Digital intelligent construction and on-site integrated finished product delivery system

Abstract

The invention relates to the technical field of construction delivery, in particular to a digital intelligent construction and on-site integrated finished product delivery system, which realizes acquisition and storage of test block quality, material consumption and key factory information in a concrete production process through a raw material quality management module, and ensures standardization and traceability of production link data. And the cast-in-place management module performs real-time monitoring, analysis and anomaly prediction on concrete state parameters, environmental data and construction behaviors by using deep learning algorithms such as a convolutional neural network and a long-short-term memory network, generates a casting management data set, and provides objective data support for subsequent quality tracing and responsibility division. On the basis of a three-dimensional modeling technology, data of the whole process of concrete construction are mapped into a unified three-dimensional model, high-precision full-life-cycle quality tracing and monitoring capacity is provided for construction parties and owners, and construction delivery transparency is greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of construction delivery technology, and in particular to a digital construction and on-site integrated finished product delivery system. Background Art

[0002] With the rapid development of the construction industry, construction quality, efficiency and safety have become core issues of concern to the industry. The traditional manual recording mode has obvious limitations when facing concrete quality problems or substandard construction quality. Since key data in the construction process often rely on manual recording or memory, the accuracy and integrity of the data cannot be guaranteed. When quality problems occur, it is impossible to trace back to the specific construction process, operating steps or material parameters, which makes it difficult to troubleshoot and identify responsibilities, increasing repair costs and time delays.

[0003] As a semi-finished product delivery model, concrete faces the challenge of not being able to meet modern construction needs. In the traditional model, the production, transportation, pouring and molding of concrete are relatively independent, and each link lacks effective connection and data sharing. This separate management model makes it difficult to achieve full-process quality traceability from the source of materials to the final product, is opaque, and easily leads to a disconnect between production and construction. Summary of the invention

[0004] In order to solve the above problems, the present invention provides a digital construction and on-site integrated finished product delivery system, which collects the status parameters of the concrete life cycle and combines deep learning for traceability and judgment analysis, and improves the delivery transparency of concrete construction through an associated visualization model.

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

[0006] A digital construction and on-site integrated finished product delivery system, including a raw material quality management module, an on-site pouring management module, and a data visualization delivery module;

[0007] The raw material quality management module is used to obtain and record the quality of test blocks, material consumption and factory information during the concrete production process, and the factory information includes factory temperature and factory slump;

[0008] The on-site pouring management module is used to perform the following steps:

[0009] Acquire the positioning information of the concrete mixer truck, the pump truck and the construction target part data, extract the features of the project positioning information and the construction target part data through the convolutional neural network, match the construction target part of the concrete mixer truck and the pump truck docking point, and generate the pouring target position data;

[0010] Collect the status data of on-site concrete pouring and use the long short-term memory network to perform time series modeling, predict quality problems during the pouring process and generate pouring anomaly scores;

[0011] Use the DeepLab segmentation algorithm to perform regional segmentation and abnormal behavior recognition on real-time construction site images and generate behavior recognition reports;

[0012] Integrate pouring target location data, pouring anomaly scores, and behavior recognition reports to generate a field pouring management data set;

[0013] The data visualization delivery module is used to perform three-dimensional modeling of the construction site through construction site images to obtain a three-dimensional model of the construction site, and to match and associate the test block quality, material consumption, factory information and pouring management data sets in the concrete production process in the three-dimensional model of the construction site to generate a delivery package.

[0014] Furthermore, the concrete mixer truck positioning information, the pump truck positioning information and the construction target location data are all acquired through GPS.

[0015] Furthermore, matching the target construction position of the concrete mixer truck and the docking point of the pump truck comprises the following steps:

[0016] Construct a matrix, standardize the concrete mixer truck positioning information, pump truck positioning information and construction target location data and map them into the matrix;

[0017] Based on the convolutional neural network, the matrix is ​​feature extracted to extract the spatial correlation features between the positioning information and the construction target part data, and a feature map is generated;

[0018] Based on the extracted feature graph, the spatial matching degree between the mixer truck and pump truck and the target construction position is calculated using the Euclidean distance to generate the pouring target position data.

[0019] Furthermore, the collecting of state data of concrete pouring on site and the use of long short-term memory network for time series modeling, predicting quality problems in the pouring process and generating pouring anomaly scores include the following steps:

[0020] Real-time collection of on-site concrete pouring status data, including pump temperature, mold temperature and slump as well as on-site environmental data;

[0021] The state data of cast-in-place concrete is modeled in multi-dimensional time series through long short-term memory network, and the concrete state data prediction set is output;

[0022] The pouring anomaly score is generated by weighting the deviation value in the concrete state data prediction set.

[0023] Furthermore, the formula of the long short-term memory network is as follows:

[0024] h t =σ(W·[h t-1 ,x t ]+b);

[0025] Among them, h t is the hidden state of the current time step; h t-1 is the hidden state of the previous time step; x t is the input data of the current time step; W is the network weight matrix; b is the bias term; σ is the activation function.

[0026] Furthermore, the on-site environmental data is acquired in real time by setting monitoring and collecting devices for temperature, humidity, light, wind speed and rainfall in the paving and leveling robots and the finishing robots at the construction site.

[0027] Furthermore, the use of the DeepLab segmentation algorithm to perform region segmentation and abnormal behavior recognition on real-time construction site images includes the following steps:

[0028] Perform noise filtering, size normalization and image enhancement on real-time construction site images;

[0029] The enhanced construction site image is divided into several pixel block areas, and the semantic segmentation training of the construction scene image is performed using the DeepLab segmentation algorithm, and the output is the segmentation mask of the construction area;

[0030] Based on the comparison between the segmentation mask and the construction planning model, abnormal behaviors in the construction process are identified and a behavior recognition report is generated.

[0031] Furthermore, in the DeepLab segmentation algorithm, the formula of the loss function is as follows:

[0032]

[0033] Among them, L is the total loss function; N is the total number of pixels; C is the number of categories; is the true category label of pixel i; is the predicted probability of pixel i; λ is the regularization coefficient; w k is the network weight; K is the total number of all parameters in the network.

[0034] Furthermore, the three-dimensional modeling of the construction site through the construction site image includes the following steps:

[0035] The camera equipment at the construction site collects image data from several angles, and uses the SIFT algorithm to extract and match the feature points in the image data from several angles. Based on the matching results, the preliminary three-dimensional point cloud data of the construction site is generated by the structure from M algorithm;

[0036] The generated 3D point cloud data is denoised and texture mapped to generate a 3D model of the construction site.

[0037] The beneficial effects of the present invention are as follows: the present invention collects and stores the quality of test blocks, material consumption and key factory information (including factory temperature and slump) in the concrete production process through the raw material quality management module, so as to realize the data standardization and traceability management of the production link; the on-site pouring management module introduces the convolutional neural network and the long short-term memory network deep learning algorithm to monitor, analyze and predict the concrete state parameters, environmental data and construction behavior at the construction site in real time, generate a data-driven pouring management data set, and provide detailed and objective data support for the subsequent quality tracing and responsibility division. Secondly, in view of the strong independence of each link and the lack of effective connection of data in the delivery mode of concrete as a semi-finished product, the present invention realizes the integration and dynamic association of the whole process data through the data visualization delivery module. The module is based on the three-dimensional modeling technology of the construction site image, and maps the data of the whole process of concrete construction into a unified three-dimensional model. The model not only covers the core indicators such as test block quality, material consumption, and construction status data, but also realizes the association display with the real-time on-site environment and construction behavior, providing the construction party and the owner with accurate quality tracing and monitoring capabilities throughout the life cycle, and significantly improving the transparency of concrete construction delivery. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a structural schematic diagram of the digital construction and on-site integrated finished product delivery system in the present invention.

[0039] Figure 2 It is a flow chart of the execution steps of the on-site pouring management module in the present invention. DETAILED DESCRIPTION

[0040] See also Figure 1-2 As shown, the present invention relates to a digital intelligent construction and on-site integrated finished product delivery system, including a raw material quality management module, an on-site pouring management module and a data visualization delivery module;

[0041] The raw material quality management module is used to obtain and record the quality of test blocks, material consumption and factory information during the concrete production process, and the factory information includes factory temperature and factory slump;

[0042] The on-site pouring management module is used to perform the following steps:

[0043] Acquire the positioning information of the concrete mixer truck, the pump truck and the construction target part data, extract the features of the project positioning information and the construction target part data through the convolutional neural network, match the construction target part of the concrete mixer truck and the pump truck docking point, and generate the pouring target position data;

[0044] Collect the status data of on-site concrete pouring and use the long short-term memory network to perform time series modeling, predict quality problems during the pouring process and generate pouring anomaly scores;

[0045] Use the DeepLab segmentation algorithm to perform regional segmentation and abnormal behavior recognition on real-time construction site images and generate behavior recognition reports;

[0046] Integrate pouring target location data, pouring anomaly scores, and behavior recognition reports to generate a field pouring management data set;

[0047] The data visualization delivery module is used to perform three-dimensional modeling of the construction site through construction site images to obtain a three-dimensional model of the construction site, and to match and associate the test block quality, material consumption, factory information and pouring management data sets in the concrete production process in the three-dimensional model of the construction site to generate a delivery package.

[0048] In some embodiments, in the raw material quality management module, the system collects and records key parameters in the concrete production process through embedded sensors and online monitoring equipment. For example, in the production stage, the system collects the compressive strength of the test block and realizes automatic docking with the quality inspection equipment through the built-in pressure sensor; for material consumption, the system records the proportion and dosage data of cement, sand, gravel and admixtures in real time, and dynamically compares it with the preset formula to ensure the consistency of concrete performance. At the same time, factory information including factory temperature and slump parameters are obtained in real time by sensors installed at the end of the production line, and transmitted to the central database through a wireless network for subsequent analysis and traceability. In the on-site pouring management module, the system first uses the device embedded with the GPS module to obtain the positioning information of the concrete mixer and pump truck, and combines the digital design model of the construction target part to extract the features of the target part through the convolutional neural network. The multi-layer structure of the convolutional neural network can accurately identify the spatial relationship of the construction area, such as the relative position of the mixer truck and the pump truck and the coordinates of the construction site. By optimizing the matching algorithm, the optimal pouring path and target position data are generated to effectively reduce construction errors. During the on-site pouring process, the system uses the long short-term memory network (LSTM) to model the concrete state data in time series. The real-time state data of concrete, including pump temperature, mold temperature and slump, are collected by sensors and uploaded to the system. The LSTM model dynamically analyzes these time series data to predict key parameters such as concrete setting time and strength development trend. When the model detects possible anomalies, such as slump exceeding the set range, the system automatically generates a pouring anomaly score and triggers an alarm to guide construction personnel to adjust construction parameters in time. At the same time, the system uses the DeepLab segmentation algorithm to process the image data of the construction site. The real-time image data collected by the high-resolution camera is input into the DeepLab model for semantic segmentation after noise filtering and image enhancement. The segmentation results can accurately identify the construction area, non-construction area and obstacles that may affect the construction. During the recognition process, the system will also generate a behavior recognition report based on the segmentation results, such as locating the deviation area of ​​the construction machinery operation or the missed construction site, and integrate the report content with the pouring target location data and anomaly score to generate a complete on-site pouring management data set. In the data visualization delivery module, the system inputs the image data of the construction site and the collected multi-dimensional construction parameters into the 3D modeling platform to generate a high-precision 3D model of the construction site. By associating with the test block quality, material consumption and factory information in the concrete production process, the system can dynamically display the data of the entire construction process in the model. For example, users can view the specific construction conditions of a certain location through an interactive interface, including the concrete batch used at that location, state parameters and related construction behavior reports. Finally, the delivery package generated by the system is output in a standardized data format for the construction party and the owner to use for quality traceability and analysis.

[0049] Furthermore, the concrete mixer truck positioning information, the pump truck positioning information and the construction target location data are all acquired through GPS.

[0050] Specifically, in the process of obtaining the positioning information of the concrete mixer truck, the pump truck and the construction target part data, the system relies on GPS equipment to achieve high-precision positioning data collection. The concrete mixer truck and the pump truck are equipped with GPS modules respectively, which are used to record the vehicle's geographic coordinates, driving path and relative position to the target construction area in real time. The system uploads the collected positioning information to the central server through wireless transmission, and matches and associates it with the digital design data of the construction target part. The construction target part data is generated by digital construction drawings and standardized through the spatial coordinate system to ensure seamless connection with the vehicle positioning data. The system uses a matching algorithm to compare the real-time position of the vehicle with the construction target coordinates, and dynamically updates the distance, angle and optimal movement path between the vehicle and the target position. For example, when the concrete mixer truck approaches the construction site, the system generates the optimal path to reach the construction target based on its real-time position, and guides the pump truck to accurately dock with the preset construction point in the construction area. The whole process not only ensures the efficient scheduling of vehicles in complex construction scenes, but also reduces construction delays caused by positioning errors, and improves the overall construction accuracy and efficiency. This solution further optimizes the association process between vehicle positioning and construction targets, ensuring standardized and intelligent management of the construction process.

[0051] Furthermore, matching the target construction position of the concrete mixer truck and the docking point of the pump truck comprises the following steps:

[0052] Construct a matrix, standardize the concrete mixer truck positioning information, pump truck positioning information and construction target location data and map them into the matrix;

[0053] Based on the convolutional neural network, the matrix is ​​feature extracted to extract the spatial correlation features between the positioning information and the construction target part data, and a feature map is generated;

[0054] Based on the extracted feature graph, the spatial matching degree between the mixer truck and pump truck and the target construction position is calculated using the Euclidean distance to generate the pouring target position data.

[0055] Specifically, first, the system obtains the real-time positioning information of the concrete mixer truck and the pump truck, as well as the spatial coordinate data of the construction target part through the embedded GPS module. After preprocessing, these data are input into the matrix construction module in a standardized form. The standardization step includes converting the positioning information into a unified coordinate format and aligning the design parameters of the construction target with its spatial coordinates to form a multidimensional matrix. The rows of the matrix represent different data types (such as vehicle position and construction target position), and the columns represent specific spatial coordinate values ​​(such as X, Y, and Z coordinates). After the matrix is ​​constructed, the system calls the convolutional neural network (CNN) model to extract features from the matrix. The convolution layer of the CNN model gradually extracts features in the matrix through a sliding window mechanism, such as the distance feature, relative direction feature, and local spatial relationship between the vehicle position and the construction target. The extracted features are processed layer by layer, and finally a high-dimensional feature map is generated, which represents the spatial correlation between the concrete mixer truck, the pump truck, and the construction target part in an encoded form. Next, based on the generated feature map, the system uses the Euclidean distance to calculate the spatial matching degree between the vehicle and the target construction position. The system calculates the Euclidean distance between the vehicle and the target point by point on the feature map, and uses the result as the basis for the matching score. The higher the matching degree, the more the spatial position relationship between the vehicle and the construction target meets the preset construction requirements. Finally, the system generates pouring target position data based on the calculation results, including the optimal path from the mixer truck to the construction location, the location of the pump truck docking point, and the dynamic relationship between the vehicle and the construction target. Taking the construction of industrial plant flooring as an example, when the mixer truck approaches the site, the system will plan the shortest path to the target for it based on real-time positioning information, and guide the pump truck to accurately dock at the designated pouring point, thereby ensuring construction efficiency and quality.

[0056] Furthermore, the collecting of state data of concrete pouring on site and the use of long short-term memory network for time series modeling, predicting quality problems in the pouring process and generating pouring anomaly scores include the following steps:

[0057] Real-time collection of on-site concrete pouring status data, including pump temperature, mold temperature and slump as well as on-site environmental data;

[0058] The state data of cast-in-place concrete is modeled in multi-dimensional time series through long short-term memory network, and the concrete state data prediction set is output;

[0059] The pouring anomaly score is generated by weighting the deviation value in the concrete state data prediction set.

[0060] Specifically, first, the system collects multi-dimensional state data of concrete in real time through the sensor network deployed at the construction site. The collected data includes key process parameters such as concrete pump temperature, mold temperature, slump, and environmental data of the construction site (such as temperature, humidity, wind speed, etc.). For example, the concrete slump data is monitored in real time by a slump tester; the pump and mold temperatures are recorded by embedded thermal sensors with an accuracy of 0.1°C. These data are uploaded to the system central database via wireless transmission to form a time series data set. Then, the long short-term memory network (LSTM) is used to model the collected multi-dimensional time series data. The core of the LSTM model lies in its gating mechanism, which can capture the short-term fluctuations and long-term trends of concrete state over time, thereby predicting future changes in key parameters. For example, the system predicts whether there is a risk of slump deviating from the process requirements in the future pouring stage by analyzing the time series of slump. At the same time, the LSTM model will comprehensively consider environmental data (such as the influence of on-site temperature and humidity on concrete rheology) to further improve the accuracy of the prediction. The model output is a set of concrete state data prediction sets, including the state parameter values ​​and their changing trends at each future time point. Finally, the system calculates the pouring anomaly score based on the deviation value in the concrete state data prediction set. The deviation value is quantified by the difference between the predicted data and the target parameter range, and a comprehensive anomaly score is generated through a weighted method. The weighting coefficient is set according to the degree of influence of different state parameters on the pouring quality. For example, slump deviation has a higher weight because it directly affects the workability and molding quality of concrete. The higher the anomaly score, the greater the risk of the pouring process deviating from the design process requirements.

[0061] Furthermore, the formula of the long short-term memory network is as follows:

[0062] h t =σ(W·[h t-1 , x t ]+b);

[0063] Among them, h t is the hidden state of the current time step; h t-1 is the hidden state of the previous time step; x t is the input data of the current time step; W is the network weight matrix; b is the bias term; σ is the activation function.

[0064] It should be noted that the system first collects the state data and environmental data of concrete in real time through a variety of sensors deployed at the construction site. For example, the pump temperature and mold temperature are collected by thermal sensors, and the slump data is recorded at second intervals by dedicated monitoring equipment. After these data are uploaded to the central database, they are preprocessed (including data cleaning and standardization) and input into the LSTM network as a time series. The LSTM network uses the weight matrix and activation function in the formula to calculate the data of the current time step and the historical time step, and outputs a hidden state, which integrates the information of the current and historical data. For example, when the slump of concrete shows a gradually decreasing trend in the first few time steps, the LSTM can capture this trend through the memory gate mechanism and combine it with the current ambient temperature to predict the slump value of the next time step. Subsequently, the system compares the time series prediction set output by the LSTM with the target state range and calculates the deviation value. According to the size of the deviation value and the importance of different parameters, the system generates a pouring anomaly score through a weighted algorithm. The weight is set according to the degree of influence of each parameter on the pouring quality. For example, the weight of slump is higher, while the weight of mold temperature is second. When the abnormal score exceeds the preset threshold, the system triggers an early warning to guide construction personnel to adjust operations in a timely manner, such as optimizing the concrete mix or changing construction parameters.

[0065] Furthermore, the on-site environmental data is acquired in real time by setting monitoring and collecting devices for temperature, humidity, light, wind speed and rainfall in the paving and leveling robots and the finishing robots at the construction site.

[0066] It should be noted that, specifically, the paving and leveling robot and the trowel robot are equipped with the following environmental monitoring equipment:

[0067] Temperature sensor: A high-precision thermocouple sensor is used to monitor the ambient temperature of the construction site in real time with an accuracy of up to 0.1°C. This data is used to evaluate the impact of the initial setting time of concrete, thereby guiding the adjustment of the pouring rhythm.

[0068] Humidity sensor: Records air humidity through a highly sensitive humidity collection device. Humidity data has an important impact on the hydration reaction and final density of the concrete surface during the finishing stage.

[0069] Light sensor: Use a photoresistor light intensity measurement module to monitor the light intensity in the construction area in real time. Light data is used to evaluate the potential impact of the photothermal effect on the concrete surface temperature at the construction site.

[0070] Wind speed sensor: Integrated ultrasonic anemometer records wind speed and direction at high frequency. This data is used to analyze the effect of wind on the evaporation rate of water on the concrete surface, thereby avoiding early cracking problems caused by excessive water loss.

[0071] Rainfall sensor: A tipping bucket rain gauge is used to record rainfall changes in real time on site, ensuring that the construction plan can be adjusted dynamically to avoid concrete construction quality problems caused by rainfall.

[0072] The above monitoring devices process the collected environmental data in real time through the embedded data processing module on the robot and transmit it to the central database of the construction management system through the wireless network. These data are not only used for real-time monitoring of the on-site pouring process, but also passed as input variables to subsequent prediction and analysis models, such as the long short-term memory network (LSTM) and the abnormal score calculation module.

[0073] Furthermore, the use of the DeepLab segmentation algorithm to perform region segmentation and abnormal behavior recognition on real-time construction site images includes the following steps:

[0074] Perform noise filtering, size normalization and image enhancement on real-time construction site images;

[0075] The enhanced construction site image is divided into several pixel block areas, and the semantic segmentation training of the construction scene image is performed using the DeepLab segmentation algorithm, and the output is the segmentation mask of the construction area;

[0076] Based on the comparison between the segmentation mask and the construction planning model, abnormal behaviors in the construction process are identified and a behavior recognition report is generated.

[0077] In some embodiments, first, the system collects real-time image data through a high-resolution camera installed at the construction site. Due to the complex on-site environment, the image data may be affected by noise interference or changes in illumination, so it needs to be preprocessed before entering the segmentation model. The image preprocessing steps include the following: Noise filtering: Use the Gaussian filtering algorithm to reduce the noise of the image, reduce the interference of external factors such as dust and light spots on the image quality, and ensure the accuracy of the segmentation results. Size normalization: The image is uniformly adjusted to a preset fixed size (for example, 1024×1024 pixels) to ensure that the image input to the DeepLab segmentation model has a consistent scale to avoid the degradation of model performance due to resolution differences. Image enhancement: Enhance the image quality through contrast stretching, histogram equalization and other technologies, so that the edges of the construction area are clearer, thereby improving the segmentation accuracy. Next, the preprocessed image is divided into several pixel block areas, which are passed to the DeepLab segmentation algorithm as input data. The DeepLab segmentation algorithm uses convolutional neural networks (CNN) and atrous convolution (Atrous Convolution) technology to extract semantic features in the image layer by layer. After training, the segmentation model can accurately identify different areas of the construction site, such as paving and leveling areas, troweling areas, unconstructed areas, and obstacle areas. Finally, the model generates a segmentation mask of the construction area, in which each pixel is assigned to the corresponding semantic category. Based on the segmentation mask, the system compares the real-time construction image with the preset construction planning model. The construction planning model contains information such as the expected construction area, equipment path, and process sequence. Through comparative analysis, the system can identify abnormal behaviors in the actual construction process, such as deviations from the trajectory of mechanical equipment, uneven paving, missing construction areas, or the presence of obstacles. For example, in a large-scale floor construction, the segmentation mask showed that the edge of the trowel area was not fully covered. The system confirmed the anomaly by comparing it with the planning model and generated a behavior recognition report. The behavior recognition report includes the specific type, location, and scope of the abnormal behavior, and is pushed to the construction team in real time through the construction management platform. For example, the report can point out that the equipment has deviated in a specific area.

[0078] Furthermore, in the DeepLab segmentation algorithm, the formula of the loss function is as follows:

[0079]

[0080] Among them, L is the total loss function; N is the total number of pixels; C is the number of categories; is the true category label of pixel i; is the predicted probability of pixel i; λ is the regularization coefficient; w k is the network weight; K is the total number of all parameters in the network.

[0081] It should be noted that in the specific implementation process, the first part of the loss function is the cross entropy loss, which is used to measure the error between the predicted category and the true category of each pixel. The overall classification error is obtained by summing up the cross entropy losses of all pixels and taking the average. For example, for a construction image, if the system predicts that a certain area is an unconstructed area, but it is actually a smoothed area, the cross entropy loss will increase significantly, guiding the model to adjust the weights in the next iteration to improve the segmentation accuracy. The second part of the loss function is a regularization term used to constrain the network weight w k The size of the model is reduced to prevent the model from overfitting in complex construction scenes. By introducing the regularization term, the model can maintain a reasonable range of weight values ​​while optimizing the segmentation accuracy, thereby improving the generalization ability of the model.

[0082] Furthermore, the three-dimensional modeling of the construction site through the construction site image includes the following steps:

[0083] The camera equipment at the construction site collects image data from several angles, and uses the SIFT algorithm to extract and match the feature points in the image data from several angles. Based on the matching results, the preliminary three-dimensional point cloud data of the construction site is generated by the structure from M algorithm;

[0084] The generated 3D point cloud data is denoised and texture mapped to generate a 3D model of the construction site.

[0085] Specifically, first, the system collects image data of the construction area through multi-angle cameras deployed at the construction site. The camera equipment shoots at preset time intervals and fixed angles to ensure that the collected images cover all viewing angles of the construction area, such as recording construction details such as concrete paving and troweling from the top and side. These image data are transmitted to the central processing unit via a wireless network as the basic input for subsequent 3D modeling. Then, the system uses the scale-invariant feature transform (SIFT) algorithm to extract and match feature points in the image data. The SIFT algorithm generates feature descriptors by detecting key points with high information density (such as edges and corners) in the image, and performs feature matching between images at different angles. For example, in a set of images, the SIFT algorithm can accurately locate the edge position of the troweling equipment or the texture details of the concrete surface. After the matching is completed, the system generates a set of data sets containing feature point coordinates and matching relationships, providing basic information of the spatial structure. Subsequently, based on the feature point matching results, the system uses the structure from motion (SfM) algorithm to generate preliminary 3D point cloud data. The SfM algorithm reconstructs the 3D spatial structure of the construction site by analyzing the position and posture of the camera and combining the matching relationship of the feature points. For example, in the construction of industrial plant floors, the SfM algorithm can generate a 3D point cloud model of the equipment and the ground based on the position information of the trowel equipment in different images. The preliminary point cloud data contains the spatial coordinates of the construction area and the dense point cloud distribution. After the preliminary point cloud is generated, the system optimizes it, including denoising and texture mapping. Denoising retains key 3D structural information by filtering out isolated points and noise data in non-construction areas; texture mapping generates a realistic construction scene model by mapping the color and texture information of the original image to the surface of the point cloud model. For example, the smoothness of the construction ground and the specific location of the mechanical equipment are clearly visible after texture mapping, providing intuitive support for subsequent data analysis and display. Finally, the optimized 3D model of the construction site is generated and uploaded to the data visualization delivery platform. Users can view the specific details of the 3D model through an interactive interface, including the spatial distribution of the construction area, the location of mechanical equipment, and the dynamics of the construction progress.

[0086] 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 digital intelligent construction and on-site integrated finished product delivery system, characterized in that: Including raw material quality management module, on-site pouring management module and data visualization delivery module; The raw material quality management module is used to obtain and record the quality of test blocks, material consumption and factory information during the concrete production process, and the factory information includes factory temperature and factory slump; The on-site pouring management module is used to perform the following steps: Acquire the positioning information of the concrete mixer truck, the pump truck and the construction target part data, extract the features of the project positioning information and the construction target part data through the convolutional neural network, match the construction target part of the concrete mixer truck and the pump truck docking point, and generate the pouring target position data; Collect the status data of on-site concrete pouring and use the long short-term memory network to perform time series modeling, predict quality problems during the pouring process and generate pouring anomaly scores; Use the DeepLab segmentation algorithm to perform regional segmentation and abnormal behavior recognition on real-time construction site images and generate behavior recognition reports; Integrate pouring target location data, pouring anomaly scores, and behavior recognition reports to generate a field pouring management data set; The data visualization delivery module is used to perform three-dimensional modeling of the construction site through construction site images to obtain a three-dimensional model of the construction site, and to match and associate the test block quality, material consumption, factory information and pouring management data sets in the concrete production process in the three-dimensional model of the construction site to generate a delivery package.

2. According to claim 1, a digital intelligent construction and on-site integrated finished product delivery system is characterized in that: The concrete mixer truck positioning information, pump truck positioning information and construction target location data are all acquired through GPS.

3. According to claim 1, a digital intelligent construction and on-site integrated finished product delivery system is characterized in that: The matching of the target construction position of the concrete mixer truck and the docking point of the pump truck comprises the following steps: Construct a matrix, standardize the concrete mixer truck positioning information, pump truck positioning information and construction target location data and map them into the matrix; Based on the convolutional neural network, the matrix is ​​feature extracted to extract the spatial correlation features between the positioning information and the construction target part data, and a feature map is generated; Based on the extracted feature graph, the spatial matching degree between the mixer truck and pump truck and the target construction position is calculated using the Euclidean distance to generate the pouring target position data.

4. According to claim 1, a digital intelligent construction and on-site integrated finished product delivery system is characterized in that: The method of collecting the state data of the concrete poured on site and performing time series modeling with a long short-term memory network, predicting the quality problems in the pouring process and generating a pouring anomaly score comprises the following steps: Real-time collection of on-site concrete pouring status data, including pump temperature, mold temperature and slump as well as on-site environmental data; The state data of cast-in-place concrete is modeled in multi-dimensional time series through long short-term memory network, and the concrete state data prediction set is output; The pouring anomaly score is generated by weighting the deviation value in the concrete state data prediction set.

5. According to claim 4, a digital intelligent construction and on-site integrated finished product delivery system is characterized in that: The formula of the long short-term memory network is as follows: h t =σ(W·[h t-1 ,x t ]+b); Among them, h t is the hidden state of the current time step; h t-1 is the hidden state of the previous time step; x t is the input data of the current time step; W is the network weight matrix; b is the bias term; σ is the activation function.

6. According to claim 3, a digital intelligent construction and on-site integrated finished product delivery system is characterized in that: The on-site environmental data is acquired in real time by setting monitoring and collecting devices for temperature, humidity, light, wind speed and rainfall in the paving and leveling robots and the trowel robots at the construction site.

7. According to claim 1, a digital intelligent construction and on-site integrated finished product delivery system is characterized in that: The use of the DeepLab segmentation algorithm to perform region segmentation and abnormal behavior recognition on real-time construction site images includes the following steps: Perform noise filtering, size normalization and image enhancement on real-time construction site images; The enhanced construction site image is divided into several pixel block areas, and the semantic segmentation training of the construction scene image is performed using the DeepLab segmentation algorithm, and the output is the segmentation mask of the construction area; Based on the comparison between the segmentation mask and the construction planning model, abnormal behaviors in the construction process are identified and a behavior recognition report is generated.

8. The digital intelligent construction and on-site integrated finished product delivery system according to claim 7 is characterized in that: In the DeepLab segmentation algorithm, the formula of the loss function is as follows: Among them, L is the total loss function; N is the total number of pixels; C is the number of categories; is the true category label of pixel i; is the predicted probability of pixel i; λ is the regularization coefficient; w k is the network weight; K is the total number of all parameters in the network.

9. According to claim 1, a digital intelligent construction and on-site integrated finished product delivery system is characterized in that: The three-dimensional modeling of the construction site by using the construction site image comprises the following steps: The camera equipment at the construction site collects image data from several angles, and uses the SIFT algorithm to extract and match the feature points in the image data from several angles. Based on the matching results, the preliminary three-dimensional point cloud data of the construction site is generated by the structure from M algorithm; The generated 3D point cloud data is denoised and texture mapped to generate a 3D model of the construction site.

Images