Artificial intelligence-based roadbed continuous compaction degree real-time monitoring method

Abstract

The invention discloses a roadbed continuous compaction degree real-time monitoring method based on artificial intelligence, and relates to the technical field of roadbed construction control, and the method comprises the following steps: constructing a multi-modal fitting model based on the correlation between roadbed compaction factors and the compaction degree; the roadbed type before compaction construction and detection data in the compaction construction process are collected and input into the multi-modal fitting model, and the real-time compaction degree of the roadbed in the current construction process is analyzed and calculated; setting a compaction qualification index based on the normative standard of the current construction area, and judging whether the current road roller meets a compaction ending condition or not; and when the compaction degree of all the compaction areas reaches a compaction qualified parameter, controlling the road roller to finish compaction. According to the method, the convergence calculation speed is accelerated by using the artificial intelligence algorithm to reach a real-time evaluation level, conditions are created for establishment of continuous compaction of the roadbed and real-time evaluation of the compaction degree, timeliness and accuracy are achieved, and the manual detection cost is greatly reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of subgrade construction control, and more specifically, to a real-time monitoring method for continuous compaction degree of subgrade based on artificial intelligence. Background Art

[0002] In the current context of the increasing development of technology, in the subgrade compaction operation process, clusterized and automated compaction machinery has been vigorously developed and promoted to meet the scientific, efficient, and automated process control and management of subgrade construction quality.

[0003] In traditional high-grade subgrade construction management methods, quality management requires on-site supervisors and construction personnel to subjectively control the vibratory roller to achieve subgrade filling compaction management. This is prone to forming human interference and influence, and the extensive management method will result in potential problems such as missed compaction, under-compaction, and over-compaction in some compaction areas. At the same time, the entire compaction process cannot record various control parameters and compaction degrees in real time, making it difficult to trace the quality of the subgrade and causing a certain waste of resources. And the quality random inspection after compaction belongs to lag control and cannot play a real-time feedback and guiding role in the quality control of the compaction process.

[0004] Therefore, in actual construction, there are defects such as low efficiency, large impact on precision evaluation, and non-real-time detection means in manual on-site control of rolling parameters and on-site test sampling.

[0005] For the problems in the related art, no effective solution has been proposed yet. Summary of the Invention

[0006] (1) Technical Problems to be Solved

[0007] Aiming at the deficiencies of the prior art, the present invention provides a real-time monitoring method for continuous compaction degree of subgrade based on artificial intelligence, which has the advantages of using artificial intelligence algorithms to accelerate the convergence calculation speed to reach the real-time evaluation level, creating conditions for continuous compaction of the subgrade and real-time evaluation of the compaction degree, being timely and accurate, and greatly reducing the cost of manual detection. Furthermore, it solves the problems of low efficiency, large impact on precision evaluation, and non-real-time detection means in manual on-site control of rolling parameters and on-site test sampling.

[0008] (2) Technical Solutions

[0009] To achieve the above advantages of using artificial intelligence algorithms to accelerate the convergence calculation speed to reach the real-time evaluation level, creating conditions for continuous compaction of the subgrade and real-time evaluation of the compaction degree, being timely and accurate, and greatly reducing the cost of manual detection, the specific technical solutions adopted by the present invention are as follows:

[0010] A real-time monitoring method for continuous compaction degree of subgrade based on artificial intelligence, the method comprising the following steps:

[0011] S1. Based on the correlation between subgrade compaction factors and compaction degree, construct a multi-modal fitting model, verify the fitting multi-modal fitting model using existing historical data, and optimize the constructed multi-modal fitting model;

[0012] S2. Collect the subgrade type before compaction construction and the detection data during the compaction construction process, input them into the multi-modal fitting model, and analyze and calculate the real-time compaction degree during the current subgrade construction process;

[0013] S3. Set the compaction qualification index based on the specification standards of the current construction area, judge whether the current roller meets the compaction end condition, and feedback the real-time compaction degree to the display device;

[0014] S4. When the compaction degree of all compaction areas reaches the compaction qualification parameters, control the roller to end compaction, record and save the data results after compaction, and the roller control parameters and subgrade parameters corresponding to the compaction degree.

[0015] Further, based on the correlation between subgrade compaction factors and compaction degree, construct a multi-modal fitting model, verify the fitting multi-modal fitting model using existing historical data, and the steps for optimizing the constructed multi-modal fitting model include:

[0016] S11. Obtain the historical data of subgrade compaction factors and the corresponding compaction degree, give priority to processing missing values, and after data format unification and normalization processing, obtain the standardized data of subgrade compaction factors. The subgrade compaction factors include subgrade type, gradation, moisture content and compaction density;

[0017] S12. Convert the processed subgrade compaction factors into input variables, sequentially establish a multi-layer model structure to obtain a multi-modal fitting model, analyze the correlation degree between subgrade compaction factors and compaction degree, extract the effective features of subgrade compaction factors, and analyze the correlation between subgrade compaction factors and compaction degree;

[0018] S13. Divide the preprocessed historical data into a training set and a test set, use the training set to train the multi-modal fitting model, adjust the parameters of the multi-modal fitting model using the optimized loss function, and then use the test set for verification to obtain the optimized multi-modal fitting model.

[0019] Further, the steps for converting the processed subgrade compaction factors into input variables, sequentially establishing a multi-layer model structure to obtain a multi-modal fitting model, analyzing the correlation degree between subgrade compaction factors and compaction degree, extracting the effective features of subgrade compaction factors, and analyzing the correlation between subgrade compaction factors and compaction degree include:

[0020] S131. Use the subgrade type, gradation, moisture content, and compaction density included in the subgrade compaction factors as the input variables of the model to construct the input layer of the multi-modal fitting model;

[0021] S132. For different types of subgrade compaction factor data, introduce different feature extraction networks, analyze the correlation degree between the subgrade compaction factors and the compaction degree, and extract the data features of the subgrade compaction factors to construct the feature extraction layer of the multi-modal fitting model;

[0022] S133. Integrate the data features extracted from different sources to generate a comprehensive feature representation, analyze the correlation between the subgrade compaction factors and the compaction degree, and construct the fusion layer of the multi-modal fitting model;

[0023] S134. Based on the fused feature vectors, use the linear regression algorithm combined with a support vector machine to predict the compaction degree corresponding to various subgrade compaction factors, and construct the prediction layer of the multi-modal fitting model;

[0024] S135. Construct the output layer of the multi-modal fitting model to output the final compaction degree prediction result.

[0025] Further, for different types of subgrade compaction factor data, introduce different feature extraction networks, analyze the correlation degree between the subgrade compaction factors and the compaction degree, and the steps of extracting the data features of the subgrade compaction factors include the following:

[0026] S1321. According to the data types of the subgrade compaction factors, divide them into categorical variables, numerical variables, and continuous numerical variables, and convert them into input variables that meet the requirements of the multi-modal fitting model respectively;

[0027] S1322. Use the embedding layer network to process the subgrade compaction factors of the categorical variable type, use the fully connected layer network to process the subgrade compaction factors of the numerical variable type, and use the convolutional neural network to process the subgrade compaction factors of the continuous numerical variable type;

[0028] S1323. Use the introduced different feature extraction networks to analyze the relationship between the subgrade compaction factors and the compaction degree, and adjust the weights and biases using backpropagation optimization according to the correlation between the input variables and the target output to obtain the effective data features of the input variables for the compaction degree.

[0029] Further, the steps of integrating the data features extracted from different sources to generate a comprehensive feature representation and analyzing the correlation between the subgrade compaction factors and the compaction degree include the following:

[0030] S1331. Assign an attention mechanism to the feature vectors of the data features of the subgrade compaction factors obtained, and calculate the feature weight values of each data feature;

[0031] S1332. Adjust the contribution of each data feature according to the weight value, perform weighted summation, and obtain a fused feature vector as a comprehensive feature representation of the roadbed compaction factor.

[0032] Furthermore, based on the fused feature vectors, the linear regression algorithm of the combined vector machine is used to predict the compaction degree corresponding to various roadbed compaction factors, including the following steps:

[0033] S1341. Introducing the kernel function of the support vector machine, mapping the original feature space to a high-dimensional feature space through the kernel function, and fitting the nonlinear relationship between the roadbed compaction factor and the compaction degree in the high-dimensional feature space;

[0034] S1342: Setting a penalty parameter and a kernel width parameter of the kernel function, optimizing the kernel function parameters using a grid search method, obtaining the optimal penalty parameter and kernel width parameter, and establishing a compaction degree prediction formula based on the kernel function;

[0035] S1343. Use the training set established by historical data to train the compaction prediction formula to obtain the optimal feature vector and the optimal weight, and use the optimized compaction prediction formula to predict the compaction corresponding to various roadbed compaction factors.

[0036] Furthermore, the compaction degree prediction formula is:

[0037]

[0038] Where, Indicates the predicted value of compaction degree; α i and All represent Lagrange multipliers; b represents the bias term; K(x i ,x new ) represents the kernel function; x new Represents the feature vector of the new input; x i represents the i-th eigenvector; N represents the total number of eigenvectors.

[0039] Furthermore, the roadbed type before compaction construction and the detection data during compaction construction are collected and input into the multimodal fitting model to analyze and calculate the real-time compaction degree of the roadbed during the current construction process, including the following steps:

[0040] S21. Test the roadbed type and gradation of the area to be compacted through indoor particle gradation analysis and wet surface vibration impact tests.

[0041] S22. Using a frequency domain reflectometer sensor to transmit electromagnetic waves of a preset frequency to the compacted area, and by measuring the propagation and reflection characteristics of the electromagnetic waves in the soil, calculating the dielectric constant of the soil and determining the moisture content in the compacted area;

[0042] S23. Taking a single-degree-of-freedom linear elastic system as the theoretical model, and based on the superposition principle, multiple-level rigid mass bodies are attached to the rockfill soil vibration system, the natural vibration frequency of the system is measured, the dynamic stiffness and participating mass of the system are solved, and then converted into the compaction density of the rockfill body, which is used as the compaction degree of the compaction area.

[0043] Further, setting the compaction qualification index based on the specification standards of the current construction area, judging whether the current roller meets the compaction end condition, and feeding back the real-time compaction degree to the display device includes the following steps:

[0044] S31. Setting the compaction qualification index according to the specification standards of the current construction area, comparing the real-time compaction degree and test data collected during the compaction process with the compaction qualification index, and judging whether the current compaction area meets the specification standards;

[0045] S32. If the detected compaction degree meets the compaction qualification index, control the roller to continue the compaction construction according to the current working parameters, record the current real-time compaction degree and the roller control parameters, and at the same time feed back a prompt message to the operator, and there is no need to adjust the current compaction operation;

[0046] S33. If the detected compaction degree does not meet the compaction qualification index, trigger an alarm and take corrective measures to adjust the roller control parameters until the compaction area meets the compaction qualification index;

[0047] S34. Real-time display the real-time compaction degree of the compaction area and the measured values during the compaction process in the form of graphics, numerical values and sound signals.

[0048] Further, the compaction qualification index includes compaction degree, flatness, density, strength and uniformity; the roller control parameters include traveling speed, compaction frequency, amplitude, compaction times and working mode.

[0049] (III) Beneficial effects

[0050] Compared with the prior art, the present invention provides a real-time monitoring method for continuous compaction degree of subgrade based on artificial intelligence, which has the following beneficial effects: by collecting data transmitted from detection devices in real time, considering the influence of multiple factors such as subgrade type, gradation, and moisture content on the compaction degree, and adopting a multimodal fitting model to supervise and adjust the correlation of multimodal parameters, it is applied to the real-time analysis and viewing of the compaction degree in the continuous subgrade compaction construction, and timely feedback opinions are given to improve the timeliness and accuracy of the evaluation of subgrade compaction effect. According to the existing detection data results, test whether the multimodal fitting model is accurate, synchronize the tested multimodal fitting model to the roller control system. Before the roller and other detection devices work, input parameters such as the subgrade type, gradation, and moisture content measured before compaction into the multimodal fitting model, and then comprehensively analyze various detection data parameters collected during the compaction process, real-time feedback the compaction degree, and make timely control adjustments to the roller. The present invention solves the problems brought by the traditional evaluation of subgrade compaction effect, uses artificial intelligence algorithms to accelerate the convergence calculation speed to reach the real-time evaluation level, creates conditions for the continuous compaction of subgrade and the real-time evaluation of compaction degree, has timeliness and accuracy, and also greatly reduces the cost of manual detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0052] Figure 1 is a flowchart of a real-time monitoring method for continuous compaction degree of subgrade based on artificial intelligence according to an embodiment of the present invention;

[0053] Figure 2 is a schematic structural diagram of a subgrade compaction degree detection system applicable to a real-time monitoring method for continuous compaction degree of subgrade based on artificial intelligence according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0054] According to an embodiment of the present invention, a real-time monitoring method for continuous compaction degree of subgrade based on artificial intelligence is provided.

[0055] Now, the present invention will be further described in combination with the drawings and specific embodiments. As Figure 1 shown, a real-time monitoring method for continuous compaction degree of subgrade based on artificial intelligence according to an embodiment of the present invention includes the following steps:

[0056] S1. Based on the correlation between subgrade compaction factors and compaction degree, construct a multi-modal fitting model, verify the fitting multi-modal fitting model using existing historical data, and optimize the constructed multi-modal fitting model.

[0057] In the description of the present invention, based on the correlation between subgrade compaction factors and compaction degree, constructing a multi-modal fitting model, verifying the fitting multi-modal fitting model using existing historical data, and optimizing the constructed multi-modal fitting model include the following steps:

[0058] S11. Obtain the historical data of subgrade compaction factors and the corresponding compaction degree, preferably perform missing value processing, and after data format unification and normalization processing, obtain the standardized data of subgrade compaction factors. The subgrade compaction factors include subgrade type, gradation, moisture content, and compaction density.

[0059] Specifically, the main influencing factors considered in the multi-modal fitting model are: subgrade type, gradation, moisture content, and compaction density. The subgrade type and gradation can be obtained before compaction construction. The subgrade type and gradation reflect the differences in the scenarios and rollers used to cope with different compaction environments; the moisture content and compaction density reflect the compaction information detected in real time during compaction construction; use the influencing factors established by this multi-modal fitting model to adjust the correlation of multi-modal parameters. The output method of the multi-modal fitting model is to be able to express the mechanical performance index of compaction, thus achieving the effect of artificial intelligence real-time evaluation of compaction degree.

[0060] S12. Convert the processed subgrade compaction factors into input variables, successively establish a multi-layer model structure to obtain a multi-modal fitting model, analyze the correlation degree between subgrade compaction factors and compaction degree, extract the effective features of subgrade compaction factors, and analyze the correlation between subgrade compaction factors and compaction degree.

[0061] S13. Divide the preprocessed historical data into a training set and a test set, use the training set to train the multi-modal fitting model, use the optimized loss function to adjust the parameters of the multi-modal fitting model, and then use the test set for verification to obtain the optimized multi-modal fitting model.

[0062] Specifically, the working principle of this multi-modal fitting model is that after a series of input variables (i.e., the factors affecting compaction degree) are given, by learning the mapping relationship between these variables and compaction degree, it predicts the possible compaction degree level under new construction conditions. Specifically, the model will try to capture the influence degree of each input variable on the final compaction degree and abstract these relationships into mathematical formulas or function expressions. When new construction conditions are input, the model can quickly calculate the expected compaction degree value.

[0063] In the description of the present invention, the processed subgrade compaction factors are converted into input variables, a multi-layer model structure is established in sequence to obtain a multi-modal fitting model, the correlation degree between the subgrade compaction factors and the degree of compaction is analyzed, the effective features of the subgrade compaction factors are extracted, and the correlation between the subgrade compaction factors and the degree of compaction includes the following steps:

[0064] S131. Taking the subgrade type, gradation, moisture content and compaction density included in the subgrade compaction factors as the model input variables, an input layer of the multi-modal fitting model is constructed.

[0065] Specifically, it is necessary to collect data on key parameters such as subgrade type, gradation, and moisture content in real time. These data can be obtained through various means such as sensors and laboratory tests. Ensure that the input data format is consistent with that during training, which includes:

[0066] 1. Missing value processing: If some data are missing, methods such as interpolation or mean filling can be used to complete them; Standardization / normalization: If the data were standardized or normalized during training, the new data also need to be processed in the same way during prediction. 2. Organize the preprocessed data into the input format required by the model, usually a multi-dimensional array or tensor. Pass the input tensor to the trained model for forward propagation. 3. The model will calculate the corresponding predicted compaction degree value based on the input data. This value is usually a floating point number representing the expected compaction percentage. If the model output has undergone a certain transformation (such as scaling, logarithmic transformation, etc.), it is necessary to perform the reverse operation to restore it to the compaction degree value in the original unit. 4. Check whether the predicted compaction degree is reasonable and meets the engineering standards and construction requirements. If possible, compare the predicted value with the actual measured value and analyze the error reasons, such as data acquisition error, model deviation, etc. Adjust the construction parameters according to the prediction results, such as increasing or decreasing the number of rolling passes, adjusting the moisture content, etc., to achieve the best compaction effect. Display the prediction results in the form of charts, such as time series charts, scatter plots, etc., for intuitive understanding and reporting.

[0067] S132. For different types of subgrade compaction factor data, different feature extraction networks are introduced to analyze the correlation degree between the subgrade compaction factors and the degree of compaction, and the data features of the subgrade compaction factors are extracted to construct a feature extraction layer of the multi-modal fitting model.

[0068] In the description of the present invention, for different types of subgrade compaction factor data, different feature extraction networks are introduced to analyze the correlation degree between the subgrade compaction factors and the degree of compaction, and the data features of the subgrade compaction factors include the following steps:

[0069] S1321. According to the data types of the subgrade compaction factors, they are divided into categorical variables, numerical variables and continuous numerical variables, and are respectively converted into input variables that meet the requirements of the multi-modal fitting model.

[0070] Specifically, 1. The subgrade type is usually a categorical variable (e.g., sandy soil, clay, gravel, etc.). It needs to be converted into an input format suitable for the model through an Embedding layer or One-Hot Encoding. 2. The gradation is generally a numerical feature, representing the particle distribution of different particle sizes in the subgrade (e.g., maximum particle size, minimum particle size, uniformity coefficient). It can be processed by normalization or standardization to make it suitable for network input. 3. The moisture content is also a numerical feature, representing the soil moisture content. Similar to the gradation, it needs to be standardized. 4. The compaction density is a continuous numerical variable, representing the degree of soil compaction, which is usually directly related to the moisture content and the compaction process. It can be standardized.

[0071] S1322. Use the Embedding layer network to process the subgrade compaction factors of the categorical variable type, use the fully connected layer network to process the subgrade compaction factors of the numerical variable type, and use the convolutional neural network to process the subgrade compaction factors of the continuous numerical variable type.

[0072] S1323. Use the introduced different feature extraction networks to analyze the relationship between the subgrade compaction factors and the compaction degree. According to the association between the input variables and the target output, use backpropagation optimization to adjust the weights and biases to obtain the effective data features of the input variables for the compaction degree.

[0073] S133. Integrate the data features extracted from different sources to generate a comprehensive feature representation, analyze the correlation between the subgrade compaction factors and the compaction degree, and construct the fusion layer of the multimodal fitting model.

[0074] In the description of the present invention, integrating the data features extracted from different sources to generate a comprehensive feature representation and analyzing the correlation between the subgrade compaction factors and the compaction degree includes the following steps:

[0075] S1331. Assign an attention mechanism to the feature vectors of the data features of the subgrade compaction factors extracted, and calculate the feature weight values of each data feature.

[0076] S1332. Adjust the contributions of each data feature according to the weight values, and through weighted summation, obtain the fused feature vector as the comprehensive feature representation of the subgrade compaction factors.

[0077] S134. Based on the fused feature vector, adopt a linear regression algorithm combined with a support vector machine to predict the compaction degrees corresponding to various subgrade compaction factors, and construct the prediction layer of the multimodal fitting model.

[0078] In the description of the present invention, based on the fused feature vectors, a linear regression algorithm combined with a support vector machine is used to predict the degree of compaction corresponding to various subgrade compaction factors, including the following steps:

[0079] S1341. Introduce the kernel function of the support vector machine, map the original feature space to a high-dimensional feature space through the kernel function, and fit the non-linear relationship between the subgrade compaction factors and the degree of compaction in the high-dimensional feature space.

[0080] S1342. Set the penalty parameter and kernel width parameter of the kernel function, optimize the kernel function parameters by means of grid search to obtain the optimal penalty parameter and kernel width parameter, and establish a compaction degree prediction formula based on the kernel function.

[0081] S1343. Use the training set established by historical data to train the compaction degree prediction formula to obtain the optimal feature vectors and optimal weights, and use the optimized compaction degree prediction formula to predict the degree of compaction corresponding to various subgrade compaction factors.

[0082] In the description of the present invention, the compaction degree prediction formula is:

[0083]

[0084] In the formula, represents the predicted value of the compaction degree; α i and both represent Lagrange multipliers; b represents the bias term; K(x i , x new ) represents the kernel function; x new represents the newly input feature vector; x i represents the i-th feature vector. N represents the total number of feature vectors.

[0085] S135. Construct the output layer of the multi-modal fitting model to output the final compaction degree prediction result.

[0086] S2. Collect the subgrade type before compaction construction and the detection data during the compaction construction, input them into the multi-modal fitting model, and analyze and calculate the real-time compaction degree during the current subgrade construction process.

[0087] In the description of the present invention, collecting the subgrade type before compaction construction and the detection data during the compaction construction, inputting them into the multi-modal fitting model, and analyzing and calculating the real-time compaction degree during the current subgrade construction process include the following steps:

[0088] S21. Detect the subgrade type and gradation of the area to be compacted through indoor particle size distribution analysis experiments and wet surface vibration compaction experiments.

[0089] S22. Use a frequency domain reflectometer sensor to emit electromagnetic waves with a preset frequency into the compacted area, calculate the dielectric constant of the soil by measuring the propagation and reflection characteristics of the electromagnetic waves in the soil, and determine the moisture content in the compacted area.

[0090] S23. Take a single-degree-of-freedom linear elastic system as the theoretical model, and according to the superposition principle, attach multiple rigid mass bodies to the vibrating system of the rockfill soil, measure the natural vibration frequency of the system, solve the dynamic stiffness and participating mass of the system, and convert them into the compaction density of the rockfill body as the compaction degree of the compacted area.

[0091] S3. Set the compaction qualification index based on the specification standards of the current construction area, judge whether the current roller meets the compaction end condition, and feedback the real-time compaction degree to the display device.

[0092] In the description of the present invention, setting the compaction qualification index based on the specification standards of the current construction area, judging whether the current roller meets the compaction end condition, and feedbacking the real-time compaction degree to the display device includes the following steps:

[0093] S31. Set the compaction qualification index according to the specification standards of the current construction area, compare the real-time compaction degree and test data collected during the compaction process with the compaction qualification index, and judge whether the current compacted area meets the specification standards.

[0094] In the description of the present invention, the compaction qualification index includes compaction degree, flatness, density, strength and uniformity. The roller control parameters include traveling speed, compaction frequency, amplitude, compaction times and working mode.

[0095] S32. If the detected compaction degree meets the compaction qualification index, control the roller to continue the compaction construction according to the current working parameters, record the current real-time compaction degree and the roller control parameters, and at the same time feedback a prompt message to the operator without adjusting the current compaction operation.

[0096] S33. If the detected compaction degree does not meet the compaction qualification index, trigger an alarm and take corrective measures to adjust the roller control parameters until the compacted area meets the compaction qualification index.

[0097] S34. Real-time display the real-time compaction degree of the compacted area and the measured values during the compaction process in the form of graphics, numerical values and sound signals.

[0098] Specifically, it can be synchronized with the roller's compaction degree information in real time to determine whether the current compaction degree meets the requirements, and then control how the roller conducts compaction construction. If the compaction degree detected in real time meets the preset standard, the roller can continue compaction construction according to the current working parameters (such as speed, frequency, etc.). The system can record the current compaction degree data and compaction construction parameters for subsequent analysis and reference. The system can send a prompt message to the operator, informing that the current compaction degree meets the requirements and no adjustment is needed. If it does not meet the requirements, that is, the compaction degree detected in real time is lower than the preset standard, the system can issue an alarm to remind the operator that the current compaction degree does not meet the requirements and measures need to be taken for adjustment. The system can automatically or prompt the operator to adjust the working parameters of the roller, such as increasing the compaction times, raising the compaction frequency, or changing the compaction speed. The system can instruct the roller to re-compact the current area until the compaction degree reaches the requirements. The system can record and mark the areas where the compaction degree does not meet the requirements for subsequent key treatment.

[0099] By comprehensively analyzing the actual operation situation of subgrade construction, the real-time subgrade compaction indicators in the entire rolling work surface area are evaluated and analyzed together with the on-site subgrade rolling construction quality results. The subgrade compaction indicators mainly include the following aspects: compaction degree, flatness, density, strength, uniformity, etc.; the construction quality results mainly include: compaction degree compliance rate, flatness qualification rate, density uniformity, strength test results, construction efficiency, etc. The specific evaluation process: collect the sensor data, video monitoring data, and environmental data of the roller. Remove invalid or incorrect data, fuse multi-source data to form comprehensive construction status information. Monitor the compaction degree, flatness, density, and strength of the subgrade in real time, compare the real-time data with the preset standard, and judge whether the current construction meets the requirements. If it does not meet the requirements, adjust the working parameters of the roller in time or re-compact. Store the real-time monitoring data in the database for subsequent analysis. Use machine learning algorithms to analyze the data to identify key construction parameters and patterns. Generate a construction report, recording the compaction degree, flatness, density, and strength data of each area. According to the construction report, evaluate the overall construction quality, calculate the compaction degree compliance rate, flatness qualification rate, density uniformity, and strength test results. For the problem areas, propose improvement measures and optimize the construction plan.

[0100] Save the roller control parameters, subgrade parameters, real-time results calculated by the multi-modal fitting model, and data obtained by the loading detection equipment, as well as the compaction degree data sampled after construction for subsequent analysis. Roller control parameters include: speed: the traveling speed of the roller; frequency: the vibration frequency of the roller; amplitude: the vibration amplitude of the roller; compaction times: the number of compaction passes in each area; working mode: the working mode of the roller (such as static pressure, vibration, etc.). Subgrade parameters refer to the physical and chemical properties of the subgrade itself, which reflect the properties of the subgrade materials and specifically include: soil type, gradation moisture content, initial density, maximum dry density, optimum moisture content, etc. The real-time results calculated by the multi-modal fitting model refer to the compaction degree data calculated in real time through mathematical models and algorithms, including: the compaction degree of the current compaction area, the change trend of the compaction degree during compaction (such as rising, falling, etc.), the predicted final compaction degree based on the current parameters, etc. The data obtained by the loading detection equipment refers to the data collected in real time through various sensors and detection equipment installed on the roller, specifically including: the pressure data of the roller on the subgrade, the acceleration data of the roller, the displacement data of the roller, the temperature data of the construction environment, the humidity data of the construction environment, etc.

[0101] S4. When the compaction degree of all compaction areas reaches the qualified compaction parameters, control the roller to end compaction, record and save the data results after compaction, as well as the roller control parameters and subgrade parameters corresponding to this compaction degree.

[0102] In addition, as Figure 2 shown, the framework structure diagram of the detection system constituted by the present invention is presented, mainly including the subgrade, loading detection equipment, and data analysis equipment. The numbers in the figure indicate the data transmission paths between each module, involving a total of seven transmission paths for realizing the real-time detection of the subgrade compaction degree.

[0103] In summary, by means of the above technical solution of the present invention, by collecting the data transmitted by the detection device in real time, considering the influence of multiple factors such as subgrade type, gradation, and moisture content on the compaction degree, and adopting a multimodal fitting model to supervise and adjust the correlation of multimodal parameters, it is applied to the real-time analysis and viewing of the compaction degree in the continuous subgrade compaction construction, and timely feedback opinions are given to improve the timeliness and accuracy of the subgrade compaction effect evaluation. According to the existing detection data results, test whether the multimodal fitting model is accurate, and synchronize the tested multimodal fitting model to the roller control system. Before the roller and other detection devices work, input the parameters such as the subgrade type, gradation, and moisture content measured before compaction into the multimodal fitting model. Subsequently, comprehensively analyze various detection data parameters collected during the compaction process, feedback the compaction degree in real time, and make timely control adjustments to the roller. The present invention solves the problems brought by the traditional subgrade compaction effect evaluation, uses artificial intelligence algorithms to accelerate the convergence calculation speed to reach the real-time evaluation level, creates conditions for the continuous compaction of the subgrade and the establishment of real-time evaluation of the compaction degree, has timeliness and accuracy, and also greatly reduces the manual detection cost.

[0104] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A real-time monitoring method for continuous compaction degree of subgrade based on artificial intelligence, characterized in that, The method includes the following steps: S1. Based on the correlation between subgrade compaction factors and compaction degree, construct a multi-modal fitting model, verify and fit the multi-modal fitting model using existing historical data, and optimize the constructed multi-modal fitting model; S2. Collect the subgrade type before compaction construction and the detection data during the compaction construction, input them into the multi-modal fitting model, and analyze and calculate the real-time compaction degree during the current subgrade construction process; S3. Set the compaction qualification index based on the specification standards of the current construction area, judge whether the current roller meets the compaction end condition, and feedback the real-time compaction degree to the display device; S4. When the compaction degrees of all compaction areas reach the compaction qualification parameters, control the roller to end compaction, record and save the data results after compaction, as well as the roller control parameters and subgrade parameters corresponding to the compaction degree.

2. The real-time monitoring method for continuous compaction degree of roadbed based on artificial intelligence according to claim 1, wherein The step of constructing a multi-modal fitting model based on the correlation between subgrade compaction factors and compaction degree, verifying and fitting the multi-modal fitting model using existing historical data, and optimizing the constructed multi-modal fitting model includes the following steps: S11. Obtain the historical data of subgrade compaction factors and corresponding compaction degrees, give priority to handling missing values, and after unified data format and normalization processing, obtain the standardized data of subgrade compaction factors. The subgrade compaction factors include subgrade type, gradation, moisture content, and compaction density; S12. Convert the processed subgrade compaction factors into input variables, successively establish a multi-layer model structure to obtain a multi-modal fitting model, analyze the correlation degree between subgrade compaction factors and compaction degree, extract the effective features of the subgrade compaction factors, and analyze the correlation between the subgrade compaction factors and compaction degree; S13. Divide the preprocessed historical data into a training set and a test set, train the multi-modal fitting model using the training set, adjust the parameters of the multi-modal fitting model using an optimized loss function, and then verify using the test set to obtain an optimized multi-modal fitting model.

3. The real-time monitoring method for the continuous compaction degree of subgrade based on artificial intelligence according to claim 2, characterized in that, The step of converting the processed subgrade compaction factors into input variables, successively establishing a multi-layer model structure to obtain a multi-modal fitting model, analyzing the correlation degree between subgrade compaction factors and compaction degree, extracting the effective features of the subgrade compaction factors, and analyzing the correlation between the subgrade compaction factors and compaction degree includes the following steps: S131. Use the subgrade type, gradation, moisture content, and compaction density included in the subgrade compaction factors as the input variables of the model to construct the input layer of the multi-modal fitting model; S132. For different types of subgrade compaction factor data, introduce different feature extraction networks, analyze the correlation degree between subgrade compaction factors and compaction degree, and extract the data features of the subgrade compaction factors to construct the feature extraction layer of the multi-modal fitting model; S133. Integrate the data features extracted from different sources to generate a comprehensive feature representation, analyze the correlation between the subgrade compaction factors and compaction degree, and construct the fusion layer of the multi-modal fitting model; S134. Based on the fused feature vectors, adopt a linear regression algorithm combining a support vector machine to predict the compaction degrees corresponding to various subgrade compaction factors, and construct the prediction layer of the multi-modal fitting model. S135. Construct the output layer of the multimodal fitting model to output the final compaction degree prediction result.

4. The real-time monitoring method for continuous compaction degree of subgrade based on artificial intelligence according to claim 3, characterized in that, For the data of different types of subgrade compaction factors, different feature extraction networks are introduced to analyze the correlation degree between the subgrade compaction factors and the compaction degree, and the data features of the subgrade compaction factors are extracted, including the following steps: S1321. Divide the data of subgrade compaction factors into categorical variables, numerical variables and continuous numerical variables according to the data type, and convert them into input variables that meet the requirements of the multimodal fitting model respectively. S1322. Use the embedding layer network to process the subgrade compaction factors of the categorical variable type, use the fully connected layer network to process the subgrade compaction factors of the numerical variable type, and use the convolutional neural network to process the subgrade compaction factors of the continuous numerical variable type. S1323. Use the introduced different feature extraction networks to analyze the relationship between the subgrade compaction factors and the compaction degree. According to the correlation between the input variables and the target output, use backpropagation optimization to adjust the weights and biases to obtain the effective data features of the input variables for the compaction degree.

5. The real-time monitoring method for continuous compaction degree of subgrade based on artificial intelligence according to claim 3, characterized in that, The data features from different sources are fused and extracted to generate a comprehensive feature representation. Analyzing the correlation between the subgrade compaction factors and the compaction degree includes the following steps: S1331. Assign an attention mechanism to the feature vectors of the data features of the subgrade compaction factors obtained, and calculate the feature weight values of each data feature. S1332. Adjust the contribution of each data feature according to the weight values, and through weighted summation, obtain the fused feature vector as the comprehensive feature representation of the subgrade compaction factors.

6. The real-time monitoring method for continuous compaction degree of roadbed based on artificial intelligence according to claim 3, characterized in that, Based on the fused feature vector, using the linear regression algorithm combined with the support vector machine to predict the compaction degree corresponding to various subgrade compaction factors includes the following steps: S1341. Introduce the kernel function of the support vector machine, and map the original feature space to a high-dimensional feature space through the kernel function to fit the non-linear relationship between the subgrade compaction factors and the compaction degree in the high-dimensional feature space. S1342. Set the penalty parameter and kernel width parameter of the kernel function, use the grid search method to optimize the kernel function parameters, obtain the optimal penalty parameter and kernel width parameter, and based on the kernel function, establish a compaction degree prediction formula. S1343. Use the training set established by historical data to train the compaction degree prediction formula to obtain the optimal feature vector and optimal weight, and use the optimized compaction degree prediction formula to predict the compaction degree corresponding to various subgrade compaction factors.

7. A real-time monitoring method for continuous compaction degree of subgrade based on artificial intelligence according to claim 6, characterized in that, The compaction degree prediction formula is: In the formula, represents the predicted value of the compaction degree; α i and both represent Lagrange multipliers; b represents the bias term; K(x i , x new ) represents the kernel function; x new represents the newly input feature vector; x i represents the i-th feature vector; N represents the total number of feature vectors.

8. A real-time monitoring method for continuous compaction degree of subgrade based on artificial intelligence according to claim 1, characterized in that, Collect the subgrade type before compaction construction and the detection data during the compaction construction process, input them into the multimodal fitting model, and analyze and calculate the real-time compaction degree during the current construction process of the subgrade, including the following steps: S21. Detect the subgrade type and gradation of the area to be compacted through indoor particle size analysis experiments and wet surface vibration compaction experiments. S22. Use the frequency domain reflectometer sensor to emit electromagnetic waves with a preset frequency to the compaction area, and calculate the dielectric constant of the soil by measuring the propagation and reflection characteristics of the electromagnetic waves in the soil, and determine the moisture content in the compaction area. S23. Taking a single-degree-of-freedom linear elastic system as the theoretical model, according to the superposition principle, multiple levels of rigid mass bodies are attached to the rockfill soil vibration system, the natural vibration frequency of the system is measured, the dynamic stiffness and the participating mass of the system are solved, and then converted into the compaction density of the rockfill body, which is used as the compaction degree of the compaction area.

9. The real-time monitoring method for continuous compaction degree of subgrade based on artificial intelligence according to claim 1, characterized in that The steps of setting the compaction qualification index based on the specification standards of the current construction area, judging whether the current roller meets the compaction end condition, and feeding back the real-time compaction degree to the display device include the following: S31. Set the compaction qualification index according to the specification standards of the current construction area, compare the real-time compaction degree and the detection data collected during the compaction process with the compaction qualification index, and judge whether the current compaction area meets the specification standards; S32. If the detected compaction degree meets the compaction qualification index, control the roller to continue the compaction construction according to the current working parameters, record the current real-time compaction degree and the roller control parameters, and at the same time feed back a prompt message to the operator, and there is no need to adjust the current compaction operation; S33. If the detected compaction degree does not meet the compaction qualification index, trigger an alarm and take corrective measures to adjust the roller control parameters until the compaction area meets the compaction qualification index; S34. Real-time display the real-time compaction degree of the compaction area and the measured values during the compaction process in the form of graphics, numerical values and sound signals.

10. A real-time monitoring method for continuous compaction degree of subgrade based on artificial intelligence according to claim 9, characterized in that, The compaction qualification index includes compaction degree, flatness, density, strength and uniformity; the roller control parameters include traveling speed, compaction frequency, amplitude, compaction times and working mode.

Images