Data fusion prediction method for hyperbolic glass curtain wall structural adhesive damage detection

Abstract

The invention provides a data fusion prediction method for hyperbolic glass curtain wall structural adhesive damage detection, and the method comprises the steps: collecting an ultrasonic reflection signal corresponding to a structural adhesive through a preset ultrasonic sensor array in a hyperbolic glass curtain wall; acquiring surface temperature distribution data of the hyperbolic glass curtain wall through an infrared thermal imager; inputting the ultrasonic reflection signal and the surface temperature distribution data into a preset hybrid neural network model, performing cross-modal fusion on the ultrasonic reflection signal and the surface temperature distribution data by the hybrid neural network model, and outputting a three-dimensional fusion feature vector; the three-dimensional fusion feature vector comprises a damage geometric feature, an energy dissipation feature and a thermodynamic feature; acquiring historical structural adhesive detection data corresponding to the hyperbolic glass curtain wall; and performing space-time correlation analysis on the three-dimensional fusion feature vector and historical structural adhesive detection data to obtain a structural adhesive detection result corresponding to the hyperbolic glass curtain wall, including damage expansion trend prediction and residual life evaluation.

Description

Technical Field

[0001] The present application relates to the technical field of glass curtain walls, and in particular to a data fusion prediction method for damage detection of structural adhesives of hyperbolic glass curtain walls. Background Art

[0002] As a building facade with a high degree of curvature, the hyperbolic curtain wall plays an important role in aesthetic design and structural function. However, the complex geometric form of the hyperbolic curtain wall makes it difficult to detect damage, and traditional damage detection methods are difficult to cope with the special requirements of the curved structure. Existing infrared thermal imaging and ultrasonic detection technologies are well applied on flat curtain walls, but in the detection of hyperbolic curtain walls, data collection and processing face more challenges, mainly including: when the existing technology is used to detect damage to the structural adhesive of the glass curtain wall, it is often necessary to detect each detection point one by one, and the detection process is cumbersome and time-consuming. This leads to low detection efficiency and difficulty in meeting the needs of large-scale detection. Ultrasonic signals and infrared thermal imaging data are both complex multidimensional data, and professional data processing and analysis technologies are required to extract useful information. At the same time, the fusion processing between the two types of data is also a technical problem, which requires advanced algorithms and computing power. Most of the existing technologies rely on manual operation and judgment, and lack intelligent data processing and prediction capabilities. This makes the interpretation of the test results and the decision-making process easily affected by human factors, reducing the efficiency and accuracy of detection.

[0003] Therefore, a method is urgently needed to solve at least one of the above problems. Summary of the invention

[0004] This application provides a data fusion prediction method for damage detection of hyperbolic glass curtain wall structural adhesive, aiming to solve the problem that most existing technologies rely on manual operation and judgment and lack intelligent data processing and prediction capabilities. This makes the interpretation of detection results and the decision-making process easily affected by human factors, reducing detection efficiency and accuracy.

[0005] In a first aspect, the present application provides a data fusion prediction method for damage detection of structural adhesives of hyperbolic glass curtain walls, comprising:

[0006] In the hyperbolic glass curtain wall, the ultrasonic reflection signal corresponding to the structural adhesive is collected through a preset ultrasonic sensor array;

[0007] Acquiring surface temperature distribution data of the hyperbolic glass curtain wall by using an infrared thermal imager;

[0008] The ultrasonic reflection signal and the surface temperature distribution data are input into a preset hybrid neural network model, and the hybrid neural network model performs cross-modal fusion on the ultrasonic reflection signal and the surface temperature distribution data, and outputs a three-dimensional fusion feature vector; the three-dimensional fusion feature vector includes damage geometry features, energy dissipation features, and thermodynamic features;

[0009] Obtain historical structural adhesive test data corresponding to the hyperbolic glass curtain wall;

[0010] The three-dimensional fusion feature vector is subjected to spatiotemporal correlation analysis with historical structural adhesive detection data to obtain structural adhesive detection results corresponding to the hyperbolic glass curtain wall; the structural adhesive detection results include damage extension trend prediction and remaining life assessment.

[0011] In some embodiments, after obtaining the structural adhesive detection result corresponding to the hyperbolic glass curtain wall, it also includes: generating a curved surface adaptive detection path corresponding to the hyperbolic glass curtain wall according to the structural adhesive detection result; dynamically adjusting the ultrasonic emission frequency corresponding to the ultrasonic sensor array and the sampling interval corresponding to the infrared thermal imager based on the reinforcement learning algorithm to achieve a balanced optimization of the accuracy and efficiency of the structural adhesive damage detection.

[0012] Exemplarily, the generating a curved surface adaptive detection path corresponding to the hyperbolic glass curtain wall according to the structural adhesive detection result includes: obtaining a damage probability and an attenuation coefficient of the damage probability corresponding to the structural adhesive detection result; generating a curved surface adaptive detection path corresponding to the hyperbolic glass curtain wall according to the structural adhesive detection result based on an ant colony optimization algorithm according to the damage probability and the attenuation coefficient; the dynamic volatility coefficient of the ant colony optimization algorithm is:

[0013] ;

[0014] in, is the dynamic volatility coefficient, is the initial volatility coefficient, is the attenuation coefficient, is the damage probability.

[0015] It should be noted that, in some embodiments, the reward function of the reinforcement learning algorithm includes the detection coverage factor, energy consumption factor and prediction confidence factor corresponding to the hyperbolic glass curtain wall.

[0016] It should be noted that, in some embodiments, the weight coefficients corresponding to the detection coverage factor, the energy consumption factor and the prediction confidence factor are dynamically adjusted according to the ambient temperature and humidity parameters corresponding to the hyperbolic glass curtain wall.

[0017] In some embodiments, the plurality of array units corresponding to the ultrasonic sensor array adopts a curved conformal array layout.

[0018] Exemplarily, the expression corresponding to the spacing of each array unit is:

[0019] ;

[0020] in is the spacing, is the propagation speed of ultrasound in the structural adhesive, is the reference transmission frequency corresponding to the array unit, is the local curvature radius of the hyperbolic glass curtain wall, is the curvature deviation allowable threshold value corresponding to the hyperbolic glass curtain wall. When the actual curvature radius of the hyperbolic glass curtain wall changes beyond the curvature deviation allowable threshold value, the spacing of the array units is automatically adjusted according to the expression.

[0021] In some embodiments, the hybrid neural network model includes a convolutional neural network model and a Transformer model; the hybrid neural network model performs cross-modal fusion on the ultrasonic reflection signal and the surface temperature distribution data, including: the Transformer model calculates the association weight matrix of the ultrasonic frequency domain features corresponding to the ultrasonic reflection signal and the infrared space features corresponding to the surface temperature distribution data according to the attention mechanism, and dynamically adjusts the contribution of the ultrasonic frequency domain features and infrared space features in the association weight matrix based on the feature gating mechanism to complete the cross-modal fusion of the ultrasonic reflection signal and the surface temperature distribution data.

[0022] In some embodiments, before inputting the ultrasonic reflection signal and the surface temperature distribution data into a preset hybrid neural network model, it also includes: preprocessing the ultrasonic reflection signal and the surface temperature distribution data based on a wavelet packet decomposition algorithm to complete denoising and filtering of the ultrasonic reflection signal and the surface temperature distribution data.

[0023] Exemplarily, the number of decomposition layers corresponding to the wavelet packet decomposition algorithm is:

[0024] ;

[0025] in, is the number of decomposition levels, is the sampling frequency corresponding to the ultrasonic reflection signal and the surface temperature distribution data, is the lower limit of the damage characteristic frequency band corresponding to the hyperbolic glass curtain wall.

[0026] In a second aspect, the present application provides a data fusion prediction device for damage detection of hyperbolic glass curtain wall structural adhesive, comprising:

[0027] A signal collection unit is used to collect ultrasonic reflection signals corresponding to the structural adhesive in the hyperbolic glass curtain wall through a preset ultrasonic sensor array;

[0028] A data acquisition unit, used to acquire surface temperature distribution data of the hyperbolic glass curtain wall through an infrared thermal imager;

[0029] A data input unit, used for inputting the ultrasonic reflection signal and the surface temperature distribution data into a preset hybrid neural network model, wherein the hybrid neural network model performs cross-modal fusion on the ultrasonic reflection signal and the surface temperature distribution data, and outputs a three-dimensional fusion feature vector; the three-dimensional fusion feature vector includes damage geometry features, energy dissipation features, and thermodynamic features;

[0030] A history acquisition unit, used to acquire historical structural adhesive detection data corresponding to the hyperbolic glass curtain wall;

[0031] A result acquisition unit is used to perform spatiotemporal correlation analysis on the three-dimensional fusion feature vector and historical structural adhesive detection data to obtain the structural adhesive detection result corresponding to the hyperbolic glass curtain wall; the structural adhesive detection result includes damage expansion trend prediction and remaining life assessment.

[0032] In a third aspect, the present application provides a computer device comprising a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and implement a method provided in any embodiment of the present application when executing the computer program.

[0033] In a fourth aspect, the present application provides a computer-readable storage medium, which stores a computer program. When the computer-readable instructions are executed by the processor, one or more processors execute a method as provided in any embodiment of the present application.

[0034] The embodiment of the present application provides a data fusion prediction method for damage detection of structural adhesive of hyperbolic glass curtain wall. By combining ultrasonic reflection signals with infrared thermal imaging data and using a hybrid neural network model for cross-modal information fusion, an efficient and accurate assessment of the state of the structural adhesive is achieved. The specific steps are as follows:

[0035] Ultrasonic signal collection: A series of ultrasonic sensor arrays are arranged at preset positions on the hyperbolic glass curtain wall to transmit ultrasonic waves to the structural adhesive area and receive reflected echoes. These signals can reflect whether there are defects inside the structure and their approximate locations. Temperature distribution measurement: An infrared thermal imager is used to scan the entire curtain wall surface and record the temperature values ​​at different points. When the structural adhesive is damaged, its thermal conductivity will change, resulting in an increase in local temperature differences, forming recognizable thermal image features.

[0036] The two different types of data (ultrasound reflection signals and infrared thermal images) are input into a pre-trained hybrid neural network model. The model is designed with a deep learning architecture and contains multiple convolutional layers to automatically extract key features in the image. At the same time, long short-term memory units (LSTM) are introduced to capture the correlation between time series data. After being processed by the model, a three-dimensional fused feature vector is output, which contains information about the geometry of the injury site, energy loss, and temperature change pattern.

[0037] Collect and organize all structural adhesive inspection records of the target building over the past period of time, including but not limited to maintenance history, changes in environmental conditions, etc. Analyze these historical data using statistical methods or machine learning algorithms to establish a benchmark reference system to facilitate subsequent comparative analysis of the differences between the current situation and past trends.

[0038] The 3D fusion feature vector obtained above is combined with the relevant information in the historical database, and the type and extent of problems that may occur in the future are predicted through spatiotemporal correlation analysis technology. The final report will cover the following aspects: the specific location and scope of the damage; the expected development trend (such as whether it will deteriorate further); the remaining service life estimation; maintenance recommendations, etc.

[0039] In order to implement the above process, the following hardware equipment and technical support are required: Ultrasonic sensor: Select products with high sensitivity and wide frequency response range to ensure that tiny cracks can be detected. Infrared thermal imager: It is required to have high resolution and fast response speed to accurately capture instantaneous temperature changes. Computing platform: A server cluster equipped with high-performance GPU accelerator cards is used to run complex deep learning model training tasks. Database management system: It is used to store massive raw data and processing results, and support efficient query operations. Cloud computing service: Provide users with a remote access interface, so that they can monitor the health of the curtain wall in real time even if they are not on site.

[0040] The provided method has the following beneficial effects:

[0041] Improved detection accuracy: Compared with traditional manual visual inspection, this solution can detect more hidden and subtle signs of damage, avoiding missing important problems due to negligence.

[0042] Improved work efficiency: The application of automation tools significantly shortens the time from problem discovery to action, reducing maintenance costs.

[0043] Provides comprehensive information support: In addition to locating potential fault points, it also provides detailed diagnostic conclusions and development forecasts to help decision makers make more scientific and reasonable plans.

[0044] Promoted optimal resource allocation: By learning from historical data, the system can, to a certain extent, foresee failure modes that are more likely to occur under certain conditions, thereby guiding the direction of preventive maintenance work and extending the overall service life of the facility.

[0045] In summary, the hyperbolic glass curtain wall structural adhesive damage detection method based on multi-source data fusion proposed in the present invention not only solves the limitations of the existing technology, but also lays a solid foundation for further improving the level of building safety management.

[0046] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0048] Figure 1 This is a schematic flow chart of the steps of a data fusion prediction method for damage detection of structural adhesives for hyperbolic glass curtain walls provided in one embodiment of the present application;

[0049] Figure 2 It is a schematic diagram of the principle of a data fusion prediction method for damage detection of structural adhesive of a hyperbolic glass curtain wall provided in one embodiment of the present application;

[0050] Figure 3 It is a schematic block diagram of the structure of a computer device provided in one embodiment of the present application.

[0051] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. DETAILED DESCRIPTION

[0052] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0053] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0054] It should be understood that, in order to facilitate the clear description of the technical solutions of the embodiments of the present invention, in the embodiments of the present invention, the words "first", "second", etc. are used to distinguish the same items or similar items with substantially the same functions and effects. Those skilled in the art can understand that the words "first", "second", etc. do not limit the quantity and execution order, and the words "first", "second", etc. do not necessarily limit the difference.

[0055] It should be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0056] It should also be understood that the term “and / or” used in the specification and appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0057] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0058] As a building facade with a high degree of curvature, the hyperbolic curtain wall plays an important role in aesthetic design and structural function. However, the complex geometric form of the hyperbolic curtain wall makes it difficult to detect damage, and traditional damage detection methods are difficult to cope with the special requirements of the curved structure. Existing infrared thermal imaging and ultrasonic detection technologies are well applied on flat curtain walls, but in the detection of hyperbolic curtain walls, data collection and processing face more challenges, mainly including: when the existing technology is used to detect damage to the structural adhesive of the glass curtain wall, it is often necessary to detect each detection point one by one, and the detection process is cumbersome and time-consuming. This leads to low detection efficiency and difficulty in meeting the needs of large-scale detection. Ultrasonic signals and infrared thermal imaging data are both complex multidimensional data, and professional data processing and analysis technologies are required to extract useful information. At the same time, the fusion processing between the two types of data is also a technical problem, which requires advanced algorithms and computing power. Most of the existing technologies rely on manual operation and judgment, and lack intelligent data processing and prediction capabilities. This makes the interpretation of the test results and the decision-making process easily affected by human factors, reducing the efficiency and accuracy of detection.

[0059] Therefore, a method is urgently needed to solve at least one of the above problems.

[0060] To solve the above problems, please refer to Figure 1 ,like Figure 1 As shown, the data fusion prediction method for damage detection of hyperbolic glass curtain wall structural adhesive includes steps S101 to S105. The data fusion prediction method for damage detection of hyperbolic glass curtain wall structural adhesive is executed by a computer device, which may be a single server or a server cluster, or may be a handheld terminal, a laptop computer, a wearable device, or a robot.

[0061] like Figure 1 As shown, the details of steps S101-S105 are as follows:

[0062] Step S101. In a hyperbolic glass curtain wall, ultrasonic reflection signals corresponding to the structural adhesive are collected by a preset ultrasonic sensor array.

[0063] Specifically, this step involves scanning the structural adhesive on the hyperbolic curtain wall using a preset array of ultrasonic sensors, which can emit ultrasonic pulses of a specific frequency and receive echo signals reflected from different layers inside the structure.

[0064] The specific implementation method of this step:

[0065] Sensor layout: a certain number of ultrasonic sensors are arranged at key locations of the curtain wall to form a network covering the entire detection area. The number and layout of sensors need to be optimized according to the specific shape and size of the curtain wall to ensure full coverage of all key areas.

[0066] Signal transmission and reception: Each sensor sends ultrasonic pulses at a predetermined time interval and records the time delay and intensity change of the return signal. By adjusting the transmission frequency and power, it can better adapt to the characteristics of different materials.

[0067] Data storage and transmission: The collected data will be temporarily stored in the local device or directly transmitted to the central processing unit. In order to ensure the integrity and real-time nature of the data, wireless transmission technology such as Wi-Fi or Bluetooth can be used to ensure fast and reliable data transmission.

[0068] Compared with the traditional point-to-point detection method, the use of sensor arrays can obtain information from multiple locations at the same time, significantly improving the detection speed. In addition, by analyzing the change pattern of the reflected signal, potential problems can be located more accurately. This multi-point synchronous detection method not only improves efficiency, but also reduces errors caused by human operation.

[0069] Step S102: Acquire surface temperature distribution data of the hyperbolic glass curtain wall by using an infrared thermal imager.

[0070] Specifically, a high-resolution infrared camera is used to photograph the curtain wall surface to obtain its temperature distribution map. Due to differences in material properties (such as thermal conductivity), defective areas will show temperature characteristics different from normal areas.

[0071] Choose appropriate weather conditions (avoid strong sunlight interference) and time period to perform the measurement. Ideally, the measurement should be carried out on cloudy days or at night to reduce the influence of external light. Ensure that the infrared camera is in the best working condition and adjust the focus to make the image clear and visible. Calibrate the camera regularly to maintain its measurement accuracy. Repeat the same area from multiple angles to ensure full coverage. Shooting at different heights and angles can be achieved by drones or mobile platforms to ensure that there are no blind spots. After shooting, perform preliminary processing on the infrared image, such as noise reduction and contrast enhancement, for subsequent analysis.

[0072] Non-contact measurement avoids physical damage to the building itself; thermal images visually display areas of thermal anomalies, helping to quickly identify suspected fault points. This method can not only detect damage to structural adhesives, but also discover other potential problems such as water leaks and insulation failure.

[0073] Step S103. Input the ultrasonic reflection signal and the surface temperature distribution data into a preset hybrid neural network model, and the hybrid neural network model performs cross-modal fusion on the ultrasonic reflection signal and the surface temperature distribution data to output a three-dimensional fusion feature vector; the three-dimensional fusion feature vector includes damage geometry features, energy dissipation features and thermodynamic features.

[0074] Specifically, the different types of data collected in the first two steps (i.e., ultrasonic signals and infrared images) are input into a pre-trained hybrid neural network model. This model has a strong ability to integrate multi-source information and can effectively extract and combine the advantages of both.

[0075] Design and optimize a deep learning architecture suitable for this application scenario, such as combining a convolutional neural network (CNN) to process visual data and a recurrent neural network (RNN) to parse time series signals. CNN can effectively extract spatial features from infrared images, while RNN can capture the temporal dynamic characteristics of ultrasonic signals.

[0076] Data preprocessing: Preprocess the raw data, including normalization, filtering, feature extraction and other steps, to improve the input quality of the model. For example, perform Fourier transform on ultrasonic signals to extract frequency domain features; perform edge detection on infrared images to highlight the boundaries of temperature changes. Use a large number of labeled samples to train the model to ensure that it has good generalization performance for different types of damage. The training data should cover various common types and degrees of damage to improve the robustness of the model. Feed the preprocessed raw data into the model and output a comprehensive feature vector containing geometry, energy consumption and temperature characteristics. Through cross-modal fusion, the model can comprehensively utilize multiple information to improve the accuracy and reliability of diagnosis.

[0077] The intelligent algorithm realizes automatic feature extraction and matching, which greatly simplifies the manual analysis process and improves the diagnostic accuracy. This method can not only identify existing damage, but also predict potential risks, providing strong support for maintenance decisions.

[0078] Step S104: Obtain historical structural adhesive detection data corresponding to the hyperbolic glass curtain wall.

[0079] Specifically, historical inspection reports saved in the database are called to provide background support for the current assessment, including but not limited to details of previously discovered problems, repair measures and their effects.

[0080] Establish a structured database to store the results of previous inspections, maintenance records, and other relevant information. The database should support efficient query and retrieval functions. Filter out the most relevant cases for reference based on factors such as geographic location and construction date. Natural language processing technology can be used to classify and summarize text data for quick search. Perform statistical analysis on historical data to identify common damage patterns and repair strategies. By comparing current inspection results with historical data, you can better understand the reasons behind the current situation.

[0081] By learning from past experience and lessons, engineers can better understand the reasons behind the current situation and develop more reasonable maintenance plans. This approach can not only improve maintenance efficiency, but also reduce maintenance costs and extend the service life of buildings.

[0082] Step S105. Performing spatiotemporal correlation analysis on the three-dimensional fusion feature vector and historical structural adhesive detection data to obtain structural adhesive detection results corresponding to the hyperbolic glass curtain wall; the structural adhesive detection results include damage extension trend prediction and remaining life assessment.

[0083] Specifically, based on all the above information, statistical principles and technical means are used to conduct a comprehensive evaluation of the target object, which is not limited to the description of the current situation, but also includes the prediction of future development trends.

[0084] Apply time series analysis technology to study the evolution of structural adhesive damage over time. Use ARIMA, LSTM and other models to predict future damage trends. Combine environmental factors (such as climate change) and usage conditions to establish mathematical models to simulate possible change paths under different scenarios. For example, use Monte Carlo simulation to predict the life of structural adhesives under different climate conditions. Consider the cost-effectiveness ratio and give the best maintenance recommendations. Use economic models to calculate the costs and benefits of different maintenance options and select the most cost-effective option. Present the analysis results in charts, reports, etc. to facilitate managers' understanding and decision-making. Use data visualization tools such as Tableau or Power BI to generate intuitive visualization reports.

[0085] Early warning of major safety risks that may occur can guide managers to take preventive measures; at the same time, it can also save companies unnecessary expenses and extend the service life of buildings. This method not only improves safety, but also improves management efficiency, achieving a dual improvement in economic and social benefits.

[0086] In summary, this data fusion prediction method for damage detection of hyperbolic glass curtain wall structural adhesive introduces advanced sensing technology and data analysis tools, such as Figure 2As shown in the figure, the computer equipment as the executor of this solution performs data collection and preprocessing through steps S101-S102, performs data fusion processing in S103 and simultaneously constructs and trains the intelligent prediction model (representing the hybrid neural network model), and finally completes the implementation effect evaluation and optimization in combination with S104-S105. It overcomes the limitations of traditional means and realizes the fundamental transformation from passive response to active management. It not only improves the accuracy, efficiency and intelligence level of detection, but also provides a scientific basis for building maintenance, which is of great significance for ensuring public safety and promoting sustainable development.

[0087] The embodiment of the present application provides a data fusion prediction method for damage detection of structural adhesive of hyperbolic glass curtain wall. By combining ultrasonic reflection signals with infrared thermal imaging data and using a hybrid neural network model for cross-modal information fusion, an efficient and accurate assessment of the state of the structural adhesive is achieved. The specific steps are as follows:

[0088] Ultrasonic signal collection: A series of ultrasonic sensor arrays are arranged at preset positions on the hyperbolic glass curtain wall to transmit ultrasonic waves to the structural adhesive area and receive reflected echoes. These signals can reflect whether there are defects inside the structure and their approximate locations. Temperature distribution measurement: An infrared thermal imager is used to scan the entire curtain wall surface and record the temperature values ​​at different points. When the structural adhesive is damaged, its thermal conductivity will change, resulting in an increase in local temperature differences, forming recognizable thermal image features.

[0089] The two different types of data (ultrasound reflection signals and infrared thermal images) are input into a pre-trained hybrid neural network model. The model is designed with a deep learning architecture and contains multiple convolutional layers to automatically extract key features in the image. At the same time, long short-term memory units (LSTM) are introduced to capture the correlation between time series data. After being processed by the model, a three-dimensional fused feature vector is output, which contains information about the geometry of the injury site, energy loss, and temperature change pattern.

[0090] Collect and organize all structural adhesive inspection records of the target building over the past period of time, including but not limited to maintenance history, changes in environmental conditions, etc. Analyze these historical data using statistical methods or machine learning algorithms to establish a benchmark reference system to facilitate subsequent comparative analysis of the differences between the current situation and past trends.

[0091] The 3D fusion feature vector obtained above is combined with the relevant information in the historical database, and the type and extent of problems that may occur in the future are predicted through spatiotemporal correlation analysis technology. The final report will cover the following aspects: the specific location and scope of the damage; the expected development trend (such as whether it will deteriorate further); the remaining service life estimation; maintenance recommendations, etc.

[0092] In order to implement the above process, the following hardware equipment and technical support are required: Ultrasonic sensor: Select products with high sensitivity and wide frequency response range to ensure that tiny cracks can be detected. Infrared thermal imager: It is required to have high resolution and fast response speed to accurately capture instantaneous temperature changes. Computing platform: A server cluster equipped with high-performance GPU accelerator cards is used to run complex deep learning model training tasks. Database management system: It is used to store massive raw data and processing results, and support efficient query operations. Cloud computing service: Provide users with a remote access interface, so that they can monitor the health of the curtain wall in real time even if they are not on site.

[0093] The provided method has the following beneficial effects:

[0094] Improved detection accuracy: Compared with traditional manual visual inspection, this solution can detect more hidden and subtle signs of damage, avoiding missing important problems due to negligence.

[0095] Improved work efficiency: The application of automation tools significantly shortens the time from problem discovery to action, reducing maintenance costs.

[0096] Provides comprehensive information support: In addition to locating potential fault points, it also provides detailed diagnostic conclusions and development forecasts to help decision makers make more scientific and reasonable plans.

[0097] Promoted optimal resource allocation: By learning from historical data, the system can, to a certain extent, foresee failure modes that are more likely to occur under certain conditions, thereby guiding the direction of preventive maintenance work and extending the overall service life of the facility.

[0098] In summary, the hyperbolic glass curtain wall structural adhesive damage detection method based on multi-source data fusion proposed in the present invention not only solves the limitations of the existing technology, but also lays a solid foundation for further improving the level of building safety management.

[0099] In some embodiments, after obtaining the structural adhesive detection result corresponding to the hyperbolic glass curtain wall, it also includes: generating a curved surface adaptive detection path corresponding to the hyperbolic glass curtain wall according to the structural adhesive detection result; dynamically adjusting the ultrasonic emission frequency corresponding to the ultrasonic sensor array and the sampling interval corresponding to the infrared thermal imager based on the reinforcement learning algorithm to achieve a balanced optimization of the accuracy and efficiency of the structural adhesive damage detection.

[0100] According to the test results of structural adhesive, the damage probability and attenuation coefficient of each test point are extracted. The adaptive detection path of the hyperbolic glass curtain wall is generated based on the ant colony optimization algorithm. The ant colony optimization algorithm optimizes the path selection by simulating the process of ants looking for food. The transmission frequency of the ultrasonic sensor array and the sampling interval of the infrared thermal imager are dynamically adjusted using the reinforcement learning algorithm. The reward function contains the following factors: Detection coverage factor: Ensure that the detection path covers all key areas. Energy consumption factor: Reduce unnecessary energy consumption and improve efficiency. Prediction confidence factor: Ensure the accuracy of the detection results. Dynamic adjustment of weight coefficients: Dynamically adjust the weight coefficients of these factors according to the temperature and humidity parameters of the environment in which the hyperbolic glass curtain wall is located. For example, in a high humidity environment, the weight of the energy consumption factor may need to be increased to ensure the normal operation of the equipment.

[0101] By dynamically adjusting the detection path and parameters, the damage of the structural adhesive can be more accurately located and evaluated. The adaptive detection path reduces redundant detection and improves the overall detection efficiency. By dynamically adjusting the parameters, energy consumption can be reduced while ensuring the detection quality. Dynamically adjusting the weight coefficient according to different environmental conditions makes the system more adaptable and robust.

[0102] Exemplarily, the generating a curved surface adaptive detection path corresponding to the hyperbolic glass curtain wall according to the structural adhesive detection result includes: obtaining a damage probability and an attenuation coefficient of the damage probability corresponding to the structural adhesive detection result; generating a curved surface adaptive detection path corresponding to the hyperbolic glass curtain wall according to the structural adhesive detection result based on an ant colony optimization algorithm according to the damage probability and the attenuation coefficient; the dynamic volatility coefficient of the ant colony optimization algorithm is: ;in, is the dynamic volatility coefficient, is the initial volatility coefficient, is the attenuation coefficient, is the damage probability.

[0103] The introduction of dynamic volatility coefficient makes path selection more flexible and can better adapt to areas with different damage probabilities. By optimizing the path, redundant detection is reduced and detection efficiency is improved. Path selection is more accurate and helps to find potential damage points.

[0104] It should be noted that, in some embodiments, the reward function of the reinforcement learning algorithm includes the detection coverage factor, energy consumption factor and prediction confidence factor corresponding to the hyperbolic glass curtain wall.

[0105] The reward function of the reinforcement learning algorithm includes the following factors: Detection coverage factor: Ensure that the detection path covers all key areas. Energy consumption factor: Reduce unnecessary energy consumption and improve efficiency. Prediction confidence factor: Ensure the accuracy of the detection results. Dynamic adjustment of weight coefficients: Dynamically adjust the weight coefficients of these factors according to the temperature and humidity parameters of the environment in which the hyperbolic glass curtain wall is located. For example, in a high humidity environment, the weight of the energy consumption factor may need to be increased to ensure the normal operation of the device.

[0106] Through the multi-factor reward function, the comprehensive optimization of detection coverage, energy consumption and prediction confidence is achieved. The weight coefficient is dynamically adjusted according to the environmental parameters to make the system more adaptable and robust.

[0107] It should be noted that, in some embodiments, the weight coefficients corresponding to the detection coverage factor, the energy consumption factor and the prediction confidence factor are dynamically adjusted according to the ambient temperature and humidity parameters corresponding to the hyperbolic glass curtain wall.

[0108] Monitor the temperature and humidity parameters of the environment where the hyperbolic glass curtain wall is located in real time. Dynamically adjust the weight coefficients of the detection coverage factor, energy consumption factor, and prediction confidence factor according to the temperature and humidity parameters. For example, in a high humidity environment, increase the weight of the energy consumption factor; in a low temperature environment, increase the weight of the prediction confidence factor.

[0109] The method can automatically adjust the strategy according to different environmental conditions, improving the adaptability and robustness of the system. The weight coefficient is dynamically adjusted so that the system can maintain optimal performance under different environmental conditions.

[0110] In some embodiments, the plurality of array units corresponding to the ultrasonic sensor array adopts a curved conformal array layout.

[0111] The spacing between array units can be automatically adjusted according to the curvature of the curtain wall to ensure uniform detection coverage. The adaptive layout enables the ultrasonic sensor array to better adapt to the complex geometric form of the curtain wall, improving detection accuracy. Automatic adjustment of spacing makes the system more robust when facing curtain walls with different curvatures.

[0112] Exemplarily, the expression corresponding to the spacing of each array unit is:

[0113] ;

[0114] in is the spacing, is the propagation speed of ultrasound in structural adhesive (the sound speed of structural adhesive is generally in the range of 1000-2500 m / s), is the reference transmission frequency corresponding to the array unit (e.g., a value within 1-5 MHz), is the local curvature radius of the hyperbolic glass curtain wall (monitored in real time by the micro-inertial measurement unit built into the array unit), is the curvature deviation allowable threshold value corresponding to the hyperbolic glass curtain wall (e.g., 0.05 ), when the actual curvature radius of the hyperbolic glass curtain wall changes beyond the curvature deviation allowable threshold, the spacing of the array units is automatically adjusted according to the expression. Automatic adjustment of spacing enables the array units to better adapt to the curvature changes of the curtain wall and ensure uniform detection coverage. Adaptive layout improves detection accuracy, especially in curtain walls with complex geometric shapes. Automatic adjustment of spacing makes the system more robust when facing curtain walls with different curvatures.

[0115] In some embodiments, the hybrid neural network model includes a convolutional neural network model and a Transformer model; the hybrid neural network model performs cross-modal fusion on the ultrasonic reflection signal and the surface temperature distribution data, including: the Transformer model calculates the association weight matrix of the ultrasonic frequency domain features corresponding to the ultrasonic reflection signal and the infrared space features corresponding to the surface temperature distribution data according to the attention mechanism, and dynamically adjusts the contribution of the ultrasonic frequency domain features and infrared space features in the association weight matrix based on the feature gating mechanism to complete the cross-modal fusion of the ultrasonic reflection signal and the surface temperature distribution data.

[0116] The hybrid neural network model includes a convolutional neural network (CNN) and a Transformer model. The Transformer model calculates the association weight matrix of the ultrasonic frequency domain features corresponding to the ultrasonic reflection signal and the infrared spatial features corresponding to the surface temperature distribution data based on the attention mechanism. Based on the feature gating mechanism, the contribution of the ultrasonic frequency domain features and the infrared spatial features is dynamically adjusted in the association weight matrix to complete the cross-modal fusion of the ultrasonic reflection signal and the surface temperature distribution data.

[0117] The attention mechanism of the Transformer model can effectively integrate the multimodal information of ultrasonic and infrared images. Cross-modal fusion improves the accuracy and reliability of detection results. The feature gating mechanism enables the model to dynamically adjust the contribution of different modal features, enhancing the robustness of the system.

[0118] In some embodiments, before inputting the ultrasonic reflection signal and the surface temperature distribution data into a preset hybrid neural network model, it also includes: preprocessing the ultrasonic reflection signal and the surface temperature distribution data based on a wavelet packet decomposition algorithm to complete denoising and filtering of the ultrasonic reflection signal and the surface temperature distribution data.

[0119] Before the ultrasonic reflection signal and surface temperature distribution data are input into the hybrid neural network model, the data is preprocessed based on the wavelet packet decomposition algorithm to complete denoising and filtering. Wavelet packet decomposition can effectively remove noise from the data and improve data quality. Feature extraction: By calculating the number of decomposition layers, the frequency band information related to the damage characteristics can be extracted. Improve detection accuracy: The preprocessed data is purer, which helps to improve the accuracy and reliability of subsequent detection.

[0120] Exemplarily, the number of decomposition layers corresponding to the wavelet packet decomposition algorithm is:

[0121] ;

[0122] in, is the number of decomposition levels, is the sampling frequency corresponding to the ultrasonic reflection signal and the surface temperature distribution data, is the lower limit of the damage characteristic frequency band corresponding to the hyperbolic glass curtain wall.

[0123] Based on the calculated number of decomposition layers, the ultrasonic reflection signal and surface temperature distribution data are decomposed by wavelet packets to complete denoising and filtering. Wavelet packet decomposition can effectively remove noise from the data and improve data quality. By calculating the number of decomposition layers, frequency band information related to damage characteristics can be extracted. The pre-processed data is purer, which helps to improve the accuracy and reliability of subsequent detection.

[0124] In order to achieve efficient, accurate and intelligent damage detection of structural adhesives of hyperbolic glass curtain walls, we integrate the above steps and embodiments to design a complete detection system.

[0125] 1. System architecture:

[0126] Data acquisition module:

[0127] Ultrasonic sensor array: arranged at key locations of the curtain wall to collect ultrasonic reflection signals.

[0128] Infrared thermal imager: captures the temperature distribution of the curtain wall surface.

[0129] Environmental monitoring sensor: real-time monitoring of environmental temperature and humidity parameters.

[0130] Data preprocessing module:

[0131] Wavelet packet decomposition: De-noise and filter the ultrasonic reflection signal and surface temperature distribution data.

[0132] Data normalization: Normalize the preprocessed data to ensure the consistency of the input data.

[0133] Feature extraction module:

[0134] Ultrasonic signal processing: Perform Fourier transform on the ultrasonic reflection signal to extract frequency domain features.

[0135] Infrared image processing: perform edge detection on infrared images and extract spatial features.

[0136] Cross-modal fusion module:

[0137] Hybrid neural network model: Combines the convolutional neural network (CNN) and the Transformer model to perform cross-modal fusion and output a three-dimensional fused feature vector.

[0138] Path planning and parameter adjustment module:

[0139] Ant Colony Optimization Algorithm: Generating Adaptive Detection Paths.

[0140] Reinforcement learning algorithm: Dynamically adjust the transmission frequency of the ultrasonic sensor array and the sampling interval of the infrared thermal imager.

[0141] Historical data analysis module:

[0142] Database management: store all previous inspection results and maintenance records.

[0143] Data Analysis: Perform statistical analysis on historical data to extract common damage patterns and repair strategies.

[0144] Spatiotemporal correlation analysis module:

[0145] Time series analysis: predicting future injury trends.

[0146] Environmental factor simulation: Combine environmental factors to simulate possible change paths under different scenarios.

[0147] Cost-benefit analysis: Give the best maintenance recommendations.

[0148] Results visualization module:

[0149] Data visualization: present the analysis results in the form of charts, reports, etc.

[0150] User interface: Provide an intuitive user interface to facilitate managers' understanding and decision-making.

[0151] 2. Workflow

[0152] Data collection: Use ultrasonic sensor arrays and infrared thermal imagers to collect ultrasonic reflection signals and surface temperature distribution data of the curtain wall. Real-time monitoring of environmental temperature and humidity parameters.

[0153] Data preprocessing: De-noise and filter the collected data through wavelet packet decomposition. Normalize the preprocessed data.

[0154] Feature extraction: Perform Fourier transform on ultrasonic reflection signals to extract frequency domain features. Perform edge detection on infrared images to extract spatial features.

[0155] Cross-modal fusion: The extracted features are input into the hybrid neural network model for cross-modal fusion, and a three-dimensional fused feature vector is output.

[0156] Path planning and parameter adjustment: Generate adaptive detection paths based on ant colony optimization algorithm. Use reinforcement learning algorithm to dynamically adjust the emission frequency of ultrasonic sensor array and the sampling interval of infrared thermal imager.

[0157] Historical data analysis: Call historical inspection results and maintenance records from the database. Perform statistical analysis on historical data to extract common damage patterns and repair strategies.

[0158] Spatiotemporal correlation analysis: Use time series analysis techniques to predict future damage trends. Combine environmental factors to simulate possible change paths under different scenarios. Perform cost-benefit analysis and provide optimal maintenance recommendations.

[0159] Result visualization: present the analysis results in the form of charts, reports, etc. Provide an intuitive user interface to facilitate managers' understanding and decision-making.

[0160] The provided fused system has the following beneficial effects:

[0161] Improve detection accuracy: Through multimodal data fusion and advanced algorithms, the accuracy and reliability of detection results are improved.

[0162] Optimize detection efficiency: Adaptive detection path and dynamic parameter adjustment reduce redundant detection and improve overall detection efficiency.

[0163] Reduce energy consumption: By dynamically adjusting parameters, energy consumption can be reduced while ensuring detection quality.

[0164] Enhanced adaptability: Dynamically adjust strategies according to different environmental conditions to make the system more adaptable and robust.

[0165] Data-driven decision-making: Combining historical data and spatiotemporal correlation analysis provides a scientific basis to help managers make more reasonable maintenance decisions.

[0166] In summary, this solution combines ultrasonic reflection signals and infrared thermal imaging data, uses a hybrid neural network model for cross-modal fusion, and combines historical data and advanced optimization algorithms to provide an efficient, accurate and intelligent method for detecting damage to hyperbolic glass curtain wall structural adhesives. The embodiments and examples further optimize the detection path and parameters, improve the adaptability and robustness of the system, and provide strong support for ensuring public safety and promoting sustainable development.

[0167] The embodiment of the present application provides a data fusion prediction device for detecting damage to the structural adhesive of a hyperbolic glass curtain wall. The data fusion prediction device for detecting damage to the structural adhesive of a hyperbolic glass curtain wall is used to execute the steps of the data fusion prediction method for detecting damage to the structural adhesive of a hyperbolic glass curtain wall shown in the above embodiments. The data fusion prediction device for detecting damage to the structural adhesive of a hyperbolic glass curtain wall can be a single server or a server cluster, or the data fusion prediction device for detecting damage to the structural adhesive of a hyperbolic glass curtain wall can be a terminal, which can be a handheld terminal, a laptop computer, a wearable device, a robot, etc.

[0168] The data fusion prediction device for damage detection of structural adhesive of hyperbolic glass curtain wall includes:

[0169] A signal collection unit is used to collect ultrasonic reflection signals corresponding to the structural adhesive in the hyperbolic glass curtain wall through a preset ultrasonic sensor array;

[0170] A data acquisition unit, used to acquire surface temperature distribution data of the hyperbolic glass curtain wall through an infrared thermal imager;

[0171] A data input unit, used for inputting the ultrasonic reflection signal and the surface temperature distribution data into a preset hybrid neural network model, wherein the hybrid neural network model performs cross-modal fusion on the ultrasonic reflection signal and the surface temperature distribution data, and outputs a three-dimensional fusion feature vector; the three-dimensional fusion feature vector includes damage geometry features, energy dissipation features, and thermodynamic features;

[0172] A history acquisition unit, used to acquire historical structural adhesive detection data corresponding to the hyperbolic glass curtain wall;

[0173] A result acquisition unit is used to perform spatiotemporal correlation analysis on the three-dimensional fusion feature vector and historical structural adhesive detection data to obtain the structural adhesive detection result corresponding to the hyperbolic glass curtain wall; the structural adhesive detection result includes damage expansion trend prediction and remaining life assessment.

[0174] In some embodiments, after obtaining the structural adhesive detection result corresponding to the hyperbolic glass curtain wall, it also includes: generating a curved surface adaptive detection path corresponding to the hyperbolic glass curtain wall according to the structural adhesive detection result; dynamically adjusting the ultrasonic emission frequency corresponding to the ultrasonic sensor array and the sampling interval corresponding to the infrared thermal imager based on the reinforcement learning algorithm to achieve a balanced optimization of the accuracy and efficiency of the structural adhesive damage detection.

[0175] Exemplarily, the generating a curved surface adaptive detection path corresponding to the hyperbolic glass curtain wall according to the structural adhesive detection result includes: obtaining a damage probability and an attenuation coefficient of the damage probability corresponding to the structural adhesive detection result; generating a curved surface adaptive detection path corresponding to the hyperbolic glass curtain wall according to the structural adhesive detection result based on an ant colony optimization algorithm according to the damage probability and the attenuation coefficient; the dynamic volatility coefficient of the ant colony optimization algorithm is:

[0176] ;

[0177] in, is the dynamic volatility coefficient, is the initial volatility coefficient, is the attenuation coefficient, is the damage probability.

[0178] It should be noted that, in some embodiments, the reward function of the reinforcement learning algorithm includes the detection coverage factor, energy consumption factor and prediction confidence factor corresponding to the hyperbolic glass curtain wall.

[0179] It should be noted that, in some embodiments, the weight coefficients corresponding to the detection coverage factor, the energy consumption factor and the prediction confidence factor are dynamically adjusted according to the ambient temperature and humidity parameters corresponding to the hyperbolic glass curtain wall.

[0180] In some embodiments, the plurality of array units corresponding to the ultrasonic sensor array adopts a curved conformal array layout.

[0181] Exemplarily, the expression corresponding to the spacing of each array unit is:

[0182] ;

[0183] in is the spacing, is the propagation speed of ultrasound in the structural adhesive, is the reference transmission frequency corresponding to the array unit, is the local curvature radius of the hyperbolic glass curtain wall, is the curvature deviation allowable threshold value corresponding to the hyperbolic glass curtain wall. When the actual curvature radius of the hyperbolic glass curtain wall changes beyond the curvature deviation allowable threshold value, the spacing of the array units is automatically adjusted according to the expression.

[0184] In some embodiments, the hybrid neural network model includes a convolutional neural network model and a Transformer model; the hybrid neural network model performs cross-modal fusion on the ultrasonic reflection signal and the surface temperature distribution data, including: the Transformer model calculates the association weight matrix of the ultrasonic frequency domain features corresponding to the ultrasonic reflection signal and the infrared space features corresponding to the surface temperature distribution data according to the attention mechanism, and dynamically adjusts the contribution of the ultrasonic frequency domain features and infrared space features in the association weight matrix based on the feature gating mechanism to complete the cross-modal fusion of the ultrasonic reflection signal and the surface temperature distribution data.

[0185] In some embodiments, before inputting the ultrasonic reflection signal and the surface temperature distribution data into a preset hybrid neural network model, it also includes: preprocessing the ultrasonic reflection signal and the surface temperature distribution data based on a wavelet packet decomposition algorithm to complete denoising and filtering of the ultrasonic reflection signal and the surface temperature distribution data.

[0186] Exemplarily, the number of decomposition layers corresponding to the wavelet packet decomposition algorithm is:

[0187] ;

[0188] in, is the number of decomposition levels, is the sampling frequency corresponding to the ultrasonic reflection signal and the surface temperature distribution data, is the lower limit of the damage characteristic frequency band corresponding to the hyperbolic glass curtain wall.

[0189] It should be noted that technical personnel in the relevant field can clearly understand that, for the convenience and simplicity of description, the specific working process of the data fusion prediction device for detecting damage to the structural adhesive of the hyperbolic glass curtain wall and each module described above can refer to the corresponding process in the data fusion prediction method embodiment for detecting damage to the structural adhesive of the hyperbolic glass curtain wall described in the above embodiments, and will not be repeated here.

[0190] The above-mentioned data fusion prediction method for damage detection of structural adhesive of hyperbolic glass curtain wall can be implemented in the form of a computer program, and the computer program can be run on the provided device.

[0191] See also Figure 3 , Figure 31 is a schematic block diagram of the structure of a computer device provided in an embodiment of the present application. The computer device includes a processor, a memory and a network interface connected via a device bus, wherein the memory may include a storage medium and an internal memory.

[0192] The storage medium can store an operating device and a computer program. The computer program includes program instructions, and when the program instructions are executed, the processor can execute any data fusion prediction method for damage detection of structural adhesive of hyperbolic glass curtain wall.

[0193] The processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.

[0194] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor, the processor can execute any data fusion prediction method for damage detection of structural adhesive of hyperbolic glass curtain wall.

[0195] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the scheme of the present application, and does not constitute a limitation on the terminal to which the scheme of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0196] It should be understood that the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0197] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0198] The preset self-propelled robot is equipped with Lamb wave detection equipment, SLAM module and adaptive surface fitting mechanism;

[0199] In the hyperbolic glass curtain wall, the ultrasonic reflection signal corresponding to the structural adhesive is collected through a preset ultrasonic sensor array;

[0200] Acquiring surface temperature distribution data of the hyperbolic glass curtain wall by using an infrared thermal imager;

[0201] The ultrasonic reflection signal and the surface temperature distribution data are input into a preset hybrid neural network model, and the hybrid neural network model performs cross-modal fusion on the ultrasonic reflection signal and the surface temperature distribution data, and outputs a three-dimensional fusion feature vector; the three-dimensional fusion feature vector includes damage geometry features, energy dissipation features, and thermodynamic features;

[0202] Obtain historical structural adhesive test data corresponding to the hyperbolic glass curtain wall;

[0203] The three-dimensional fusion feature vector is subjected to spatiotemporal correlation analysis with historical structural adhesive detection data to obtain structural adhesive detection results corresponding to the hyperbolic glass curtain wall; the structural adhesive detection results include damage extension trend prediction and remaining life assessment.

[0204] In some embodiments, after obtaining the structural adhesive detection result corresponding to the hyperbolic glass curtain wall, it also includes: generating a curved surface adaptive detection path corresponding to the hyperbolic glass curtain wall according to the structural adhesive detection result; dynamically adjusting the ultrasonic emission frequency corresponding to the ultrasonic sensor array and the sampling interval corresponding to the infrared thermal imager based on the reinforcement learning algorithm to achieve a balanced optimization of the accuracy and efficiency of the structural adhesive damage detection.

[0205] Exemplarily, the generating a curved surface adaptive detection path corresponding to the hyperbolic glass curtain wall according to the structural adhesive detection result includes: obtaining a damage probability and an attenuation coefficient of the damage probability corresponding to the structural adhesive detection result; generating a curved surface adaptive detection path corresponding to the hyperbolic glass curtain wall according to the structural adhesive detection result based on an ant colony optimization algorithm according to the damage probability and the attenuation coefficient; the dynamic volatility coefficient of the ant colony optimization algorithm is:

[0206] ;

[0207] in, is the dynamic volatility coefficient, is the initial volatility coefficient, is the attenuation coefficient, is the damage probability.

[0208] It should be noted that, in some embodiments, the reward function of the reinforcement learning algorithm includes the detection coverage factor, energy consumption factor and prediction confidence factor corresponding to the hyperbolic glass curtain wall.

[0209] It should be noted that, in some embodiments, the weight coefficients corresponding to the detection coverage factor, the energy consumption factor and the prediction confidence factor are dynamically adjusted according to the ambient temperature and humidity parameters corresponding to the hyperbolic glass curtain wall.

[0210] In some embodiments, the plurality of array units corresponding to the ultrasonic sensor array adopts a curved conformal array layout.

[0211] Exemplarily, the expression corresponding to the spacing of each array unit is:

[0212] ;

[0213] in is the spacing, is the propagation speed of ultrasound in the structural adhesive, is the reference transmission frequency corresponding to the array unit, is the local curvature radius of the hyperbolic glass curtain wall, is the curvature deviation allowable threshold value corresponding to the hyperbolic glass curtain wall. When the actual curvature radius of the hyperbolic glass curtain wall changes beyond the curvature deviation allowable threshold value, the spacing of the array units is automatically adjusted according to the expression.

[0214] In some embodiments, the hybrid neural network model includes a convolutional neural network model and a Transformer model; the hybrid neural network model performs cross-modal fusion on the ultrasonic reflection signal and the surface temperature distribution data, including: the Transformer model calculates the association weight matrix of the ultrasonic frequency domain features corresponding to the ultrasonic reflection signal and the infrared space features corresponding to the surface temperature distribution data according to the attention mechanism, and dynamically adjusts the contribution of the ultrasonic frequency domain features and infrared space features in the association weight matrix based on the feature gating mechanism to complete the cross-modal fusion of the ultrasonic reflection signal and the surface temperature distribution data.

[0215] In some embodiments, before inputting the ultrasonic reflection signal and the surface temperature distribution data into a preset hybrid neural network model, it also includes: preprocessing the ultrasonic reflection signal and the surface temperature distribution data based on a wavelet packet decomposition algorithm to complete denoising and filtering of the ultrasonic reflection signal and the surface temperature distribution data.

[0216] Exemplarily, the number of decomposition layers corresponding to the wavelet packet decomposition algorithm is:

[0217] ;

[0218] in, is the number of decomposition levels, is the sampling frequency corresponding to the ultrasonic reflection signal and the surface temperature distribution data, is the lower limit of the damage characteristic frequency band corresponding to the hyperbolic glass curtain wall.

[0219] It should be noted that technical personnel in the relevant field can clearly understand that, for the convenience and simplicity of description, the specific working process of the computer equipment and each module described above can refer to the corresponding process in the data fusion prediction method embodiment for damage detection of hyperbolic glass curtain wall structural adhesive described in the above embodiments, and will not be repeated here.

[0220] The present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the processor implements the steps of a data fusion prediction method for damage detection of hyperbolic glass curtain wall structural adhesive as provided in any embodiment of the present application.

[0221] The computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiment, such as a hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a smart memory card (SmartMedia Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc., equipped on the computer device.

[0222] Exemplarily, the medium is used to implement the following steps:

[0223] In the hyperbolic glass curtain wall, the ultrasonic reflection signal corresponding to the structural adhesive is collected through a preset ultrasonic sensor array;

[0224] Acquiring surface temperature distribution data of the hyperbolic glass curtain wall by using an infrared thermal imager;

[0225] The ultrasonic reflection signal and the surface temperature distribution data are input into a preset hybrid neural network model, and the hybrid neural network model performs cross-modal fusion on the ultrasonic reflection signal and the surface temperature distribution data, and outputs a three-dimensional fusion feature vector; the three-dimensional fusion feature vector includes damage geometry features, energy dissipation features, and thermodynamic features;

[0226] Obtain historical structural adhesive test data corresponding to the hyperbolic glass curtain wall;

[0227] The three-dimensional fusion feature vector is subjected to spatiotemporal correlation analysis with historical structural adhesive detection data to obtain structural adhesive detection results corresponding to the hyperbolic glass curtain wall; the structural adhesive detection results include damage extension trend prediction and remaining life assessment.

[0228] In some embodiments, after obtaining the structural adhesive detection result corresponding to the hyperbolic glass curtain wall, it also includes: generating a curved surface adaptive detection path corresponding to the hyperbolic glass curtain wall according to the structural adhesive detection result; dynamically adjusting the ultrasonic emission frequency corresponding to the ultrasonic sensor array and the sampling interval corresponding to the infrared thermal imager based on the reinforcement learning algorithm to achieve a balanced optimization of the accuracy and efficiency of the structural adhesive damage detection.

[0229] Exemplarily, the generating a curved surface adaptive detection path corresponding to the hyperbolic glass curtain wall according to the structural adhesive detection result includes: obtaining a damage probability and an attenuation coefficient of the damage probability corresponding to the structural adhesive detection result; generating a curved surface adaptive detection path corresponding to the hyperbolic glass curtain wall according to the structural adhesive detection result based on an ant colony optimization algorithm according to the damage probability and the attenuation coefficient; the dynamic volatility coefficient of the ant colony optimization algorithm is:

[0230] ;

[0231] in, is the dynamic volatility coefficient, is the initial volatility coefficient, is the attenuation coefficient, is the damage probability.

[0232] It should be noted that, in some embodiments, the reward function of the reinforcement learning algorithm includes the detection coverage factor, energy consumption factor and prediction confidence factor corresponding to the hyperbolic glass curtain wall.

[0233] It should be noted that, in some embodiments, the weight coefficients corresponding to the detection coverage factor, the energy consumption factor and the prediction confidence factor are dynamically adjusted according to the ambient temperature and humidity parameters corresponding to the hyperbolic glass curtain wall.

[0234] In some embodiments, the plurality of array units corresponding to the ultrasonic sensor array adopts a curved conformal array layout.

[0235] Exemplarily, the expression corresponding to the spacing of each array unit is:

[0236] ;

[0237] in is the spacing, is the propagation speed of ultrasound in the structural adhesive, is the reference transmission frequency corresponding to the array unit, is the local curvature radius of the hyperbolic glass curtain wall, is the curvature deviation allowable threshold value corresponding to the hyperbolic glass curtain wall. When the actual curvature radius of the hyperbolic glass curtain wall changes beyond the curvature deviation allowable threshold value, the spacing of the array units is automatically adjusted according to the expression.

[0238] In some embodiments, the hybrid neural network model includes a convolutional neural network model and a Transformer model; the hybrid neural network model performs cross-modal fusion on the ultrasonic reflection signal and the surface temperature distribution data, including: the Transformer model calculates the association weight matrix of the ultrasonic frequency domain features corresponding to the ultrasonic reflection signal and the infrared space features corresponding to the surface temperature distribution data according to the attention mechanism, and dynamically adjusts the contribution of the ultrasonic frequency domain features and infrared space features in the association weight matrix based on the feature gating mechanism to complete the cross-modal fusion of the ultrasonic reflection signal and the surface temperature distribution data.

[0239] In some embodiments, before inputting the ultrasonic reflection signal and the surface temperature distribution data into a preset hybrid neural network model, it also includes: preprocessing the ultrasonic reflection signal and the surface temperature distribution data based on a wavelet packet decomposition algorithm to complete denoising and filtering of the ultrasonic reflection signal and the surface temperature distribution data.

[0240] Exemplarily, the number of decomposition layers corresponding to the wavelet packet decomposition algorithm is:

[0241] ;

[0242] in, is the number of decomposition levels, is the sampling frequency corresponding to the ultrasonic reflection signal and the surface temperature distribution data, is the lower limit of the damage characteristic frequency band corresponding to the hyperbolic glass curtain wall.

[0243] It should be noted that technical personnel in the relevant field can clearly understand that, for the convenience and conciseness of description, the specific working process of the storage medium and each module described above can refer to the corresponding process in the data fusion prediction method embodiment for damage detection of hyperbolic glass curtain wall structural adhesive described in the above embodiments, and will not be repeated here.

[0244] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A data fusion prediction method for damage detection of structural adhesive of hyperbolic glass curtain wall, characterized in that: include: In the hyperbolic glass curtain wall, the ultrasonic reflection signal corresponding to the structural adhesive is collected through a preset ultrasonic sensor array; Acquiring surface temperature distribution data of the hyperbolic glass curtain wall by using an infrared thermal imager; The ultrasonic reflection signal and the surface temperature distribution data are input into a preset hybrid neural network model, and the hybrid neural network model performs cross-modal fusion on the ultrasonic reflection signal and the surface temperature distribution data, and outputs a three-dimensional fusion feature vector; the three-dimensional fusion feature vector includes damage geometry features, energy dissipation features, and thermodynamic features; Obtain historical structural adhesive test data corresponding to the hyperbolic glass curtain wall; The three-dimensional fusion feature vector is subjected to spatiotemporal correlation analysis with historical structural adhesive detection data to obtain structural adhesive detection results corresponding to the hyperbolic glass curtain wall; the structural adhesive detection results include damage extension trend prediction and remaining life assessment.

2. The method according to claim 1, characterized in that: After obtaining the structural adhesive test result corresponding to the hyperbolic glass curtain wall, the method further includes: Generating a curved surface adaptive detection path corresponding to the hyperbolic glass curtain wall according to the structural adhesive detection result; The ultrasonic emission frequency corresponding to the ultrasonic sensor array and the sampling interval corresponding to the infrared thermal imager are dynamically adjusted based on the reinforcement learning algorithm to achieve a balanced optimization of the accuracy and efficiency of the structural adhesive damage detection.

3. The method according to claim 2, characterized in that The step of generating a curved surface adaptive detection path corresponding to the hyperbolic glass curtain wall according to the structural adhesive detection result includes: Obtain the damage probability and the attenuation coefficient of the damage probability corresponding to the structural adhesive test result; According to the damage probability and the attenuation coefficient, an adaptive curved surface detection path corresponding to the hyperbolic glass curtain wall is generated based on the structural adhesive detection result based on the ant colony optimization algorithm; the dynamic volatility coefficient of the ant colony optimization algorithm is: ； in, is the dynamic volatility coefficient, is the initial volatility coefficient, is the attenuation coefficient, is the damage probability.

4. The method according to claim 3, characterized in that The reward function of the reinforcement learning algorithm includes a detection coverage factor, an energy consumption factor, and a prediction confidence factor corresponding to the hyperbolic glass curtain wall.

5. The method according to claim 4, characterized in that The weight coefficients corresponding to the detection coverage factor, energy consumption factor and prediction confidence factor are dynamically adjusted according to the environmental temperature and humidity parameters corresponding to the hyperbolic glass curtain wall.

6. The method according to claim 1, characterized in that The plurality of array units corresponding to the ultrasonic sensor array adopt a curved surface conformal array layout.

7. The method according to claim 6, characterized in that The expression corresponding to the spacing of each array element is: ； in is the spacing, is the propagation speed of ultrasound in the structural adhesive, is the reference transmission frequency corresponding to the array unit, is the local curvature radius of the hyperbolic glass curtain wall, is the curvature deviation allowable threshold value corresponding to the hyperbolic glass curtain wall. When the actual curvature radius of the hyperbolic glass curtain wall changes beyond the curvature deviation allowable threshold value, the spacing of the array units is automatically adjusted according to the expression.

8. The method according to claim 1, characterized in that: The hybrid neural network model includes a convolutional neural network model and a Transformer model; the hybrid neural network model performs cross-modal fusion on the ultrasonic reflection signal and the surface temperature distribution data, including: The Transformer model calculates the association weight matrix of the ultrasonic frequency domain features corresponding to the ultrasonic reflection signal and the infrared space features corresponding to the surface temperature distribution data according to the attention mechanism, and completes the cross-modal fusion of the ultrasonic reflection signal and the surface temperature distribution data by dynamically adjusting the contribution of the ultrasonic frequency domain features and infrared space features in the association weight matrix based on the feature gating mechanism.

9. The method according to claim 1, characterized in that: Before inputting the ultrasonic reflection signal and the surface temperature distribution data into the preset hybrid neural network model, the method further includes: The ultrasonic reflection signal and the surface temperature distribution data are preprocessed based on a wavelet packet decomposition algorithm to complete denoising and filtering of the ultrasonic reflection signal and the surface temperature distribution data.

10. The method according to claim 9, characterized in that The number of decomposition layers corresponding to the wavelet packet decomposition algorithm is: ； in, is the number of decomposition levels, is the sampling frequency corresponding to the ultrasonic reflection signal and the surface temperature distribution data, is the lower limit of the damage characteristic frequency band corresponding to the hyperbolic glass curtain wall.

Images