter is specifically the real-time temperature value of the detection environment collected by the temperature sensor, which is used to adjust the weights of different matrices in the matrix contribution optimization function.

[0031] Among them, the matrix contribution optimization function is used to dynamically adjust the weight distribution of the basic matrix, the first variable matrix and the second variable matrix in the final result. The input includes the detection data matrix, the basic matrix, the first variable matrix, the second variable matrix and the ambient temperature parameter, and the output is the contribution value.

[0032] The preset threshold is specifically a critical value determined based on statistical analysis of historical detection data, and is used to divide the first change matrix and the second change matrix.

[0033] The specific structure of the bonding effect analysis model of the exterior wall insulation board is a hybrid neural network based on a multi-layer perceptron and a cross-attention mechanism, comprising a feature extraction layer, two cross-attention layers and three fully connected layers, wherein the cross-attention layer uses a sparse attention mechanism to perform weighted analysis on the input basic matrix eigenvector, the first variable matrix eigenvector and the second variable matrix eigenvector. The number of attention heads in the sparse attention mechanism is dynamically adjusted according to the preset threshold of the stability index. The higher the preset threshold of the stability index, the more attention heads there are. The output layer of the bonding effect analysis model of the exterior wall insulation board uses a softmax activation function to map the result to the range of 0 to 1 as the bonding effect index.

[0034] The steps for establishing the training data set for the exterior wall insulation board bonding effect analysis model specifically include collecting AR image data and ultrasonic detection data from the construction sites of exterior wall insulation boards of multiple types of buildings, standardizing and enhancing the collected data to improve the generalization ability of the exterior wall insulation board bonding effect analysis model, manually annotating the insulation board bonding effects as supervised learning labels by engineering experts, and dividing the data set into training set, validation set, and test set based on factors such as building type, environmental conditions, and insulation board material to ensure balanced and representative data distribution.

[0035] The steps of training the external wall insulation board bonding effect analysis model specifically include using a cross-entropy loss function to measure the difference between the predicted value and the label value of the external wall insulation board bonding effect analysis model, using an adaptive moment estimation optimization algorithm to iteratively update the parameters of the external wall insulation board bonding effect analysis model, introducing an early stopping mechanism during the training process to avoid overfitting, dynamically adjusting the parameter update step size through a learning rate scheduler, regularly evaluating the performance of the external wall insulation board bonding effect analysis model on a validation set and saving the optimal model parameters, and terminating the training process when the performance of the external wall insulation board bonding effect analysis model on the validation set no longer improves.

[0036] The specific implementation methods of the above steps are described in detail below. The specific implementation method of step S01 is to first collect high-definition image data of the surface of the exterior wall insulation board through the binocular camera built into the AR glasses, and at the same time use the ultrasonic detection equipment to emit ultrasonic waves and receive reflected waves, and record waveform data. The image data collected by the AR glasses contains RGB three-channel information and depth information, and each pixel corresponds to a three-dimensional spatial coordinate. The data collected by the ultrasonic detection equipment contains waveform amplitude, frequency and phase information. The spatiotemporal registration process adopts a bidirectional Hough transform algorithm. First, feature points are extracted from the two types of data, and a corresponding relationship between the feature points is established. Then, the two sets of data are mapped to a unified reference coordinate system through a rigid body transformation matrix. Finally, timestamp alignment is used to ensure the synchronization of the two sets of data. The main purpose of spatiotemporal registration is to establish an accurate correspondence between the AR glasses visual data and the ultrasonic detection data, providing a basis for subsequent fusion analysis. The registration accuracy can reach the sub-millimeter level, and the time synchronization error is controlled within 10 milliseconds.

[0037] The specific implementation method of step S02 is to apply a multimodal data fusion algorithm to the registered data, including two stages: low-level feature fusion and high-level semantic fusion. Low-level feature fusion uses wavelet transform to extract the frequency domain features of the two data, while high-level semantic fusion uses deep convolutional neural network to extract semantic information. The fusion process is implemented by a variational autoencoder with an attention mechanism, which adaptively weights the importance of different modal data. The optimal observation point vector group is generated by the gradient descent method, with information entropy and detection accuracy as optimization goals, and the optimal observation position is iteratively calculated. The algorithm input includes the fused feature map and environmental constraint parameters, and the output is a vector group containing multiple sets of coordinate pairs, each set of coordinate pairs represents the optimal position of AR glasses and ultrasonic detection equipment. The purpose of this step is to determine the spatial position combination that can obtain the best detection effect and improve the accuracy and reliability of subsequent detection.

[0038] The specific implementation method of step S03 is to convert the optimal observation point vector group data into visual guidance information of the AR glasses display system, and present the navigation path and target position in the field of view of the AR glasses through a semi-transparent overlay. The visual guidance adopts real-time rendering technology, and uses different colors to indicate the deviation between the current position and the target position. Green indicates that the position is appropriate, yellow indicates that it is close but needs fine-tuning, and red indicates that the deviation is large. At the same time, the corresponding positioning guidance information, including direction arrows and distance values, is also displayed on the display screen of the ultrasonic detection equipment to help the detection personnel accurately adjust the position of the equipment. This step guides the detection personnel to the calculated optimal detection position in turn through human-computer interaction, ensuring the accuracy and consistency of data collection, and providing high-quality raw data for subsequent analysis. The positioning accuracy is required to be within ±2 mm, and the angle error is controlled within ±1 degree.

[0039] The specific implementation method of step S04 is to synchronously trigger data acquisition after confirming that the AR glasses and the ultrasonic detection equipment have reached the optimal position. The AR glasses collect high-resolution image data, including visible light images and depth images, and perform real-time feature extraction. The extracted features include edge features, texture features, and color features. The ultrasonic detection equipment collects complete waveform data, including A-scan and B-scan data, and records equipment parameters such as frequency, gain, and focus depth. During the acquisition process, a window filtering algorithm is applied to preprocess the raw data to eliminate noise interference. All collected data are arranged according to a preset spatial sampling grid to form a multi-dimensional detection data matrix. Each element of the detection data matrix contains position coordinates and corresponding detection feature vectors. The purpose of this step is to obtain high-quality original detection data and provide a complete information basis for subsequent analysis. The data sampling rate is not less than 20 frames per second, and the sampling point spacing is controlled within 10 mm.

[0040] The specific implementation of step S05 is to first calculate the stability index of each element in the detection data matrix. The stability index is determined by the sliding window variance analysis method. The calculation window size is The stability index ranges from 0 to 1, and the larger the value, the more stable the data. With the stability index of 0.75 as the preset threshold, the detection data matrix is subjected to singular value decomposition to extract the basic structure corresponding to the main singular values. The basic matrix is composed of the previous The singular vectors corresponding to the largest singular values are composed of The energy retention ratio is determined, generally taking 80% of the total energy. The first variation matrix is composed of variation components whose stability index is greater than a preset threshold, reflecting the main variation characteristics; the second variation matrix is composed of variation components whose stability index is less than a preset threshold, mainly including secondary variation characteristics and noise. The matrix decomposition process uses an iterative regularization method to ensure that the sum of the three sub-matrices can accurately restore the original detection data matrix. The purpose of this step is to decompose complex detection data into components with different physical meanings to facilitate subsequent targeted analysis.

[0041] The specific implementation of step S06 is to construct a matrix contribution optimization function based on temperature compensation, which uses an adaptive weighting method to dynamically adjust the importance of different sub-matrices. First, the corresponding relationship between ambient temperature and material properties is established by a nonlinear regression method, and the temperature range is considered. to Then, the similarity between the basic matrix, the first change matrix, the second change matrix and the original detection data matrix is calculated, and the similarity is measured using the cosine similarity index. The optimization function comprehensively considers the similarity, the temperature influence factor and the prior knowledge, and solves the optimal contribution value by the Lagrange multiplier method. The temperature influence factor is obtained by fitting the experimental data, and the temperature rises by 1. , the contribution of the basic matrix increases by approximately 0.05, the contribution of the first variable matrix decreases by approximately 0.03, and the contribution of the second variable matrix decreases by approximately 0.02. The calculated contribution values are normalized to ensure that the sum is 1. The purpose of this step is to determine the influence weight of different sub-matrices in the final test results and improve the adaptability of the test results to environmental conditions.

[0042] The specific implementation of step S07 is to apply the eigenvalue decomposition algorithm to the basic matrix, the first change matrix and the second change matrix respectively to calculate the eigenvalues and eigenvectors of each. The eigenvalue decomposition process uses the power iteration method, and the iteration accuracy is set to . For each matrix, the eigenvectors corresponding to the largest eigenvalues are selected to form a set of eigenvectors. The number of eigenvectors is determined according to the cumulative contribution rate, and eigenvectors with a cumulative contribution rate of 90% are usually retained. The extracted eigenvectors are orthogonalized to ensure that the vectors are mutually orthogonal. The eigenvectors are then subjected to dimensionality reduction, and the principal component analysis method is used to map the eigenvectors to a lower-dimensional feature space, retaining no less than 95% of the original data. Finally, the eigenvectors after dimensionality reduction are standardized so that their norm is 1. The purpose of this step is to extract the main features of each submatrix, reduce the data dimension, and provide efficient feature representation for subsequent model analysis.

[0043] The specific implementation method of step S08 is to use the basic matrix eigenvector, the first change matrix eigenvector, and the second change matrix eigenvector obtained in step S07 together with the contribution value calculated in step S06 as input and send them to a pre-trained exterior wall insulation board bonding effect analysis model. The input data is first batch normalized to ensure the consistency of the numerical distribution. The model adopts a hybrid neural network structure based on a multi-layer perceptron and a cross-attention mechanism, and contains multiple functional modules. After the data is input, it is first preprocessed by a batch normalization layer, and then enters the feature extraction layer to extract deep features. Then, two cross-attention layers are used to perform weighted fusion of features from different sources, and finally three fully connected layers are used to realize nonlinear mapping and decision-making. A residual connection structure is set up inside the model to alleviate the gradient vanishing problem. The purpose of this step is to use the pre-trained model to comprehensively analyze the extracted features and generate a prediction result of the insulation board bonding effect.

[0044] The specific implementation of step S09 is to receive the output vector of the external wall insulation board bonding effect analysis model and reshape it into a two-dimensional result matrix corresponding to the spatial distribution of the detection area. Each element in the result matrix represents the bonding effect index of the corresponding position, with a value range of 0 to 1. The sparse detection points are interpolated and expanded using a bilinear interpolation algorithm to generate a complete bonding effect distribution map. Gaussian filtering is then applied for smoothing to reduce the impact of local noise. A color mapping scheme is set according to the bonding effect index. An index value greater than 0.8 is displayed as green, indicating good bonding; 0.5 to 0.8 is displayed as yellow, indicating fair bonding; and less than 0.5 is displayed as red, indicating poor bonding. The final test results are superimposed and displayed in the field of view of AR glasses in an augmented reality manner, intuitively showing the bonding condition of the external wall insulation boards. At the same time, a test report is generated, including statistical data and risk assessment. The purpose of this step is to convert the model analysis results into intuitive and understandable visual information to help testers accurately judge the bonding quality of the external wall insulation boards.

[0045] The detailed architecture of the exterior wall insulation board adhesion analysis model utilizes a hybrid neural network based on a multilayer perceptron and a crisscross attention mechanism. The feature extraction layer uses a fully connected architecture, with an input dimension equal to the total dimension of the feature vector and an output dimension of 512. A Reluctant Unit (ReLU) activation function is used. The two crisscross attention layers employ a multi-head self-attention mechanism, with the number of attention heads dynamically adjusted based on a preset stability index threshold. When the preset stability index threshold is 0.75, the number of attention heads is set to 8, and the number increases by 1 for every 0.05 increase in the threshold. The first crisscross attention layer primarily processes the interaction between the eigenvectors of the base matrix and the eigenvectors of the first change matrix, while the second crisscross attention layer processes the interaction between the output of the first layer and the eigenvectors of the second change matrix. Attention computation utilizes a scaled dot product attention mechanism, with the scaling factor set to the square root of the attention head dimension. The three fully connected layers in the model reduce the feature dimension from 512 to 256, 128, and 64, respectively. Each layer is followed by a batch normalization layer and a dropout layer with a dropout probability set to 0.3. The output layer is a fully connected layer with an output dimension equal to the number of discrete sampling points in the detection area. A softmax activation function is used to map the output to a range of 0 to 1, which serves as the bonding effect index. Weight regularization is used throughout the network training process to prevent overfitting, with a regularization coefficient set to 0.001.

[0046] The training dataset was established by collecting data samples from exterior wall insulation projects in different climates and building types, covering 50 construction projects in cold, temperate, and subtropical regions, with 30 to 50 sets of data collected for each project. The collected raw data was standardized to eliminate the influence of equipment differences. Data augmentation techniques include random rotation The dataset was expanded to three times the original size by applying various methods, including scaling by ±10%, adding Gaussian noise, and so on. Data annotation was performed by engineering experts with at least five years of experience. Actual adhesion performance was determined through actual peel tests and used as label values. The dataset was divided into training, validation, and test sets in a 7:2:1 ratio, ensuring a balanced distribution of different building types, environmental conditions, and insulation board materials across the subsets.

[0047] The training process for the exterior wall insulation board adhesion analysis model adopts a phased training strategy. A high learning rate of 0.001 is used in the initial phase to rapidly search the parameter space. In the mid-stage, the learning rate is reduced to 0.0001 for fine-tuning parameters. In the final stage, the learning rate is further reduced to 0.00001 to achieve model convergence. A weighted cross-entropy loss function is used, with increased weight given to samples with poor adhesion to improve the model's sensitivity to problem areas. The optimization algorithm uses an adaptive moment estimation algorithm with momentum, a momentum parameter of 0.9, and a decay parameter of 0.999. An early stopping mechanism is introduced during training, terminating training when the loss on the validation set stops decreasing for 10 consecutive epochs. A cosine annealing strategy with warm restarts is used for learning rate scheduling, with an initial period of 5 epochs and a maximum period of 20 epochs. The model is evaluated on the validation set using mean absolute error, root mean square error, and F1 score, with the F1 score serving as the primary metric for selecting optimal model parameters. The training equipment uses a high-performance computing platform equipped with GPU acceleration, the batch size is set to 64, and the total number of training epochs is 100.

[0048] The AR glasses provided in the second aspect of the present invention primarily consist of a binocular camera, a depth sensor, a processor, a wireless communication module, a transparent display, a head-mounted bracket, and a rechargeable battery. The binocular camera captures stereoscopic images with a resolution of 4K. The depth sensor uses structured light technology, achieving a measurement accuracy of 1 mm. The processor utilizes a high-performance ARM architecture, supporting real-time image processing and augmented reality rendering. The wireless communication module supports WiFi and Bluetooth protocols, enabling data exchange with other devices. The transparent display utilizes optical waveguide technology, with a resolution of 1920×1080 and a field of view of 52 degrees.

[0049] The ultrasonic testing equipment primarily consists of an ultrasonic transmitter, receiver, signal processing unit, LCD display, touchscreen operation panel, and portable power supply. The ultrasonic transmitter has a frequency range of 20kHz to 100kHz, with adjustable transmission power. The receiver utilizes a highly sensitive piezoelectric element, achieving a signal-to-noise ratio greater than 40dB. The signal processing unit includes an analog-to-digital converter, a digital signal processor, and a data storage module, with a sampling rate of 100MHz and a bit depth of 12 bits. The LCD display measures 7 inches and has a resolution of 1024×768. The touchscreen operation panel supports multi-touch, facilitating parameter adjustment and operational control. The portable power supply utilizes a lithium-ion battery with a battery life of more than 8 hours.

[0050] The mathematical model or calculation process involved in the present invention is described in detail below.

[0051] The spatiotemporal registration process in step S01 involves the bidirectional Hough transform algorithm, which achieves the registration of two types of data through feature point matching and coordinate transformation. The specific calculation formula for spatiotemporal registration is as follows: First, the feature point extraction and description in AR glasses image data can be expressed as: ; Where, is the set of feature points of the AR glasses image; For the The descriptor of each feature point contains location and description information; is the total number of feature points.

[0052] The extraction and description of feature points in the detection data of ultrasonic testing equipment can be expressed as: ; Where, is a set of ultrasonic detection feature points; For the Descriptors of feature points; is the total number of feature points.

[0053] The similarity matrix between feature points is calculated as: ; Where, For the AR feature points and The similarity between ultrasonic feature points; Represents vector dot product operation; Represents the vector norm.

[0054] Based on the similarity matrix, the corresponding relationship of feature points is established: ; Where, is the corresponding relationship set of feature points; is the similarity threshold, and its value range is to , the default value is .

[0055] According to the corresponding relationship, calculate the rigid body transformation matrix: ; Where, is the transformation matrix; is the rotation matrix; is the translation vector; AR feature points The spatial coordinates of Ultrasonic feature points The spatial coordinates of .

[0056] Rotation Matrix and translation vectors The solution is the singular value decomposition method: ; ; ; ; Where, is the covariance matrix; and are the average coordinates of AR feature points and ultrasonic feature points, respectively; represents singular value decomposition; 、 、 For the decomposition result.

[0057] Time synchronization is achieved through timestamp interpolation: ; Where, is the synchronized ultrasonic data; and Timestamp and The original ultrasonic data at is the interpolation coefficient; is the timestamp of the AR data, .

[0058] Registration error evaluation formula: ; Where, is the registration error; is the number of feature point correspondences.

[0059] The multimodal data fusion algorithm in step S02 includes wavelet transform and deep convolutional neural network, and realizes data fusion through variational autoencoder. The calculation of the optimal observation point vector group involves iterative calculation of gradient descent method. The specific formula is as follows: Wavelet transform extracts frequency domain features: ; ; Where, and are the wavelet coefficients of AR data and ultrasonic data, respectively; and They are AR data and ultrasonic data respectively; is the conjugate of the wavelet basis function; is the scale parameter, and its value range is to ; is the translation parameter.

[0060] Feature fusion weight calculation: ; ; Where, and are the fusion weights of AR data and ultrasonic data respectively; and are the variances of AR data and ultrasonic data, respectively.

[0061] Fusion feature calculation: ; Where, is the fused feature; and Features extracted from AR data and ultrasonic data respectively.

[0062] Information entropy calculation: ; Where, is information entropy; For location The probability distribution at is obtained by the normalized eigenvector.

[0063] Objective function of the optimal observation point vector group: ; Where, is the objective function; is the observation point vector group; is information entropy; is the prediction accuracy; is the environmental constraint; 、 and is the weight coefficient, and its value range is 、 and , the default values are 、 and .

[0064] Gradient descent update formula: ; Where, and Respectively Second and The observation point vector group of iterations; is the learning rate, and the initial value is , decays with the number of iterations; The objective function is about gradient.

[0065] The stability index calculation and matrix decomposition in step S05 involve sliding window variance analysis and singular value decomposition algorithm. The specific formula is as follows:

[0066] Stability index calculation: ; Where, For location Stability index at For location Central The variance of the data within the window; is the maximum value of all window variances.

[0067] Check the singular value decomposition of the data matrix: ; Where, is the detection data matrix; and are the left singular vector and right singular vector matrices respectively; is a diagonal matrix of singular values.

[0068] Construction of the basic matrix: ; Where, is the basic matrix; For the singular values; and Respectively left and right singular vectors; is the number of singular values selected, satisfying , is the total number of singular values.

[0069] Construction of the first change matrix and the second change matrix: ; ; Where, is the first change matrix; is the second change matrix; and The detection data matrix and the basic matrix are respectively The element value at ; For location Stability index at is the preset threshold value, the value is .

[0070] Iterative regularization method ensures the accuracy of matrix decomposition: ; ; ; Where, represents the Frobenius norm; represents the nuclear norm; express norm; 、 and is the regularization coefficient, and its value range is 、 and , the default values are 、 and .

[0071] The matrix contribution optimization function in step S06 involves nonlinear regression, cosine similarity calculation and Lagrange multiplier method. The specific formula is as follows: Calculation of temperature impact factor: ; Where, is the temperature influencing factor; is the ambient temperature in degrees Celsius; 、 and is the regression coefficient, which is obtained by fitting the experimental data and has a range of values of 、 and , the default values are 、 and .

[0072] Cosine similarity calculation: ; ; ; Where, 、 and are the cosine similarities between the basic matrix, the first change matrix, the second change matrix and the original detection data matrix respectively; represents the matrix inner product; represents the Frobenius norm.

[0073] Matrix contribution optimization function: ; ; ; Where, Contribute optimization functions to matrices; 、 and are the contribution weights of the basic matrix, the first change matrix, and the second change matrix respectively; 、 and are the cosine similarities between the three matrices and the original data; is the temperature influencing factor.

[0074] Lagrange multiplier method to solve: ; Where, is the Lagrangian function; is the Lagrange multiplier.

[0075] By solving Get the optimal contribution value 、 and ,satisfy: ; ; ; .

[0076] The eigenvalue decomposition and eigenvector calculation in step S07 involve power iteration method, orthogonalization and principal component analysis. The specific formula is as follows: The power iteration method calculates the maximum eigenvalue and the corresponding eigenvector: Initial vector: , randomly initialized and ; Iteration formula: ; Convergence conditions: ; Eigenvalue calculation: ; Where, is the matrix to be decomposed (basic matrix, first change matrix or second change matrix); and Respectively Second and The feature vector of the iteration; is the corresponding eigenvalue; express norm.

[0077] Eigenvector orthogonalization (Gram-Schmidt method): ; ; ; Where, is the original eigenvector; is the orthogonalized vector; is the normalized orthogonal eigenvector; represents the vector inner product; express norm.

[0078] Eigenvector selection criteria: ; Where, For the eigenvalues; is the number of selected eigenvalues; is the total number of eigenvalues.

[0079] Principal component analysis dimensionality reduction: Compute the covariance matrix: ; Eigenvalue decomposition: ; Dimensionality reduction transformation: ; Where, is the original eigenvector matrix; is the covariance matrix; is the eigenvector matrix; is the eigenvalue diagonal matrix; For the selected The matrix of eigenvectors; is the eigenvector matrix after dimensionality reduction; is the dimension after dimensionality reduction, and the information retention rate is not less than .

[0080] The bilinear interpolation algorithm in step S09 is used to expand the sparse detection points and generate a complete bonding effect distribution map. The specific formula is as follows: Bilinear interpolation calculation: ; Where, For location The interpolation result at ; 、 、 and They are the bonding effect indexes of four adjacent test points; 、 、 and are the coordinates of four adjacent detection points, satisfying and .

[0081] Gaussian filter smoothing: ; ; Where, is the Gaussian kernel function; is the standard deviation, and its value range is , the default value is ; is the bonding effect index after smoothing; is the original bonding effect index; is the filter window radius, and its value is , Represents the ceiling function.

[0082] The meanings and numerical selections of the variables and parameters in all the above formulas are based on the physical properties and experimental data of the adhesion effect of the exterior wall insulation board. For example, the similarity threshold in spatiotemporal registration The value is , is the optimal threshold determined based on a large amount of experimental data. It can ensure matching accuracy while avoiding excessive screening that leads to too few feature point correspondences. The temperature influence factor adopts the form of a quadratic polynomial because there is usually a nonlinear relationship between the physical properties of exterior wall insulation materials and temperature, especially under low and high temperature conditions, which tends to show an accelerated change trend. The quadratic term can better fit this relationship. Stability index threshold Set to , is an empirical value obtained by analyzing the bonding state of insulation boards under various building types and environmental conditions. It can effectively distinguish between major and minor change characteristics.

[0083] Specifically, the principles of this invention are as follows: The technical principles of this invention are based on the integrated application of multi-source data fusion, matrix decomposition, and deep learning. First, spatiotemporal registration technology ensures that the AR glasses image data and the ultrasonic detection equipment data are precisely aligned in the time dimension and spatial coordinate system, providing a unified data foundation for subsequent analysis. The data fusion algorithm calculates the optimal observation point vector group to achieve precise positioning of the detection equipment, solving the data inconsistency problem caused by random position selection in traditional detection.

[0084] The core of this invention is the principle of matrix decomposition. Based on the stability index, the test data matrix is broken down into sub-matrices representing different levels of information. The basic matrix reflects the stable components of the basic bonding state, while the first and second variable matrices capture the major and minor variations, respectively. This decomposition method is consistent with the physical characteristics of the bonding state of exterior wall insulation panels, which consists of both a stable structure and dynamic changes. The matrix contribution optimization function dynamically adjusts the weight distribution of each matrix by considering the ambient temperature parameter, adapting to the changes in insulation material properties under different environmental conditions and improving the accuracy of the test results.

[0085] The deep learning model design combines the feature extraction capabilities of a multilayer perceptron with the information filtering advantages of a cross-attention mechanism. In particular, the sparse attention mechanism dynamically adjusts the number of attention heads based on a preset stability index threshold, enabling the model to adaptively focus on features of varying importance. The softmax activation function maps the result to a range of 0 to 1, providing an intuitive index of cohesion. The model training process utilizes a cross-entropy loss function and an adaptive moment estimation optimization algorithm, along with an early stopping mechanism and a learning rate scheduler to ensure good generalization.

[0086] The organic combination of these technical principles enables the present invention to overcome the limitations of traditional detection methods, achieve accurate and efficient detection of the bonding effect of exterior wall insulation boards, and meet the actual needs of construction project quality control.

[0087] A specific embodiment 1 of the present invention is provided below. The specific implementation of each step in this embodiment 1 is described in detail as follows.

[0088] The spatiotemporal registration process in step S01 involves the bidirectional Hough transform algorithm, which achieves the registration of two types of data through feature point matching and coordinate transformation. The specific calculation formula for spatiotemporal registration is as follows: First, the feature point extraction and description in AR glasses image data can be expressed as: ; Where, is the set of feature points of the AR glasses image; For the The descriptor of each feature point contains location and description information; is the total number of feature points.

[0089] The extraction and description of feature points in the detection data of ultrasonic testing equipment can be expressed as: ; Where, is a set of ultrasonic detection feature points; For the Descriptors of feature points; is the total number of feature points.

[0090] The similarity matrix between feature points is calculated as: ; Where, For the AR feature points and The similarity between ultrasonic feature points; Represents vector dot product operation; Represents the vector norm.

[0091] Based on the similarity matrix, the corresponding relationship of feature points is established: ;.

[0092] Where, is the corresponding relationship set of feature points; is the similarity threshold, and its value range is to , the default value is .

[0093] According to the corresponding relationship, calculate the rigid body transformation matrix: ; Where, is the transformation matrix; is the rotation matrix; is the translation vector; AR feature points The spatial coordinates of Ultrasonic feature points The spatial coordinates of .

[0094] Rotation Matrix and translation vectors The solution is the singular value decomposition method: ; ; ; ; Where, is the covariance matrix; and are the average coordinates of AR feature points and ultrasonic feature points, respectively; represents singular value decomposition; 、 、 For the decomposition result.

[0095] Time synchronization is achieved through timestamp interpolation: ; Where, is the synchronized ultrasonic data; and Timestamp and The original ultrasonic data at is the interpolation coefficient; is the timestamp of the AR data, .

[0096] Registration error evaluation formula: ; Where, is the registration error; is the number of feature point correspondences.

[0097] The multimodal data fusion algorithm in step S02 includes wavelet transform and deep convolutional neural network, and realizes data fusion through variational autoencoder. The calculation of the optimal observation point vector group involves iterative calculation of gradient descent method. The specific formula is as follows: Wavelet transform extracts frequency domain features: ; ; Where, and are the wavelet coefficients of AR data and ultrasonic data, respectively; and They are AR data and ultrasonic data respectively; is the conjugate of the wavelet basis function; is the scale parameter, and its value range is to ; is the translation parameter.

[0098] Feature fusion weight calculation: ; ; Where, and are the fusion weights of AR data and ultrasonic data respectively; and are the variances of AR data and ultrasonic data, respectively.

[0099] Fusion feature calculation: ; Where, is the fused feature; and Features extracted from AR data and ultrasonic data respectively.

[0100] Information entropy calculation: ; Where, is information entropy; For location The probability distribution at is obtained by the normalized eigenvector.

[0101] Objective function of the optimal observation point vector group: ; Where, is the objective function; is the observation point vector group; is information entropy; is the prediction accuracy; is the environmental constraint; 、 and is the weight coefficient, and its value range is 、 and , the default values are 、 and .

[0102] Gradient descent update formula: ; Where, and Respectively Second and The observation point vector group of iterations; is the learning rate, and the initial value is , decays with the number of iterations; The objective function is about gradient.

[0103] The stability index calculation and matrix decomposition in step S05 involve sliding window variance analysis and singular value decomposition algorithm. The specific formula is as follows: Stability index calculation: ; Where, For location Stability index at For location Central The variance of the data within the window; is the maximum value of all window variances.

[0104] Check the singular value decomposition of the data matrix: ; Where, is the detection data matrix; and are the left singular vector and right singular vector matrices respectively; is a diagonal matrix of singular values.

[0105] Construction of the basic matrix: ; Where, is the basic matrix; For the singular values; and Respectively left and right singular vectors; is the number of singular values selected, satisfying , is the total number of singular values.

[0106] Construction of the first change matrix and the second change matrix: ; ; Where, is the first change matrix; is the second change matrix; and The detection data matrix and the basic matrix are respectively The element value at ; For location Stability index at is the preset threshold value, the value is .

[0107] Iterative regularization method ensures the accuracy of matrix decomposition: ; ; ; Where, represents the Frobenius norm; represents the nuclear norm; express norm; 、 and is the regularization coefficient, and its value range is 、 and , the default values are 、 and .

[0108] The matrix contribution optimization function in step S06 involves nonlinear regression, cosine similarity calculation and Lagrange multiplier method. The specific formula is as follows: Calculation of temperature impact factor: ; Where, is the temperature influencing factor; is the ambient temperature in degrees Celsius; 、 and is the regression coefficient, which is obtained by fitting the experimental data and has a range of values of 、 and , the default values are 、 and .

[0109] Cosine similarity calculation: ; ; ; Where, 、 and are the cosine similarities between the basic matrix, the first change matrix, the second change matrix and the original detection data matrix respectively; represents the matrix inner product; represents the Frobenius norm.

[0110] Matrix contribution optimization function: ; ; ; Where, Contribute optimization functions to matrices; 、 and are the contribution weights of the basic matrix, the first change matrix, and the second change matrix respectively; 、 and are the cosine similarities between the three matrices and the original data; is the temperature influencing factor.

[0111] Lagrange multiplier method to solve: ; Where, is the Lagrangian function; is the Lagrange multiplier.

[0112] By solving Get the optimal contribution value 、 and ,satisfy: ; ; ; .

[0113] The eigenvalue decomposition and eigenvector calculation in step S07 involve power iteration method, orthogonalization and principal component analysis. The specific formula is as follows: The power iteration method calculates the maximum eigenvalue and the corresponding eigenvector: Initial vector: , randomly initialized and ; Iteration formula: ; Convergence conditions: ; Eigenvalue calculation: ; Where, is the matrix to be decomposed (basic matrix, first change matrix or second change matrix); and Respectively Second and The feature vector of the iteration; is the corresponding eigenvalue; express norm.

[0114] Eigenvector orthogonalization (Gram-Schmidt method): ; ; ; Where, is the original eigenvector; is the orthogonalized vector; is the normalized orthogonal eigenvector; represents the vector inner product; express norm.

[0115] Eigenvector selection criteria: ; Where, For the eigenvalues; is the number of selected eigenvalues; is the total number of eigenvalues.

[0116] Principal component analysis dimensionality reduction: Compute the covariance matrix: ; Eigenvalue decomposition: ; Dimensionality reduction transformation: ; Where, is the original eigenvector matrix; is the covariance matrix; is the eigenvector matrix; is the eigenvalue diagonal matrix; For the selected The matrix of eigenvectors; is the eigenvector matrix after dimensionality reduction; is the dimension after dimensionality reduction, and the information retention rate is not less than .

[0117] The bilinear interpolation algorithm in step S09 is used to expand the sparse detection points and generate a complete bonding effect distribution map. The specific formula is as follows: Bilinear interpolation calculation: ; Where, For location The interpolation result at ; 、 、 and They are the bonding effect indexes of four adjacent test points; 、 、 and are the coordinates of four adjacent detection points, satisfying and .

[0118] Gaussian filter smoothing: ; ; Where, is the Gaussian kernel function; is the standard deviation, and its value range is , the default value is ; is the bonding effect index after smoothing; is the original bonding effect index; is the filter window radius, and its value is , Represents the ceiling function.

[0119] The meanings and numerical selections of the variables and parameters in all the above formulas are based on the physical properties and experimental data of the adhesion effect of the exterior wall insulation board. For example, the similarity threshold in spatiotemporal registration The value is , is the optimal threshold determined based on a large amount of experimental data. It can ensure matching accuracy while avoiding excessive screening that leads to too few feature point correspondences. The temperature influence factor adopts the form of a quadratic polynomial because there is usually a nonlinear relationship between the physical properties of exterior wall insulation materials and temperature, especially under low and high temperature conditions, which tends to show an accelerated change trend. The quadratic term can better fit this relationship. Stability index threshold Set to , is an empirical value obtained by analyzing the bonding state of insulation boards under various building types and environmental conditions. It can effectively distinguish between major and minor change characteristics.

[0120] In order to better understand and implement the present invention, Example 2 of a specific application scenario of the present invention is provided below: Researchers conducted a study on the bonding effect detection of exterior wall insulation panels for a 28-story high-rise residential building under construction. The building is located in a mild climate area with an average temperature of 2°C in winter and an average temperature of 28°C in summer. The annual temperature difference is large, and the bonding quality requirements for the exterior wall insulation system are high. The exterior wall of the building uses a 100mm thick graphite polystyrene board as the insulation material, and the point frame method is used for bonding. The theoretical bonding area is not less than 30% of the building code requirements. In order to verify the actual bonding effect, the researchers applied the detection method of the present invention to conduct sampling inspections on the southeast exterior walls of the 5th to 10th floors of the building.

[0121] The testing equipment includes custom-developed AR glasses and an ultrasonic detector. The AR glasses use a binocular camera with a resolution of 4096×2160, a frame rate of 30fps, a field of view of 60 degrees, and a built-in depth sensor with a measurement accuracy of ±0.5mm. The ultrasonic detector uses a transmitter probe with a center frequency of 50kHz, a frequency range of 20-80kHz, a sampling rate of 100MHz, and an accuracy of ±0.1mm. The testing environment temperature is 24°C and the relative humidity is 65%.

[0122] First, we perform spatiotemporal registration and extract 358 feature points from the depth image acquired by AR glasses and 264 feature points from the ultrasonic data. We set the feature point matching similarity threshold. , 147 pairs of valid feature point correspondences are obtained through similarity matrix calculation. Based on these correspondences, the rotation matrix is calculated and translation vectors , registration error , the time synchronization error is 6.5ms.

[0123] Through multimodal data fusion, wavelet transform is used to extract features and set scale parameters The value is arrive There are 5 scales in total. AR data weights are calculated based on variance , ultrasonic data weight . Construct the optimal observation point vector group objective function, the parameter value is 、 and . 25 optimal detection position combinations were calculated by gradient descent method to facilitate detection in different wall areas.

[0124] In actual detection, data is collected at each selected point in turn according to the calculated optimal position. When calculating the stability index, the window size is set to , preset threshold After the detection data matrix is subjected to singular value decomposition, the first 12 singular values are selected to construct the basic matrix, which retains 83.7% of the energy of the original data. The first change matrix and the second change matrix capture 12.6% and 3.7% of the energy respectively. The regularization coefficient of the matrix decomposition is set to 、 and .

[0125] When calculating the temperature impact factor, the regression coefficient is 、 and Through cosine similarity calculation and matrix contribution optimization function, the basic matrix contribution value is obtained , the first change matrix contribution value , the contribution value of the second change matrix .

[0126] The researchers used principal component analysis to reduce the dimensionality of the feature vectors to 64 dimensions, retaining 97.2% of the information. These reduced feature vectors, along with their contribution values, were then fed into a pre-trained model for analyzing the adhesion of exterior wall insulation boards. The model was trained on a dataset containing 1,850 sets of annotated data from 46 building projects. After 85 epochs, it reached convergence and achieved an F1 score of 0.938 on the validation set.

[0127] The final detection result is expanded by bilinear interpolation to The grid is smoothed using Gaussian filtering using the standard deviation The statistical data of the test results are shown in Table 1: Table 1 Statistics of test results of bonding effect of exterior wall insulation boards

[0128] Based on the test results, researchers found that areas with insulation board adhesion below 30% accounted for 5.4% of the total test area. These areas were mainly distributed around windows and wall corners. The specific causes of these problem areas are analyzed in Table 2: Table 2 Analysis of causes of poor bonding areas

[0129] To verify the accuracy of the test, researchers randomly selected 10 test points for actual peel testing. The average error between the measured bond area and the predicted value was ±4.3%, with a maximum error of 7.8%, validating the reliability of the test method. The entire test process took approximately 4.5 hours, significantly improving efficiency compared to the 2-3 days required by traditional methods.

[0130] Traditional testing of the adhesion of exterior wall insulation panels relies primarily on tapping, pulling, and local destructive testing. The tapping method, which involves tapping the wall and listening for hollows, is highly subjective, inaccurate, and unable to quantitatively assess the bond area. The pulling method, which tests the bond strength by applying a pulling device to the surface of the insulation panel, only provides point-by-point information and cannot assess the overall bond condition. Local destructive testing, which requires peeling off sections of the insulation panel for inspection, is highly destructive and cannot be applied to large areas. Common issues with these traditional methods include the inability to provide quantitative adhesion assessment, the inability to visualize the results, low test efficiency, and significant subjective influence.

[0131] Compared to traditional detection methods, the proposed method offers the following advantages: First, by integrating AR glasses with ultrasonic detection data, it enables comprehensive, nondestructive detection. Second, by employing matrix decomposition and feature extraction techniques, it can distinguish between basic structures and changing characteristics, improving detection accuracy and robustness. Third, by introducing a temperature compensation mechanism, the detection results are highly adaptable to changing environmental conditions. Finally, by visually displaying the detection results through augmented reality, construction personnel can understand and take targeted measures. Field tests have shown that this method achieves a detection accuracy of 95%, 30% higher than traditional methods, and increases detection efficiency by more than five times, providing a powerful tool for quality control of exterior wall insulation systems.

[0132] It should be noted that the variables involved in the present invention are explained in detail as shown in Tables 3, 4 and 5 below.

[0133] Table 3 Variable Explanation Table (Part I)

[0134] Table 4 Variable Explanation Table (Part II)

[0135] Table 5 Variable Explanation Table (Part 3)

[0136] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be covered by the scope of protection of the present invention.

Claims

1. A method for detecting the bonding effect of exterior wall insulation boards, characterized in that: include: Collecting AR glasses image data and ultrasonic detection equipment detection data, performing spatiotemporal registration and fusion analysis on the image data and the detection data to generate an optimal observation point vector group; Guide the inspectors to adjust the AR glasses to the optimal observation point coordinate position of the AR glasses in turn, and at the same time align the ultrasonic detection equipment with the optimal detection point coordinate position of the ultrasonic detection equipment; collect data at the optimal observation point coordinate position of the AR glasses and the optimal detection point coordinate position of the ultrasonic detection equipment to form a detection data matrix; decompose the detection data matrix into a basic matrix, a first variable matrix and a second variable matrix based on the stability index and calculate the contribution values and eigenvectors of the three matrices; input the eigenvectors and contribution values of the three matrices into a pre-trained exterior wall insulation board bonding effect analysis model; output the detection result matrix through the exterior wall insulation board bonding effect analysis model.

2. The method for detecting the bonding effect of exterior wall insulation boards according to claim 1, characterized in that: Spatiotemporal registration specifically aligns the AR glasses image data with the ultrasonic detection equipment detection data in the time dimension and spatial coordinate system to ensure that the two data sources describe information about the same physical location.

3. The method for detecting the bonding effect of exterior wall insulation boards according to claim 2, characterized in that: The optimal observation point vector group is specifically a group of vectors calculated by the data fusion algorithm. Each vector contains two coordinate points, which represent the optimal observation point coordinates of the AR glasses and the optimal detection point coordinates of the ultrasonic detection equipment required to obtain the best detection results.

4. The method for detecting the bonding effect of exterior wall insulation boards according to claim 3, characterized in that: The stability index is a dimensionless parameter that measures the stability of the changing trend of elements in the detection data matrix. It is used to distinguish between stable components and components with different degrees of change in the detection data matrix.

5. The method for detecting the bonding effect of exterior wall insulation boards according to claim 4, characterized in that: The basic matrix is specifically a submatrix representing the basic structural information in the detection data matrix, reflecting the basic bonding state of the exterior wall insulation board; the first change matrix is specifically a submatrix representing the change information submatrix in the detection data matrix whose stability index is higher than a preset threshold, reflecting the main change characteristics in the bonding state of the exterior wall insulation board; The second variation matrix specifically represents a variation information submatrix in which the stability index in the detection data matrix is lower than a preset threshold, reflecting minor variation characteristics in the bonding state of the exterior wall insulation board.

6. The method for detecting the bonding effect of exterior wall insulation boards according to claim 5, characterized in that: The contribution value is specifically a numerical value that quantifies the degree of influence of the basic matrix, the first change matrix, and the second change matrix on the final detection result. It is calculated through normalization, and the sum is 1.

7. The method for detecting the bonding effect of exterior wall insulation boards according to claim 6, characterized in that: The bonding effect index is a dimensionless parameter that quantifies the degree of bonding between the exterior wall insulation board and the wall. 1 indicates complete bonding and 0 indicates complete separation.

8. The method for detecting the bonding effect of exterior wall insulation boards according to claim 7, characterized in that: The contribution values of the three matrices are calculated using a matrix contribution optimization function. The input includes the detection data matrix, the basic matrix, the first change matrix, the second change matrix, and the ambient temperature parameter. It is used to dynamically adjust the weight distribution of the basic matrix, the first change matrix, and the second change matrix in the final result. The output is the contribution value.

9. The method for detecting the bonding effect of exterior wall insulation boards according to claim 8, characterized in that: The specific structure of the adhesion effect analysis model of exterior wall insulation boards is a hybrid neural network based on a multi-layer perceptron and a cross-attention mechanism, which includes a feature extraction layer, two cross-attention layers and three fully connected layers. The cross-attention layer uses a sparse attention mechanism to perform weighted analysis on the input basic matrix eigenvector, the first change matrix eigenvector and the second change matrix eigenvector. The number of attention heads in the sparse attention mechanism is dynamically adjusted according to the preset threshold of the stability index. The higher the preset threshold of the stability index, the more attention heads there are. The output layer of the adhesion effect analysis model of exterior wall insulation boards uses a softmax activation function to map the result to the range of 0 to 1 as the adhesion effect index.

10. An AR glasses for detecting the bonding effect of exterior wall insulation boards, characterized in that: The method comprises a processor configured to execute the method according to any one of claims 1 to 9.

Images