Cross-business-line bridge construction monitoring method based on machine vision recognition algorithm

Through the cross-business line bridge construction monitoring method based on machine vision recognition algorithm, construction data is collected and analyzed in real time, combined with deep learning and image processing technology, a multi-model fusion early warning system was established, which solved the safety and quality management problems in bridge construction and realized real-time Early warning and efficient monitoring.

CN119043405APending Publication Date: 2024-11-29CHINA RAILWAY CONSTRUCTION BRIDGE ENGINEERING BUREAU GROUP FOURTH ENGINEERING CO LTD
View PDF 0 Cites 7 Cited by

Patent Information

Application Number
CN202410846074.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-06-27
Publication Date
2024-11-29

AI Technical Summary

Technical Problem

During the bridge construction process, it is difficult for the existing technology to effectively monitor the risk factors and business line operating status during the construction process, resulting in poor safety and quality management, visual blind spots, and difficulty in real-time early warning.

Method used

Using a cross-business line bridge construction monitoring method based on machine vision recognition algorithm, stress, strain and displacement data are collected in real time through wireless transmission technology, combined with deep learning and image processing technology, a multi-model fusion early warning system is established to automatically trigger early warning and visualize output. Report.

Benefits of technology

Real-time monitoring of the bridge construction process is realized, safety and management efficiency are improved, structural safety and construction quality are ensured, and the impact on the business line is reduced.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119043405A_ABST
    Figure CN119043405A_ABST
Patent Text Reader

Abstract

The invention provides a cross-business-line bridge construction monitoring method based on a machine vision recognition algorithm, and the method comprises the steps: placing a wireless monitoring device at a designated position in a bridge construction process to obtain monitoring data in real time, and transmitting the monitoring data obtained by the monitoring device in real time to a receiving terminal based on a repeater; information from wireless monitoring equipment is collected, multi-source heterogeneous information is obtained, and data service is provided for a receiving terminal through a processing layer and a service layer. The receiving terminal comprises a structural stress monitoring module, a displacement monitoring module, a crack monitoring module, a bridge construction evolution analysis module and a safety early warning module, related safety thresholds are set, and a multi-model fusion early warning system is established. Based on the automation degree and the informatization integration capability of machine vision, effective management of cross-business-line bridge construction safety is realized, so that the safety of personnel and properties is ensured.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of construction safety management, and in particular to a method for monitoring the construction of a bridge across an operating line based on a machine vision recognition algorithm. The method can monitor the dangerous factors in the bridge construction process and the operating status of the operating line in real time, and effectively manage the safety quality of the bridge based on the automation degree and information integration capability of machine vision. Background Art

[0002] In recent years, with the continuous expansion and improvement of the domestic transportation network, the intersection between existing railway lines and newly built highways, railways and municipal roads has become more and more frequent. During the construction of bridges, it is easy to cause injuries or damage to the quality of the project due to failure to timely monitor the dangerous factors in the construction process and the operating status of the operating line.

[0003] The advantages of machine vision are reflected in its degree of automation and information integration capabilities. It has the characteristics of fast speed, high precision, and strong anti-interference ability, making it suitable for dangerous working environments and occasions where artificial vision cannot meet the requirements. Automated and intelligent detection technology based on machine vision has been successfully applied on roads and tunnels, and has also been initially applied on bridges, but it mainly focuses on the surface image acquisition technology of high-altitude concrete components with a wide line of sight, and the automatic identification of quality and safety issues is still in the theoretical research stage. In the current construction process of cross-operating line bridges, the safety and quality management means are single, and there are visual blind spots in monitoring, making it difficult to effectively manage the safety and quality of bridges. Therefore, it is of great significance to propose a cross-operating line bridge construction monitoring method based on machine vision recognition algorithm to ensure the quality and safety of bridge construction. Summary of the invention

[0004] To solve the above problems, the present invention provides a cross-operating line bridge construction monitoring method based on machine vision recognition algorithm, which can realize real-time collection of stress, strain, displacement data and on-site monitoring images during bridge construction, and quickly transmit the data to the receiving terminal through wireless transmission technology. Through the comprehensive analysis of data by each module and the early warning system with multi-model fusion, the early warning can be automatically triggered, and the relevant early warning reports can be visually output to promptly notify on-site management personnel and relevant departments.

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

[0006] A method for detecting construction of a bridge across an operating line based on a machine vision recognition algorithm, comprising:

[0007] Place wireless monitoring equipment at designated locations to obtain real-time monitoring data during bridge construction;

[0008] Based on the repeater, the monitoring data acquired by the monitoring device in real time is transmitted to the receiving terminal, the detection data from the wireless monitoring device is collected, multi-source heterogeneous information is acquired, and data services are provided to the receiving terminal through the processing layer and the service layer;

[0009] The receiving terminal receives the monitoring data and processes and outputs the data;

[0010] The wireless monitoring equipment includes: a wireless vibrating-wire strain gauge, a wireless pressure sensor, a displacement meter and a monitoring device;

[0011] The monitoring data include: stress, strain and displacement of key parts during bridge construction, and monitoring images of the bridge construction site;

[0012] The receiving terminal includes: a structural stress monitoring module, a displacement monitoring module, a crack monitoring module, a bridge construction evolution analysis module and a safety early warning module.

[0013] Furthermore, the monitoring data obtained during the bridge construction process include:

[0014] Obtain stress and pressure data of various parts of the bridge during the bridge construction process, and install multiple wireless vibrating wire strain gauges and wireless pressure sensors at the bridge structure connection and bridge foundation;

[0015] Obtaining displacement change data of the bridge structure during the bridge construction process, wherein the displacement change data includes horizontal displacement, vertical displacement and torsional displacement, and setting a plurality of displacement meters at the connection between the bridge pier and the bridge deck, the mid-span position of the bridge, the bridge support and the expansion joint;

[0016] Acquire the construction site monitoring images during the bridge construction process, and set up multiple monitoring devices in the areas that need to be monitored to obtain different viewing angles of the monitoring images.

[0017] Furthermore, the bridge structure connection includes: the connection seams between precast beams, the connection between the beam body and the top of the pier, the intersection between the bottom of the pier and the top surface of the foundation, the connection surface between the top of the pile foundation and the abutment, the connection between the abutment and the beam body, and the transition section connection between the continuous beam and the simply supported beam;

[0018] Furthermore, the bridge foundation parts include: sections of different heights of the pier body, the top surface of the pier foundation, the wall of the caisson foundation, and the pile body of the pile foundation;

[0019] The areas that need to be monitored include the construction area and the non-construction area of ​​the construction site.

[0020] Furthermore, the multi-source heterogeneous information is obtained, and a stream processing engine is used to process the multi-source heterogeneous data at the processing layer, and the processed data is stored in a stream processing database at the service layer to provide data services for the receiving terminal;

[0021] The multi-source heterogeneous information includes: model information, monitoring information, forecast result information, disease monitoring information, maintenance measure information, quality-related file data and on-site acceptance data.

[0022] Furthermore, the structural stress monitoring module is used to obtain stress and strain monitoring data, and analyze the obtained stress and strain monitoring data through a time series analysis method based on deep learning, so as to monitor and predict the structural stress of the bridge.

[0023] Furthermore, the displacement monitoring module monitors the displacement of the bridge based on deep learning image recognition technology, monitors the displacement of the bridge by comparing images at different time points, and combines three-dimensional laser scanning technology and uses deep learning algorithms to perform three-dimensional reconstruction to monitor the three-dimensional displacement of the bridge.

[0024] Furthermore, the crack detection module uses image processing and machine learning technology to identify the characteristics of bridge construction defects, identifies the shape, size and distribution of cracks through a deep learning model, and predicts the development trend of cracks.

[0025] Furthermore, the bridge construction evolution analysis module uses a convolutional neural network to extract and classify features from video data to identify the construction area of ​​the bridge construction site, trains the model to identify different types of construction activities, combines the optical flow method with the background subtraction method to track the dynamic changes of the construction site, and distinguishes between the construction area and the non-construction area; by combining AI S and VHF data, a machine learning algorithm is used to analyze and predict the progress of bridge construction to evaluate the impact of bridge construction on the operating line, and a clustering algorithm is used to classify vehicle traffic patterns to identify possible traffic hotspots;

[0026] The different types of construction activities include: lifting, welding, concrete pouring, grouting reinforcement and grouting.

[0027] Furthermore, the safety warning module comprehensively analyzes the monitoring data of structural stress, displacement, cracks and traffic flow through a preset multi-model fusion warning system, and uses a random forest algorithm to formulate a safety threshold. When the detection index exceeds the formulated threshold, the system will automatically trigger a warning, realize dynamic monitoring and visualize the status of related areas, and output the warning information and related decision-making suggestions to on-site managers and relevant departments based on the monitoring results.

[0028] The beneficial effects of the present invention are:

[0029] The present invention can collect stress, strain and displacement data of the bridge, as well as on-site monitoring images, in real time by installing wireless monitoring equipment at key parts of bridge construction. The acquisition of these data is crucial to bridge safety. It can help engineers to timely understand the status of the bridge during construction and prevent potential safety risks. The repeater collects monitoring data and obtains multi-source heterogeneous data and processes the data through a stream processing engine, so that it can quickly respond to and process real-time generated data to ensure the timeliness of the data. The data is stored in a stream processing database to meet the high-speed inflow and reliable storage of stream data, so that the data can be quickly transmitted to the receiving terminal, greatly improving the timeliness of data processing.

[0030] The application of each module in the receiving terminal of the present invention further improves the monitoring efficiency and accuracy. The structural stress monitoring module adopts a time series analysis method based on deep learning, which makes the monitoring and prediction of bridge structural stress more accurate, helping to ensure the load-bearing capacity and structural safety of the bridge. The displacement monitoring module uses deep learning image recognition technology to monitor bridge displacement, combined with three-dimensional laser scanning technology and three-dimensional reconstruction by deep learning algorithm, which can monitor the displacement of the bridge in three dimensions and provide all-round displacement data.

[0031] The crack detection module of the present invention adopts image processing and machine learning technology, which can accurately identify cracks and predict their development trend, effectively avoiding structural damage caused by crack expansion. The application of video surveillance in the bridge construction evolution analysis module improves the management efficiency of the construction site. Through the intelligent analysis of video data by convolutional neural network, not only can the construction area be automatically identified, but also the dynamic changes of the construction site can be tracked, thereby improving the intelligent level of site management. Combining AI S and VHF data, the construction progress is analyzed and predicted by machine learning algorithm, which helps to optimize the construction progress and reduce the impact on the business line. The classification of vehicle traffic patterns by clustering algorithm K-means helps to discover and prevent the emergence of traffic hot spots, thereby ensuring traffic safety. The establishment of the safety warning module monitors the structural stress, displacement, cracks and traffic flow data in real time through the multi-model fusion warning system. When the monitoring indicators exceed the safety threshold, the warning can be automatically triggered and visualized, and the on-site management personnel and relevant departments can be notified in time, thereby ensuring the safety of personnel and property.

[0032] In summary, the cross-operating line bridge construction monitoring method based on machine vision recognition algorithm not only improves the safety and efficiency of bridge construction, but also plays a positive role in improving the level of intelligent construction management, preventing safety accidents, and reducing the impact on social traffic. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1It is a schematic diagram of a cross-operating line bridge construction monitoring algorithm based on a machine vision recognition algorithm provided by this application;

[0034] Figure 2 It is a schematic diagram of the process of detecting and predicting bridge structure stress;

[0035] Figure 3 It is a flowchart of combining optical flow method and background subtraction method to track the dynamic changes of the construction site;

[0036] Figure 4 It is a schematic diagram of the process of evaluating the construction status of a bridge by combining AIS and VHF data;

[0037] Figure 5 It is a flowchart of the repeater processing multi-source heterogeneous information. DETAILED DESCRIPTION

[0038] See also Figures 1 to 5 As shown, the present invention provides a cross-operating line bridge construction monitoring method based on machine vision recognition algorithm, which can realize real-time collection of stress, strain, displacement data and on-site monitoring images during bridge construction, and quickly transmit the data to the receiving terminal through wireless transmission technology. Through the comprehensive analysis of data by each module and the early warning system of multi-model fusion, the early warning can be automatically triggered, and the relevant early warning reports can be visually output to promptly notify on-site management personnel and relevant departments.

[0039] The method for monitoring construction of a bridge across an operating line based on a machine vision recognition algorithm comprises:

[0040] Place wireless monitoring equipment at designated locations to obtain real-time monitoring data during bridge construction;

[0041] The wireless monitoring equipment includes: a wireless vibrating-wire strain gauge, a wireless pressure sensor, a displacement meter and a monitoring device;

[0042] The monitoring data include: stress, strain and displacement of key parts during bridge construction, and monitoring images of the bridge construction site;

[0043] Furthermore, obtaining monitoring data during bridge construction includes:

[0044] Obtain stress and pressure data of various parts of the bridge during the bridge construction process, and install multiple wireless vibrating wire strain gauges and wireless pressure sensors at the bridge structure connection and bridge foundation;

[0045] Obtaining displacement change data of the bridge structure during the bridge construction process, wherein the displacement change data includes horizontal displacement, vertical displacement and torsional displacement, and setting a plurality of displacement meters at the connection between the bridge pier and the bridge deck, the mid-span position of the bridge, the bridge support and the expansion joint;

[0046] Acquire the construction site monitoring images during the bridge construction process, and set up multiple monitoring devices in the areas that need to be monitored to obtain different viewing angles of the monitoring images.

[0047] Furthermore, the bridge structure connection includes: the connection seams between precast beams, the connection between the beam body and the top of the pier, the junction area between the bottom of the pier and the top surface of the foundation, the connection surface between the top of the pile foundation and the abutment, the connection between the abutment and the beam body, and the transition section connection between the continuous beam and the simply supported beam;

[0048] Furthermore, the bridge foundation parts include: sections of different heights of the pier body, the top surface of the pier foundation, the wall of the caisson foundation, and the pile body of the pile foundation;

[0049] The areas that need to be monitored include the construction area and the non-construction area of ​​the construction site.

[0050] Specifically, according to the design plan, the wireless vibrating wire strain gauge is firmly installed at the key monitoring part of the bridge to ensure that there is no gap between the strain gauge and the measured structure to ensure the accuracy of the measurement. The impact of temperature changes on the measurement results needs to be fully considered, and temperature compensation or correction should be performed when necessary, and the strain gauge should be calibrated at preset time intervals.

[0051] Under the influence of temperature, the volume of any material will change with temperature. Different materials have different linear expansion coefficients. The calculation formula of the wireless vibrating wire strain gauge under the action of temperature is as follows:

[0052] ε=k×(F-F0)+(b-α)×(T-T0);

[0053] Among them, ε is the strain change of the measured structure, k is the sensitivity coefficient of the sensor, b is the temperature correction coefficient of the sensor, α is the linear expansion coefficient of the measured structure, F is the measured modulus value of the sensor, T is the measured modulus value of the sensor, F0 is the reference value of the modulus, and T0 is the standard value of the temperature.

[0054] The stress can be calculated based on the strain. The calculation formula of stress is as follows:

[0055] σ=ε×E;

[0056] Among them, σ is stress, ε is strain, and E is the elastic modulus of the object being measured.

[0057] Displacement meters are installed according to the characteristics of the bridge structure and construction requirements. The installation of the displacement meter must ensure accurate positioning and reliable fixation. At the same time, the installation direction of the displacement meter must also be considered to ensure that the displacement changes of the bridge structure in a specific direction can be accurately measured. After installation, the displacement meter needs to be initialized and calibrated, usually at the initial position of the bridge structure without external force. Calibration is to verify the accuracy and reliability of the displacement meter by comparing the known displacement with the displacement meter reading.

[0058] According to the scale, complexity and safety risk level of bridge construction, corresponding monitoring equipment is set up. Considering the coverage and clarity of the monitoring angle, multiple monitoring equipment is set up in the construction area, which includes: foundation construction area, pier construction area, beam hoisting area, etc. For example, in the pier construction area, the monitoring equipment is installed on a nearby tower crane or a temporary device at a high place to obtain comprehensive monitoring of the pier construction; multiple monitoring equipment is set up in the foundation construction area to cover different working surfaces and depths.

[0059] Based on the repeater, the monitoring data acquired by the monitoring device in real time is transmitted to the receiving terminal, the monitoring data from the wireless monitoring device is collected, multi-source heterogeneous information is acquired, and data services are provided to the receiving terminal through the processing layer and the service layer;

[0060] Furthermore, the multi-source heterogeneous information is obtained, and a stream processing engine is used to process the multi-source heterogeneous data at the processing layer, and the processed data is stored in a stream processing database at the service layer to provide data services for the receiving terminal;

[0061] The multi-source heterogeneous information includes: model information, monitoring information, forecast result information, disease monitoring information, maintenance measure information, quality-related file data and on-site acceptance data.

[0062] Specifically, the repeater receives real-time monitoring data and multi-source heterogeneous data from different data sources, which have different formats, structures, and transmission rates. When the data enters the processing layer, the stream processing engine parses and classifies the data in real time, for example, distinguishing between real-time monitoring data from sensors, monitoring equipment data, and event data sent by other multi-source heterogeneous data sources. Each type of data is processed according to preset rules and logic, including data cleaning, noise removal, outliers, and duplicate data removal, to ensure data quality.

[0063] The stream processing engine will convert data and convert data of different formats into a unified annotated format for subsequent analysis and processing, such as converting text data of various encoding formats into a unified encoding format. While converting data, it will aggregate and associate data, for example, aggregating relevant data from different sensors in the same time period, and associating information about the same entity in different data sources to form a more comprehensive view.

[0064] The data processed by the stream processing engine is transmitted to the stream processing database of the service layer. The stream processing database automatically allocates appropriate storage space and storage structure according to the data type. For example, a special storage method is used for time series data to facilitate subsequent query and analysis based on the time dimension. During the data storage process, data indexes are created and optimized to enable fast retrieval and acquisition of data. At the same time, a data backup and recovery mechanism is set up to run in the background.

[0065] When the data is successfully stored in the stream processing database, it starts to provide data services to the receiving terminal. During the process, the stream processing database always maintains efficient data management and service provision, provides accurate and timely data support for each module of the receiving terminal, and ensures smooth system data flow.

[0066] The receiving terminal receives the monitoring data and processes and outputs the data;

[0067] The receiving terminal includes: a structural stress monitoring module, a displacement monitoring module, a crack monitoring module, a bridge construction evolution analysis module and a safety early warning module.

[0068] Furthermore, the structural stress monitoring module is used to obtain stress and strain monitoring data, and analyze the acquired stress and strain monitoring data through a time series analysis method based on deep learning, so as to monitor and predict the structural stress of the bridge.

[0069] Optionally, the time series analysis method based on deep learning uses an LSTM model to monitor and predict bridge structure stress. The steps for building the LSTM model are as follows:

[0070] Data collection: historical stress data, environmental data and traffic data of the bridge structure are collected from the stream processing system. The environmental data includes temperature, humidity and wind speed, and the traffic data includes traffic flow data, vehicle load data and special event data.

[0071] Data preprocessing: Clean the collected data, remove outliers and noise, and normalize the data for model training.

[0072] Feature engineering: Analyze the relationship between bridge structure stress and other factors, and extract features related to prediction, including the periodicity, trend, and seasonality of the time series.

[0073] Data segmentation: The dataset is divided into training set, validation set, and test set in a 6:2:2 ratio to facilitate model training and evaluation.

[0074] Model building and training: Determine the structure of the LSTM model, including the number of layers, the number of elements in each layer, and hyperparameters such as the activation function; use the training set to train the LSTM model, adjust the model parameters and supervise the training process, and use the validation set to evaluate the model performance to prevent overfitting.

[0075] Model verification and adjustment: Use the test set to evaluate the performance of the trained LSTM model, and evaluate and adjust the model parameters according to the evaluation indicators, which include prediction accuracy, mean square error (MSE), and correlation coefficient, etc. The model parameters include network structure, learning rate, and regularization parameters, etc.

[0076] The trained model is deployed to the receiving terminal to monitor and predict the bridge structure stress collected during the bridge construction process in real time, and the prediction results are output. The prediction results of the model are analyzed and compared with the actual stress values ​​to evaluate the accuracy of the model, and the monitoring strategy or model parameters are adjusted according to the feedback. As the monitoring data accumulates, the model is continuously optimized to improve the accuracy and adaptability of the prediction.

[0077] Furthermore, the displacement monitoring module monitors the displacement of the bridge based on deep learning image recognition technology, monitors the displacement of the bridge by comparing images at different time points, combines three-dimensional laser scanning technology, and uses deep learning algorithms to perform three-dimensional reconstruction to monitor the three-dimensional displacement of the bridge.

[0078] Specifically, the displacement monitoring value is obtained in real time through the displacement meter and transmitted to the repeater, and a large number of image sequences at different time points during the bridge construction process are obtained by monitoring equipment at multiple fixed positions. The collected images are preprocessed, including: denoising, filtering, image enhancement, etc. The image recognition model based on deep learning is used to extract and match the features of the images at different time points to calculate the bridge displacement information reflected by the image.

[0079] The training of the image recognition model is carried out by using bridge data with known displacement states. These data include image data collected by monitoring equipment and corresponding labels. The corresponding labels are the displacement states of the bridges. Through training, the image recognition model learns how to extract key features from image data and predict the displacement states of bridges.

[0080] Preferably, a high-precision 3D laser scanner is used to scan the bridge from multiple angles and distances to obtain 3D point cloud data of the bridge surface. Image data of the bridge at different positions and different time points are obtained through monitoring equipment, and these images are used for subsequent deep learning feature extraction and displacement monitoring.

[0081] Remove noise, redundant points and outliers from 3D point cloud data, perform denoising, contrast enhancement, color correction and other operations on the acquired image data to improve the quality of image data and the accuracy of feature extraction. Coordinate transformation and calibration technology ensure that data collected at different time points and different devices can be compared in the same coordinate system.

[0082] A convolutional neural network model is selected to extract bridge features. A large amount of annotated bridge image data is obtained to train the convolutional neural network model. The annotated data includes information such as the location of key points and dimensions of the bridge. The annotated data is used for model training. The accuracy and robustness of feature extraction are improved by adjusting model parameters and optimizing algorithms. The three-dimensional point cloud data is processed using a point cloud segmentation algorithm to extract bridge structure information. Based on the processed point cloud data, a deep learning surface reconstruction algorithm is used to generate a three-dimensional model of the bridge. The algorithm can learn the shape and structure of the bridge and generate a high-resolution three-dimensional surface.

[0083] The three-dimensional models of the bridge at different time points are registered to ensure alignment in the same coordinate system. The position information of key points is extracted from the registered three-dimensional model and matched. These key points can be specific parts or structural elements of the bridge. By comparing the position changes of key points at different time points, the three-dimensional displacement of the bridge is calculated, including the translation and rotation displacement of the X-axis, Y-axis, and Z-axis, and the calculated displacement data is output to the repeater.

[0084] The repeater receives the precise displacement value transmitted by the displacement meter and the displacement information calculated by the image recognition system, integrates the data, and transmits the complete displacement monitoring data to the receiving terminal.

[0085] Furthermore, the crack detection module uses image processing and machine learning technology to identify the characteristics of bridge construction defects, identifies the shape, size and distribution of cracks through a deep learning model, and predicts the development trend of cracks.

[0086] Specifically, high-quality bridge construction images containing defective areas in historical data are obtained, and operations such as denoising, contrast enhancement, and cropping are performed on the images to improve image quality and highlight defect features.

[0087] Manually annotate the large amount of image data obtained to clarify information such as the shape, size, and distribution of cracks, select a suitable deep learning model for image segmentation and feature recognition, train the model using the annotated data so that the model learns the crack characteristics, evaluate the model performance through the validation set, and make adjustments and optimizations based on the results.

[0088] The characteristic data of the cracks are extracted from the continuous images of the historical data to form a time series, and a time series prediction model is adopted. In this embodiment, a recurrent neural network model is adopted, and the model is trained using the historical crack characteristic sequence data.

[0089] Inputting the current fracture characteristics collected in real time enables the model to output future development trends.

[0090] After preprocessing the bridge construction images, they are input into the trained deep learning model, which can accurately segment the crack area and identify the shape and size of the cracks. Then, using the extracted crack feature time series, the LSTM model is used to predict the possible expansion of the cracks in the future. With the continuous accumulation of data and continuous improvement of the model, the identification of crack features and the prediction of development trends will be more accurate and reliable.

[0091] Furthermore, the bridge construction evolution analysis module uses a convolutional neural network to extract and classify features from video data to identify the construction area of ​​the bridge construction site, trains the model to identify different types of construction activities, combines the optical flow method with the background subtraction method to track the dynamic changes of the construction site, and distinguishes between the construction area and the non-construction area; by combining AIS and VHF data, a machine learning algorithm is used to analyze and predict the progress of bridge construction to evaluate the impact of bridge construction on the operating line, and a clustering algorithm is used to classify vehicle traffic patterns to identify possible traffic hotspots;

[0092] The different types of construction activities include: lifting, welding, concrete pouring, grouting reinforcement and grouting.

[0093] Specifically, we collect historical detection data, perform preprocessing operations such as cropping and scaling on relevant video data to make it meet the input requirements of the convolutional neural network, and design a convolutional neural network architecture containing multiple convolutional layers. These convolutional layers use convolution kernels of different sizes to extract features at different levels in the video, such as edges, shapes, etc., and use the ReLU activation function after the convolutional layer to introduce nonlinearity and enhance the feature expression ability. Use the pooling layer to downsample the convolutional features, reduce the amount of data and extract the main features. Add a fully connected layer to map the extracted features to different categories, and set the softmax classifier to output the probability corresponding to each category.

[0094] A large number of video clips were annotated to clarify the construction area and type of construction activity corresponding to each clip. The annotated data was divided into a training set and a validation set in a ratio of 7:3. The training set was used for model training, and the model parameters were continuously adjusted through the back propagation algorithm. The validation set was used to evaluate the model performance, and necessary adjustments and optimizations were made based on the results, such as modifying the network structure and adjusting hyperparameters. The trained model was then deployed to the module.

[0095] Combining optical flow and background subtraction to track dynamic changes at the construction site, distinguishing between construction and non-construction areas includes:

[0096] Equipment installation: Install cameras at the construction site to ensure that the cameras can cover the area that needs to be monitored. Multiple cameras can be set up to obtain different perspectives of the construction site;

[0097] Background modeling: Using background subtraction, we first need to record areas of the construction site where there is no construction activity for a period of time to build a background model. The adaptive background model is used for subsequent image processing to identify changing areas to adapt to the dynamic changes of the construction site.

[0098] Application of optical flow method: At the construction site, the displacement variables are calculated by the optical flow algorithm to track the moving construction equipment and workers.

[0099] Image processing and analysis: Get the images captured by the camera in real time. By applying background subtraction, compare the current frame with the background model to identify the changed area. Apply optical flow to track the trajectory of moving objects.

[0100] Area classification: Use computer vision technology and deep learning classification models to classify the changed areas and distinguish between construction areas and non-construction areas. Specifically, the trained convolutional neural network model deployed in the module is trained using labeled image data to improve the accuracy of classification.

[0101] Combining AI S and VHF data with vehicle traffic pattern data classified by cluster analysis, machine learning algorithms are used to analyze and predict the progress of bridge construction to evaluate the impact of bridge construction on the operating line. The image information is enhanced by YOLOv5s and Cutpaste algorithms, and the backbone network based on the CBAM attention mechanism and FPN (feature pyramid network) + PAN (path aggregation network) are introduced to extract target features to achieve monitoring of the bridge installation and construction status.

[0102] The module collects AIS and VHF data during bridge construction. AIS data can provide information such as the ship's location, speed, and heading. VHF data can provide communication content between ships, which is crucial for analyzing the activity patterns of ships during construction. At the same time, image data of the bridge construction process obtained by monitoring equipment uploaded by the repeater is collected for training machine learning models to identify and detect the installation and construction status of the bridge.

[0103] By using a high-precision 3D laser scanner, the bridge is holographically scanned to obtain 3D point cloud data of the bridge surface. The bridge-related image data is obtained through monitoring equipment, and the point cloud data is processed, including denoising, fusion, thinning and registration, etc. The image data is grayed and denoised, texture features are extracted, and the data is annotated to mark key information such as the various components of the bridge, installation location, and construction status.

[0104] Clean the collected AIS data, remove duplicate, erroneous and irrelevant data, and correct missing data to maintain the integrity and consistency of the data set. Convert AIS data into a unified format to ensure that data fields, data types and units meet the requirements. Extract data-related features, including location, speed and direction.

[0105] The VHF data is formatted and the data containing voice content is converted into text format using speech recognition technology, which helps to extract text features for analysis. The converted text is cleaned to remove irrelevant words, punctuation marks, and special characters. Natural language processing (NLP) technology is used to perform text segmentation and part-of-speech tagging, and relevant features such as keyword frequency and sentiment analysis are extracted from the processed data.

[0106] At the same time, data related to vehicle traffic patterns are collected, including vehicle trajectories, speeds, travel times, stops, and other data. These data can be obtained through traffic monitoring systems, vehicle sensors, GPS positioning, and other methods. The collected raw data needs to be cleaned and preprocessed to remove noise and outliers and converted into a format suitable for clustering algorithm processing. This may include operations such as standardization, normalization, or dimensionality reduction of the data to ensure that the dimensions and ranges of different features are consistent.

[0107] Extract features related to vehicle traffic patterns from the preprocessed data. These features can include driving speed, acceleration, driving distance, driving time, driving path, etc. These features will be used to describe the vehicle traffic patterns and provide input for the clustering algorithm.

[0108] The steps to build the K-means algorithm are as follows:

[0109] Initialize the cluster center and randomly select K points from the data set as the cluster center;

[0110] Assign data points to the nearest cluster center. For each point in the data set, calculate the Euclidean distance to each cluster center and assign each point to the cluster represented by the nearest cluster center.

[0111] For each cluster, calculate the mean of all points within it and use this mean as the new cluster center;

[0112] Repeatedly calculate the Euclidean distance between all data points and the new cluster center, divide each data point into the new cluster center with the smallest Euclidean distance to form a new cluster; calculate the average value of each new cluster as the new cluster center again, until the new cluster center no longer changes, and determine the final cluster;

[0113] Output the final clustering features and the cluster label to which each data point belongs.

[0114] The vehicle traffic patterns are divided into different categories according to the K-means algorithm, and the clustering results are analyzed to determine possible traffic hotspots. The areas with high density or abnormal traffic patterns are identified by analyzing the feature distribution and quantity of each category, and the analysis results are output.

[0115] YOLOv5s and Cutpaste algorithms are used for feature enhancement. As the basic model for detection, YOLOv5s has a fast detection speed and high accuracy, and is suitable for real-time monitoring tasks. During the training process, the Cutpaste algorithm is used to enhance the image data. This algorithm increases the generalization ability of the model by randomly cropping the target area in the image and pasting it to other images.

[0116] Introduce the CBAM attention mechanism and introduce the CBAM attention mechanism into the backbone network of YOLOv5s. CBAM improves the model's feature representation ability through two dimensions: channel attention and spatial attention. Among them, channel attention helps the model focus on important channel features and ignore unimportant channel information; spatial attention enables the model to focus on key areas in the image and improve the ability to locate the target position. Add a CBAM module after each residual unit and SPP layer of YOLOv5s to enhance the model's ability to capture the dependencies between features.

[0117] FPN (feature pyramid) + PAN (path aggregation network) is used for target feature extraction. A feature pyramid network is constructed to improve the model's detection ability for targets of different sizes by fusing features of different scales. FPN performs feature fusion from top to bottom and from bottom to top, so that features of each scale are integrated with information of other scales. On the basis of FPN, a bottom-up path aggregation network is added. PAN further enhances the information transmission of low-level features, so that the model can consider both semantic information and positioning information when extracting target features. FPN and PAN are used together to form a more powerful feature extractor. In this way, the model can make full use of feature information of different scales and levels to improve the accuracy and robustness of bridge installation and construction status monitoring.

[0118] Use the labeled data to train the model that integrates YOLOv5s, CBAM, FPN, and PAN. During the training process, continuously adjust the model parameters and learning rate to achieve the best performance. Use the validation set to evaluate the trained model, and the evaluation indicators include accuracy, recall, F1 score, etc. Based on the evaluation results, further adjust and optimize the model.

[0119] The trained model is used to monitor the images during the bridge construction process in real time, identify the various components, installation locations and construction status of the bridge, and predict the progress of the bridge construction based on the results of real-time monitoring, including construction progress, installation quality and other aspects. Combined with AIS and VHF data and cluster analysis results, the activity patterns of ships and vehicle traffic patterns during construction are analyzed, the development trend of bridge construction is analyzed, and the impact of bridge construction on the operating line is evaluated. The evaluation results are output and uploaded to the bridge monitoring file corresponding to the bridge for storage. Combined with the above processing, the full life cycle management of bridge construction is achieved.

[0120] Furthermore, the safety warning module comprehensively analyzes the monitoring data of structural stress, displacement, cracks and traffic flow through a preset multi-model fusion warning system, and uses a random forest algorithm to formulate a safety threshold. When the detection index exceeds the formulated threshold, the system will automatically trigger a warning, realize dynamic monitoring and visualize the status of related areas, and output the warning information and related decision-making suggestions to on-site managers and relevant departments based on the monitoring results.

[0121] Specifically, the monitoring data output by the structural stress monitoring module, displacement monitoring module, crack monitoring module and bridge construction evolution analysis module are received and preprocessed. Key features related to bridge structure safety and traffic flow are extracted, including stress peak value, displacement length, crack width and time distribution of traffic flow, etc. New features are created through feature combination or conversion to better reflect the safety status of the bridge structure.

[0122] The integrated data set is used to train the random forest model, and multiple decision trees are constructed to improve the stability and accuracy of the prediction. During the training process, the model learns the impact of each feature on the safety of the bridge structure and makes predictions based on these features. The difference between the predicted results and the actual safety status is analyzed to confirm which features or feature combinations have the greatest impact on bridge safety. The safety threshold is preliminarily determined based on the analysis results and the safety standards for rain leakage.

[0123] For example, in bridge construction, stress monitoring is mainly to evaluate the safety performance of the bridge structure when bearing loads. In a certain key section of the bridge, such as the mid-span section of the main beam, the maximum tensile stress should not exceed the allowable tensile stress of the section material. Assuming that C30 concrete is used in this section, the maximum tensile stress safety threshold of this section can be set to 1.4MPa based on the prediction and analysis results of the random forest model. For a certain support point of the bridge, such as the top of the pier, its vertical displacement should be controlled within a certain range to ensure the stability and normal use of the bridge. Assuming that according to the prediction and analysis results of the random forest model combined with the design documents, the maximum vertical displacement of the support point should not exceed 10mm, then we can set the vertical displacement safety threshold of the support point to 8mm to leave a certain safety margin.

[0124] The safety warning module visualizes the regional status according to the regional division information of the bridge construction evolution analysis module. When the monitoring data detection index in the monitoring data output by each module exceeds the set safety threshold, the detection index and warning report are output at the corresponding position in the visualized regional status, and decision-making suggestions are provided based on the processing method and processing effect of the most similar related events in the historical related events, thereby improving the efficiency of solving safety hazards in the bridge construction process and ensuring the safety of personnel and property.

[0125] The above implementation modes are merely descriptions of the preferred implementation modes of the present invention, and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary engineering and technical personnel in the field shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for monitoring construction of bridges across operating lines based on machine vision recognition algorithm, characterized in that: include: Place wireless monitoring equipment at designated locations to obtain real-time monitoring data during bridge construction; Based on the repeater, the monitoring data acquired by the monitoring device in real time is transmitted to the receiving terminal, the monitoring data from the wireless monitoring device is collected, multi-source heterogeneous information is acquired, and data services are provided to the receiving terminal through the processing layer and the service layer; The receiving terminal receives the monitoring data and processes and outputs the data; The wireless monitoring equipment includes: a wireless vibrating-wire strain gauge, a wireless pressure sensor, a displacement meter and a monitoring device; The monitoring data include: stress, strain and displacement of key parts during bridge construction, and monitoring images of the bridge construction site; The receiving terminal includes: a structural stress monitoring module, a displacement monitoring module, a crack monitoring module, a bridge construction evolution analysis module and a safety early warning module.

2. The method for monitoring construction of bridges across operating lines based on machine vision recognition algorithm according to claim 1 is characterized in that: Obtaining monitoring data during bridge construction includes: Obtain stress and pressure data of various parts of the bridge during the bridge construction process, and install multiple wireless vibrating wire strain gauges and wireless pressure sensors at the bridge structure connection and bridge foundation; Obtaining displacement change data of the bridge structure during the bridge construction process, wherein the displacement change data includes horizontal displacement, vertical displacement and torsional displacement, and setting a plurality of displacement meters at the connection between the bridge pier and the bridge deck, the mid-span position of the bridge, the bridge support and the expansion joint; Acquire the construction site monitoring images during the bridge construction process, and set up multiple monitoring devices in the areas that need to be monitored to obtain different viewing angles of the monitoring images.

3. The method for monitoring construction of bridges across operating lines based on machine vision recognition algorithm according to claim 2 is characterized in that: The bridge structure connection includes: the connection seams between precast beams, the joints between the beam body and the top of the pier, the intersection area between the bottom of the pier and the top surface of the foundation, the connection surface between the top of the pile foundation and the pedestal, the connection between the pier and the beam body, and the transition section connection between the continuous beam and the simply supported beam.

4. The method for monitoring construction of bridges across operating lines based on machine vision recognition algorithm according to claim 2 is characterized in that: The bridge foundation parts include: sections of the pier body at different heights, the top surface of the pier foundation, the wall of the caisson foundation, and the pile body of the pile foundation; The areas that need to be monitored include the construction area and the non-construction area of ​​the construction site.

5. The method for monitoring construction of bridges across operating lines based on machine vision recognition algorithm according to claim 1 is characterized in that: The method of acquiring multi-source heterogeneous information, using a stream processing engine to process the multi-source heterogeneous data at the processing layer, storing the processed data in a stream processing database at the service layer, and providing data services for receiving terminals; The multi-source heterogeneous information includes: model information, monitoring information, forecast result information, disease monitoring information, maintenance measure information, quality-related file data and on-site acceptance data.

6. The method for monitoring construction of bridges across operating lines based on machine vision recognition algorithm according to claim 1 is characterized in that: The structural stress monitoring module is used to obtain stress and strain monitoring data, and analyze the obtained stress and strain monitoring data through a time series analysis method based on deep learning to monitor and predict the structural stress of the bridge.

7. The method for monitoring construction of bridges across operating lines based on machine vision recognition algorithm according to claim 1 is characterized in that: The displacement monitoring module monitors the displacement of the bridge based on deep learning image recognition technology, monitors the displacement of the bridge by comparing images at different time points, and combines three-dimensional laser scanning technology and uses deep learning algorithms to perform three-dimensional reconstruction to monitor the three-dimensional displacement of the bridge.

8. The method for monitoring construction of bridges across operating lines based on machine vision recognition algorithm according to claim 1 is characterized in that: The crack detection module uses image processing and machine learning technology to identify the characteristics of bridge construction defects, identifies the shape, size and distribution of cracks through a deep learning model, and predicts the development trend of cracks.

9. The method for monitoring construction of bridges across operating lines based on machine vision recognition algorithm according to claim 1 is characterized in that: The bridge construction evolution analysis module uses a convolutional neural network to extract and classify features from video data to identify the construction area of ​​the bridge construction site, trains the model to identify different types of construction activities, combines the optical flow method with the background subtraction method to track the dynamic changes of the construction site, and distinguishes between the construction area and the non-construction area; by combining AIS and VHF data, a machine learning algorithm is used to analyze and predict the progress of bridge construction to evaluate the impact of bridge construction on the operating line, and a clustering algorithm is used to classify vehicle traffic patterns to identify possible traffic hotspots; The different types of construction activities include: lifting, welding, concrete pouring, grouting reinforcement and grouting.

10. The method for monitoring construction of bridges across operating lines based on machine vision recognition algorithm according to claim 1 is characterized in that: The safety warning module comprehensively analyzes the monitoring data of structural stress, displacement, cracks and traffic flow through a preset multi-model fusion warning system, and uses a random forest algorithm to formulate a safety threshold. When the detection index exceeds the formulated threshold, the system will automatically trigger a warning, realize dynamic monitoring and visualize the status of related areas, and output the warning information and related decision-making suggestions to on-site managers and relevant departments based on the monitoring results.

Citation Information

Cited By

  • Intelligent bridge support detection and maintenance management system and method

    CN119714433A

  • Construction road crack prediction method based on time sequence monitoring data

    CN119723448A

  • Concrete construction quality monitoring platform based on Internet of Things

    CN119990920A

  • High-strength concrete construction crack detection system and method

    CN120352608A

  • A high-strength concrete construction crack detection system and method

    CN120352608B