Building structure health real-time monitoring method based on Internet of Things
Patent Information
- Application Number
- CN202510828662.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-20
- Publication Date
- 2025-10-17
- Estimated Expiration
- Not applicable · inactive patent
Smart Images
Figure CN120808137A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of Internet of Things, and particularly relates to a building structure health real-time monitoring method based on Internet of Things. BACKGROUND
[0002] With the acceleration of urbanization and the rapid development of infrastructure construction, the safety and durability of building structures have increasingly become the focus of society. Traditional building structure health monitoring methods mainly rely on periodic manual inspection and local sensor data collection, which has many limitations. First, manual inspection is time-consuming and labor-intensive, and it is difficult to fully cover complex building structures, which may miss potential risk points. Second, although local sensors can provide accurate data, the installation and maintenance costs are high, making it difficult to be widely applied to the entire building structure.
[0003] In recent years, with the rapid development of Internet of Things technology, some researchers have tried to apply it to the field of building structure health monitoring. These methods usually arrange a large number of sensors on the building to collect various physical parameters such as stress, vibration, temperature, etc. in real time, and then evaluate the health status of the building structure through data analysis. However, this type of method still faces several key problems: First, the arrangement and maintenance of sensors are still a huge challenge. For large and complex building structures, to achieve full coverage requires the installation of a large number of sensors, which not only increases the initial investment cost, but also puts a lot of pressure on subsequent maintenance. Second, traditional data analysis methods often have difficulty effectively processing massive multi-source heterogeneous data, resulting in the inability to fully utilize the collected information. Third, existing methods mostly focus on a single type of data or local structure analysis, making it difficult to make a comprehensive and accurate assessment of the overall health status of the building structure.
[0004] In addition, the existing technology also generally has the problem of insufficient real-time performance. Due to the complexity of data processing and analysis, many methods cannot achieve real-time monitoring, which may result in missing the best intervention opportunity when problems are discovered. At the same time, existing methods lack adaptability in dealing with unexpected abnormal situations, making it difficult to cope with complex and variable actual monitoring scenarios.
[0005] In view of the above problems, there is an urgent need for a new method that can comprehensively, accurately, and in real time monitor the health status of building structures. This method should be able to overcome the limitations of traditional sensor arrangement, effectively integrate multi-source data, achieve comprehensive assessment of the overall health status of the building structure, and have strong real-time performance and adaptability. SUMMARY
[0006] The building structure health real-time monitoring method based on the Internet of Things is designed for the above technical problems. The method realizes comprehensive, accurate and real-time monitoring of the building structure by innovatively combining the Internet of Things technology, computer vision and deep learning algorithm.
[0007] The building structure health real-time monitoring method based on the Internet of Things comprises: The acquisition step comprises: Acquiring a building image data set to be monitored; Acquiring physical parameter data of the building structure; The processing step comprises: Building a deep residual network model based on the building image data set and the physical parameter data; Extracting spatial features and physical features of the building structure based on the deep residual network model; Calculating health indicators of the building structure according to the spatial features and the physical features; The output step comprises: Generating a health condition report and a risk prediction result of the building structure; Visualizing the health condition report and the risk prediction result.
[0008] As a preferred embodiment, the acquisition of the building image data set to be monitored comprises: Using a drone or a aerial vehicle to shoot the building image to be monitored; Pretreating the building image to be monitored to obtain a pretreated image data set; The pretreatment comprises image alignment and noise removal.
[0009] As a preferred embodiment, the deep residual network model comprises: Building a feature fusion module, the feature fusion module comprising a multi-scale fusion module and a channel fusion module; The feature fusion module is used to complete feature fusion of the output of the double-branch dense convolution network; The multi-scale fusion module uses 1x1, 3x3 and 5x5 convolution operations to realize feature dimension reduction and channel number maintenance.
[0010] As a preferred embodiment, the method further comprises: Building a data generation module to perform data enhancement on the pretreated building image to generate a training set for model training; Using the training set to train the deep residual network model; The data enhancement comprises image rotation, scaling and noise addition.
[0011] As preferred, the extracting the spatial features and physical features of the building structure specifically comprises: constructing a feature extractor, training the feature extractor by using the training method of the data generation module; analyzing the feature vector output by the feature extractor by using a principal component analysis algorithm to obtain a structural health state feature vector; wherein the feature extractor adopts a full convolutional network structure.
[0012] As preferred, the calculating the health index of the building structure specifically comprises: constructing an LSTM artificial neural network to train the structural health state feature vector; using the trained LSTM network to make real-time judgments on the health state of the building structure; wherein the input of the LSTM network is a time-series structural health state feature vector, and the output is a health index.
[0013] As preferred, it further comprises: dividing the building structure into different regions; performing building parameter testing on each region to obtain a regional building data set; inputting the regional building data set into a data-driven model to obtain the building structure performance index of each region; wherein the data-driven model is constructed by using a full convolutional network.
[0014] As preferred, the calculating the health index of the building structure further comprises: calculating a structural level optimization index of the data-driven model; calculating a total performance index of the building structure based on the structural level optimization index; wherein the structural level optimization index is the ratio of the optimization result of the data-driven model in each level to the overall optimization result.
[0015] As preferred, the obtaining the building image data set further comprises: obtaining an initial image of the current bridge; based on the initial image, identifying the target monitoring object on the bridge by using a deep learning algorithm; calculating and obtaining the optimal gray level of the target monitoring object on the initial bridge; wherein the target monitoring object includes bridge piles, pile caps, piers, and bridge decks.
[0016] As preferred, it further comprises: obtaining the current gray level of the current bridge image; comparing the gray values of the current gray level and the optimal gray level to obtain a difference value; When the difference value is greater than a preset standard value, the current image is marked as an abnormal image; Screening the abnormal image to obtain an abnormal image identical to the target monitoring object in the initial image; When multiple abnormal images are obtained through screening, an abnormal image with the minimum optimal gray level layer difference value is selected as the final gray image of the target monitoring object.
[0017] The method of the present application has the following remarkable beneficial effects: Firstly, the present application breaks through the limitations of traditional sensor arrangement. By using unmanned aerial vehicles or aerial photography equipment to collect building image data, combined with existing physical sensor data, it realizes all-around and multi-angle monitoring of building structures. This method greatly reduces the hardware cost, while improving the flexibility and coverage of data collection.
[0018] Secondly, the present application uses a deep residual network model to effectively solve the problem of multi-source heterogeneous data fusion. This model can process image data and physical parameter data simultaneously, extract spatial and physical features of building structures, and realize deep fusion and full utilization of data. This method significantly improves the accuracy and comprehensiveness of monitoring.
[0019] Thirdly, the present application realizes comprehensive assessment of the overall health status of building structures by constructing a complex data-driven model. This model not only considers single parameters or local structures, but also comprehensively analyzes various aspects of the building, including spatial structure, physical parameters, boundary conditions and load conditions, etc., thus giving more comprehensive and accurate health assessment results.
[0020] In addition, the method of the present application has strong real-time performance. Through efficient data processing and analysis algorithms, it can quickly generate health status reports and risk prediction results, and present them in a visual way. This enables managers to discover potential problems in a timely manner and take preventive measures, greatly improving the efficiency of building safety management.
[0021] Finally, the method of the present application has good adaptability and scalability. Through data enhancement and deep learning technology, the method can continuously learn and adapt to new monitoring scenarios, handle various unexpected abnormal situations. At the same time, the modular design of the method makes it easy to extend to different types of building structure monitoring.
[0022] In summary, the proposed method, through the innovative combination of IoT technology, computer vision, and deep learning algorithms, effectively addresses the challenges of existing building structural health monitoring methods, achieving comprehensive, accurate, and real-time monitoring. This not only improves the efficiency of building safety management but also provides strong technical support for preventing major accidents and extending the service life of buildings, possessing significant practical application value and social significance. BRIEF DESCRIPTION OF THE DRAWINGS
[0023] Figure 1 It is the main flow chart of the present invention; Figure 2 A detailed diagram of the acquisition steps of the present invention; Figure 3 Detailed diagram of the processing steps of the present invention; Figure 4 LSTM network and region partition diagram of the present invention; Figure 5 This is a flow chart of bridge monitoring according to the present invention; Figure 6 Detailed diagram of the output steps of the present invention. DETAILED DESCRIPTION
[0024] Please refer to the attached Figures 1-6 The present invention provides a real-time monitoring method for building structure health based on the Internet of Things. This method can effectively monitor and evaluate the health status of building structures in real time. The present invention will be described in detail below with reference to specific implementation methods.
[0025] The monitoring method of the present invention includes an acquisition step, a processing step, and an output step. In the acquisition step, an image dataset of the building to be monitored is first acquired. Preferably, this image data can be collected using drones or aerial photography equipment to ensure comprehensiveness and accuracy. Simultaneously, this method also acquires physical parameter data of the building structure, such as key indicators like stress and deformation. This data is typically collected in real time by various sensors deployed on the building structure.
[0026] During the processing step, the method of the present invention constructs a deep residual network model based on the acquired building image dataset and physical parameter data. A deep residual network is an advanced deep learning model that can effectively process complex image features. In one embodiment of the present invention, the network model can include multiple residual blocks, each consisting of a convolutional layer, a batch normalization layer, and an activation function. This structure enables the model to learn the deep features of the building structure.
[0027] Next, based on the constructed deep residual network model, the method extracts the spatial features and physical features of the building structure. Spatial features mainly reflect the geometric shape and structural layout of the building, while physical features include material properties, stress distribution, and other information. The combination of these two types of features can comprehensively describe the health status of the building structure.
[0028] According to the extracted spatial features and physical features, the method calculates the health index of the building structure. This process usually involves complex mathematical models and algorithms. For example, the health index can be calculated using the following formula: , where, is the health index, is the spatial feature value, is the physical feature value, and are weight coefficients. The selection of weight coefficients is usually based on empirical values and specific building types, for example, for high-rise buildings, may be slightly larger, and the typical value may be 0.6, while is 0.4. In the output step, the method generates a health status report and risk prediction result of the building structure. These results are presented through visualization techniques, allowing managers to visually understand the health status of the building structure. Visualization methods can include color-coded 3D models, time series charts, and other methods. Further, the method of the present invention preprocesses the acquired images of the building to be monitored. This preprocessing process includes image alignment and noise removal. Image alignment ensures that images taken at different times and angles can be accurately corresponded, while noise removal improves the quality of the images. In a specific embodiment, a Gaussian filter can be used for noise removal, and its mathematical expression is: , where, is the Gaussian function, is the pixel coordinate, is the standard deviation. Usually, the value of
[0029] The method of the present invention also constructs a feature fusion module, including a multi-scale fusion module and a channel fusion module. The multi-scale fusion module uses 1x1, 3x3, and 5x5 convolution operations to realize feature dimension reduction and channel number preservation. This design can capture building structure features of different scales, improving the comprehensiveness of feature extraction. For example, 1x1 convolution can be used to reduce the number of channels of the feature map, 3x3 convolution can capture local features, and 5x5 convolution can obtain more context information in a larger range.
[0030] Through the above steps, the method of the present application can comprehensively and accurately monitor the health status of building structures, providing strong support for building safety management. The innovation of this method mainly lies in the application of deep learning technology, the fusion of multi-source data, and the comprehensiveness of feature extraction and analysis. Compared with traditional methods, this method can better handle complex building structures and provide more accurate health assessment and risk prediction.
[0031] The method of the present application also includes a data generation module that performs data augmentation on the pre-processed building images to generate a training set for model training. The introduction of this step significantly improves the generalization ability and robustness of the model. In a preferred embodiment, the data augmentation techniques include image rotation, scaling, and adding noise. For example, the image can be randomly rotated by an angle between -15° and 15°, scaled by a factor between 0.8 and 1.2, and added with Gaussian noise with a standard deviation usually set between 0.01 and 0.05. The selection of these parameters is based on the characteristics of building images and practical application experience.
[0032] Training the deep residual network model using the augmented training set is a key step in the method. During training, the batch gradient descent method can be used, with a learning rate usually set to 0.001 and a batch size of 32 or 64, depending on the available computing resources. Preferably, a learning rate decay strategy can be used to reduce the learning rate to 0.1 of the original every 50 epochs, which helps the model to converge better.
[0033] In terms of feature extraction, the method of the present application constructs a feature extractor and trains it using the training method of the data generation module. The feature extractor uses a fully convolutional network structure, which can maintain spatial information and is particularly suitable for processing image data of building structures. In a specific embodiment, the feature extractor can contain 5 convolutional layers, each followed by a ReLU activation function and a max-pooling layer. The size of the convolution kernel can be selected as 3x3, which is a good balance between computational efficiency and feature extraction capability.
[0034] The method of the present application also innovatively introduces the Principal Component Analysis (PCA) algorithm to analyze the feature vectors output by the feature extractor, thereby obtaining the structural health state feature vector. The application of PCA can effectively reduce the dimensionality of the data and extract the most representative features. In practical applications, usually the principal components that can explain 90% to 95% of the variance are retained. The mathematical expression of PCA is as follows: , where, is the original data matrix, and are the left and right singular vectors, respectively, is a singular value matrix. By selecting the singular vectors corresponding to the largest singular values, the reduced dimension features can be obtained.
[0035] To realize the real-time judgment of the health state of the building structure, the method constructs an LSTM (Long Short-Term Memory) artificial neural network, and trains the structural health state feature vector. The advantage of the LSTM network is that it can effectively process time series data, which is very suitable for the scene of long-term monitoring of building structures. In a preferred embodiment, the LSTM network can contain two LSTM layers, and the number of neurons in each layer can be set to 128 or 256, depending on the complexity of the input data. The input of the network is the time series structural health state feature vector, and the output is the health index.
[0036] When training the LSTM network, a sequence-to-sequence training method can be used, that is, a feature sequence in a time window is input to predict the health index at the next time step. The size of the time window can usually be selected as 30 to 60, which can capture the short-term changes of the building structure state. Preferably, the Adam optimizer can be used, and the initial learning rate is set to 0.001, and gradually reduced during the training process.
[0037] The method of the present application also innovatively divides the building structure into different regions, and tests the building parameters of each region to obtain the regional building data set. This regional processing method can more accurately locate the potential problem area. In a specific embodiment, the building can be divided into foundation area, main structure area, roof area, etc. according to its structural characteristics. The parameter test of each region may include stress test, vibration test, temperature test, etc., and the specific test items and frequency need to be determined according to the building type and use.
[0038] The regional building data set is input into the data-driven model to obtain the building structure performance index of each region. The data-driven model here uses a fully convolutional network, which has the advantage of preserving the spatial structure information of the input data. In a preferred embodiment, the fully convolutional network can contain multiple convolutional layers and skip connections to better fuse features of different scales. The last layer of the network can use 1x1 convolution to map the feature map to the required performance index dimension.
[0039] Through the above steps, the method of the present application can comprehensively and accurately evaluate the health state of the building structure, and provide reliable data support for building management and maintenance. This method based on deep learning and Internet of Things technology has higher accuracy and real-time performance than traditional methods, and can better meet the complex and variable building structure monitoring requirements.
[0040] In a preferred embodiment of the present invention, calculating the health index of the building structure also includes calculating the structural hierarchical optimization index of the data-driven model. This index reflects the optimization effect of the model at different levels and is of great significance for evaluating model performance and guiding model optimization. The calculation method of the structural hierarchical optimization index is as follows: , in, Optimize indicators for the structural level, is the total number of layers in the model, For the The number of nodes in the layer, and Respectively Tier The output and input values of each node, is the total number of nodes in the model. The larger the value of this indicator, the better the optimization effect of the model. In practical applications, it is usually The threshold is set between 0.5 and 0.8, and the specific value needs to be determined according to the complexity of the building structure and the monitoring requirements. Based on the structural hierarchical optimization index, this method further calculates the overall performance index of the building structure. The overall performance index comprehensively considers the optimization effects of each level and can more comprehensively reflect the overall health status of the building structure. Its calculation formula is as follows: , in, is the overall performance index, For the The weights of the layers, For the The structural hierarchical optimization index of the layer. The setting of weights needs to take into account the degree of influence of different levels on the health status of the building structure, and can usually be determined by expert scoring or data-driven methods. For example, for a typical high-rise building, the weights of the bottom and middle layers may be slightly higher, and can be set to 0.3 and 0.4, while the weight of the top layer can be set to 0.3. The method of the present invention also includes obtaining the initial image of the current bridge when acquiring the building image dataset. This step is very important for establishing baseline data and is helpful for subsequent comparative analysis. Preferably, the initial image should be collected when the bridge is just built or just after a comprehensive overhaul to ensure that the image data of the bridge is in the best state. Based on the initial image, the method identifies the target monitoring objects on the bridge through a deep learning algorithm. These objects typically include key structural components such as bridge piles, abutments, piers, and bridge decks. In a specific embodiment, the Faster R-CNN (Region-based Convolutional Neural Networks) algorithm can be used for target detection. The loss function of the algorithm can be expressed as: , in, is the classification loss, is the bounding box regression loss. In the training "pen", you can use cross entropy loss as the classification loss and smooth L1 loss as the regression loss.
[0041] After identifying the target monitoring object, this method calculates and obtains the optimal grayscale layer of the target monitoring object on the initial bridge. The selection of the optimal grayscale layer is crucial for subsequent image analysis, as it can maximize the characteristics of the target monitoring object. In a preferred embodiment, the Otsu algorithm can be used to determine the optimal grayscale threshold. The goal of the Otsu algorithm is to maximize the between-class variance, and its mathematical expression is: , , in, is the between-class variance, and are the weights of foreground and background respectively, and are the average grayscale values of the foreground and background respectively. The algorithm will traverse all possible thresholds and select The maximum threshold is used as the optimal grayscale threshold. In the actual monitoring process, this method also includes obtaining the current grayscale layer of the current bridge image and comparing it with the grayscale value of the optimal grayscale layer to obtain a difference value. This step can quickly determine whether the current image has an abnormality. The calculation formula of the difference value can be expressed as: , wherein, is a difference value, is a gray value of the current gray layer, is a gray value of the optimal gray layer. When the difference value is greater than a preset standard value, the current image is marked as an abnormal image. The selection of the preset standard value needs to balance the detection sensitivity and the false positive rate, and can be usually set to 10% to 20% of the optimal gray value. For example, if the optimal gray value is 128, the preset standard value can be set to between 12 and 25. For the image marked as abnormal, the method further screens to obtain an abnormal image that is completely the same as the target monitoring object in the initial image. This step can effectively reduce false positives and improve the accuracy of monitoring. In the screening process, a structural similarity index (SSIM) can be used to measure the similarity of images, and the calculation formula is: , wherein, and represent two images, and represent the mean and standard deviation, respectively, and are constants. The closer the value of SSIM is to 1, the more similar the two images are.
[0042] In the case of obtaining multiple abnormal images, the method selects an abnormal image with the smallest difference value of the optimal gray layer as the final gray image of the target monitoring object. This selection strategy can ensure the abnormal detection sensitivity while minimizing false positives.
[0043] Through the above steps, the method of the present application realizes efficient and accurate monitoring of complex building structures such as bridges. The method not only can detect structural abnormalities in real time, but also can provide detailed health status evaluation, providing strong technical support for the maintenance and management of building structures.
[0044] It should be noted that the above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the principles of the present application shall be included in the protection scope of the present application.
Claims
1. A real-time monitoring method for building structure health based on the Internet of Things, characterized in that: include: The acquisition steps include: Obtaining a dataset of images of buildings to be monitored; Obtain physical parameter data of building structures; Processing steps include: Constructing a deep residual network model based on the building image dataset and the physical parameter data; Extracting spatial and physical features of building structures based on the deep residual network model; Calculating health indicators of the building structure based on the spatial characteristics and physical characteristics; Output steps include: Generate health status reports and risk prediction results of building structures; The health status report and risk prediction results are visually presented.
2. The method according to claim 1, characterized in that The step of obtaining the image dataset of the building to be monitored specifically includes: Use drones or aerial vehicles to capture images of the buildings to be monitored; Preprocessing the image of the building to be monitored to obtain a preprocessed image data set; Wherein, the preprocessing includes image alignment and noise removal.
3. The method according to claim 1, characterized in that The construction of the deep residual network model specifically includes: Constructing a feature fusion module, wherein the feature fusion module includes a multi-scale fusion module and a channel fusion module; The feature fusion module completes the feature fusion of the dual-branch dense convolutional network output; The multi-scale fusion module uses 1x1, 3x3 and 5x5 convolution operations to achieve feature dimensionality reduction and channel number preservation.
4. The method according to claim 1, wherein Also includes: Build a data generation module to perform data augmentation on preprocessed building images and generate a training set for model training; Using the training set to train the deep residual network model; The data enhancement includes image rotation, scaling and adding noise.
5. The method according to claim 1, wherein The extraction of spatial features and physical features of the building structure specifically includes: Constructing a feature extractor and training the feature extractor using the training method of the data generation module; The principal component analysis algorithm is used to analyze the feature vector output by the feature extractor to obtain the structural health status feature vector; Wherein, the feature extractor adopts a fully convolutional network structure.
6. The method according to claim 1, characterized in that The calculation of the health index of the building structure specifically includes: Construct an LSTM artificial neural network to train the structural health status feature vector; Use the trained LSTM network to make real-time judgments on the health status of building structures; The input of the LSTM network is the time series structural health status feature vector, and the output is the health indicator.
7. The method according to claim 1, characterized in that Also includes: Divide the building structure into different areas; Conduct building parameter tests on each region to obtain regional building datasets; Inputting the regional building dataset into a data-driven model to obtain building structure performance indicators for each region; Among them, the data-driven model is constructed using a fully convolutional network.
8. The method according to claim 7, characterized in that The calculation of the health index of the building structure also includes: Calculating structural hierarchical optimization indicators of the data-driven model; Calculating the overall performance index of the building structure based on the structural hierarchical optimization index; The structural level optimization index is the ratio of the optimization results of the data-driven model in each level to the optimization results of the entire level.
9. The method according to claim 1, characterized in that The acquiring of the building image dataset further comprises: Get the initial image of the current bridge; Based on the initial image, identifying the target monitoring object on the bridge through a deep learning algorithm; Calculate and obtain the optimal grayscale layer of the target monitoring object on the initial bridge; The target monitoring objects include bridge piles, abutments, piers and bridge decks.
10. The method according to claim 9, characterized in that Also includes: Get the current grayscale layer of the current bridge image; Comparing the grayscale values of the current grayscale layer with those of the optimal grayscale layer to obtain a difference value; When the difference value is greater than a preset standard value, marking the current image as an abnormal image; Screening the abnormal image to obtain an abnormal image that is completely identical to the target monitoring object in the initial image; Among them, when multiple abnormal images are screened, the abnormal image with the smallest optimal grayscale layer difference is selected as the final target monitoring object grayscale image.