Method, medium and system for detecting flatness in concrete solidification process

Abstract

The invention provides a concrete solidification process flatness detection method, medium and system, and belongs to the technical field of concrete construction, and the method comprises the steps: continuously obtaining a laser point height matrix collected by a flatness laser collection device, the collection frequency being once every 5-20 minutes; the laser point height matrix is used as the concrete flatness at the acquisition moment; according to the concrete solidification stage, the continuously obtained laser point height matrix is marked according to the concrete solidification stage, and the concrete solidification stage comprises an initial solidification stage, a final solidification stage, a concrete early strength development stage and a concrete later strength development stage; using a pre-trained concrete flatness prediction model to input continuously acquired laser point height matrixes marked with all concrete solidification stages, and outputting a concrete flatness prediction matrix at a sequence moment after the final acquisition moment, the technical problem that the flatness of the concrete solidification process is difficult to predict in the prior art is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of concrete construction, and in particular, relates to a method, medium and system for detecting the flatness of concrete during solidification. Background Art

[0002] As an important building material widely used in the fields of construction and infrastructure, concrete's structural performance and surface flatness directly affect the quality and service life of the entire project. Reasonable concrete flatness can not only improve the bearing capacity of the structure, but also reduce energy consumption and maintenance costs during use. Therefore, how to effectively monitor and control the flatness of the concrete surface has always been a key issue of concern to the industry.

[0003] The traditional method of detecting the flatness of concrete mainly relies on manual measurement, which detects the height difference of the concrete surface by laying a ruler and placing steel balls. This method is simple to operate, but the measurement accuracy is low, and it is impossible to continuously monitor the flatness changes of concrete during pouring and solidification. With the development of detection technology, some new measurement methods have gradually emerged, such as the use of laser scanners and image measurement, which realize the accurate acquisition of the three-dimensional shape of the concrete surface. However, these technologies often require professional measurement equipment and complex data processing processes, and have certain limitations for their application on ordinary construction sites.

[0004] In addition, simple flatness monitoring is not enough. How to combine the actual solidification characteristics of concrete materials to predict the future trend of flatness changes is equally important for construction management and quality control. However, most of the existing flatness prediction technologies rely on the experience of construction personnel, and it is difficult to capture the influence mechanism of the complex concrete solidification process on flatness. The prediction accuracy and robustness need to be further improved. In other words, the existing technology has the problem of difficulty in predicting the flatness of the concrete solidification process. Summary of the invention

[0005] In view of this, the present invention provides a method, medium and system for detecting the flatness of concrete during solidification, which can solve the technical problem that it is difficult to predict the flatness of concrete during solidification in the prior art.

[0006] The present invention is achieved in that:

[0007] A first aspect of the present invention provides a method for detecting the flatness of concrete during solidification, comprising the following steps:

[0008] S10, continuously acquiring a laser point height matrix collected by a flatness laser collection device, with a collection frequency of once every 5 minutes to 20 minutes; and using the laser point height matrix as the concrete flatness at the time of collection;

[0009] S20, marking the continuously acquired laser point height matrix according to the concrete solidification stage, wherein the concrete solidification stage includes an initial setting stage, a final setting stage, an early strength development stage of concrete, and a late strength development stage of concrete;

[0010] S30, using a pre-trained concrete flatness prediction model, inputting a continuously acquired laser point height matrix of all marked concrete solidification stages, and outputting a concrete flatness prediction matrix of a sequence moment after the last acquisition moment, which is used as a flatness prediction result of the concrete solidification process.

[0011] The steps of collecting the laser point height matrix are as follows:

[0012] S11, placing the flatness laser collection device on one side of the concrete;

[0013] S12, controlling the stepper motor to drive the reflector to rotate, so that the lasers emitted by the multiple laser emitters are irradiated onto the concrete surface after the reflector rotates to form multiple laser spots; synchronously using the camera to collect an image of the concrete surface, which is recorded as a first image;

[0014] S13, preprocessing the first image to obtain a preprocessed first image, establishing a coordinate axis on the preprocessed first image, where the horizontal axis is a direction perpendicular to the crossbar, and the vertical axis is a direction parallel to the crossbar;

[0015] S14, extracting all laser points in the pre-processed first image, and generating coordinate vectors containing all laser points, recorded as a laser point vector group; and generating an average horizontal coordinate of the laser point vector group;

[0016] S15, continuously executing steps S12-S14 to obtain multiple groups of laser point vector groups, wherein the average horizontal coordinates of the multiple groups of laser point vector groups are an arithmetic sequence and at least include the horizontal coordinates of points close to both ends of the concrete on the horizontal axis;

[0017] S16, merging the multiple groups of laser point vectors into a laser point matrix;

[0018] S17, calculating the relative height of the laser point matrix according to the central axis height of the reflector and the angle between the normal axis of the reflector and the longitudinal axis to obtain a laser point height matrix.

[0019] Among them, the step S10 specifically includes: continuously acquiring the laser point height matrix collected by the flatness laser collection device, with the collection frequency being once every 5 minutes to 20 minutes, and using the laser point height matrix as the concrete flatness at the time of collection.

[0020] Wherein, the step S11 specifically comprises: setting the flatness laser acquisition device on one side of the concrete; fixing the flatness laser acquisition device to ensure that its position and angle do not change; and appropriately adjusting the parameters of the flatness laser acquisition device, such as laser emission power, camera focal length, etc., according to the on-site environment to obtain the best acquisition effect.

[0021] Wherein, the step S12 specifically comprises: controlling the stepper motor to drive the reflector to rotate, so that the lasers emitted by the multiple laser emitters are irradiated on the concrete surface after the reflector rotates to form multiple laser points; and synchronously using the camera to collect an image of the concrete surface, which is recorded as the first image.

[0022] Wherein, the step S13 specifically comprises: preprocessing the first image, such as filtering, brightness adjustment, etc., to obtain a preprocessed first image; establishing a coordinate axis on the preprocessed first image, wherein the horizontal axis is a direction perpendicular to the crossbar, and the vertical axis is a direction parallel to the crossbar.

[0023] Wherein, the step S14 specifically comprises: extracting all laser points in the preprocessed first image, and generating coordinate vectors containing all laser points, recorded as a laser point vector group; and generating an average horizontal coordinate of the laser point vector group at the same time.

[0024] Wherein, the step S15 is specifically: continuously executing steps S12-S14 to obtain multiple groups of laser point vector groups, wherein the average horizontal coordinates of the multiple groups of laser point vector groups are an arithmetic sequence and at least include the horizontal coordinates of points close to the two ends of the concrete on the horizontal axis.

[0025] Wherein, the step S16 specifically comprises: merging the multiple groups of laser point vectors into a laser point matrix.

[0026] Wherein, the step S17 specifically comprises: performing relative height calculation on the laser point matrix according to the center axis height of the reflector and the angle between the normal axis of the reflector and the longitudinal axis to obtain the laser point height matrix.

[0027] Wherein, the step S20 specifically includes: marking the continuously acquired laser point height matrix according to the concrete solidification stage according to the concrete solidification stage, and the concrete solidification stage includes the initial setting stage, the final setting stage, the early strength development stage of concrete, and the late strength development stage of concrete.

[0028] Furthermore, the flatness laser acquisition device includes a bracket, multiple laser emitters, a rotating shaft, a reflector, a stepper motor, and a camera. The bracket includes an upper rod, two parallel vertical rods and a base. The multiple laser emitters are equidistantly arranged below the upper rod. The emitting ends of the multiple laser emitters are arranged in a straight line parallel to the upper rod. Through holes are provided at relative positions of the two vertical rods. The rotating shaft is provided between the two vertical rods through the through hole. One end of the rotating shaft passes through the corresponding vertical rod and extends outside the vertical rod to form an axis end. The axis end is fixedly connected to the output shaft of the stepper motor. The stepper motor is fixedly set on the corresponding vertical rod; the reflector is fixedly set on the rotating shaft, and the camera is also set on the top of the upper rod.

[0029] Furthermore, when the stepping motor is started, the rotating shaft drives the reflector to rotate around the rotating shaft.

[0030] Furthermore, when the reflector is in a vertical state, the top end of the reflector is lower than the bottom ends of the multiple laser emitters.

[0031] Furthermore, the multiple laser emitters are equidistantly arranged below the upper rod in a suspended manner.

[0032] Furthermore, the camera is arranged at the center of the vertical pole through a telescopic rod, and the telescopic rod is used to adjust the height of the camera above the vertical pole.

[0033] Furthermore, the concrete flatness prediction model adopts a multi-branch pyramid structure, including an initial setting subnetwork, a final setting subnetwork, an early subnetwork, a late subnetwork, a short-term prediction subnetwork, a medium-term prediction subnetwork, a long-term prediction subnetwork and a fusion subnetwork.

[0034] The initial setting subnetwork is used to process the flatness characteristics of the initial setting stage of concrete. The input is the laser point height matrix marked as the initial setting stage, and the output is the flatness feature vector of the initial setting stage. The structure is a 3-layer convolutional neural network plus a 1-layer fully connected layer.

[0035] The final setting subnetwork is used to process the flatness change characteristics of concrete from the initial setting stage to the final setting stage. The input is the laser point height matrix marked as the initial setting stage and the final setting stage. The output is the flatness feature vector of the final setting stage. The structure is a 4-layer convolutional neural network plus a 2-layer fully connected layer.

[0036] The early sub-network is used to process the flatness change characteristics of concrete from the initial setting stage to the early strength development stage. The input is the laser point height matrix marked as the initial setting stage, the final setting stage and the early strength development stage of concrete. The output is the flatness feature vector of the early strength development stage. The structure is a 5-layer convolutional neural network plus a 2-layer fully connected layer.

[0037] The late sub-network is used to process the flatness change characteristics of the whole process of concrete. The input is the laser point height matrix marked as all, and the output is the flatness feature vector of the late strength development stage. The structure is a 6-layer convolutional neural network plus a 3-layer fully connected layer.

[0038] The short-term prediction subnetwork is used to predict the concrete flatness in the next 1-6 hours. The input is the output feature vector of each stage subnetwork, and the output is the flatness prediction matrix in the next 1-6 hours. The structure is 1 LSTM layer plus 1 fully connected layer.

[0039] The medium-term prediction subnetwork is used to predict the concrete flatness in the next 6-24 hours. The input is the output feature vector of each stage subnetwork, and the output is the flatness prediction matrix in the next 6-24 hours. The structure is 2 layers of LSTM plus 1 layer of fully connected layer.

[0040] The long-term prediction subnetwork is used to predict the concrete flatness in the next 1-7 days. The input is the output feature vector of each stage subnetwork, and the output is the flatness prediction matrix in the next 1-7 days. The structure is 3 layers of LSTM plus 2 layers of fully connected layers.

[0041] The fusion sub-network is used to fuse the output results of each prediction sub-network. The input is the output matrix of the short-term, medium-term and long-term prediction sub-networks. The output is the concrete flatness prediction matrix at the final sequence moment. The structure is 1 layer of attention mechanism plus 1 layer of fully connected layer.

[0042] The multi-branch pyramid structure has a specific hierarchical relationship or association relationship as follows: the initial condensation, final condensation, early and late sub-networks process data at different stages in parallel, and their output feature vectors are used as inputs to the short-term, medium-term and long-term prediction sub-networks, and the output of the prediction sub-network is then fed into the fusion sub-network to obtain the final prediction result;

[0043] The training steps of the concrete flatness prediction model are specifically:

[0044] Constructing a training data set, including collecting a large number of time-series laser point height matrices continuously collected by the flatness laser acquisition device during the concrete solidification process with different mix ratios and different environmental conditions;

[0045] Model training: dividing the training data set into a training set and a validation set to train a neural network constructed according to the multi-branch pyramid structure to obtain a concrete flatness prediction model.

[0046] The training steps of the concrete flatness prediction model are specifically:

[0047] Step 1: Initialize the parameters of each sub-network;

[0048] Step 2: Use the training set data to train the sub-networks of each stage according to the solidification stage;

[0049] Step 3: Fix the sub-network parameters and train three prediction sub-networks;

[0050] Step 4: Fix the parameters of the three prediction sub-networks and train the fusion sub-network;

[0051] Step 5: Use the validation set to evaluate model performance and adjust hyperparameters;

[0052] Step 6: Repeat steps 2-5 until the model converges;

[0053] Step 7: Use the test set for final evaluation and optimization.

[0054] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the above-mentioned method for detecting flatness of concrete solidification process.

[0055] A third aspect of the present invention provides a system for detecting the flatness of concrete during solidification, which comprises the above-mentioned flatness laser acquisition device and a host computer, wherein the above-mentioned computer-readable storage medium is provided in the host computer.

[0056] Compared with the prior art, the method, medium and system for detecting the flatness of concrete during solidification provided by the present invention have the following beneficial effects:

[0057] First of all, the present invention provides a flatness laser acquisition device with a simple structure, easy on-site deployment, and meeting the convenience requirements of engineering applications. Compared with traditional manual measurement methods, the device can provide higher measurement accuracy and time resolution, effectively solving the problem of inability to continuously monitor.

[0058] Secondly, after obtaining high-quality flatness data, the present invention further matches it with the actual solidification stage of concrete and establishes a deep learning prediction model based on a multi-branch pyramid structure. The model can capture the characteristics of flatness changes in different solidification stages, and based on this, a reliable prediction of the flatness change trend in the future is achieved. Compared with the existing technology that only relies on empirical models or a single analysis method, the prediction method of the present invention is closer to the actual solidification mechanism of concrete materials, and the prediction accuracy and robustness are greatly improved.

[0059] In general, the present invention combines the powerful modeling capabilities of deep learning while ensuring high-precision flatness measurement, realizes real-time monitoring of the flatness of the concrete solidification process and accurate prediction of future change trends, and solves the technical problem that the existing technology is difficult to predict the flatness of the concrete solidification process. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 A flow chart of the method provided by the present invention;

[0061] Figure 2 It is a structural schematic diagram of a flatness laser acquisition device;

[0062] Figure 3 A flow chart of the steps for laser point height matrix acquisition;

[0063] Figure 4 is a schematic diagram of the laser dot matrix;

[0064] Figure 5 A visualization chart of the experimental results;

[0065] In the accompanying drawings, the components represented by the reference numerals are listed as follows:

[0066] 10. Bracket; 11. Upper pole; 12. Vertical pole; 13. Base; 14. Telescopic pole; 20. Laser transmitter; 30. Rotating shaft; 31. Reflector; 32. Stepping motor; 40. Camera. DETAILED DESCRIPTION

[0067] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0068] like Figure 1 FIG. 1 is a flow chart of a method for detecting the flatness of concrete during solidification provided by the present invention. The method comprises the following steps:

[0069] S10, continuously acquiring a laser point height matrix collected by a flatness laser collection device, with a collection frequency of once every 5 minutes to 20 minutes; and using the laser point height matrix as the concrete flatness at the time of collection;

[0070] S20, marking the continuously acquired laser point height matrix according to the concrete solidification stage, wherein the concrete solidification stage includes an initial setting stage, a final setting stage, an early strength development stage of concrete, and a late strength development stage of concrete;

[0071] S30, using a pre-trained concrete flatness prediction model, inputting a continuously acquired laser point height matrix of all marked concrete solidification stages, and outputting a concrete flatness prediction matrix of a sequence moment after the last acquisition moment, which is used as a flatness prediction result of the concrete solidification process.

[0072] like Figure 2As shown, it is a structural schematic diagram of a flatness laser acquisition device, which includes a bracket 10, multiple laser emitters 20, a rotating shaft 30, a reflector 31, a stepper motor 32, and a camera 40. The bracket includes an upper rod 11, two parallel vertical rods 12 and a base 13. The multiple laser emitters are equidistantly arranged below the upper rod, and the emission ends of the multiple laser emitters are arranged in a straight line parallel to the upper rod. Through holes are provided at relative positions of the two vertical rods, and the rotating shaft is provided between the two vertical rods through the through hole. One end of the rotating shaft passes through the corresponding vertical rod and extends outside the vertical rod to form an axis end, and the axis end is fixedly connected to the output shaft of the stepper motor, and the stepper motor is fixedly arranged on the corresponding vertical rod; the reflector is fixedly arranged on the rotating shaft, and the camera is also arranged on the top of the upper rod.

[0073] Furthermore, when the stepping motor is started, the rotating shaft drives the reflector to rotate around the rotating shaft.

[0074] Furthermore, when the reflector is in a vertical state, the top end of the reflector is lower than the bottom ends of the multiple laser emitters.

[0075] Furthermore, the multiple laser emitters are equidistantly arranged below the upper rod in a suspended manner.

[0076] Furthermore, the camera is arranged at the center of the vertical pole through a telescopic rod 14, and the telescopic rod is used to adjust the height of the camera above the vertical pole.

[0077] like Figure 3 FIG. 1 is a flowchart of the steps of collecting the laser point height matrix, specifically:

[0078] S11, placing the flatness laser collection device on one side of the concrete;

[0079] S12, controlling the stepper motor to drive the reflector to rotate, so that the lasers emitted by the multiple laser emitters are irradiated onto the concrete surface after the reflector rotates to form multiple laser spots; synchronously using the camera to collect an image of the concrete surface, which is recorded as a first image;

[0080] S13, preprocessing the first image to obtain a preprocessed first image, establishing a coordinate axis on the preprocessed first image, where the horizontal axis is a direction perpendicular to the crossbar, and the vertical axis is a direction parallel to the crossbar;

[0081] S14, extracting all laser points in the pre-processed first image, and generating coordinate vectors containing all laser points, recorded as a laser point vector group; and generating an average horizontal coordinate of the laser point vector group;

[0082] S15, continuously executing steps S12-S14 to obtain multiple groups of laser point vector groups, wherein the average horizontal coordinates of the multiple groups of laser point vector groups are an arithmetic sequence and at least include the horizontal coordinates of points close to both ends of the concrete on the horizontal axis;

[0083] S16, merging the multiple groups of laser point vectors into a laser point matrix, such as Figure 4 As shown, Figure 4 The blue dots in the middle represent the laser points that are in a straight line, and the red dots represent the laser points that are not in this straight line;

[0084] S17, calculating the relative height of the laser point matrix according to the central axis height of the reflector and the angle between the normal axis of the reflector and the longitudinal axis to obtain a laser point height matrix.

[0085] The specific implementation methods of the above steps are described in detail below:

[0086] The specific implementation of step S10 is: continuously acquiring the laser point height matrix collected by the flatness laser acquisition device. The purpose of this step is to obtain the height information of the concrete surface and provide data support for the subsequent flatness analysis. Specifically, this step includes the following sub-steps:

[0087] Sub-step 1: Set the acquisition frequency. The acquisition frequency of the laser point height matrix is ​​once every 5 minutes to 20 minutes. This frequency has been verified in practice and can ensure the continuity of data while avoiding the processing difficulties caused by excessive data volume.

[0088] Sub-step 2: Collect laser point height matrix. Through the flatness laser acquisition device, continuously collect the laser point height information of the concrete surface to form a time-series laser point height matrix. Each element in the matrix represents the concrete surface height value at a specific position and time.

[0089] Sub-step 3: Use the laser point height matrix as the concrete flatness. Since the laser point height matrix collected at each moment can reflect the ups and downs of the concrete surface at that time, it can be directly regarded as the concrete flatness at that moment.

[0090] In general, step S10 continuously collects laser point height information on the concrete surface by setting a reasonable collection frequency, and directly uses it as the flatness value at that moment, which provides basic data for subsequent flatness analysis and prediction.

[0091] The specific implementation of step S11 is: the flatness laser acquisition device is set on one side of the concrete. The purpose of this step is to determine the installation position of the flatness laser acquisition device, laying the foundation for subsequent data acquisition. Specifically, it includes the following sub-steps:

[0092] Sub-step 1: Choose a suitable installation location. The flatness laser acquisition device should be as close to the concrete surface as possible to ensure that clear laser point height information is collected. At the same time, the stability and safety of the equipment should also be considered, and a location that will not interfere with concrete pouring and solidification should be selected.

[0093] Sub-step 2: Fix the flatness laser acquisition device. Fix the device firmly to the side of the concrete, making sure its position and angle do not change. This can be achieved using a special bracket or fixture.

[0094] Sub-step 3: Adjust equipment parameters. According to the site environment, appropriately adjust some parameters of the flatness laser acquisition device, such as laser emission power, camera focal length, etc., to obtain the best acquisition effect.

[0095] Through the above sub-steps, the flatness laser acquisition device is stably installed on one side of the concrete, providing a solid foundation for subsequent data acquisition.

[0096] The specific implementation of step S12 is: control the stepper motor to drive the reflector to rotate, so that the lasers emitted by the multiple laser emitters are irradiated on the concrete surface after the reflector rotates to form multiple laser points; synchronously use the camera to collect the concrete surface image, which is recorded as the first image. The purpose of this step is to obtain the laser dot matrix image of the concrete surface to prepare for subsequent image processing and height extraction. It specifically includes the following sub-steps:

[0097] Sub-step 1: Start the stepper motor. The stepper motor is driven by a control signal to drive the reflector mounted on the shaft to rotate at a certain speed.

[0098] Sub-step 2: Laser emission and irradiation. Multiple laser emitters emit laser beams simultaneously, which pass through the rotating reflector to form a linear laser dot matrix and irradiate on the concrete surface.

[0099] Sub-step 3: Image acquisition. The camera synchronously captures the concrete surface and records the distribution of the laser dot matrix at this moment to form the first image.

[0100] Through the above sub-steps, the flatness laser acquisition device generates a laser dot matrix distribution image covering the entire concrete surface, laying the foundation for subsequent height information extraction.

[0101] The specific implementation of step S13 is: preprocessing the first image to obtain a preprocessed first image, establishing a coordinate axis on the preprocessed first image, with the horizontal axis being the direction perpendicular to the crossbar and the vertical axis being the direction parallel to the crossbar. The purpose of this step is to process the original image and define the coordinate system to prepare for subsequent data extraction and calculation. Specifically, it includes the following sub-steps:

[0102] Sub-step 1: Image preprocessing. Perform some basic preprocessing operations on the first image, such as filtering, brightness adjustment, etc., to enhance the image quality and improve the applicability and accuracy of subsequent algorithms. Common image processing algorithms can be used here, such as Gaussian filtering, histogram equalization, etc.

[0103] Sub-step 2: Coordinate system establishment. On the preprocessed image, a rectangular coordinate system is established. The horizontal axis (x-axis) is perpendicular to the crossbar of the flatness laser acquisition device, and the vertical axis (y-axis) is parallel to the crossbar. Such a coordinate system definition can better correspond to the spatial distribution of the laser points.

[0104] Through the above sub-steps, the original first image is preprocessed and a suitable coordinate system is established, laying the foundation for subsequent laser point information extraction and calculation.

[0105] The specific implementation of step S14 is: extract all laser points in the preprocessed first image, and generate coordinate vectors containing all laser points, recorded as a laser point vector group; and generate the average horizontal coordinate of the laser point vector group. The purpose of this step is to extract the coordinate information of all laser points from the preprocessed image and calculate their average value on the horizontal axis to prepare for subsequent time series analysis. Specifically, it includes the following sub-steps:

[0106] Sub-step 1: Laser point extraction. In the pre-processed first image, use image processing algorithms (such as threshold segmentation, connected domain analysis, etc.) to extract all clearly visible laser points. For each laser point, record its x and y coordinate values ​​in the preset coordinate system.

[0107] Sub-step 2: Generate a laser point vector group. The coordinate values ​​of all the extracted laser points are combined into a coordinate vector , where n is the total number of laser points.

[0108] Sub-step 3: Average horizontal coordinate calculation. Calculate the laser point vector group The average value of all x-coordinates in , as the average position of the laser point on the horizontal axis at that moment.

[0109] Through the above sub-steps, the coordinate information of all laser points is extracted from the preprocessed first image, and their average value on the horizontal axis is calculated, laying the foundation for subsequent timing analysis.

[0110] The specific implementation of step S15 is: continuously execute steps S12-S14 to obtain multiple groups of laser point vector groups, the average horizontal coordinates of the multiple groups of laser point vector groups are an arithmetic sequence, and at least contain the horizontal coordinates of the points close to the concrete at both ends on the horizontal axis. The purpose of this step is to obtain a series of time-continuous laser point coordinate information to prepare for subsequent height calculation and flatness prediction. Specifically, it includes the following sub-steps:

[0111] Sub-step 1: Loop through steps S12-S14. Repeat the image acquisition, laser point extraction and average horizontal coordinate calculation operations to obtain multiple groups of laser point vectors and corresponding average horizontal coordinate values ​​within a continuous time period.

[0112] Sub-step 2: Average horizontal coordinate sequence analysis. Check the average horizontal coordinates of multiple groups of laser point vectors. Whether it forms an arithmetic sequence. The existence of an arithmetic sequence indicates that the distribution of the laser points in the horizontal direction is uniform and can cover the entire concrete surface.

[0113] Sub-step 3: Boundary point confirmation. Further analyze this arithmetic sequence to confirm whether it contains the laser points closest to the two ends of the concrete surface. The coordinate information of these boundary points is very important for the subsequent height calculation.

[0114] Through the above sub-steps, we obtained a series of time-continuous laser point coordinate information and confirmed that it contained representative data of the entire concrete surface, laying the foundation for subsequent height calculations.

[0115] The specific implementation of step S16 is: merging the multiple groups of laser point vectors into a laser point matrix. The purpose of this step is to integrate the scattered laser point coordinate information into a unified data structure to facilitate subsequent height calculation. It specifically includes the following sub-steps:

[0116] Sub-step 1: Integration of laser point vector groups. The multiple groups of laser point vectors obtained in step S15 are integrated. Integrate into a large coordinate vector in time order middle.

[0117] Sub-step 2: Laser dot matrix construction. The distribution of coordinate values ​​in the matrix is ​​reorganized into a two-dimensional matrix The number of rows in the matrix corresponds to the distribution of the laser points in the longitudinal direction, and the number of columns corresponds to the distribution of the laser points in the transverse direction.

[0118] Through the above sub-steps, we integrate the scattered laser point coordinate information into a unified two-dimensional matrix , which provides a convenient data structure for subsequent height calculations.

[0119] The specific implementation of step S17 is: according to the central axis height of the reflector and the angle between the normal axis of the reflector and the longitudinal axis, the relative height of the laser point matrix is ​​calculated to obtain the laser point height matrix. The purpose of this step is to convert the planar laser point distribution into actual three-dimensional height information based on known geometric parameters to provide basic data for subsequent flatness analysis. It specifically includes the following sub-steps:

[0120] Sub-step 1: Determine the reflector parameters. According to the design parameters of the flatness laser acquisition device, determine the center axis height of the reflector. and the normal axis and the longitudinal axis of the reflector Angle These two geometric parameters play a key role in the subsequent height calculation.

[0121] Sub-step 2: Height calculation. For the laser point matrix Each element in , according to its coordinate values ​​and the parameters of the reflector and , calculate the height of the point on the z-axis (perpendicular to the concrete surface) The following formula can be used here:

[0122]

[0123] Sub-step 3: Height matrix construction. The height values ​​of all laser points calculated are Reorganize into a two-dimensional matrix , compared with the original laser dot matrix One-to-one correspondence forms the final laser point height matrix.

[0124] Through the above sub-steps, we convert the original planar laser point distribution into actual three-dimensional height information according to the geometric parameters of the flatness laser acquisition device, providing basic data for subsequent flatness analysis. .

[0125] The specific implementation of step S20 is: according to the solidification stage of concrete, the continuously acquired laser point height matrix is ​​marked according to the concrete solidification stage, and the concrete solidification stage includes the initial solidification stage, the final solidification stage, the early strength development stage of concrete, and the late strength development stage of concrete. The purpose of this step is to correspond the acquired height matrix data with the actual solidification process of concrete, in preparation for the subsequent flatness prediction model training. Specifically, it includes the following sub-steps:

[0126] Sub-step 1: Division of solidification stages. According to the actual solidification process of concrete, the whole process is divided into four main stages: initial solidification stage, final solidification stage, early strength development stage of concrete and late strength development stage of concrete. These four stages correspond to different development states of concrete performance.

[0127] Sub-step 2: Data marking. The continuous laser point height matrix obtained in step S17 , marked according to the above four solidification stages. That is, for each moment Matrix, with a label attached to it, indicating which specific solidification stage it belongs to.

[0128] Through the above sub-steps, we matched the acquired concrete surface height data with the actual solidification process, laying the foundation for the subsequent flatness prediction model training.

[0129] The specific implementation of step S30 is: using a pre-trained concrete flatness prediction model, inputting the laser point height matrix of all the continuously acquired marked concrete solidification stages, and outputting the concrete flatness prediction matrix of the sequence moments after the last acquisition moment, which is used as the flatness prediction result of the concrete solidification process. The purpose of this step is to use the prediction model to predict the change of the flatness of the concrete surface in the future period of time based on the acquired laser point height data. It specifically includes the following sub-steps:

[0130] Sub-step 1: Prediction model selection. This method uses a complex neural network model called a multi-branch pyramid structure as a concrete flatness prediction model. The model includes multiple sub-networks that specialize in different solidification stages, as well as sub-networks for short-term, medium-term, and long-term predictions.

[0131] Sub-step 2: Model input preparation. The laser point height matrix obtained in step S20 and marked with the solidification stage is , as the input of the prediction model.

[0132] Sub-step 3: Model inference. The prepared input data is passed into the pre-trained concrete flatness prediction model, and the flatness prediction matrix corresponding to the "sequence moment after the last acquisition moment" is obtained through the forward calculation of the model.

[0133] Sub-step 4: Output of prediction results. The output of the prediction model, that is, the concrete flatness prediction matrix for a period of time in the future, is output as the final prediction result.

[0134] Among them, the "interval of the sequence moments is equal to the time interval of the frequency of the acquisition". That is to say, the time series of the output of the prediction result is consistent with the time interval of the original data acquisition.

[0135] Through the above sub-steps, we used the pre-trained complex neural network model to predict the concrete flatness in the future based on the acquired laser point height data, providing valuable information for engineering practice.

[0136] In general, this invention proposes a method for detecting the flatness of concrete during solidification based on laser scanning. The core idea is:

[0137] 1. Use the flatness laser acquisition device to continuously collect the laser point height information on the concrete surface to form a time-series laser point height matrix.

[0138] 2. Mark the acquired laser point height matrix data corresponding to the actual solidification stage of concrete.

[0139] 3. Based on the pre-trained complex neural network model, the marked laser point height data is input and the concrete flatness prediction results for a period of time in the future are output.

[0140] The entire method makes full use of laser measurement technology and deep learning algorithms to achieve real-time monitoring and future prediction of the flatness of the concrete solidification process, providing timely and effective data support for engineering construction.

[0141] Specifically, the principle of the present invention is to make full use of laser scanning technology and deep learning algorithms to achieve accurate measurement of the flatness of the concrete surface and reliable prediction of future change trends.

[0142] First, the present invention designs a special flatness laser collection device, which is composed of a bracket, a laser transmitter, a reflector, a stepper motor, a camera and other components. By controlling the stepper motor to drive the reflector to rotate, the laser beams emitted by multiple laser transmitters are reflected to form a laser dot matrix covering the entire concrete surface. The camera synchronously captures the laser dot matrix distribution image, laying the foundation for subsequent height information extraction. Compared with the traditional manual measurement method, the device can continuously collect the height data of the concrete surface, greatly improving the time resolution and spatial coverage of the measurement.

[0143] After obtaining high-quality flatness data, the present invention further marks the laser point height matrix of these time series with the actual solidification process of concrete. The solidification process is divided into several main stages, such as the initial setting stage, the final setting stage, the early strength development stage, and the late strength development stage. This correspondence between data and the actual solidification process provides valuable supervision information for the subsequent deep learning prediction model training.

[0144] Based on the above-mentioned rich training data, the present invention adopts a complex neural network model called multi-branch pyramid structure as a concrete flatness predictor. The model consists of multiple sub-networks that specialize in processing different solidification stages and sub-networks for short-term, medium-term and long-term predictions. Through this multi-branch structure, the model can fully capture the complex characteristics of flatness changes in different solidification stages and integrate them into the final prediction output. Compared with a single data analysis method or empirical model, this prediction method based on deep learning is closer to the actual solidification mechanism of concrete materials, greatly improving the accuracy and robustness of the prediction.

[0145] In general, the technical idea of ​​the present invention is: first, use a specially designed laser acquisition device to obtain high-quality three-dimensional data of the concrete surface, then annotate these data according to the actual solidification process, and finally use a complex deep learning model to predict the future trend of flatness changes. This method, which combines advanced measurement technology and powerful modeling algorithms, solves the shortcomings of existing technologies in terms of measurement accuracy, field applicability, and prediction reliability, and provides valuable technical support for the quality control of concrete engineering.

[0146] In order to better understand and implement the present invention, a specific embodiment 1 of the present invention is provided below. The specific steps of embodiment 1 are described as follows:

[0147] The specific implementation of step S10 is to continuously obtain the laser point height matrix collected by the flatness laser collection device. This step is to obtain the height information of the concrete surface and provide basic data for subsequent flatness analysis.

[0148] Specifically, this step includes the following sub-steps:

[0149] Sub-step 1: Set the acquisition frequency. The acquisition frequency of the laser point height matrix is ​​every Minutes~ Once every minute. This frequency has been verified in practice and can ensure the continuity of data while avoiding the processing difficulties caused by excessive data volume.

[0150] Sub-step 2: Collect the laser point height matrix. Through the flatness laser acquisition device, continuously collect the laser point height information of the concrete surface to form a time-series laser point height matrix . Each element in this matrix Representative moments Location The concrete surface height value.

[0151] Sub-step 3: Use the laser point height matrix as the concrete flatness. They can reflect the ups and downs of the concrete surface at that time, so they can be directly regarded as the flatness of the concrete at that moment.

[0152] In general, step S10 continuously collects laser point height information on the concrete surface by setting a reasonable collection frequency, and directly uses it as the flatness value at that moment, which provides basic data for subsequent flatness analysis and prediction.

[0153] The specific implementation of step S11 is to set the flatness laser acquisition device on one side of the concrete. The purpose of this step is to determine the installation position of the flatness laser acquisition device to lay the foundation for subsequent data acquisition. Specifically, it includes the following sub-steps:

[0154] Sub-step 1: Choose a suitable installation location. The flatness laser acquisition device should be as close to the concrete surface as possible to ensure that clear laser point height information is collected. At the same time, the stability and safety of the equipment should also be considered, and a location that will not interfere with concrete pouring and solidification should be selected.

[0155] Sub-step 2: Fix the flatness laser acquisition device. Fix the device firmly to the side of the concrete, making sure its position and angle do not change. This can be achieved using a special bracket or fixture.

[0156] Sub-step 3: Adjust equipment parameters. According to the site environment, appropriately adjust some parameters of the flatness laser acquisition device, such as laser emission power, camera focal length, etc., to obtain the best acquisition effect.

[0157] Through the above sub-steps, the flatness laser acquisition device is stably installed on one side of the concrete, providing a solid foundation for subsequent data acquisition.

[0158] The specific implementation of step S12 is to control the stepper motor to drive the reflector to rotate, so that the lasers emitted by the multiple laser emitters are irradiated on the concrete surface after the reflector rotates to form multiple laser points; the camera is used to collect the concrete surface image at the same time, which is recorded as the first image. The purpose of this step is to obtain the laser dot matrix image of the concrete surface to prepare for subsequent image processing and height extraction. It specifically includes the following sub-steps:

[0159] Sub-step 1: Start the stepper motor. Drive the stepper motor at an angular velocity Drive the reflector mounted on the rotating shaft to rotate.

[0160] Sub-step 2: Laser emission and irradiation. Multiple laser emitters emit laser beams simultaneously. ,After passing through the rotating reflector, a linear laser dot array is formed and irradiated on the concrete surface.

[0161] Sub-step 3: Image acquisition. The camera synchronously captures the concrete surface and records the distribution of the laser dot matrix at this moment to form the first image. .

[0162] Through the above sub-steps, the flatness laser acquisition device generates a laser dot matrix distribution image covering the entire concrete surface. , which laid the foundation for the subsequent extraction of height information.

[0163] The specific implementation method of step S13 is to process the first image Perform preprocessing to obtain the first preprocessed image , in the preprocessed first image The coordinate axes are established on the image, the horizontal axis is the direction perpendicular to the crossbar, and the vertical axis is the direction parallel to the crossbar. The purpose of this step is to process the original image and define the coordinate system to prepare for subsequent data extraction and calculation. It specifically includes the following sub-steps:

[0164] Sub-step 1: Image preprocessing. For the first image Perform some basic preprocessing operations, such as Gaussian filtering and histogram equalization, to enhance image quality and improve the applicability and accuracy of subsequent algorithms. The following Gaussian filtering algorithm can be used:

[0165]

[0166] in, is the Gaussian kernel function, is the convolution operation.

[0167] Sub-step 2: Coordinate system establishment. After preprocessing the image , establish a rectangular coordinate system Among them, the horizontal axis The vertical axis is perpendicular to the horizontal bar of the flatness laser acquisition device. This coordinate system definition can better correspond to the spatial distribution of the laser points.

[0168] Through the above sub-steps, the original first image The image was preprocessed and a suitable coordinate system was established, which laid the foundation for the subsequent extraction and calculation of laser point information.

[0169] The specific implementation of step S14 is to extract the pre-processed first image All laser points in the image are generated, and the coordinate vector containing all laser points is generated. , and generate a laser point vector group The average horizontal coordinate of The purpose of this step is to extract the coordinate information of all laser points from the preprocessed image and calculate their average value on the horizontal axis to prepare for subsequent time series analysis. It includes the following sub-steps:

[0170] Sub-step 1: Laser point extraction. After preprocessing, the first image In the image processing algorithm (such as Otsu's method), all clearly visible laser points are extracted. , record its coordinate value in the preset coordinate system.

[0171] Sub-step 2: Generate a laser point vector group. The coordinate values ​​of all the extracted laser points are combined into a coordinate vector ,in is the total number of laser points.

[0172] Sub-step 3: Average horizontal coordinate calculation. Calculate the laser point vector group All The average value of the coordinates , as the average position of the laser point on the horizontal axis at that moment.

[0173] Through the above sub-steps, from the first image after preprocessing Extract the coordinate information of all laser points and calculate their average value on the horizontal axis , which laid the foundation for the subsequent timing analysis.

[0174] The specific implementation of step S15 is to continuously execute steps S12-S14 to obtain multiple groups of laser point vector groups. , the average horizontal coordinate of the multiple groups of laser point vectors It is an arithmetic sequence and at least contains the horizontal coordinates of the points close to the two ends of the concrete on the horizontal axis. The purpose of this step is to obtain a series of time-continuous laser point coordinate information to prepare for subsequent height calculation and flatness prediction. It specifically includes the following sub-steps:

[0175] Sub-step 1: Loop through steps S12-S14. Repeat the image acquisition, laser point extraction and average horizontal coordinate calculation operations to obtain multiple groups of laser point vectors in a continuous time period. and the corresponding average horizontal coordinate .

[0176] Sub-step 2: Average horizontal coordinate sequence analysis. Check the average horizontal coordinates of multiple groups of laser point vectors. Whether it forms an arithmetic sequence. The existence of an arithmetic sequence indicates that the distribution of the laser points in the horizontal direction is uniform and can cover the entire concrete surface.

[0177] Sub-step 3: Boundary point confirmation. Further analysis of this arithmetic sequence , confirm whether it contains the coordinates of the laser points closest to the two ends of the concrete surface and The coordinate information of these boundary points is very important for the subsequent height calculation.

[0178] Through the above sub-steps, we obtain a series of time-continuous laser point coordinate information , and confirmed that it contained representative data of the entire concrete surface, laying the foundation for subsequent height calculations.

[0179] The specific implementation of step S16 is to convert the multiple groups of laser point vectors Merge into laser dot matrix The purpose of this step is to integrate the scattered laser point coordinate information into a unified data structure to facilitate subsequent height calculations. It specifically includes the following sub-steps:

[0180] Sub-step 1: Integration of laser point vector groups. The multiple groups of laser point vectors obtained in step S15 are integrated. Integrate into a large coordinate vector in time order middle.

[0181] Sub-step 2: Laser dot matrix construction. The distribution of coordinate values ​​in the matrix is ​​reorganized into a two-dimensional matrix The number of rows in the matrix Corresponding to the longitudinal distribution of laser points, number of columns The corresponding laser point distribution in the lateral direction is .

[0182] Through the above sub-steps, we integrate the scattered laser point coordinate information into a unified two-dimensional matrix , which provides a convenient data structure for subsequent height calculations.

[0183] The specific implementation of step S17 is based on the central axis height of the reflector , the normal axis and the longitudinal axis of the reflector Angle , for the laser dot matrix Calculate the relative height and get the laser point height matrix The purpose of this step is to convert the planar laser point distribution into actual three-dimensional height information based on known geometric parameters, providing basic data for subsequent flatness analysis. It specifically includes the following sub-steps:

[0184] Sub-step 1: Determine the reflector parameters. According to the design parameters of the flatness laser acquisition device, determine the center axis height of the reflector. and the normal axis and the longitudinal axis of the reflector Angle These two geometric parameters play a key role in the subsequent height calculation.

[0185] Sub-step 2: Height calculation. For the laser point matrix Each element in , according to its coordinate values ​​and the parameters of the reflector and , calculate the point at Height value on the axis (perpendicular to the concrete surface) The following formula can be used here:

[0186]

[0187] Sub-step 3: Height matrix construction. The height values ​​of all laser points calculated are Reorganize into a two-dimensional matrix , compared with the original laser dot matrix One-to-one correspondence forms the final laser point height matrix .

[0188] Through the above sub-steps, we can calculate the geometric parameters of the flatness laser acquisition device. and , the original plane laser point distribution Convert to actual 3D height information , which provides basic data for subsequent flatness analysis.

[0189] The specific implementation method of step S20 is to obtain the laser point height matrix continuously according to the setting stage of concrete. Mark according to the concrete solidification stage. The concrete solidification stage includes the initial setting stage, the final setting stage, the early strength development stage of concrete, and the late strength development stage of concrete. The purpose of this step is to correspond the acquired height matrix data with the actual solidification process of concrete, in preparation for the subsequent flatness prediction model training. It includes the following sub-steps:

[0190] Sub-step 1: Division of solidification stages. According to the actual solidification process of concrete, the whole process is divided into four main stages: initial solidification stage, final solidification stage, early strength development stage of concrete and late strength development stage of concrete. These four stages correspond to different development states of concrete performance.

[0191] Sub-step 2: Data marking. The continuous laser point height matrix sequence obtained in step S17 , marked according to the above four solidification stages. That is, for each moment Matrix, with a label , indicating which specific solidification stage it belongs to.

[0192] Through the above sub-steps, we will obtain the concrete surface height data It corresponds to the actual solidification process, laying the foundation for the subsequent flatness prediction model training.

[0193] The specific implementation method of step S30 is to use the pre-trained concrete flatness prediction model, input the laser point height matrix of all the marked concrete solidification stages obtained continuously, and , output the last acquisition time The subsequent sequence time Concrete flatness prediction matrix , which is used as the flatness prediction result of the concrete solidification process. The purpose of this step is to use the prediction model to predict the change of the flatness of the concrete surface in the future period of time based on the acquired laser point height data. It specifically includes the following sub-steps:

[0194] Sub-step 1: Prediction model selection. This method uses a complex neural network model called a multi-branch pyramid structure as a concrete flatness prediction model. The model includes multiple sub-networks that specialize in different solidification stages, as well as sub-networks for short-term, medium-term, and long-term predictions.

[0195] Sub-step 2: Model input preparation. The laser point height matrix sequence obtained in step S20 and marked with the solidification stage is and the corresponding label sequence , as the input of the prediction model.

[0196] Sub-step 3: Model inference. The prepared input data is passed into the pre-trained concrete flatness prediction model, and the corresponding "last acquisition time" is obtained through the forward calculation of the model. The subsequent sequence time "The flatness prediction matrix .

[0197] Sub-step 4: Output of prediction results. The output of the prediction model, that is, the concrete flatness prediction matrix for a period of time in the future, is output as the final prediction result.

[0198] Among them, the interval of the sequence time That is to say, the time series of the forecast result output is consistent with the time interval of the original data collection.

[0199] Through the above sub-steps, we used the pre-trained complex neural network model to predict the concrete flatness in the future based on the acquired laser point height data, providing valuable information for engineering practice.

[0200] In general, this invention proposes a method for detecting the flatness of concrete during solidification based on laser scanning. The core idea is:

[0201] 1. Use the flatness laser acquisition device to continuously collect the laser point height matrix of the concrete surface .

[0202] 2. Obtain the laser point height matrix data , corresponding to the actual setting stage of concrete .

[0203] 3. Based on the pre-trained complex neural network prediction model, input the marked laser point height data and output the concrete flatness prediction matrix for a period of time in the future .

[0204] The entire method makes full use of laser measurement technology and deep learning algorithms to achieve real-time monitoring and future prediction of the flatness of the concrete solidification process, providing timely and effective data support for engineering construction.

[0205] In order to further better understand and implement the present invention, the following provides Example 2 of a specific application scenario of the present invention: A large-scale infrastructure project in a certain city adopts the concrete solidification process flatness detection method proposed in the present invention to ensure the quality and progress of the project construction. The project is located in the central area of ​​the city, and the concrete pouring operation needs to be completed on a limited construction site, which has high requirements for the portability and installation location of the flatness detection equipment.

[0206] According to the actual situation of the project, the engineering party chose to use the flatness laser acquisition device of the present invention to continuously acquire the height information of the concrete surface at about 1.5 meters on one side of the concrete pouring surface. The specific structure of the device is as follows:

[0207] The bracket consists of an upper rod, two parallel vertical rods and a base, with an overall size of 2.5 meters long, 1.5 meters wide and 2 meters high. The upper rod is about 1.8 meters above the ground, and the distance between the two vertical rods is 1 meter. The base adopts a thickened steel structure to ensure the stability of the device.

[0208] The laser emission part has a total of 9 infrared laser emitters, which are installed equidistantly about 0.2 meters below the upper pole. The power of each laser emitter is 5mW, the emission wavelength is 650nm, and they are arranged linearly to form a laser line about 1.8 meters long.

[0209] The rotating shaft is set between the two vertical rods, and one end of the shaft extends out of the vertical rod to the outside and is fixedly connected to the output shaft of the stepper motor. The stepper motor is installed on the vertical rod about 1.2 meters above the ground, and can drive the rotating shaft and the reflector installed on it to rotate at a speed of 60 revolutions per minute. The reflector is 0.5 meters long and 0.1 meters wide and is installed at the center of the rotating shaft.

[0210] The camera part uses a high-definition digital camera, which is installed on the bracket at the top of the upper pole. The installation height of the camera can be adjusted by the telescopic rod to keep it about 1.9 meters above the ground. The camera parameters are set as: resolution 1920×1080, frame rate 30fps, exposure time 5ms.

[0211] The overall weight of the flatness laser acquisition device is about 120kg, and it can be portable and mobile by adding universal wheels to the base. After the device is installed, the automatic acquisition of concrete surface height data can be achieved by connecting the stepper motor to the power supply and the digital camera to the control computer.

[0212] The specific data collection process is as follows:

[0213] 1. Install the flatness laser acquisition device securely about 1.5 meters from one side of the concrete casting surface, ensuring that the position and angle of the device do not change. Adjust the installation height of the camera so that the area it captures can cover the entire concrete surface.

[0214] 2. Start the stepper motor to drive the reflector to rotate at a speed of 60 revolutions per minute. At the same time, 9 laser emitters work simultaneously, emitting infrared laser beams, which pass through the rotating reflector to form a laser line about 1.8 meters long, shining on the concrete surface.

[0215] 3. The digital camera captures the laser dot distribution images on the concrete surface synchronously at a frame rate of 30fps. These images are recorded as the first image ,in is the total number of images acquired every 5 minutes.

[0216] 4. For the first image sequence collected Preprocessing is performed, including Gaussian filtering, histogram equalization and other operations, to improve the image quality. The processed image is recorded as .

[0217] 5. Image after preprocessing On the image, Otsu's image segmentation algorithm is used to extract clearly visible laser points and record these laser points in the preset coordinate system. The position information in the image is used to form a series of laser point vector groups. . At the same time, calculate the average horizontal coordinate of each group of laser point vectors .

[0218] 6. Check these average horizontal coordinates Whether it forms an arithmetic sequence, if so, it means that the distribution of laser points in the horizontal direction is uniform and can cover the entire concrete surface. Further analyze the arithmetic sequence to confirm whether it contains the coordinates of the laser points closest to the two ends of the concrete surface. and .

[0219] 7. Set of laser point vectors Integrate into a large coordinate vector in time order In, and according to The distribution of coordinate values ​​in the middle is reorganized into a two-dimensional laser point matrix .

[0220] 8. Determine the center axis height of the reflector according to the design parameters of the flatness laser acquisition device Meters, and the normal axis and longitudinal axis of the reflector Angle . Using the following formula, the laser point matrix Each element in Perform height calculation to obtain the final laser point height matrix :

[0221]

[0222] 9. According to the actual solidification process of concrete, the whole process is divided into four main stages: initial setting stage (0-3 hours), final setting stage (3-6 hours), early strength development stage (6-24 hours) and late strength development stage (more than 24 hours). , according to these four solidification stages, the corresponding label sequence is obtained .

[0223] Through the above steps, we obtained a set of time series concrete surface height data , and the corresponding solidification stage labels Next, we use these data and apply the deep learning prediction model proposed in this invention to predict the concrete flatness in the future.

[0224] The specific structure of the prediction model is as follows:

[0225] 1. Initial setting subnetwork: This network is used to process the flatness characteristics of concrete in the initial setting stage (0-3 hours). The input is the laser point height matrix marked as the initial setting stage. , the output is the flatness feature vector of the initial setting stage The network structure is a 3-layer convolutional neural network plus a fully connected layer.

[0226] 2. Final setting subnetwork: This network is used to process the flatness change characteristics of concrete from the initial setting to the final setting stage (0-6 hours). The input is the laser point height matrix marked as the initial setting stage and the final setting stage. , the output is the flatness feature vector of the final setting stage The network structure is a 4-layer convolutional neural network plus 2 fully connected layers.

[0227] 3. Early sub-network: This network is used to process the flatness change characteristics of concrete from initial setting to early strength development stage (0-24 hours). The input is the laser point height matrix marked as initial setting stage, final setting stage and early strength development stage. , the output is the flatness feature vector of the early strength development stage The network structure is a 5-layer convolutional neural network plus 2 fully connected layers.

[0228] 4. Late sub-network: This network is used to process the flatness change characteristics of the entire concrete process (0-7 days), and the input is the laser point height matrix marked as all stages , the output is the flatness feature vector of the later strength development stage The network structure is a 6-layer convolutional neural network plus 3 fully connected layers.

[0229] 5. Short-term prediction subnetwork: This network is used to predict the concrete flatness in the next 1-6 hours. The input is the output feature vector of each subnetwork. , the output is the flatness prediction matrix for the next 1-6 hours The network structure is 1 LSTM layer plus 1 fully connected layer.

[0230] 6. Medium-term prediction subnetwork: This network is used to predict the concrete flatness in the next 6-24 hours. The input is the output feature vector of each subnetwork. , the output is the flatness prediction matrix for the next 6-24 hours The network structure is 2 layers of LSTM plus 1 fully connected layer.

[0231] 7. Long-term prediction subnetwork: This network is used to predict the concrete flatness in the next 1-7 days. The input is the output feature vector of each stage subnetwork. , the output is the flatness prediction matrix for the next 1-7 days The network structure is 3 layers of LSTM plus 2 layers of fully connected layers.

[0232] 8. Fusion sub-network: This network is used to fuse the output results of each prediction sub-network. The input is the output matrix of the short-term, medium-term and long-term prediction sub-networks, and the output is the final concrete flatness prediction matrix at the sequence moment. The network structure is 1 layer of attention mechanism plus 1 layer of fully connected layer.

[0233] When training this complex neural network model, the parameters of each sub-network are initialized first, and then each sub-network is trained separately according to the solidification stage; next, the sub-network parameters are fixed, and the short-term, medium-term and long-term prediction sub-networks are trained; finally, the prediction sub-network parameters are fixed, and the fusion sub-network is trained. The validation set is used to evaluate the model performance throughout the training process, and the hyperparameters are continuously adjusted until the model converges. Finally, an independent test set is used for final evaluation and optimization.

[0234] In this project, we collected continuous monitoring data of different concrete mixes and pouring environments in the construction site over the past three months as the training set. The validation set and test set accounted for 20% and 10% of the total samples respectively. Through the above phased training strategy, we successfully trained a concrete flatness prediction model with excellent performance.

[0235] In practical applications, the input of the prediction model is:

[0236] 1. Continuously collected laser point height matrix sequence , time interval minute.

[0237] 2. Corresponding solidification stage label sequence .

[0238] The output is:

[0239] 1. Flatness prediction matrix sequence for the next 7 days ,in minute.

[0240] These prediction results can intuitively reflect the trend of the flatness of concrete in the next week, providing valuable decision support for project managers. For example, if the prediction shows that the flatness will deviate greatly within a certain period of time, the project can take corresponding remedial measures, such as adjusting the pouring process parameters, making flatness corrections in a timely manner, etc., to ensure the flatness performance of the final concrete structure.

[0241] In order to verify the practical application effect of the method of the present invention, we conducted a series of comparative tests in the project. First, the traditional manual measurement method was used to test the surface flatness of the concrete on the 3rd, 7th and 14th days after the concrete was poured. The results showed that the average height difference was between 3.2mm and 5.6mm, which was somewhat different from the industry acceptance standard.

[0242] Then, we deployed the flatness laser acquisition device of the present invention at the same location to continuously monitor the concrete surface height information and input the pre-trained deep learning prediction model. Table 1 shows some of the results of this experiment:

[0243] Table 1 Experimental results

[0244]

[0245] Converted into a visual chart, such as Figure 5 As shown in the figure, it can be seen that in the initial setting stage (0-3 hours), the surface flatness of the concrete fluctuates greatly, but as the subsequent solidification process proceeds, the flatness gradually stabilizes. Compared with the static data obtained by manual measurement, the method of the present invention can not only continuously monitor the change of flatness, but also the predicted results are consistent with the measured values, and the maximum error does not exceed 2.5%. This fully demonstrates the superiority of the present invention in terms of measurement accuracy, dynamic monitoring and future prediction.

[0246] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for detecting the flatness of concrete during solidification, characterized in that: The following steps are involved: S10, continuously acquiring a laser point height matrix collected by a flatness laser collection device, with a collection frequency of once every 5 minutes to 20 minutes; and using the laser point height matrix as the concrete flatness at the time of collection; S20, marking the continuously acquired laser point height matrix according to the concrete solidification stage, wherein the concrete solidification stage includes an initial setting stage, a final setting stage, an early strength development stage of concrete, and a late strength development stage of concrete; S30, using a pre-trained concrete flatness prediction model, inputting a continuously acquired laser point height matrix of all marked concrete solidification stages, and outputting a concrete flatness prediction matrix of a sequence moment after the last acquisition moment, which is used as a flatness prediction result of the concrete solidification process.

2. A method for detecting flatness during concrete solidification according to claim 1, characterized in that: The steps of collecting the laser point height matrix are specifically: S11, placing the flatness laser collection device on one side of the concrete; S12, controlling the stepper motor to drive the reflector to rotate, so that the lasers emitted by the multiple laser emitters are irradiated onto the concrete surface after the reflector rotates to form multiple laser spots; synchronously using the camera to collect an image of the concrete surface, which is recorded as a first image; S13, preprocessing the first image to obtain a preprocessed first image, establishing a coordinate axis on the preprocessed first image, where the horizontal axis is a direction perpendicular to the crossbar, and the vertical axis is a direction parallel to the crossbar; S14, extracting all laser points in the pre-processed first image, and generating coordinate vectors containing all laser points, recorded as a laser point vector group; and generating an average horizontal coordinate of the laser point vector group; S15, continuously executing steps S12-S14 to obtain multiple groups of laser point vector groups, wherein the average horizontal coordinates of the multiple groups of laser point vector groups are an arithmetic sequence and at least include the horizontal coordinates of points close to both ends of the concrete on the horizontal axis; S16, merging the multiple groups of laser point vectors into a laser point matrix; S17, calculating the relative height of the laser point matrix according to the central axis height of the reflector and the angle between the normal axis of the reflector and the longitudinal axis to obtain a laser point height matrix.

3. A method for detecting flatness during concrete solidification process according to claim 2, characterized in that: The flatness laser acquisition device includes a bracket, multiple laser emitters, a rotating shaft, a reflector, a stepper motor, and a camera. The bracket includes an upper rod, two parallel vertical rods and a base. The multiple laser emitters are equidistantly arranged below the upper rod. The emission ends of the multiple laser emitters are arranged in a straight line parallel to the upper rod. Through holes are arranged at opposite positions of the two vertical rods. The rotating shaft is arranged between the two vertical rods through the through hole. One end of the rotating shaft passes through the corresponding vertical rod and extends outside the vertical rod to form an axis end. The axis end is fixedly connected to the output shaft of the stepper motor. The stepper motor is fixedly arranged on the corresponding vertical rod; the reflector is fixedly arranged on the rotating shaft, and the camera is also arranged on the top of the upper rod.

4. A method for detecting flatness during concrete solidification process according to claim 3, characterized in that: When the stepper motor is started, the rotating shaft drives the reflector to rotate around the rotating shaft.

5. A method for detecting flatness during concrete solidification according to claim 4, characterized in that: When the reflector is in a vertical state, the top end of the reflector is lower than the bottom ends of the multiple laser emitters.

6. A method for detecting flatness during concrete solidification process according to claim 5, characterized in that: The multiple laser transmitters are equidistantly arranged below the upper rod in a suspended manner.

7. A method for detecting flatness during concrete solidification process according to claim 6, characterized in that: The camera is arranged at the center of the vertical pole through a telescopic rod, and the telescopic rod is used to adjust the height of the camera above the vertical pole.

8. A method for detecting flatness during concrete solidification process according to claim 7, characterized in that: The concrete flatness prediction model adopts a multi-branch pyramid structure, including an initial setting sub-network, a final setting sub-network, an early sub-network, a late sub-network, a short-term prediction sub-network, a medium-term prediction sub-network, a long-term prediction sub-network and a fusion sub-network.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the method for detecting flatness during concrete solidification process according to any one of claims 1 to 8.

10. A system for detecting the flatness of concrete during solidification, characterized in that: It comprises the flatness laser acquisition device involved in any one of claims 2-6, and a host computer, wherein the host computer is provided with the computer-readable storage medium as claimed in claim 9.

Images