Building seepage diagnosis method based on AI

Abstract

The invention relates to the technical field of data processing, in particular to an AI-based building seepage diagnosis method, which comprises the steps of S1, a data acquisition and preprocessing stage, S2, an AI model construction and training stage, S3, a seepage diagnosis and positioning stage and S4, a result visualization and early warning stage. According to the AI-based building seepage diagnosis method, multi-source data are collected in real time, and an AI model is used for real-time analysis, so that seepage information of a building can be obtained in time, real-time diagnosis and early warning are realized, and the hysteresis defect existing in a traditional detection method is avoided; data such as an infrared image is corrected through an environment correction algorithm, the influence of environmental factors on the data is reduced, and the accuracy of the data is improved based on the influence; meanwhile, the AI model is used for automatic diagnosis, manual intervention is reduced, and the diagnosis efficiency and accuracy are improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to an AI-based method for diagnosing building seepage. Background Technology

[0002] Building seepage refers to the leakage of liquid inside or outside a building, which poses a serious threat to the structural safety and service life of the building. Therefore, seepage diagnosis is necessary. Existing diagnostic methods include water injection testing, visual inspection, and core drilling. For basement walls, the water injection test involves setting up a dike of a certain height on the inner side of the wall, injecting water into the dike, and observing the drop in water level and whether water seeps out from the outer side of the wall. For roofs, water can be injected into gutters or eaves troughs, and the water level changes over a certain period of time and whether there is seepage on the roof.

[0003] Visual inspection involves inspectors conducting a thorough and detailed examination of all parts of the building to check for water stains, mold, discoloration, peeling, and flaking. For example, if obvious water stains are found on the basement walls, and the stains are dark in color and gradually expand, it may indicate a seepage problem in the walls.

[0004] Core sampling is a method where inspectors drill holes at suspected seepage locations using drilling equipment to extract core samples. These core samples are then observed and analyzed to check for water stains, cracks, looseness, and whether the thickness and quality of the waterproofing layer meet requirements.

[0005] The existing diagnostic methods rely on manual inspections and instruments, and suffer from problems such as data lag, significant environmental interference, and low detection rate of hidden leaks. Therefore, they need to be improved. Summary of the Invention

[0006] To achieve the above objectives, the present invention provides the following technical solution: an AI-based method for diagnosing building seepage, comprising the following steps;

[0007] S1. Data Acquisition and Preprocessing Stage: Collect multi-source data of buildings and perform environmental correction, standardization, and blockchain-based evidence storage on the data.

[0008] S2, AI Model Building and Training Phase: Select algorithms to build models, train through multiple samples, introduce technologies to enhance generalization and interpretability, and use blockchain to record the training process.

[0009] S3, Seepage Diagnosis and Location Stage: Automatically extract infrared image features and perform diagnosis, record the diagnosis process with the help of blockchain, accurately locate the seepage location and analyze the cause;

[0010] S4. Results Visualization and Early Warning Stage: Construct a 3D model to visualize the seepage situation, dynamically adjust the early warning threshold for timely warning, and use blockchain to record relevant information to present problems intuitively.

[0011] Furthermore, the S1 multi-source data includes infrared image data, humidity sensor data, pressure sensor data, and text data of building design drawings, construction records, and historical maintenance records;

[0012] Environmental correction processing includes developing environmental correction algorithms to correct infrared thermal imaging data based on real-time environmental parameters;

[0013] Unified standard processing includes developing standards and specifications for multi-source data fusion, and developing data preprocessing and cleaning tools to ensure that data from different sources can be accurately and efficiently fused.

[0014] Blockchain-based evidence storage processing involves uploading the collected raw data and pre-processed data to a blockchain platform. This involves adding timestamps and digital signatures to each data block, and automatically verifying the source and integrity of the data through smart contracts during the upload process. Only data that passes verification can be stored on the blockchain.

[0015] Furthermore, the preprocessing of the S1 data also includes:

[0016] Data augmentation: Rotate, flip, scale, and translate infrared images, and use generative adversarial networks and variational autoencoders to generate simulated data to expand the training dataset;

[0017] Data quality assessment: Establish a data quality assessment index system to evaluate the quality of multi-source data. The assessment index includes data completeness, accuracy, and consistency. For infrared image data, check for image blurring and noise issues. For sensor data, check for missing data or outliers.

[0018] Furthermore, S2 constructs a model by combining a convolutional neural network with a decision tree, and the multi-sample training measures include collecting labeled building data under normal conditions and various seepage conditions as training samples.

[0019] Explanation techniques are manifested in feature importance analysis techniques and visualization techniques;

[0020] The blockchain records information including the usage of training data, model parameter tuning records, training iteration count, and loss function values.

[0021] Furthermore, S3 divides the building seepage diagnosis process into three stages: preliminary screening, detailed diagnosis, and verification diagnosis.

[0022] In the initial screening phase, a threshold judgment method is used to quickly identify building areas that may have seepage problems;

[0023] In the detailed diagnostic phase, deep learning models are used to accurately diagnose the selected regions;

[0024] During the review and diagnosis phase, final confirmation is made by combining expert experience with more multi-source data.

[0025] Furthermore, the detailed diagnostic phase utilizes a trained CNN to automatically extract features from the infrared image data. The feature extraction formula includes:

[0026] Convolutional layer formula:

[0027] Let the input feature map be X, the convolution kernel be W, the bias be b, and the output feature map be Y. Then the convolution operation is represented as:

[0028]

[0029] Where i and j are the coordinates of the output feature map, and m and n are the coordinates of the convolution kernel.

[0030] Pooling layer formula:

[0031] Let the pooling window size be k×k, the input feature map be X, and the output feature map be Y. Then the max pooling operation can be expressed as:

[0032]

[0033] Fully connected layer formula:

[0034] Let the input vector be x, the weight matrix be W, the bias vector be b, and the output vector be y. Then the operation of a fully connected layer can be represented as:

[0035] y = Wx + b.

[0036] Furthermore, S3 employs a decision tree algorithm to analyze and diagnose the extracted features. The analysis and diagnosis formula includes:

[0037] Information gain formula:

[0038] Let D be the dataset and A be the features. The information gain g(D,A) is defined as:

[0039] g(D,A)=H(D)-H(D∣A)

[0040] Where H(D) is the information entropy of dataset D, and H(D|A) is the conditional entropy of D given feature A;

[0041] Information entropy formula:

[0042]

[0043] Where K is the number of categories, P k It is the probability that the k-th type of sample appears in D.

[0044] Conditional entropy formula:

[0045]

[0046] Where, D i Let D take the value α on feature A. i The sample subset is n, where n is the number of values ​​for feature A.

[0047] Furthermore, in addition to diagnosis based on extracted image and sensor data features, S3 also includes comprehensive analysis combining building design drawings and construction record text information, and locating the specific location of seepage by combining the analysis results with the three-dimensional model of the building. S3 records the diagnosis time, results, features used and algorithms through blockchain.

[0048] Furthermore, the visualization in S4 is manifested in the three-dimensional model, where different colors or markers are used to represent the location, degree, and range of seepage. The basis for dynamically adjusting the warning threshold is historical data, and the warning methods include SMS and email. In the S4 stage, the visualization results, warning information, and processing suggestions are synchronously recorded on the blockchain.

[0049] Furthermore, S4 also includes integrating the AI-based building seepage diagnosis method with the building maintenance management system to achieve automatic push of diagnosis results and automatic allocation of maintenance tasks. At the same time, an access control mechanism is established to set different access permissions for different levels of users to ensure that only authorized personnel can access and process relevant data and diagnosis results.

[0050] Compared with existing technologies, the beneficial effects of this invention are: this AI-based building seepage diagnosis method...

[0051] (1) By collecting multi-source data in real time and using AI models for real-time analysis, the seepage information of buildings can be obtained in a timely manner, enabling real-time diagnosis and early warning, thereby avoiding the lag defects of traditional detection methods. Secondly, in the data acquisition stage, the infrared image data is corrected by environmental correction algorithms to reduce the impact of environmental factors on the data, thereby improving the accuracy of diagnostic data.

[0052] (2) The use of AI models for automatic diagnosis reduces human intervention and improves diagnostic efficiency and accuracy. In addition, the application of blockchain technology ensures the immutability and traceability of data and diagnostic process, enhancing the credibility of the results. Attached Figure Description

[0053] Figure 1 This is a flowchart of an AI-based building seepage diagnosis method. Detailed Implementation

[0054] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] Please see Figure 1 ,

[0056] This embodiment provides an AI-based method for diagnosing building seepage, comprising the following steps;

[0057] S1. Data Acquisition and Preprocessing Stage: Collect multi-source data of buildings and perform environmental correction, standardization, and blockchain-based evidence storage on the data.

[0058] S1 multi-source data includes infrared image data, humidity sensor data, pressure sensor data, and text data such as building design drawings, construction records, and historical maintenance records;

[0059] Environmental correction processing includes developing environmental correction algorithms to correct infrared thermal imaging data based on real-time environmental parameters;

[0060] Unified standard processing includes developing standards and specifications for multi-source data fusion, and developing data preprocessing and cleaning tools to ensure that data from different sources can be accurately and efficiently fused.

[0061] Blockchain-based evidence storage processing involves uploading the collected raw data and pre-processed data to a blockchain platform. This involves adding timestamps and digital signatures to each data block, and automatically verifying the source and integrity of the data through smart contracts during the upload process. Only data that passes verification can be stored on the blockchain.

[0062] For example, in a seepage monitoring project for a large commercial building, an infrared thermal imager is used to scan the building's exterior walls weekly to acquire infrared image data. Simultaneously, humidity sensors are installed inside and around the building to collect humidity data in real time. The infrared images are corrected for environmental temperature and humidity at the time of acquisition to remove environmental interference. All collected data is stored in a database in a standardized format and cleaned to remove outliers. Finally, the raw data and processing information are recorded on a blockchain, including details such as the time and location of data collection and the equipment used.

[0063] Preprocessing of S1 data also includes:

[0064] Data augmentation: Perform operations such as rotation (e.g., random rotation of 90 degrees, 180 degrees, 270 degrees), flipping (horizontal and vertical flipping), scaling, and translation on infrared images, and use generative adversarial networks (GANs) and variational autoencoders (VAEs) to generate simulated data to expand the training dataset.

[0065] Data quality assessment: Establish a data quality assessment index system to evaluate the quality of multi-source data. The assessment index includes data completeness, accuracy, and consistency. For infrared image data, check for image blurring and noise issues. For sensor data, check for missing data or outliers. If data quality issues are found, take corresponding measures, such as sharpening blurry images and interpolating missing data.

[0066] S2, AI Model Building and Training Phase: Select algorithms to build models, train through multiple samples, introduce technologies to enhance generalization and interpretability, and use blockchain to record the training process.

[0067] S2 constructs a model by combining convolutional neural networks with decision trees. The multi-sample training measures include collecting labeled building data under normal conditions and various seepage conditions as training samples.

[0068] Explanation techniques are manifested in feature importance analysis techniques and visualization techniques;

[0069] The blockchain records information including the usage of training data, model parameter tuning records, training iteration count, and loss function values.

[0070] For example, a CNN-based model can be built to process infrared image data, trained using a dataset containing 1000 infrared images with and 2000 without percolation. During training, data augmentation techniques are used to preprocess the images, and importance analysis is used to interpret the model's output. Information such as data usage and model parameter adjustments during training is recorded on the blockchain, including the time of each training round, the batches of training data used, and changes in the model's loss function value.

[0071] S3, Seepage Diagnosis and Location Stage: Automatically extract infrared image features and perform diagnosis, record the diagnosis process with the help of blockchain, accurately locate the seepage location and analyze the cause;

[0072] S3 divides the building seepage diagnosis process into three stages: preliminary screening, detailed diagnosis, and verification diagnosis.

[0073] In the initial screening phase, a threshold judgment method is used to quickly identify building areas that may have seepage problems;

[0074] In the detailed diagnostic phase, deep learning models are used to accurately diagnose the selected regions;

[0075] During the review and diagnosis phase, final confirmation is made by combining expert experience with more multi-source data.

[0076] For example, suppose we need to conduct a seepage diagnosis on a large commercial complex. This building contains multiple functional areas, such as a shopping mall, office space, and parking lots, and has a complex structure with walls and floors made of different materials. We have already collected multi-source data on the building using equipment such as infrared thermal imagers, humidity sensors, and pressure sensors.

[0077] Preliminary screening phase

[0078] Method: A threshold judgment method is used to quickly screen out building areas that may have seepage problems.

[0079] For example:

[0080] For infrared thermal imaging data, a temperature difference threshold is set. Under normal circumstances, the temperature difference between different areas of a building's exterior wall should be within ±2℃. If the temperature of a certain area is more than 3℃ lower than the surrounding area, it is considered that there may be a seepage problem in that area. For example, in an infrared image of a section of a shopping mall's exterior wall, if an area is found to have a significantly lower temperature than the surrounding area, with a temperature difference reaching 4℃, then this area would be marked as a potential seepage area.

[0081] For humidity sensor data, a humidity threshold of 70% is set. If the humidity sensor detects humidity exceeding 70% in a certain area, that area is also included in the scope of potential seepage problems. For example, in a corner of a parking lot, if the humidity sensor shows a humidity level of 75%, this area will be initially identified.

[0082] By using this threshold judgment method, we can quickly filter out a small number of areas that may have seepage problems from many areas of the entire building, reducing the workload of subsequent detailed diagnosis.

[0083] Detailed diagnostic phase

[0084] Method: A deep learning model is used to accurately diagnose the selected regions.

[0085] For example:

[0086] Infrared images, humidity data, pressure data, and other multi-source data from the initially screened areas are input into a pre-trained deep learning model. For example, a CNN model, trained on a large number of building seepage data samples, is used to automatically extract features from the data.

[0087] For the area of ​​the exterior wall in the shopping mall with abnormal temperature, the infrared image of that area is input into a CNN model. The model analyzes features such as texture, edges, and temperature distribution in the image, and combines this with humidity and pressure data to make a comprehensive judgment. The model may output the probability that there is a seepage problem in that area, such as 80%, and give a rough assessment of the degree of seepage, such as mild, moderate, or severe.

[0088] For areas in parking lots with excessive humidity, the relevant data is also input into the model. The model will determine the type of seepage in the area based on the learned feature patterns, such as seepage caused by wall cracks or seepage caused by rising groundwater levels.

[0089] Review and diagnosis stage

[0090] Method: Final confirmation was made by combining expert experience with more multi-source data.

[0091] For example:

[0092] Building structure experts and hydraulic engineering experts are invited to evaluate the results of the detailed diagnosis. The experts will conduct on-site inspections of the suspected areas identified in the initial screening and detailed diagnosis, making judgments based on their professional knowledge and experience. For example, when inspecting the exterior wall area of ​​the shopping mall, the experts will observe the wall's appearance for signs such as cracks and mold, while also referring to the building's design drawings to understand the area's waterproofing and drainage system.

[0093] Collect more multi-source data, such as the building's historical maintenance records and recent meteorological data. If the area has not undergone waterproofing repairs in the past few years and has recently experienced severe weather such as heavy rain, combining this information can more accurately determine the cause and severity of the seepage problem.

[0094] Based on expert opinions and more data, the detailed diagnosis results were revised and confirmed to ultimately determine which areas of the building had seepage problems, as well as the specific nature and severity of the problems, providing an accurate basis for subsequent maintenance and treatment.

[0095] Through the above three-stage diagnostic process, leakage problems in buildings can be identified efficiently and accurately, making full use of the speed of threshold judgment, the accuracy of deep learning models, and the professionalism of expert experience.

[0096] In the detailed diagnostic phase, a pre-trained CNN is used to automatically extract features from the infrared image data. The feature extraction formula includes:

[0097] Convolutional layer formula:

[0098] Let the input feature map be X, the convolution kernel be W, the bias be b, and the output feature map be Y. Then the convolution operation is represented as:

[0099]

[0100] Where i and j are the coordinates of the output feature map, and m and n are the coordinates of the convolution kernel.

[0101] Pooling layer formula:

[0102] Let the pooling window size be k×k, the input feature map be X, and the output feature map be Y. Then the max pooling operation can be expressed as:

[0103]

[0104] Fully connected layer formula:

[0105] Let the input vector be x, the weight matrix be W, the bias vector be b, and the output vector be y. Then the operation of a fully connected layer can be represented as:

[0106] y = Wx + b

[0107] For example, in a CNN model, the input is a 32×32 infrared image (which can be regarded as a 32×32 matrix). Assuming the convolution kernel size is 3×3, the stride is 1, and the padding is 0, in the convolution layer, the input image and the convolution kernel are processed according to the convolution formula to obtain a new feature map.

[0108] If a 3×3 region of the input image is multiplied by the corresponding element of the convolution kernel and then summed, and a bias is added, an element of the output feature map is obtained. After multiple convolutional, pooling, and fully connected layers, the model outputs a judgment result, such as 0 indicating no percolation and 1 indicating percolation.

[0109] S3 uses a decision tree algorithm to analyze and diagnose the extracted features. The analysis and diagnosis formulas include:

[0110] Information gain formula:

[0111] Let D be the dataset and A be the features. The information gain g(D,A) is defined as:

[0112] g(D,A)=H(D)-H(D∣A)

[0113] Where H(D) is the information entropy of dataset D, and H(D|A) is the conditional entropy of D given feature A;

[0114] Information entropy formula:

[0115]

[0116] Where K is the number of categories, P k It is the probability that the k-th type of sample appears in D.

[0117] Conditional entropy formula:

[0118]

[0119] Where, D i Let D take the value α on feature A. i The sample subset is n, where n is the number of values ​​for feature A.

[0120] For example, suppose we have a dataset D about building seepage, containing 100 samples, of which 50 show seepage and 50 do not. Feature A represents the building material (there are two materials, α1 and α2). Under material α1, 30 show seepage and 20 do not; under material α2, 20 show seepage and 30 do not. Substituting the above data, we can calculate the information entropy. This indicates that the original dataset has a relatively uniform class distribution and high uncertainty;

[0121] Then calculate the conditional entropy:

[0122] This indicates that even with known building material characteristics A, the dataset still has a certain degree of uncertainty.

[0123] Finally, the information gain g(D,A) was calculated as H(D) - H(D|A) = 1 - 0.971 = 0.029;

[0124] The results show that using building material feature A to split dataset D can achieve some information gain, but the gain is relatively small. Decision trees select the optimal feature for splitting based on metrics such as information gain. In the above example, because the information gain is small, it indicates that the building material feature does not contribute significantly to distinguishing whether a building has seepage. Decision trees may prioritize other features with greater information gain for splitting to more effectively classify the dataset.

[0125] In addition to diagnosis based on extracted image and sensor data features, S3 also includes comprehensive analysis combining building design drawings and construction record text information, and uses the analysis results to locate the specific location of seepage by combining the building's three-dimensional model. S3 records the diagnosis time, results, features used and algorithms through blockchain.

[0126] S4. Results Visualization and Early Warning Stage: Construct a 3D model to visualize the seepage situation, dynamically adjust the early warning threshold for timely warning, and use blockchain to record relevant information to present problems intuitively.

[0127] The visualization in S4 is manifested in the use of different colors or markers to represent the location, degree and range of seepage in a 3D model. The basis for dynamically adjusting the warning threshold is historical data, and the warning methods include SMS and email. In the S4 stage, visualization results, warning information and handling suggestions are recorded on the blockchain.

[0128] For example, using BIM technology to construct a 3D model of a building, the results of seepage diagnosis are displayed in the model as red markers, with the size and color intensity of the markers indicating the degree of seepage. Warning thresholds are set for seepage areas exceeding 1 square meter or seepage temperatures exceeding 5°C. When seepage in a monitored area exceeds the threshold, the system automatically sends an SMS warning to the property management personnel's mobile phones, providing possible causes and handling suggestions. Simultaneously, the visual model, warning information, and handling suggestions are recorded on the blockchain for easy future viewing and management.

[0129] S4 also includes integrating AI-based building seepage diagnosis methods with building maintenance management systems to achieve automatic push of diagnostic results and automatic allocation of maintenance tasks. For example, when a seepage problem is diagnosed in a certain area of ​​a building, the system automatically pushes the diagnostic results to the relevant maintenance personnel and automatically allocates maintenance tasks to specific personnel or teams according to the severity and location of the problem, while tracking the maintenance progress and providing feedback on the maintenance results.

[0130] The established access control mechanism sets different access permissions for different levels of users, ensuring that only authorized personnel can access and process relevant data and diagnostic results. For example, for building seepage diagnostic data and results, different roles are set up for administrators, engineers, and ordinary users. Administrators have the highest privileges and can perform operations such as data management and model training; engineers can view and analyze diagnostic results; ordinary users can only view some publicly available information, such as the overall health status of the building.

[0131] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An AI-based method for diagnosing building seepage, characterized in that, Includes the following steps: S1. Data Acquisition and Preprocessing Stage: Collect multi-source data of buildings and perform environmental correction, standardization, and blockchain-based evidence storage on the data. S2, AI Model Building and Training Phase: Select algorithms to build models, train through multiple samples, introduce technologies to enhance generalization and interpretability, and use blockchain to record the training process. S3, Seepage Diagnosis and Location Stage: Automatically extract infrared image features and perform diagnosis, record the diagnosis process with the help of blockchain, accurately locate the seepage location and analyze the cause; S4. Results Visualization and Early Warning Stage: Construct a 3D model to visualize the seepage situation, dynamically adjust the early warning threshold for timely warning, and use blockchain to record relevant information to present problems intuitively.

2. The AI-based building seepage diagnosis method according to claim 1, characterized in that: The S1 multi-source data includes infrared image data, humidity sensor data, pressure sensor data, and text data of building design drawings, construction records, and historical maintenance records; Environmental correction processing includes developing environmental correction algorithms to correct infrared thermal imaging data based on real-time environmental parameters; Unified standard processing includes developing standards and specifications for multi-source data fusion, and developing data preprocessing and cleaning tools to ensure that data from different sources can be accurately and efficiently fused. Blockchain-based evidence storage processing involves uploading the collected raw data and pre-processed data to a blockchain platform. This involves adding timestamps and digital signatures to each data block, and automatically verifying the source and integrity of the data through smart contracts during the upload process. Only data that passes verification can be stored on the blockchain.

3. The AI-based building seepage diagnosis method according to claim 1, characterized in that: The preprocessing of the S1 data also includes: Data augmentation: Rotate, flip, scale, and translate infrared images, and use generative adversarial networks and variational autoencoders to generate simulated data to expand the training dataset; Data quality assessment: Establish a data quality assessment index system to evaluate the quality of multi-source data. The assessment index includes data completeness, accuracy, and consistency. For infrared image data, check for image blurring and noise issues. For sensor data, check for missing data or outliers.

4. The AI-based building seepage diagnosis method according to claim 1, characterized in that: The S2 constructs a model by combining a convolutional neural network with a decision tree. The multi-sample training measures include collecting labeled building data under normal conditions and various seepage conditions as training samples. Explanation techniques are manifested in feature importance analysis techniques and visualization techniques; The blockchain records information including the usage of training data, model parameter tuning records, training iteration count, and loss function values.

5. The AI-based building seepage diagnosis method according to claim 1, characterized in that: S3 divides the building seepage diagnosis process into three stages: preliminary screening, detailed diagnosis, and verification diagnosis. In the initial screening phase, a threshold judgment method is used to quickly identify building areas that may have seepage problems; In the detailed diagnostic phase, deep learning models are used to accurately diagnose the selected regions; During the review and diagnosis phase, final confirmation is made by combining expert experience with more multi-source data.

6. The AI-based building seepage diagnosis method according to claim 5, characterized in that: The detailed diagnostic phase uses a trained CNN to automatically extract features from the infrared image data. The feature extraction formula includes: Convolutional layer formula: Let the input feature map be X, the convolution kernel be W, the bias be b, and the output feature map be Y. Then the convolution operation is represented as: Where i and j are the coordinates of the output feature map, and m and n are the coordinates of the convolution kernel. Pooling layer formula: Let the pooling window size be k×k, the input feature map be X, and the output feature map be Y. Then the max pooling operation is represented as: Fully connected layer formula: Let the input vector be x, the weight matrix be W, the bias vector be b, and the output vector be y. Then the operation of a fully connected layer can be represented as follows: y = Wx + b.

7. The AI-based building seepage diagnosis method according to claim 1, characterized in that: S3 uses a decision tree algorithm to analyze and diagnose the extracted features. The analysis and diagnosis formula includes: Information gain formula: Let D be the dataset and A be the features. The information gain g(D,A) is defined as: g(D,A)=H(D)-H(D∣A) Where H(D) is the information entropy of dataset D, and H(D|A) is the conditional entropy of D given feature A; Information entropy formula: Where K is the number of categories, P k It is the probability that the k-th type of sample appears in D; Conditional entropy formula: Where, D i Let D take the value α on feature A. i The sample subset is n, where n is the number of values ​​for feature A.

8. The AI-based building seepage diagnosis method according to claim 1, characterized in that: In addition to diagnosis based on extracted image and sensor data features, S3 also includes comprehensive analysis combining building design drawings and construction record text information, and uses the analysis results to locate the specific location of seepage in combination with the building's three-dimensional model. S3 records the diagnosis time, results, features used and algorithms through blockchain.

9. The AI-based building seepage diagnosis method according to claim 1, characterized in that: The visualization in S4 is manifested in the three-dimensional model, where different colors or markers are used to represent the location, degree and range of seepage. The basis for dynamically adjusting the warning threshold is historical data, and the warning methods include SMS and email. In the S4 stage, the visualization results, warning information and handling suggestions are synchronously recorded on the blockchain.

10. The AI-based building seepage diagnosis method according to claim 1, characterized in that: The S4 also includes integrating the AI-based building seepage diagnosis method with the building maintenance management system to achieve automatic push of diagnosis results and automatic allocation of maintenance tasks. At the same time, an access control mechanism is established to set different access permissions for different levels of users to ensure that only authorized personnel can access and process relevant data and diagnosis results.

Images