Intelligent quality detection system and method for roller compacted concrete

Abstract

The invention discloses an intelligent quality detection system and method for roller compacted concrete, and the system comprises a data collection module which is used for obtaining underground data through the measurement of a ground penetrating radar on a to-be-detected land parcel, inputting the obtained underground data into a machine learning model, and outputting the obtained actually measured compactness; the underground model making module is used for acquiring an underground image output by the ground penetrating radar on the to-be-detected land parcel and making an underground model according to the underground image; and the compaction degree correction module is used for detecting the underground model by using a nucleon density instrument to obtain the detected compaction degree, and obtaining and outputting the corrected compaction degree according to the actually measured compaction degree. According to the method, the underground data output by the ground penetrating radar are converted into the compaction degree in a mode of establishing the model, the underground model is made through the underground image output by the ground penetrating radar, and the compaction degree is corrected by using the nuclear density instrument, so that the compaction degree is corrected under the condition that the built roller compacted concrete is not damaged. And the compaction degree of the roller compacted concrete is detected.

Description

Technical Field

[0001] The present invention relates to the technical field of roller compacted concrete construction, and in particular to an intelligent roller compacted concrete quality detection system and method. Background Art

[0002] The core testing method for quality control and acceptance of roller-compacted concrete projects generally relies on traditional on-site sampling methods. Specifically, the two most commonly used sampling techniques in engineering practice are the sand injection method and the nuclear density meter method. However, this sampling-based quality assessment model has exposed a series of significant limitations in practice.

[0003] First, random sampling is generally inefficient. Whether it's the tedious on-site steps of digging, filling, and weighing required for the sand filling method, or point-by-point measurement using a nuclear density meter, both require dedicated manpower and time to perform each operation at specific locations on site. This point-by-point testing approach, when applied to large-scale RCC paving, lags far behind the pace of construction, making it difficult to meet the quality control requirements of rapid construction.

[0004] Secondly, and more critically, the results of random sampling tests are severely delayed, preventing real-time quality feedback during construction. Inspectors can only draw conclusions about quality at specific points after sampling or measuring, analyzing and processing the data. This prevents construction managers from instantly understanding whether key indicators like compaction and moisture content are meeting standards during rolling operations, hindering their ability to quickly respond to and adjust to potential quality fluctuations.

[0005] Furthermore, particularly for destructive testing methods like sand injection, the very act of testing physically damages the completed RCC. To obtain specimens or conduct density tests, holes must be drilled or sample locations excavated on the surface of the structure. These artificially created holes disrupt the integrity and continuity of the RCC layer, compromising its structural integrity. Therefore, after completing the sampling test, the construction company must immediately perform specialized repair work on these damaged areas, known as supplemental construction. This not only incurs additional material and labor costs but also introduces new quality risks, as the quality of the repaired areas may differ from that of the original RCC.

[0006] In summary, although sampling inspection is currently the main basis for evaluating the quality of RCC, its inherent inefficiency, delayed feedback of results, and the inevitable damage to completed structures that require subsequent repairs have become bottlenecks restricting construction efficiency and the immediacy of quality monitoring. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to address the deficiencies in the above-mentioned prior art and provide an intelligent quality detection system and method for roller-compacted concrete. By establishing a model, the underground data output by the ground-penetrating radar is converted into compaction degree, and an underground model is made using the underground image output by the ground-penetrating radar. The compaction degree is corrected using a nuclear density meter, so that the compaction degree of roller-compacted concrete can be detected without destroying the roller-compacted concrete that has been constructed.

[0008] To solve the above technical problems, the present invention adopts a technical solution: an intelligent quality detection system and method for roller compacted concrete, wherein the intelligent quality detection system for roller compacted concrete comprises: The data model construction module uses ground-penetrating radar to obtain underground data on the test plot, and measures the compaction degree by taking geological samples on the test plot. The corresponding underground data and compaction degree are used to train the machine learning model to obtain a trained machine learning model. A data acquisition module, which uses a ground penetrating radar to measure and obtain underground data on the land to be tested, and inputs the obtained underground data into the trained machine learning model to output the measured compaction degree; An underground model making module obtains underground images output by a ground-penetrating radar on the plot to be inspected, and uses 3D printing technology to make an underground model based on the underground images; The compaction correction module detects the underground model using a nuclear density meter to obtain a detected compaction, obtains a corrected compaction based on the measured compaction and the detected compaction, and outputs the corrected compaction.

[0009] Furthermore, the data model construction module includes: The test area division module divides the test plot into multiple test areas, and each test area is evenly distributed; The training data acquisition module uses ground penetrating radar to obtain underground data in each test area, takes geological samples from the test area, and obtains the compaction of the test area through experimental measurement. The underground data and compaction of each test area are matched one by one; The data model training module establishes a machine learning model, takes the underground data of each test area as output, and its corresponding compaction degree as output, trains the established machine learning model, and obtains a trained machine learning model.

[0010] Furthermore, the underground model making module includes: The underground image acquisition module acquires the underground image output by the ground penetrating radar on the land to be detected; An underground image 3D restoration module performs 3D restoration on the underground image through image processing technology to obtain an underground 3D image; The underground model making module makes an underground model using 3D printing technology according to the underground 3D image.

[0011] Furthermore, the underground image 3D restoration module performs 3D restoration on the underground image using image processing technology to obtain an underground 3D image, including: Extracting the color value of each pixel of the underground image; Establishing a two-dimensional coordinate system to represent the position of each pixel point of the underground image in the form of two-dimensional coordinates; Expanding the two-dimensional coordinate system into a three-dimensional coordinate system, and rotating the two-dimensional coordinate system once on the expanded three-dimensional axis to obtain three-dimensional coordinates; Each color value is assigned to a point corresponding to a three-dimensional coordinate, and the coordinate value of the assigned three-dimensional coordinate includes the two-dimensional coordinate corresponding to the color value, and an underground 3D image is output.

[0012] Furthermore, the method of expanding the two-dimensional coordinate system into a three-dimensional coordinate system includes: Setting the two-dimensional coordinate system longitudinally as the coordinates of the YOZ plane of the three-dimensional coordinate system; The three-dimensional coordinate system is expanded in the X-axis direction of the two-dimensional coordinate system to obtain an expanded three-dimensional coordinate system.

[0013] Furthermore, after the underground 3D image is obtained, the output includes: Smoothing the color value of each pixel in the underground 3D image; Output smoothed 3D underground image.

[0014] Furthermore, the compaction correction module obtains a corrected compaction according to the measured compaction and the detected compaction, including: According to the difference between the measured compaction degree and the detected compaction degree, using a pre-trained neural network model to obtain weight factors corresponding to the measured compaction degree and the detected compaction degree respectively; According to the weight factors corresponding to the measured compaction degree and the detected compaction degree respectively, the measured compaction degree and the detected compaction degree are combined to obtain a corrected compaction degree.

[0015] The neural network model takes the difference between the measured compaction degree and the detected compaction degree as input, and outputs a weight factor corresponding to the measured compaction degree and a weight factor corresponding to the detected compaction degree.

[0016] An intelligent quality detection method for roller compacted concrete, comprising: Step 1: Use ground penetrating radar to obtain underground data on the test plot, and measure the compaction by taking geological samples on the test plot. The corresponding underground data and compaction are used to train the machine learning model to obtain a trained machine learning model; Step 2: Using ground penetrating radar to measure underground data on the plot to be tested, and inputting the obtained underground data into the trained machine learning model to output the measured compaction degree; Step 3: Obtain an underground image output by a ground-penetrating radar on the plot to be inspected, and use 3D printing technology to produce an underground model based on the underground image; Step 4: The underground model is tested using a nuclear density meter to obtain a tested compaction degree, and a corrected compaction degree is obtained based on the measured compaction degree and the tested compaction degree, and the corrected compaction degree is output.

[0017] Compared with the prior art, the present invention has the following advantages: The present invention provides an intelligent quality inspection system and method for roller-compacted concrete (RCC). Its core advantage lies in completely overcoming the significant drawbacks inherent in traditional sampling inspection methods (such as sand filling or nuclear density meter sampling), enabling efficient, non-destructive, and real-time testing of the compaction of RCC. Specifically, the system first uses ground-penetrating radar to collect subsurface data at a test site. Combined with geological sampling, it obtains actual compaction data. This data is then used to train a machine learning model, establishing a precise mapping relationship between subsurface data and compaction. During the engineering application phase, GPR is simply used to rapidly and continuously acquire large-scale subsurface data from the test site. This data is then fed into the trained model to generate preliminary "actual compaction" results in real time. This process completely avoids the tedious on-site pit excavation, sand filling, weighing, or point-by-point nuclear density meter measurements required by traditional methods, significantly improving inspection efficiency. Its coverage and speed match the rapid pace of construction, and it provides near-real-time feedback on compaction information, enabling construction managers to instantly assess quality during rolling operations and quickly respond to potential issues. Crucially, the system further utilizes image processing and 3D printing technology to create a precise underground physical model based on the underground images generated by the ground-penetrating radar. This model fully reproduces the subsurface structural characteristics of the area being tested, enabling non-destructive transfer of nuclear density meter testing, typically performed on the physical structure, to the model (i.e., "testing the subsurface model with the nuclear density meter"), thereby obtaining "tested compaction" data. Finally, the system combines the "measured compaction" predicted by the model with the "tested compaction" obtained through non-destructive testing on the model, calculating the final "corrected compaction" using a specific correction algorithm (such as a neural network model based on difference weighting factors). This innovative process fundamentally eliminates physical damage to completed RCC structures (such as the excavation sampling required by the sand injection method), completely avoiding the subsequent repair work required for damage detection, with its associated additional costs, delays, and quality risks to the repaired area. This truly achieves non-destructive compaction testing of RCC throughout the entire process.

[0018] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a schematic diagram of the overall structure of an intelligent quality detection system for roller compacted concrete provided by the present invention.

[0020] Figure 2 This is a schematic diagram of the overall process of an intelligent quality detection method for roller compacted concrete provided by the present invention. DETAILED DESCRIPTION

[0021] like Figure 1-2 As shown, the present invention provides an intelligent quality detection system and method for roller compacted concrete, which is described in two embodiments below. Example 1: Specific implementation of the intelligent quality detection system for roller-compacted concrete like Figure 1 As shown, an intelligent quality detection system for roller compacted concrete includes: The data model construction module uses ground-penetrating radar to obtain underground data on the test plot, and measures the compaction degree by taking geological samples on the test plot. The corresponding underground data and compaction degree are used to train the machine learning model to obtain a trained machine learning model. A data acquisition module, which uses a ground penetrating radar to measure and obtain underground data on the land to be tested, and inputs the obtained underground data into the trained machine learning model to output the measured compaction degree; An underground model making module obtains underground images output by a ground-penetrating radar on the plot to be inspected, and uses 3D printing technology to make an underground model based on the underground images; The compaction correction module detects the underground model using a nuclear density meter to obtain a detected compaction, obtains a corrected compaction based on the measured compaction and the detected compaction, and outputs the corrected compaction.

[0022] The following is a detailed introduction to each of the above modules: The core goal of this module is to establish a reliable mapping between subsurface data acquired by GPR and the actual compaction of RCC. This module is based on supervised learning using machine learning algorithms (e.g., support vector machines (SVRs), random forests (Random Forests), or deep neural networks (DNNs).

[0023] The data model construction module includes: The test area division module divides the test plot into multiple test areas, and each test area is evenly distributed; The training data acquisition module uses ground penetrating radar to obtain underground data in each test area, takes geological samples from the test area, and obtains the compaction of the test area through experimental measurement. The underground data and compaction of each test area are matched one by one; The data model training module establishes a machine learning model, takes the underground data of each test area as output, and its corresponding compaction degree as output, trains the established machine learning model, and obtains a trained machine learning model.

[0024] The test area division module (test area division module) involves regularly dividing a representative test plot into multiple test areas (for example, into 1m x 1m or 2m x 2m grids). The key requirement is that these areas should be evenly distributed across the test plot to ensure that the collected training data covers different geological and compaction conditions.

[0025] Training data collection (training data collection module): In each divided test area, two key operations are performed: * Use ground penetrating radar to scan and measure the underground data of the area (usually including multi-dimensional signals such as electromagnetic wave reflection intensity, time delay, waveform characteristics, etc.).

[0026] * Geological sampling is carried out in the test area (usually using sand filling or core sampling). The samples are sent to the laboratory for experimental measurement (such as drying and weighing) to obtain the precise compaction value of the area.

[0027] The underground data obtained by ground penetrating radar measurement in each test area are matched one-to-one with the corresponding compaction value obtained through experimental measurement to form a set of training data pairs (underground data->compactness).

[0028] Data Model Training (Data Model Training Module): Build a selected machine learning model (e.g., initialize an SVR model). Subsurface data collected from all test areas serves as the model's input feature vectors, and the corresponding compaction values ​​serve as the model's output target values. These training data pairs (input-output pairs) are used to train the established machine learning model. The training process uses optimization algorithms (such as gradient descent) to adjust the model's internal parameters, enabling it to learn the complex nonlinear mapping relationship from subsurface data features to compaction values. Upon completion, a trained machine learning model is obtained.

[0029] The technical effect achieved by this invention is that the established model serves as the foundation for subsequent rapid nondestructive testing. By combining destructive sampling (the method described in the background art) in a typical test area to obtain accurate "true values," and leveraging the powerful pattern recognition capabilities of machine learning, the non-intuitive signals from ground-penetrating radar are converted into a model that can predict compaction. This enables subsequent large-scale testing at the construction site without requiring damage.

[0030] A data acquisition module that uses a trained machine learning model to perform rapid, large-scale preliminary compaction assessments of actual RCC layers.

[0031] In this module, operators use ground-penetrating radar (GPR) to continuously scan the plot of land (i.e., the actual RCC construction area) requiring quality inspection along a pre-defined path (such as a serpentine or grid pattern), obtaining subsurface data covering the entire area. This newly acquired subsurface data is fed into the machine learning model trained in the previous step. Based on the learned mapping relationship, the model outputs the measured compaction value (i.e., the model-predicted compaction value) at the corresponding scan location.

[0032] The resulting effect directly addresses the core issues of low efficiency and delayed results in existing technologies. Ground-penetrating radar can scan rapidly and continuously, covering an area far greater than traditional spot checks. Model predictions are nearly real-time (in seconds or milliseconds), providing near-real-time feedback on compaction levels. This allows construction managers to instantly assess the quality of the rolling operation and quickly respond to potential issues (e.g., immediately recompacting areas of insufficient compaction). This significantly improves inspection efficiency, and its coverage and speed match the rapid pace of construction.

[0033] The underground model making module is designed to use the imaging capability of ground penetrating radar to non-destructively construct a physical solid model to carry the nuclear density meter detection that originally needed to be performed on the real concrete structure.

[0034] The underground model making module includes: The underground image acquisition module acquires the underground image output by the ground penetrating radar on the land to be detected; An underground image 3D restoration module performs 3D restoration on the underground image through image processing technology to obtain an underground 3D image; The underground model making module makes an underground model using 3D printing technology according to the underground 3D image.

[0035] The underground image acquisition module uses ground-penetrating radar (GPR) equipment or its supporting software to acquire underground images of the ground to be inspected. These images are typically 2D cross-sections or 3D volume data, reflecting the distribution of media at different depths (e.g., reflection intensity maps).

[0036] 3D Restoration of Subsurface Images (3D Restoration Module): This module uses image processing technology to restore the acquired subsurface images (usually 2D or stacked 2D slices) into 3D to obtain a 3D image of the subsurface. This process involves more sophisticated algorithms, as follows: The underground image 3D restoration module performs 3D restoration on the underground image using image processing technology to obtain an underground 3D image, including: Extracting the color value of each pixel of the underground image; Establishing a two-dimensional coordinate system to represent the position of each pixel point of the underground image in the form of two-dimensional coordinates; Expanding the two-dimensional coordinate system into a three-dimensional coordinate system, and rotating the two-dimensional coordinate system once on the expanded three-dimensional axis to obtain three-dimensional coordinates; Each color value is assigned to a point corresponding to a three-dimensional coordinate, and the coordinate value of the assigned three-dimensional coordinate includes the two-dimensional coordinate corresponding to the color value, and an underground 3D image is output.

[0037] The method of expanding the two-dimensional coordinate system into a three-dimensional coordinate system includes: Setting the two-dimensional coordinate system longitudinally as the coordinates of the YOZ plane of the three-dimensional coordinate system; The three-dimensional coordinate system is expanded in the X-axis direction of the two-dimensional coordinate system to obtain an expanded three-dimensional coordinate system.

[0038] After the underground 3D image is obtained, the output includes: Smoothing the color value of each pixel in the underground 3D image; Output smoothed 3D underground image.

[0039] The above algorithm is explained as follows: * Extract the color value (or grayscale value, reflection intensity value) of each pixel in the underground image.

[0040] * Establish a two-dimensional coordinate system (for example, the row and column coordinates U, V of the image) to represent the position of each pixel point of the underground image in the form of two-dimensional coordinates.

[0041] Expanding the 2D coordinate system into a 3D coordinate system: Specifically, the 2D coordinate system is arranged vertically (assuming each row of the image represents a depth profile) and used as the coordinates of the YOZ plane of the 3D coordinate system (where the Y axis represents the horizontal direction and the Z axis represents the depth direction). Then, the 3D coordinate system is expanded along the X axis (usually perpendicular to the scan profile) within the 2D coordinate system to obtain the expanded 3D coordinate system (X, Y, Z).

[0042] * Rotate the 2D coordinate system about the extended 3D axis: The original YOZ plane (i.e., each depth profile) is rotated 360 degrees (or an angle determined by the actual scan coverage) around the extended X axis (or the principal axis determined by the actual data organization) to obtain the 3D coordinates of each point. This is essentially the process of reconstructing 2D profile data into 3D volume data.

[0043] Assign each color value to the point corresponding to the 3D coordinate. Specifically, the assigned 3D coordinates include the 2D coordinates corresponding to the color value (i.e., the original U and V coordinates are mapped to a location in 3D space and the color value is assigned to that point). The output is a 3D image of the subsurface (typically a 3D voxel model).

[0044] Optionally, after the underground 3D image is output, the color values ​​of each pixel (voxel) in the underground 3D image may be smoothed (for example, using a Gaussian filter, a median filter, or other algorithm) to eliminate image noise or reconstruction artifacts. The smoothed underground 3D image is then output.

[0045] Subsurface Model Production (Subsurface Model Production Module): Based on the processed subsurface 3D image (voxel model), a physical subsurface model is produced using 3D printing technologies (such as FDM fused deposition modeling and SLA stereolithography). This model is printed using a material that simulates the electromagnetic properties (or at least the density properties) of concrete, and its internal structure (such as layers and areas of density variation) accurately replicates the subsurface characteristics of the area being measured.

[0046] This module of the present invention represents a key innovation in addressing destructive testing issues. Through advanced image processing and 3D printing technologies, it transforms invisible radar images into tangible physical models. This model fully reproduces the subsurface structural characteristics of the area being tested, providing a physical foundation for the subsequent non-destructive use of nuclear density meters. This fundamentally eliminates the need for physical damage to completed RCC structures (as required for excavation sampling during sand injection).

[0047] The compaction correction module is designed to integrate the preliminary model prediction results ("measured compaction") and the non-destructive testing results performed on the physical model ("tested compaction"), and use the correction algorithm to obtain a more accurate and reliable final compaction assessment value.

[0048] The specific steps are as follows: The operator places the underground model prepared as described above in a safe location, and uses a nuclear density meter to detect the density of the model (which is then converted into compaction) in accordance with standard operating procedures to obtain the detected compaction value.

[0049] Based on the measured compaction (the model prediction value from step 2) and the detected compaction (the value measured by the nuclear density meter on the model), the corrected compaction is obtained. The specific correction algorithm is as follows: Calculate the difference: Calculate the difference between the measured compaction and the tested compaction.

[0050] Obtaining weight factors: A pre-trained neural network model (a small network specifically designed for correction) is used, with the calculated difference as input. This neural network is trained to output weight factors corresponding to the measured compaction (W_m) and the detected compaction (W_d). The learning goal of this pre-trained small neural network (correction model) is to adjust the weights of the predicted value and the model measurement value when there is a significant difference, so that the final result is more likely to be reliable. (Typically, model measurements are closer to the "true value" when the model accuracy is high enough, but the correction model learns the optimal weight distribution through historical data.)

[0051] Calculate the correction value: Based on the weighting factors (W_m, W_d) corresponding to the measured and tested compactions, respectively, combined with the measured compaction (P_m) and the tested compaction (P_d), a corrected compaction value (P_corr) is calculated. This is typically calculated as a weighted average: P_corr = W_m * P_m + W_d * P_d, where W_m + W_d = 1.

[0052] The corrected compaction degree is output as the final quality assessment result of the detection area.

[0053] That is, the compaction correction module obtains a corrected compaction according to the measured compaction and the detected compaction, including: According to the difference between the measured compaction degree and the detected compaction degree, using a pre-trained neural network model to obtain weight factors corresponding to the measured compaction degree and the detected compaction degree respectively; According to the weight factors corresponding to the measured compaction degree and the detected compaction degree respectively, the measured compaction degree and the detected compaction degree are combined to obtain a corrected compaction degree.

[0054] The neural network model takes the difference between the measured compaction degree and the detected compaction degree as input, and outputs a weight factor corresponding to the measured compaction degree and a weight factor corresponding to the detected compaction degree.

[0055] The technical effect achieved by this module of the present invention is to overcome the limitations of single-method approaches (pure model predictions may have errors, and nuclear density meter testing based on the model depends on model accuracy). By fusing the "measured compaction" predicted by the model with the "tested compaction" obtained through nondestructive testing on the model, and utilizing a neural network model based on differential weighting factors for intelligent correction, the accuracy and reliability of the final compaction assessment results are significantly improved. Furthermore, this entire correction process completely avoids the subsequent repair work required by damage detection, with its attendant additional costs, project delays, and quality risks to the repaired area, thus achieving truly nondestructive compaction testing of roller-compacted concrete.

[0056] Example 1: Specific implementation of the intelligent quality detection method for roller compacted concrete like Figure 2 As shown, an intelligent quality detection method for roller compacted concrete includes: Step 1: Use ground penetrating radar to obtain underground data on the test plot, and measure the compaction by taking geological samples on the test plot. The corresponding underground data and compaction are used to train the machine learning model to obtain a trained machine learning model; Step 2: Using ground penetrating radar to measure underground data on the plot to be tested, and inputting the obtained underground data into the trained machine learning model to output the measured compaction degree; Step 3: Obtain an underground image output by a ground-penetrating radar on the plot to be inspected, and use 3D printing technology to produce an underground model based on the underground image; Step 4: The underground model is tested using a nuclear density meter to obtain a tested compaction degree, and a corrected compaction degree is obtained based on the measured compaction degree and the tested compaction degree, and the corrected compaction degree is output.

[0057] The steps of this method correspond to the functional modules of the system described in Example 1.

[0058] Step 1: Model training Specific steps: On a representative test plot: A systematic scan was performed using ground penetrating radar to obtain underground data covering the plot.

[0059] By taking geological samples on the test plot (at selected evenly distributed points or small areas, as described in Example 1), the precise compaction values ​​of these sampling points / areas are measured (laboratory measurement).

[0060] The corresponding subsurface data (radar data corresponding to each sampling point / area) and compaction values ​​are used to train the machine learning model.

[0061] Select an appropriate machine learning algorithm (such as SVR, RF, DNN), input underground data, output target compaction, and perform model training.

[0062] After the training is completed, a trained machine learning model is obtained.

[0063] Principle and Effect: Same as the "Data Model Construction Module" in Example 1. A mapping relationship between ground penetrating radar signals and compaction degree is established, laying the foundation for non-destructive and rapid detection.

[0064] Step 2: Obtaining the measured compaction Specific steps: On the plot to be tested (actual RCC construction surface): Use ground penetrating radar to perform continuous, large-area scanning to measure the underground data of the land.

[0065] Input the obtained underground data into the machine learning model trained in step 1.

[0066] The model output is the measured compaction value (model predicted value) corresponding to the scanning position.

[0067] Principle and Effect: Same as the "Data Acquisition Module" in Example 1. It realizes efficient and near real-time preliminary evaluation of compaction degree over a large area, solving the problems of low efficiency and delayed feedback of traditional methods.

[0068] Step 3: Underground Model Making Specific steps: Obtain underground images (two-dimensional profiles or original three-dimensional data) output by ground penetrating radar scanning on the land to be inspected.

[0069] 3D printing technology is used to create a subsurface model based on the subsurface image. This process typically includes: Perform necessary processing on the original image (such as 3D restoration and smoothing) to generate a three-dimensional digital model that accurately describes the underground structure.

[0070] Import this three-dimensional digital model file into the 3D printer, select the appropriate material (which can simulate the density characteristics of concrete), and print out the physical entity model.

[0071] Principle and Effect: Same as the "underground model making module" in Example 1. Create a physical model as a non-destructive testing carrier to completely avoid damage to the actual concrete structure and eliminate repair costs and risks.

[0072] Step 4: Compaction correction and output Specific steps: The underground model produced in step 3 is tested using a nuclear density meter (measurements are performed on the model according to the operating specifications of the nuclear density meter) to obtain the tested compaction value (based on the model density conversion).

[0073] According to the measured compaction degree (result of step 2) and the detected compaction degree (result of this step), the modified compaction degree is obtained. The modification algorithm preferably adopts: Calculate the difference between the measured compaction and the tested compaction.

[0074] This difference is input into the pre-trained neural network model (corrected model).

[0075] The model outputs the weight factor corresponding to the measured compaction degree and the weight factor corresponding to the detected compaction degree.

[0076] The two weight factors are used to perform weighted average calculation on the measured compaction degree and the detected compaction degree to obtain the corrected compaction degree.

[0077] The corrected compaction degree is output as the final detection result.

[0078] Principle and Effect: Same as the "Compactness Correction Module" in Example 1. By integrating the preliminary prediction results with the non-destructive measurement results on the physical model and using an intelligent correction algorithm, the accuracy and reliability of the final compaction assessment results are significantly improved while maintaining the advantage of full non-destructiveness.

[0079] As can be seen from the above-mentioned specific embodiments, the system and method provided by the present invention systematically integrate multiple technologies, including nondestructive ground-penetrating radar (GPR), intelligent prediction using machine learning, 3D printing physical modeling, and nondestructive testing and correction using nuclear density meters. The workflow is clear: first, a radar data-compactness model is established at the test site; then, this model is used at the construction site to quickly and in real time obtain a preliminary compaction distribution map through radar scanning; then, the same radar data is used to create an accurate 3D-printed physical model; finally, a nuclear density meter test is performed on the model, and the results are intelligently integrated and corrected with the model predictions to produce a highly accurate final compaction assessment. This solution completely overcomes the three inherent drawbacks of traditional sampling testing methods (sand filling and nuclear density meter spot checks) in the prior art: low efficiency, delayed result feedback, and physical damage to completed structures. It achieves efficient (matching construction speed), nondestructive (non-destructive throughout the process), and real-time (near-instant feedback) testing of RCC compaction, significantly improving the level and efficiency of construction quality control.

[0080] In summary, the present invention converts the underground data output by the ground-penetrating radar into compaction by establishing a model, and makes an underground model through the underground image output by the ground-penetrating radar, and uses a nuclear density meter to correct the compaction, so that the compaction of the roller-compacted concrete can be detected without destroying the roller-compacted concrete that has been built.

[0081] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any way. Any simple modification, change and equivalent structural change made to the above embodiment based on the technical essence of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. An intelligent quality detection system for roller compacted concrete, characterized in that: include: The data model construction module uses ground-penetrating radar to obtain underground data on the test plot, and measures the compaction degree by taking geological samples on the test plot. The corresponding underground data and compaction degree are used to train the machine learning model to obtain a trained machine learning model. A data acquisition module, which uses a ground penetrating radar to measure and obtain underground data on the land to be tested, and inputs the obtained underground data into the trained machine learning model to output the measured compaction degree; An underground model making module obtains underground images output by a ground-penetrating radar on the plot to be inspected, and uses 3D printing technology to make an underground model based on the underground images; The compaction correction module detects the underground model using a nuclear density meter to obtain a detected compaction, obtains a corrected compaction based on the measured compaction and the detected compaction, and outputs the corrected compaction.

2. The intelligent quality detection system for roller compacted concrete according to claim 1, characterized in that: The data model construction module includes: The test area division module divides the test plot into multiple test areas, and each test area is evenly distributed; The training data acquisition module uses ground penetrating radar to obtain underground data in each test area, takes geological samples from the test area, and obtains the compaction of the test area through experimental measurement. The underground data and compaction of each test area are matched one by one; The data model training module establishes a machine learning model, takes the underground data of each test area as output, and its corresponding compaction degree as output, trains the established machine learning model, and obtains a trained machine learning model.

3. The intelligent quality detection system for roller compacted concrete according to claim 1, characterized in that: The underground model making module includes: The underground image acquisition module acquires the underground image output by the ground penetrating radar on the land to be detected; An underground image 3D restoration module performs 3D restoration on the underground image through image processing technology to obtain an underground 3D image; The underground model making module makes an underground model using 3D printing technology according to the underground 3D image.

4. The intelligent quality detection system for roller compacted concrete according to claim 3, characterized in that: The underground image 3D restoration module performs 3D restoration on the underground image using image processing technology to obtain an underground 3D image, including: Extracting the color value of each pixel of the underground image; Establishing a two-dimensional coordinate system to represent the position of each pixel point of the underground image in the form of two-dimensional coordinates; Expanding the two-dimensional coordinate system into a three-dimensional coordinate system, and rotating the two-dimensional coordinate system once on the expanded three-dimensional axis to obtain three-dimensional coordinates; Each color value is assigned to a point corresponding to a three-dimensional coordinate, and the coordinate value of the assigned three-dimensional coordinate includes the two-dimensional coordinate corresponding to the color value, and an underground 3D image is output.

5. The intelligent quality detection system for roller compacted concrete according to claim 4, characterized in that: The method of expanding the two-dimensional coordinate system into a three-dimensional coordinate system includes: Setting the two-dimensional coordinate system longitudinally as the coordinates of the YOZ plane of the three-dimensional coordinate system; The three-dimensional coordinate system is expanded in the X-axis direction of the two-dimensional coordinate system to obtain an expanded three-dimensional coordinate system.

6. The intelligent quality inspection system for roller compacted concrete according to claim 5, characterized in that: After the underground 3D image is obtained, the output includes: Smoothing the color value of each pixel in the underground 3D image; Output smoothed 3D underground image.

7. The intelligent quality inspection system for roller compacted concrete according to claim 1, characterized in that: The compaction correction module obtains a corrected compaction according to the measured compaction and the detected compaction, including: According to the difference between the measured compaction degree and the detected compaction degree, using a pre-trained neural network model to obtain weight factors corresponding to the measured compaction degree and the detected compaction degree respectively; According to the weight factors corresponding to the measured compaction degree and the detected compaction degree respectively, the measured compaction degree and the detected compaction degree are combined to obtain a corrected compaction degree. The neural network model takes the difference between the measured compaction degree and the detected compaction degree as input, and outputs a weight factor corresponding to the measured compaction degree and a weight factor corresponding to the detected compaction degree.

8. An intelligent quality detection method for roller compacted concrete, characterized in that: include: Step 1: Use ground penetrating radar to obtain underground data on the test plot, and measure the compaction by taking geological samples on the test plot. The corresponding underground data and compaction are used to train the machine learning model to obtain a trained machine learning model; Step 2: Using ground penetrating radar to measure underground data on the plot to be tested, and inputting the obtained underground data into the trained machine learning model to output the measured compaction degree; Step 3: Obtain an underground image output by a ground-penetrating radar on the plot to be inspected, and use 3D printing technology to produce an underground model based on the underground image; Step 4: The underground model is tested using a nuclear density meter to obtain a tested compaction degree, and a corrected compaction degree is obtained based on the measured compaction degree and the tested compaction degree, and the corrected compaction degree is output.

Images