Engineering management system based on digital twinning

Abstract

The engineering management system based on digital twinning comprises a data acquisition synchronization module which acquires construction site data by using an Internet of Things device and a sensor; a digital twinborn model: constructing a digital twinborn model corresponding to an actual project based on the collected data; the twinborn model management module is used for constructing and maintaining a digital twinborn model of an engineering project; the data digitalization module is used for digitally processing the collected data, and comparing and analyzing the obtained point cloud data with a digital twinborn model to verify the integrating degree; the data analysis and decision-making module is used for evaluating project progress deviation by predicting project risks and generating corresponding decision-making suggestions; and the visual interaction module displays the digital twinborn model through a three-dimensional visual interface, and views an engineering state, historical data and a prediction result in real time through interactive operation, and aims to solve the problems of low data acquisition and processing efficiency, unsmooth information transmission and poor management collaboration in a traditional engineering management mode.

Description

Technical Field

[0001] The present invention belongs to the technical field of engineering management and relates to an engineering management system based on digital twins. Background Art

[0002] With the booming construction industry and the continuous advancement of science and technology, the scale and complexity of engineering projects are increasing day by day. In the field of engineering construction, ensuring that the construction process meets the design requirements and grasping the construction progress in real time is crucial to ensuring project quality, controlling costs and delivering on time. Digital twin technology came into being. With the help of advanced modeling, simulation and data processing technologies, it builds a digital model that completely corresponds to the real project in the virtual space, which can reflect the actual status of the project in real time and bring new ideas and methods to project management. Through the digital twin model, engineering managers can intuitively understand the progress of the project, simulate the effects of different construction plans, and make more scientific decisions. However, in actual applications, there are still many challenges to achieve accurate reflection of the construction progress and effective comparison of twin data with design data.

[0003] The current problems in production and actual use are that the construction progress is not accurately reflected. Although digital twin technology can build virtual models, in actual construction scenarios, due to the complex construction site environment and various interference factors, the collected data may be inaccurate or missing. For example, some sensors may be affected by bad weather, electromagnetic interference from construction equipment, etc., resulting in the collected data being unable to accurately reflect the actual construction situation, which in turn causes the digital twin model to have deviations in reflecting the construction progress and cannot provide a reliable basis for project management.

[0004] It is difficult to compare twin data with design data: Twin data and design data come from different sources, in various formats, and have different semantics and structures. Design data usually exists in the form of drawings, documents, etc., and contains a large number of professional symbols, annotations, and text descriptions; while twin data is real-time data obtained through sensor collection, simulation, etc., such as point cloud data, equipment operating parameters, etc. The comparison of these two types of non-similar data requires solving a series of complex problems such as data fusion, feature extraction, and matching. Traditional methods are difficult to conduct efficient and accurate comparative analysis, resulting in the difficulty in timely discovery and quantification of construction and design differences, and the difficulty in effectively ensuring the quality of the project.

[0005] Lack of effective decision support: Even if it is possible to obtain construction progress information and discover differences between construction and design, the existing project management system is still insufficient in providing decision support for project managers. Due to the lack of in-depth data mining and analysis capabilities, it is impossible to provide highly targeted and operational decision-making suggestions based on the difference analysis results and the actual project situation. Project managers can only make decisions based on experience, which not only increases decision-making risks, but may also lead to adverse consequences such as project delays and increased costs.

[0006] At present, in some projects, manual inspections are still relied on regularly to inspect the construction site, judge the construction progress through observation and recording, and compare the actual situation with the design drawings. This method has a certain effect when discovering obvious construction deviations; the point is that it is intuitive and flexible, and does not require complex technical equipment and professional knowledge. But the disadvantages are also obvious. Manual inspections are inefficient and limited in frequency, making it difficult to achieve real-time monitoring; and human judgment is easily affected by subjective factors. Some subtle differences between construction and design may not be accurately identified, and the requirements of modern engineering management for high precision and timeliness cannot be met; some projects use some simple data processing software to organize and analyze the collected construction data, and display the construction progress in the form of charts and other forms. For the comparison of twin data and design data, these software mainly processes through simple data format conversion and basic data analysis functions. Compared with manual methods, they can process and display some data more quickly, improving work efficiency. However, this type of software has relatively single functions and limited processing capabilities for complex data. It cannot effectively solve the problem of deep comparison between twin data and design data. When facing large-scale and complex engineering projects, the accuracy and reliability of its analysis results are low.

[0007] The patent number is 202411020624.X, and the patent name is a construction site supervision data analysis method and system based on digital twins. The solution uses digital models and data processing technology to digitally present the actual scene of the construction site through the digital twin module, and compare and analyze data from different time periods. It can detect data anomalies in a timely manner. However, only data with large contrast differences are visualized, and there is a lack of deeper mining and analysis of the data. It is difficult to discover the potential rules and trends behind the data, which is not conducive to forward-looking management decisions. In addition, different construction site scenes vary greatly, and the existing modeling method may not be able to adapt to all scenes, resulting in the digital twin module The degree of restoration of the actual scene is insufficient.

[0008] In view of the above problems, the present invention proposes an engineering management system based on digital twins. The system uses Internet of Things devices and sensors through a data acquisition and synchronization module to realize real-time acquisition of construction site data, and ensures that the data is accurately transmitted to the twin model, so as to more accurately reflect the construction progress, convert various types of information in the design stage into digital format and effectively manage them, and standardize and associate the construction site data. The data analysis and decision-making module adopts data mining and machine learning algorithms, especially for the comparison between point cloud data and digital twin models, the feature matching algorithm is used to accurately calculate the deviation value between the construction and design models, and construct a difference analysis model. Combined with the engineering experience knowledge base and the decision support model, scientific and accurate decision-making suggestions are provided for engineering managers. The visualization interaction module uses 3D visualization and VR / AR technology to enable engineering managers to intuitively view the real-time status of the project, construction progress and difference analysis results. Through interactive operations, historical data and prediction results can also be obtained, which improves the efficiency and accuracy of engineering management, and effectively solves the technical problem of reflecting the construction progress by fitting the real scene through digital twins, and comparing the twin data with the design data to reflect the difference between construction and design. Summary of the invention

[0009] The present invention provides an engineering management system based on digital twins to solve the problem of reflecting the construction progress by fitting real-life scenarios through digital twins, while comparing twin data and design data, two types of non-similar data, to reflect the differences between construction and design.

[0010] In order to solve the above problems, the technical solution adopted by the invention is:

[0011] Engineering management system based on digital twin, including:

[0012] Data acquisition and synchronization module: uses IoT devices and sensors to collect construction site data and transmits it to the twin model in real time;

[0013] Digital twin model: Based on the collected data, three-dimensional modeling, simulation algorithms and machine learning technology are used to build a digital twin model that corresponds to the actual project.

[0014] Twin model management module: According to the engineering design and construction plan, the digital twin model is decomposed into subtasks and distributed according to the construction team and stage; the task status in the digital twin model is updated according to the collected construction progress data; the status changes in different stages are recorded to facilitate retrospective comparison. When the design changes, the model can be quickly modified and the tasks can be redistributed;

[0015] Data digitization module: convert paper drawings and documents in the design phase into digital text and image data, convert drawing images into editable vector graphics through drawing vectorization software, extract geometry and annotation information and store them in the design database; digitize the collected construction site data, convert analog signals into digital signals, store and manage them in a unified format, and perform standardized processing; establish a data association mechanism to associate design and construction data according to engineering objects and construction processes

[0016] The data analysis and decision-making module uses data mining and machine learning algorithms to compare the twin data generated by the digital twin model with the design data processed by the data digitization module, build a difference analysis model, and identify the differences between construction and design. For the comparison between point cloud data and digital twin models, a feature matching algorithm is used to extract the geometric features of point cloud and digital twin models for matching, which is used to calculate the deviation value of construction and design models. Based on the difference analysis results, combined with the engineering experience knowledge base and decision support model, decision suggestions are provided to engineering managers.

[0017] The visualization interaction module uses 3D visualization and virtual reality (VR) / augmented reality (AR) technology to present the digital twin model to project managers. Users can view the real-time status of the project, construction progress and equipment operation status, and display the difference reports and decision-making recommendations generated by the data analysis and decision-making modules in visual charts and text.

[0018] The digital twin model is displayed through a 3D visualization interface, and the project status, historical data, and prediction results can be viewed in real time through interactive operations;

[0019] The specific method of data processing by the data digitization module is as follows:

[0020] S01 uses block-based density clustering and direction vector analysis. It first divides the point cloud data into small blocks, clusters them by calculating the density of the points in each small block, and then determines the boundary and connection relationship based on the direction vector between the points, thereby extracting various structural features in the point cloud. At the same time, it uses a denoising autoencoder based on deep learning to denoise the point cloud data obtained by 3D laser scanning. At the same time, it uses a feature-based point cloud enhancement algorithm, edge detection and curvature calculation algorithms to highlight the structural features in the point cloud data.

[0021] S02 For project-related materials, first perform image scanning or text recognition to convert them into digital formats. For design drawings, use image vectorization technology to convert scanned raster images into vector graphics. For text materials, use optical character recognition technology to convert them into editable text. Then use natural language processing technology to extract information and build a structured knowledge graph. Then, associate the information in the project materials with the digital twin model and point cloud data.

[0022] S03 uses an algorithm based on normal vectors and curvature features to calculate the local features of each point in the point cloud data, quickly finds the corresponding point pairs between the point cloud data and the digital twin model, uses a random sampling consistency algorithm to remove mismatched point pairs, and defines an energy function that includes the distance error and geometric constraints between the point cloud data and the digital twin model, and uses the gradient descent method to solve the minimum value of the energy function to obtain a registration result;

[0023] S04 compares the registered point cloud data with the digital twin model point by point in three-dimensional space, calculates the Euclidean distance between each corresponding point, and constructs a three-dimensional space error field;

[0024] S05 slices the three-dimensional error field into multiple two-dimensional images, and then trains the CNN model to enable the CNN model to learn the characteristic patterns of the error field under different degrees of fit. The error field data of the current construction is classified through the trained model, and the degree of fit between the actual construction and the design requirements is output as "high fit", "medium fit" and "low fit".

[0025] The principles and advantages of this solution are:

[0026] By constructing a digital twin model that closely corresponds to the actual project and realizing comprehensive management and monitoring of the project through the collaborative operation of multiple modules, the data acquisition and synchronization module uses IoT devices and sensors to obtain construction site data in real time; the digital twin model converts the collected data into a virtual model. The twin model management module decomposes, distributes and updates the status of the digital twin model according to the project design and construction plan. The data digitization module digitizes the paper materials in the design phase, processes and standardizes the construction site data, and establishes a linkage mechanism between design and construction data. The data analysis and decision-making module compares the twin data with the design data, identifies differences and provides decision-making suggestions. The visualization interaction module uses 3D visualization and VR / AR technology to intuitively present the digital twin model and analysis results to project managers, facilitating their operation and decision-making.

[0027] Compared with the existing technology, in terms of data processing, the data digitization module uses a denoising autoencoder based on deep learning and a point cloud enhancement algorithm to process point cloud data, which can improve data quality, highlight structural features, and provide a data basis for subsequent analysis. When digitizing the data, a structured knowledge graph is constructed and associated with the digital twin model, which enhances the relevance and availability of the data. For the comparison between point cloud data and the digital twin model, corresponding point pairs can be found quickly and accurately and registered, which improves the accuracy and efficiency of difference analysis. In terms of difference analysis and decision support, by constructing a difference analysis model and using the engineering experience knowledge base, more targeted and reliable decision-making suggestions are provided to managers. The application of the visual interaction module enables engineering managers to view the project status, historical data and prediction results more intuitively and conveniently, effectively improving the visualization and interactivity of engineering management, which helps to improve the efficiency and quality of engineering management and reduce construction risks.

[0028] Furthermore, the method for constructing a digital twin model by the twin model management module includes:

[0029] S011 uses laser radar to obtain 3D point cloud data of engineering entities, including the geometric shape of objects; uses cameras to collect texture information, including color and material details; deploys sensors to collect temperature, humidity, stress and strain related physical quantity data, and monitors engineering environment and structural status; at the same time, extracts text data information from engineering design documents, construction records, and operation and maintenance logs;

[0030] S012 performs corresponding preprocessing for different types of data. For point cloud data, a statistical filtering algorithm is used to remove outliers. Suppose each point P in the point cloud data set i The neighborhood point set is N i , calculate the neighborhood point to P i The distance d ij (j∈N i ), if d i If it is greater than the mean μ plus k times the standard deviation σ, then P i To remove outliers, grayscale, denoise and enhance image data to improve image quality. Natural language processing technology is used for text data to perform word segmentation, part-of-speech tagging and named entity recognition, extract key information and convert it into structured data.

[0031] S013 uses the Poisson reconstruction algorithm to construct the initial geometric model based on the preprocessed point cloud data. The algorithm converts the discrete point cloud data into a continuous triangular mesh model by solving the Poisson equation. The normal vector field of the point cloud data is set to n, and a scalar function S is defined to satisfy the Poisson equation. The function S is obtained by iterative solution, and then a geometric model represented by a triangular mesh is generated;

[0032] S014 combines the semantic information extracted from the text data to semantically annotate the geometric model. Different components in the building model are annotated according to the names and functions in the design documents. The Apriori algorithm is used to mine the potential relationship between geometric features and semantic information. Let I be the set of geometric features and semantic labels of all projects, and transaction T is a subset of I. By setting the support S and confidence C thresholds, association rules of the form X→Y (X, Y∈I) are mined to achieve deep fusion of semantics and geometry.

[0033] S015 uses the Kalman filter algorithm to process the dynamic data collected by the sensor, predict the model status and update it; when a specific event occurs in the project, the event-driven update mechanism is triggered. Based on the knowledge graph technology, according to the event type and related knowledge rules, the model parts related to it are automatically searched and updated;

[0034] S016 uses a deep autoencoder to detect anomalies in model data. The geometric features and physical quantity data under normal conditions are used as training sets to train the deep autoencoder. The deep autoencoder learns the feature representation of normal data through the encoding and decoding process. When new data is input, the reconstruction error is calculated. If the reconstruction error is greater than the set threshold, it is judged as abnormal data. Suppose the input data is x and the reconstructed data is Reconstruction error n is the data dimension. By analyzing the characteristics and location of abnormal data and combining methods such as fault tree analysis, the cause and impact range of potential problems can be diagnosed;

[0035] S017 uses long short-term memory network to predict the future state of engineering equipment or structure, providing a basis for predictive maintenance. Taking historical data as input, the LSTM model is trained to learn the time series characteristics of the data. Assuming the input sequence is x1, x2, ...x t , the LSTM unit passes through the forget gate f t , input gate i t , output gate o t and memory unit c t The collaborative work of calculating the hidden state h t and the predicted output Based on the prediction results, a maintenance plan is made in advance, and the maintenance information is fed back to the twin model to optimize the model parameters and structure.

[0036] Furthermore, in S01, the denoising autoencoder based on deep learning adopts a convolutional neural network structure, wherein the encoder part is composed of multiple convolutional layers and pooling layers, which are used to extract deep feature representations of point cloud data; the decoder part is composed of deconvolution layers and upsampling layers, which restores the encoded feature vector to denoised point cloud data.

[0037] Furthermore, in the feature-based point cloud enhancement algorithm in S01, edge detection uses the Canny edge detection algorithm, which extracts edge points in the point cloud data through steps such as smoothing the image by Gaussian filtering, calculating the gradient amplitude and direction, refining the edge by non-maximum suppression, and detecting and connecting the edges by double thresholds. The curvature calculation uses the principal component analysis (PCA) method. For each P, a PCA analysis is performed in its neighborhood to calculate the principal curvatures K1 and K2. The curvature type of the point is determined based on the principal curvature to highlight different types of structural features.

[0038] Furthermore, in S04, when constructing the three-dimensional space error field, for each point p in the point cloud data, its nearest neighbor point q is found in the digital twin model, and the Euclidean distance is calculated.

[0039] ,

[0040] In order to improve the efficiency of searching for nearest neighbor points, the KD tree data structure is used to organize the points in the digital twin model. The KD tree is a binary tree that divides the space into multiple subspaces by dividing the data points in different dimensions, making the time complexity of searching for the nearest neighbor points from o n Reduce to o logn , where n is the number of points in the digital twin model; after calculating the Euclidean distance, in order to more accurately evaluate the distribution of the error field, the error field is normalized, assuming that the maximum distance in the error field is d max , the minimum distance is d min , then the normalized error value

[0041] ,

[0042] Where d is the original Euclidean distance, and the normalized error value ranges between [0,1]. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 is a flow chart of the present invention;

[0044] Figure 2 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0045] Embodiment 1 is basically as attached Figure 1 As shown in the figure, the engineering management system based on digital twin includes:

[0046] Data acquisition and synchronization module: uses IoT devices and sensors to collect construction site data and transmits it to the twin model in real time;

[0047] Digital twin model: Based on the collected data, three-dimensional modeling, simulation algorithms and machine learning technology are used to build a digital twin model that corresponds to the actual project.

[0048] Twin model management module: According to the engineering design and construction plan, the digital twin model is decomposed into subtasks and distributed according to the construction team and stage; the task status in the digital twin model is updated according to the collected construction progress data; the status changes in different stages are recorded to facilitate retrospective comparison. When the design changes, the model can be quickly modified and the tasks can be redistributed;

[0049] Data digitization module: convert paper drawings and documents in the design phase into digital text and image data, convert drawing images into editable vector graphics through drawing vectorization software, extract geometry and annotation information and store them in the design database; digitize the collected construction site data, convert analog signals into digital signals, store and manage them in a unified format, and perform standardized processing; establish a data association mechanism to associate design and construction data according to engineering objects and construction processes

[0050] The data analysis and decision-making module uses data mining and machine learning algorithms to compare the twin data generated by the digital twin model with the design data processed by the data digitization module, build a difference analysis model, and identify the differences between construction and design. For the comparison between point cloud data and digital twin models, a feature matching algorithm is used to extract the geometric features of point cloud and digital twin models for matching, which is used to calculate the deviation value of construction and design models. Based on the difference analysis results, combined with the engineering experience knowledge base and decision support model, decision suggestions are provided to engineering managers.

[0051] The visualization interaction module uses 3D visualization and virtual reality (VR) / augmented reality (AR) technology to present the digital twin model to project managers. Users can view the real-time status of the project, construction progress and equipment operation status, and display the difference reports and decision-making recommendations generated by the data analysis and decision-making modules in visual charts and text.

[0052] The digital twin model is displayed through a 3D visualization interface, and the project status, historical data, and prediction results can be viewed in real time through interactive operations;

[0053] The specific method of data processing by the data digitization module is as follows:

[0054] S01 uses block-based density clustering and direction vector analysis. It first divides the point cloud data into small blocks, clusters the points by calculating the density of each small block, and then determines the boundary and connection relationship based on the direction vector between points, thereby extracting various structural features in the point cloud.

[0055] S02 For project-related materials, first perform image scanning or text recognition to convert them into digital formats. For design drawings, use image vectorization technology to convert scanned raster images into vector graphics. For text materials, use optical character recognition technology to convert them into editable text. Then use natural language processing technology to extract information and build a structured knowledge graph. Then, associate the information in the project materials with the digital twin model and point cloud data.

[0056] S03 uses an algorithm based on normal vectors and curvature features to calculate the local features of each point in the point cloud data, quickly finds the corresponding point pairs between the point cloud data and the digital twin model, uses a random sampling consistency algorithm to remove mismatched point pairs, and defines an energy function that includes the distance error and geometric constraints between the point cloud data and the digital twin model, and uses the gradient descent method to solve the minimum value of the energy function to obtain a registration result;

[0057] S04 compares the registered point cloud data with the digital twin model point by point in three-dimensional space, calculates the Euclidean distance between each corresponding point, and constructs a three-dimensional space error field;

[0058] S05 slices the three-dimensional error field into multiple two-dimensional images, and then trains the CNN model to enable the CNN model to learn the characteristic patterns of the error field under different degrees of fit. The error field data of the current construction is classified through the trained model, and the degree of fit between the actual construction and the design requirements is output as "high fit", "medium fit" and "low fit".

[0059] By constructing a digital twin model that closely corresponds to the actual project and realizing comprehensive management and monitoring of the project through the collaborative operation of multiple modules, the data acquisition and synchronization module uses IoT devices and sensors to obtain construction site data in real time; the digital twin model converts the collected data into a virtual model. The twin model management module decomposes, distributes and updates the status of the digital twin model according to the project design and construction plan. The data digitization module digitizes the paper materials in the design phase, processes and standardizes the construction site data, and establishes a linkage mechanism between design and construction data. The data analysis and decision-making module compares the twin data with the design data, identifies differences and provides decision-making suggestions. The visualization interaction module uses 3D visualization and VR / AR technology to intuitively present the digital twin model and analysis results to project managers, facilitating their operation and decision-making.

[0060] Compared with the existing technology, in terms of data processing, the data digitization module uses a denoising autoencoder based on deep learning and a point cloud enhancement algorithm to process point cloud data, which can improve data quality, highlight structural features, and provide a data basis for subsequent analysis. When digitizing the data, a structured knowledge graph is constructed and associated with the digital twin model, which enhances the relevance and availability of the data. For the comparison between point cloud data and the digital twin model, corresponding point pairs can be found quickly and accurately and registered, which improves the accuracy and efficiency of difference analysis. In terms of difference analysis and decision support, by constructing a difference analysis model and using the engineering experience knowledge base, more targeted and reliable decision-making suggestions are provided to managers. The application of the visual interaction module enables engineering managers to view the project status, historical data and prediction results more intuitively and conveniently, effectively improving the visualization and interactivity of engineering management, which helps to improve the efficiency and quality of engineering management and reduce construction risks.

[0061] The method for constructing a digital twin model by the twin model management module includes:

[0062] S011 uses laser radar to obtain 3D point cloud data of engineering entities, including the geometric shape of objects; uses cameras to collect texture information, including color and material details; deploys sensors to collect temperature, humidity, stress and strain related physical quantity data, and monitors engineering environment and structural status; at the same time, extracts text data information from engineering design documents, construction records, and operation and maintenance logs;

[0063] S012 performs corresponding preprocessing for different types of data. For point cloud data, a statistical filtering algorithm is used to remove outliers. Suppose each point P in the point cloud data set i The neighborhood point set is N i , calculate the neighborhood point to P i The distance d ij (j∈N i ), if d i If it is greater than the mean μ plus k times the standard deviation σ, then P i To remove outliers, grayscale, denoise and enhance image data to improve image quality. Natural language processing technology is used for text data to perform word segmentation, part-of-speech tagging and named entity recognition, extract key information and convert it into structured data.

[0064] S013 uses the Poisson reconstruction algorithm to construct the initial geometric model based on the preprocessed point cloud data. The algorithm converts the discrete point cloud data into a continuous triangular mesh model by solving the Poisson equation. The normal vector field of the point cloud data is set to n, and a scalar function S is defined to satisfy the Poisson equation. The function S is obtained by iterative solution, and then a geometric model represented by a triangular mesh is generated;

[0065] S014 combines the semantic information extracted from the text data to semantically annotate the geometric model. Different components in the building model are annotated according to the names and functions in the design documents. The Apriori algorithm is used to mine the potential relationship between geometric features and semantic information. Let I be the set of geometric features and semantic labels of all projects, and transaction T is a subset of I. By setting the support S and confidence C thresholds, association rules of the form X→Y (X, Y∈I) are mined to achieve deep fusion of semantics and geometry.

[0066] S015 uses the Kalman filter algorithm to process the dynamic data collected by the sensor, predict the model status and update it; when a specific event occurs in the project, the event-driven update mechanism is triggered. Based on the knowledge graph technology, according to the event type and related knowledge rules, the model parts related to it are automatically searched and updated;

[0067] S016 uses a deep autoencoder to detect anomalies in model data. The geometric features and physical quantity data under normal conditions are used as training sets to train the deep autoencoder. The deep autoencoder learns the feature representation of normal data through the encoding and decoding process. When new data is input, the reconstruction error is calculated. If the reconstruction error is greater than the set threshold, it is judged as abnormal data. Suppose the input data is x and the reconstructed data is Reconstruction error n is the data dimension. By analyzing the characteristics and location of abnormal data and combining methods such as fault tree analysis, the cause and impact range of potential problems can be diagnosed;

[0068] S017 uses long short-term memory network to predict the future state of engineering equipment or structure, providing a basis for predictive maintenance. Taking historical data as input, the LSTM model is trained to learn the time series characteristics of the data. Assuming the input sequence is x1, x2, ...x t , the LSTM unit passes through the forget gate f t , input gate i t , output gate o t and memory unit c t The collaborative work of calculating the hidden state h t and the predicted output According to the prediction results, a maintenance plan is made in advance, and the maintenance information is fed back to the twin model to optimize the model parameters and structure. The above scheme obtains comprehensive and rich data through various means, including the three-dimensional point cloud of the lidar, the texture of the camera, the physical quantity of the sensor, and text data, etc., which provides a sufficient information basis for building an accurate and comprehensive digital twin model, and can truly and comprehensively reflect the status and characteristics of the engineering entity. The preprocessing steps for different types of data effectively improve the quality and availability of the data. Removing outliers in the point cloud data improves the accuracy of the geometric model; processing image data enhances its quality, which is convenient for subsequent analysis and application; processing text data extracts key information and converts it into structured data, so that text information can be more effectively integrated with other data.

[0069] The Poisson reconstruction algorithm builds an initial geometric model and converts discrete point cloud data into a continuous triangular mesh model, which provides a good foundation for subsequent semantic labeling and fusion and can more accurately present the geometric shape of the object.

[0070] By combining semantic information to annotate the geometric model and exploring the potential relationship between the two, a deep fusion of semantics and geometry is achieved, so that the model not only has shape features, but also contains rich functional and meaningful information, which is more in line with the needs and understanding of actual engineering.

[0071] In the S01, the denoising autoencoder based on deep learning adopts a convolutional neural network structure, and its encoder part is composed of multiple convolutional layers and pooling layers, which are used to extract the deep feature representation of point cloud data; the decoder part is composed of deconvolution layers and upsampling layers, which restore the encoded feature vector to the denoised point cloud data. In the above scheme, the encoder part is composed of multiple convolutional layers and pooling layers, which can effectively extract the deep feature representation of point cloud data. The convolutional layer can automatically learn the local patterns and features in the data, while the pooling layer can reduce the dimension of the features, reduce the amount of calculation and extract the main features, thereby capturing the complex structure and hidden information in the point cloud data, and providing more accurate and valuable features for subsequent denoising processing. The denoising autoencoder using the convolutional neural network structure can effectively extract the deep features of point cloud data, improve the efficiency and accuracy of data processing, and provide data support for engineering management systems based on digital twins.

[0072] In the S01, in the feature-based point cloud enhancement algorithm, the Canny edge detection algorithm is used for edge detection, which extracts edge points in the point cloud data through steps such as smoothing the image by Gaussian filtering, calculating the gradient amplitude and direction, refining the edge by non-maximum suppression, and detecting and connecting the edges by double thresholds. The curvature calculation uses the principal component analysis (PCA) method. For each P, PCA analysis is performed in its neighborhood to calculate the principal curvatures K1 and K2. The curvature type of the point is determined according to the principal curvature to highlight different types of structural features. The Canny edge detection algorithm can effectively remove noise interference, and the image is smoothed by Gaussian filtering to provide a purer data basis for subsequent edge extraction, thereby more accurately extracting the real edge points in the point cloud data.

[0073] Calculating the gradient magnitude and direction helps to clearly define the position and direction of the edge, making the extracted edge information more accurate and complete.

[0074] The step of non-maximum suppression and edge refinement can make the extracted edges thinner and more accurate, reduce the width of the edges, and highlight the detailed features of the edges.

[0075] Dual threshold detection and edge connection can build a more complete and continuous edge structure by connecting weak edges while retaining strong edges, thus avoiding edge breakage and loss.

[0076] The use of principal component analysis (PCA) method to calculate curvature can deeply explore the local geometric characteristics of point cloud data.

[0077] For each point, PCA analysis is performed in its neighborhood, which can fully consider the local structural information of the point, and the calculated principal curvature can accurately reflect the curvature of the point.

[0078] Determining the curvature type of a point based on the principal curvature can highlight different types of structural features, such as plane, convex, concave, etc., which helps to more carefully analyze and understand the shape and structure of the object represented by the point cloud.

[0079] Furthermore, in S04, when constructing the three-dimensional space error field, for each point p in the point cloud data, its nearest neighbor point q is found in the digital twin model, and the Euclidean distance is calculated.

[0080] ,

[0081] In order to improve the efficiency of searching for nearest neighbor points, the KD tree data structure is used to organize the points in the digital twin model. The KD tree is a binary tree that divides the space into multiple subspaces by dividing the data points in different dimensions, making the time complexity of searching for the nearest neighbor points from o n Reduce to o logn, where n is the number of points in the digital twin model; after calculating the Euclidean distance, in order to more accurately evaluate the distribution of the error field, the error field is normalized, assuming that the maximum distance in the error field is d max , the minimum distance is d min , then the normalized error value

[0082]

[0083] Where d is the original Euclidean distance, and the normalized error value range is between [0,1]. The KD tree data structure is used to organize the points in the digital twin model to find the nearest neighbor points, which improves the computational efficiency. When processing large-scale digital twin models, the time complexity of finding the nearest neighbor points is reduced to o n Reduce to o logn , which reduces the calculation time, enables faster calculation results in practical applications, and enhances the real-time and responsiveness of the system.

[0084] The normalized error value range is between [0,1], which makes error fields of different scales and complexities have a unified measurement standard, which is convenient for comparison and analysis between different models or scenarios. The distribution of the error field can be evaluated more accurately, which helps to intuitively understand the concentration and change trend of the error.

[0085] The normalization process eliminates the influence of the dimension and numerical range in the original data, so that when analyzing the error field, we can focus more on the distribution pattern and characteristics of the error without being disturbed by the specific numerical value.

[0086] For subsequent data analysis and decision-making processes, the normalized error values ​​are easier to be processed and understood by algorithms and models, which improves the accuracy and reliability of data analysis.

[0087] In actual engineering projects, the data acquisition synchronization module is deployed first, and the Internet of Things devices such as various sensors are used to collect physical quantity data such as temperature, humidity, stress and strain at the construction site in real time. At the same time, the three-dimensional point cloud data of the engineering entity is obtained through the laser radar, and the color, material and other texture information of the object are collected by the camera, and the text data information such as engineering design documents, construction records, and operation and maintenance logs are extracted. These data will be transmitted to the digital twin model in real time.

[0088] Then, the twin model management module starts working to pre-process the different types of collected data. For point cloud data, a statistical filtering algorithm is used to remove outliers; for image data, grayscale, noise reduction and enhancement processing are performed; for text data, natural language processing technology is used to perform word segmentation, part-of-speech tagging and named entity recognition, extract key information and convert it into structured data.

[0089] Based on the preprocessed point cloud data, the Poisson reconstruction algorithm is used to build the initial geometric model. Combined with the semantic information extracted from the text data, the geometric model is semantically annotated, and the potential relationship between geometric features and semantic information is explored to achieve deep fusion of semantics and geometry.

[0090] In the data digitization module, the point cloud data obtained by 3D laser scanning is denoised using a deep learning-based denoising autoencoder. At the same time, a feature-based point cloud enhancement algorithm is used, and edge detection and curvature calculation algorithms are used to highlight the structural features in the point cloud data. For project-related materials, image scanning or text recognition is performed to convert them into digital formats. For design drawings, image vectorization technology is used to convert them into vector graphics. For text materials, optical character recognition technology is used to convert them into editable text. Then, a structured knowledge graph is constructed with the help of natural language processing technology, and the information in the project materials is associated with the digital twin model and point cloud data.

[0091] Next, the data analysis and decision-making module conducts in-depth mining and analysis of the data in the digital twin model. Through preset algorithms and models, it predicts project risks, evaluates project progress deviations, and generates corresponding decision-making recommendations.

[0092] Finally, the digital twin model is displayed through the three-dimensional visualization interface of the visualization interaction module. Users can view the project status, historical data and prediction results in real time through interactive operations.

[0093] When constructing the three-dimensional space error field, for each point p in the point cloud data, the KD tree data structure is used to quickly find its nearest neighbor point q in the digital twin model, calculate the Euclidean distance, and then normalize the error field to more accurately evaluate the distribution of the error field. Specific embodiment 2

[0095] like Figure 2 Shown

[0096] Collection and archiving stage

[0097] Data collection: At the beginning of the project, drones equipped with high-precision optical cameras and lidar were used for frequent low-altitude flight photography to obtain high-resolution images and point cloud data of the project area and its surroundings. Data collection was carried out at least once a week to capture changes in topography in a timely manner. Using satellite remote sensing technology, a large range of geological and hydrological information, including groundwater level changes and soil type distribution, was obtained regularly once a month to provide basic data for subsequent design and construction. The construction site was scanned in detail using a ground scanner, and a comprehensive scan was performed every two weeks to obtain high-precision topographic data with an accuracy of up to centimeters. Temperature, humidity, light intensity, sound, vibration and other sensors were deployed at various key locations on the construction site, such as tower cranes, construction equipment, and the main structure of the building. A total of more than 500 sensors were deployed, and the data was connected to the cloud in real time through IoT devices to ensure the real-time and accuracy of the data.

[0098] Establishing a database: After evaluation, the project selected MySQL as a relational database to store structured data, such as real-time data collected by sensors, construction progress data, personnel information, etc.; MongoDB was also used to store unstructured data, such as drawings, documents, and image data. By writing ETL (Extract, Transform, Load) scripts, all collected data was integrated into the central database, and data integration and cleaning were performed once a day to ensure data consistency and accuracy.

[0099] IoT deployment: Various sensors and high-definition cameras are installed at the construction site, covering all entrances and exits, main construction areas and material storage areas of the construction site, with a total of 30 cameras installed. RESTful API technology is used to connect all devices to a unified data management platform, and a unique API interface is assigned to each device to ensure the stability and security of data transmission.

[0100] Digital twin initialization: The development team creates a preliminary digital twin model in the Java Web application, and sets the initial parameters of the model based on the collected topographic data, project design planning and other information, including the geographic location, site boundaries, and preliminary design outlines of the building. Invite all parties involved in the project to participate in the confirmation of the model parameters to ensure that the initial state of the model conforms to the actual situation of the project.

[0101] Planning Phase

[0102] Land use analysis: Import the basic data obtained in the collection and archiving stage into the JavaWeb application, and use professional GIS tools to perform land use analysis. By analyzing factors such as the land's topography, traffic convenience, and surrounding supporting facilities, combined with the functional requirements of the project, evaluate the suitability of each area of ​​the land, and determine the best layout plan for functional areas such as shopping centers, office buildings, and hotels. During the analysis process, use the three-dimensional visualization function of GIS to display the effects of different layout plans in an intuitive way, and invite planning experts and project investors to discuss and make decisions.

[0103] Environmental impact assessment: With the help of simulation tools in Java Web applications, we simulate and analyze the noise, dust, wastewater and other pollutants that may be generated during the construction process. By establishing an environmental impact model and combining local meteorological data and terrain conditions, we can predict the spread and impact of pollutants. For example, through simulation, we found that the noise in a certain area during the construction process may have a greater impact on the surrounding residential areas. The project team adjusted the construction plan accordingly and arranged the construction time reasonably to avoid high-noise operations during residents' rest time.

[0104] Design plan discussion: Display the preliminary design plan in the Java Web application, including the building exterior design, interior space layout, infrastructure planning, etc. Use charts and tables to display the key indicators of the design plan, such as building area, volume ratio, greening rate, number of parking spaces, etc. Organize online and offline discussion meetings for the design team, construction party, investment party, etc. All parties can view and mark the design plan in real time in the application, and put forward modification opinions and suggestions. Based on the discussion results, the design team optimizes and improves the plan to form the final design plan.

[0105] Design review stage

[0106] Scheme refinement: The design team further refined the design scheme in the Java Web application, ensuring that every detail was fully considered from various professional perspectives such as building structure design, electrical system design, and water supply and drainage system design. The refined design scheme was imported into the digital twin model using BIM software to achieve three-dimensional visualization and digital simulation of the design scheme. Through the digital twin model, the project team can conduct virtual construction, discover problems in the design scheme in advance, such as pipeline collisions and unreasonable spatial layout, and make timely adjustments.

[0107] Compliance check: Develop a detailed compliance checklist in the Java Web application, covering requirements in multiple aspects such as building regulations, environmental regulations, and fire regulations. Use the form function in the application to check the design plans one by one and record the inspection results. For example, when checking the design of fire protection facilities, check the width of fire passages and the setting of evacuation signs according to regulatory requirements. For parts that do not meet regulatory requirements, mark them and clarify the rectification requirements and responsible persons to ensure that the design plan fully complies with relevant regulatory standards while meeting functional requirements.

[0108] Expert review: The final design plan is displayed in the Java Web application, and experts in multiple fields such as architecture, structure, environment, and fire protection are invited to conduct online reviews. Experts can view the detailed content of the design plan, the simulation effect of the digital twin model, and the compliance check results in the application, and put forward modification opinions through online comments and annotations. Based on the expert review opinions, the project team will make the final optimization and improvement of the design plan to ensure the scientificity and rationality of the design plan.

[0109] Construction Phase

[0110] Data collection and transmission: During the construction process, IoT devices are continuously used for real-time data collection, including the operating status of construction equipment, such as the lifting weight of the tower crane, the driving speed of the construction vehicle, etc., the environmental parameters of the construction site such as temperature, humidity, noise, etc., and the construction progress data such as the amount of concrete poured and the progress of wall masonry. The collected data is uploaded to the cloud in real time through the wireless network, and the data is updated every 5 minutes to ensure the timeliness of the data. The on-site data is updated in real time in the JavaWeb application, and the construction management personnel can view the latest situation of the construction site anytime and anywhere.

[0111] Construction progress monitoring: Develop a dedicated progress management module in the Java Web application, deeply integrate it with the digital twin model, set the time and progress targets for each construction node in the digital twin model according to the construction plan, and automatically update the display of the construction progress in the digital twin model through real-time collected data. For example, when the concrete pouring progress lags behind, the digital twin model will prompt the management personnel with eye-catching colors, and find out the reasons for the progress lag through data analysis, such as insufficient material supply and shortage of construction personnel. Management personnel will adjust the construction plan and allocate resources in a timely manner based on the analysis results to ensure that the construction tasks are completed on time.

[0112] Quality control: By analyzing the data collected by sensors, the construction process is simulated in the Java Web application to monitor the construction quality in real time. For example, the temperature sensor and humidity sensor data inside the concrete are used to simulate the solidification process of the concrete to determine whether the quality of the concrete meets the requirements. Once a quality problem is found, the location and type of the problem are marked in the Java Web application, and the quality rectification process is initiated to clarify the person responsible for rectification, the rectification deadline, and the rectification requirements. Through the digital twin model, the rectification progress is tracked to ensure that the quality problem is completely resolved.

[0113] Safety management: Use vibration sensors, tilt sensors and other equipment to monitor construction safety, and simulate possible safety hazards in Java Web applications, such as crane tilt and deformation of the main structure of the building. When the sensor detects abnormal data, the system automatically conducts risk assessment and issues early warnings for potential safety risks. For example, when the tilt angle of the crane exceeds the safety threshold, the system immediately issues an alarm and marks the location of the crane in the digital twin model, prompting managers to take emergency measures. At the same time, the accident response plan is recorded in the Java Web application, including personnel evacuation routes, rescue equipment deployment, etc., to ensure that a quick and effective response can be made in the event of a safety accident.

[0114] Operation and maintenance stage

[0115] Facility management: After the commercial complex is put into use, the IoT technology is used to continuously monitor the status of the building, including the operation status of the elevator, the energy consumption of the air conditioning system, the working status of the lighting system, etc. By installing smart sensors on the equipment, the data is transmitted to the Java Web application in real time for display. Remote monitoring and intelligent maintenance functions are realized. When the equipment fails, the system automatically issues an alarm and arranges maintenance personnel for maintenance according to the preset maintenance strategy. For example, when the elevator fails, the system automatically notifies the maintenance personnel and provides the elevator's fault code and location information. The maintenance personnel can view the relevant information through the mobile device, prepare maintenance tools and accessories in advance, improve maintenance efficiency, and reduce equipment downtime.

[0116] Energy consumption monitoring: Develop energy consumption monitoring modules in Java Web applications to track the energy consumption of buildings, including electricity, water, gas and other energy consumption. By analyzing energy consumption patterns, identify areas and equipment with high energy consumption, such as the lighting system of shopping malls and the air conditioning system of hotels. Use big data analysis technology to develop personalized energy optimization plans, such as adjusting the brightness and switching time of the lighting system, optimizing the operating parameters of the air conditioning system, etc.

[0117] Repair and maintenance: Based on the equipment operation data, use machine learning algorithms in Java Web applications to predict equipment failures. For example, by analyzing the number of elevator operations, operating time, fault history and other data, predict the time and type of possible elevator failures. According to the prediction results, arrange preventive maintenance plans and replace wearing parts in advance to avoid equipment failures. Record maintenance history in Java Web applications, including maintenance time, maintenance content, maintenance personnel and other information, to facilitate the query and management of equipment maintenance status and improve equipment reliability and service life.

[0118] Regulatory Phase

[0119] Compliance supervision: Establish a compliance supervision module in the Java Web application to record various activities during the project operation, such as regularly checking whether the fire protection facilities of the commercial complex are intact and effective, and whether the environmental protection measures meet the requirements. Mark the behaviors that do not comply with the regulations, start the rectification process, track the rectification progress, and ensure that the problems are solved in a timely manner. By connecting with the information systems of relevant regulatory departments, the real-time sharing of regulatory data is realized to improve regulatory efficiency.

[0120] Environmental monitoring: Continuously monitor environmental changes around the project, including air quality, water quality, noise and other indicators. Simulate environmental impacts in Java Web applications and analyze the long-term impact of building operations on the environment. For example, through monitoring, it was found that the heat emitted by the air conditioning system of the commercial complex has a certain impact on the surrounding temperature. The project team optimized the heat dissipation method of the air conditioning system to reduce the negative impact on the environment. According to the results of environmental monitoring, take corresponding environmental protection measures, such as increasing green area, installing air purification equipment, etc., to achieve harmonious coexistence between the project and the environment.

[0121] Social feedback: Collect public opinions by setting up suggestion boxes in commercial complexes and opening online feedback channels. The collected feedback results will be displayed in JavaWeb applications, including consumers' evaluation of the shopping environment and service quality of the commercial complex, and feedback from surrounding residents on project noise, traffic and other issues.

[0122] The above are only embodiments of the present invention. Common knowledge such as the known specific structures and characteristics in the scheme is not described in detail here. Ordinary technicians in the relevant field are aware of all the common technical knowledge in the technical field to which the invention belongs before the application date or priority date, can obtain all the existing technologies in the field, and have the ability to apply conventional experimental means before that date. Ordinary technicians in the relevant field can improve and implement this scheme in combination with their own abilities under the enlightenment given by this application. Some typical known structures or known methods should not become obstacles for ordinary technicians in the relevant field to implement this application. It should be pointed out that for those skilled in the art, several deformations and improvements can be made without departing from the structure of the present invention, which should also be regarded as the scope of protection of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. Engineering management system based on digital twin, including: Data acquisition and synchronization module: uses IoT devices and sensors to collect construction site data and transmits it to the twin model in real time; Digital twin model: Based on the collected data, three-dimensional modeling, simulation algorithms and machine learning technology are used to build a digital twin model that corresponds to the actual project. Twin model management module: according to the engineering design and construction plan, the digital twin model is decomposed into subtasks and distributed according to the construction team and stage; the task status in the digital twin model is updated according to the collected construction progress data; Record status changes at different stages to facilitate retrospective comparison. When the design changes, the model can be quickly modified and tasks redistributed. Data digitization module: convert paper drawings and documents in the design phase into digital text and image data, convert drawing images into editable vector graphics through drawing vectorization software, extract geometry and annotation information and store them in the design database; digitize the collected construction site data, convert analog signals into digital signals, store and manage them in a unified format, and perform standardized processing; establish a data association mechanism to associate design and construction data according to engineering objects and construction processes; The data analysis and decision-making module uses data mining and machine learning algorithms to compare the twin data generated by the digital twin model with the design data processed by the data digitization module, build a difference analysis model, and identify the differences between construction and design. For the comparison between point cloud data and digital twin models, a feature matching algorithm is used to extract the geometric features of point cloud and digital twin models for matching, which is used to calculate the deviation value of construction and design models. Based on the difference analysis results, combined with the engineering experience knowledge base and decision support model, decision suggestions are provided to engineering managers. The visualization interaction module uses 3D visualization and virtual reality (VR) / augmented reality (AR) technology to present the digital twin model to project managers. Users can view the real-time status of the project, construction progress and equipment operation status, and display the difference reports and decision-making recommendations generated by the data analysis and decision-making modules in visual charts and text. The digital twin model is displayed through a 3D visualization interface, and the project status, historical data, and prediction results can be viewed in real time through interactive operations; The specific method of data processing by the data digitization module is as follows: S01 uses block-based density clustering and direction vector analysis. It first divides the point cloud data into small blocks, clusters them by calculating the density of the points in each small block, and then determines the boundary and connection relationship based on the direction vector between the points, thereby extracting various structural features in the point cloud. At the same time, it uses a denoising autoencoder based on deep learning to denoise the point cloud data obtained by 3D laser scanning. At the same time, it uses a feature-based point cloud enhancement algorithm, edge detection and curvature calculation algorithms to highlight the structural features in the point cloud data. S02 For project-related materials, first perform image scanning or text recognition to convert them into digital formats. For design drawings, use image vectorization technology to convert scanned raster images into vector graphics. For text materials, use optical character recognition technology to convert them into editable text. Then use natural language processing technology to extract information and build a structured knowledge graph. Then, associate the information in the project materials with the digital twin model and point cloud data. S03 uses an algorithm based on normal vectors and curvature features to calculate the local features of each point in the point cloud data, finds the corresponding point pairs between the point cloud data and the digital twin model, uses a random sampling consistency algorithm to remove mismatched point pairs, and defines an energy function that includes the distance error and geometric constraints between the point cloud data and the digital twin model. The gradient descent method is used to solve the minimum value of the energy function to obtain a registration result. S04 compares the registered point cloud data with the digital twin model point by point in three-dimensional space, calculates the Euclidean distance between each corresponding point, and constructs a three-dimensional space error field; S05 slices the three-dimensional error field into multiple two-dimensional images, and then trains the CNN model to enable the CNN model to learn the characteristic patterns of the error field under different degrees of fit. The error field data of the current construction is classified through the trained model, and the degree of fit between the actual construction and the design requirements is output as "high fit", "medium fit" or "low fit".

2. The engineering management system based on digital twin according to claim 1 is characterized in that: The method for constructing a digital twin model by the twin model management module includes: S011 uses laser radar to obtain 3D point cloud data of engineering entities, including the geometric shape of objects; uses cameras to collect texture information, including color and material details; deploys sensors to collect temperature, humidity, stress and strain related physical quantity data, and monitors engineering environment and structural status; at the same time, extracts text data information from engineering design documents, construction records, and operation and maintenance logs; S012 performs corresponding preprocessing for different types of data. For point cloud data, a statistical filtering algorithm is used to remove outliers. Suppose each point P in the point cloud data set i The neighborhood point set is N i , calculate the neighborhood point to P i The distance d ij (j∈N i ), if d i If it is greater than the mean μ plus k times the standard deviation σ, then P i To remove outliers, grayscale, denoise and enhance image data to improve image quality. Natural language processing technology is used for text data to perform word segmentation, part-of-speech tagging and named entity recognition, extract key information and convert it into structured data. S013 uses the Poisson reconstruction algorithm to construct the initial geometric model based on the preprocessed point cloud data. The algorithm converts the discrete point cloud data into a continuous triangular mesh model by solving the Poisson equation. The normal vector field of the point cloud data is set to n, and a scalar function S is defined to satisfy the Poisson equation. The function S is obtained by iterative solution, and then a geometric model represented by a triangular mesh is generated; S014 combines the semantic information extracted from the text data to semantically annotate the geometric model. Different components in the building model are annotated according to the names and functions in the design documents. The Apriori algorithm is used to mine the potential relationship between geometric features and semantic information. Let I be the set of geometric features and semantic labels of all projects, and transaction T is a subset of I. By setting the support S and confidence C thresholds, association rules of the form X→Y (X, Y∈I) are mined to achieve deep fusion of semantics and geometry. S015 uses the Kalman filter algorithm to process the dynamic data collected by the sensor, predict the model status and update it; when a specific event occurs in the project, the event-driven update mechanism is triggered. Based on the knowledge graph technology, according to the event type and related knowledge rules, the model parts related to it are automatically searched and updated; S016 uses a deep autoencoder to detect anomalies in model data. The geometric features and physical quantity data under normal conditions are used as training sets to train the deep autoencoder. The deep autoencoder learns the feature representation of normal data through the encoding and decoding process. When new data is input, the reconstruction error is calculated. If the reconstruction error is greater than the set threshold, it is judged as abnormal data. Suppose the input data is x and the reconstructed data is Reconstruction error n is the data dimension. By analyzing the characteristics and location of abnormal data and combining methods such as fault tree analysis, the cause and impact range of potential problems can be diagnosed; S017 uses long short-term memory network to predict the future state of engineering equipment or structure, providing a basis for predictive maintenance. Taking historical data as input, the LSTM model is trained to learn the time series characteristics of the data. Assuming the input sequence is x1, x2, ...x t , the LSTM unit passes through the forget gate f t , input gate i t , output gate o t and memory unit c t The collaborative work of calculating the hidden state h t and the predicted output Based on the prediction results, a maintenance plan is made in advance, and the maintenance information is fed back to the twin model to optimize the model parameters and structure.

3. The engineering management system based on digital twin according to claim 1, characterized in that: In S01, the denoising autoencoder based on deep learning adopts a convolutional neural network structure, and its encoder part is composed of multiple convolutional layers and pooling layers, which is used to extract deep feature representation of point cloud data; The decoder part consists of a deconvolution layer and an upsampling layer, which restores the encoded feature vector to denoised point cloud data.

4. The engineering management system based on digital twin according to claim 1, characterized in that: In the S01, in the feature-based point cloud enhancement algorithm, the Canny edge detection algorithm is used for edge detection, which extracts edge points in the point cloud data through steps such as smoothing the image by Gaussian filtering, calculating the gradient amplitude and direction, refining the edge by non-maximum suppression, and detecting and connecting the edges by double thresholds. The curvature calculation uses the principal component analysis (PCA) method. For each P, a PCA analysis is performed in its neighborhood to calculate the principal curvatures K1 and K2. The curvature type of the point is determined based on the principal curvature to highlight different types of structural features.

5. The engineering management system based on digital twin according to claim 1 is characterized in that In S04, when constructing the three-dimensional space error field, for each point p in the point cloud data, find its nearest neighbor point q in the digital twin model and calculate the Euclidean distance ， In order to improve the efficiency of searching for nearest neighbor points, the KD tree data structure is used to organize the points in the digital twin model. The KD tree is a binary tree that divides the space into multiple subspaces by dividing the data points in different dimensions, making the time complexity of searching for the nearest neighbor points from o n Reduce to o logn , where n is the number of points in the digital twin model; after calculating the Euclidean distance, in order to more accurately evaluate the distribution of the error field, the error field is normalized, assuming that the maximum distance in the error field is d max , the minimum distance is d min , then the normalized error value ， Where d is the original Euclidean distance, and the normalized error value ranges between [0,1].

Images