An asphalt pavement compaction degree monitoring system based on artificial intelligence

Abstract

The application discloses an asphalt pavement compactness monitoring system based on artificial intelligence, and relates to the technical field of road construction quality monitoring.The system collects pavement images, temperature, vibration acceleration and base type and other working condition data, extracts effective image blocks from the images and obtains an initial feature matrix, and the core of the system is that the system searches historical similar working conditions based on temperature and vibration data, constructs a target function containing a distance constraint term to cooperatively optimize the feature matrix, obtains initial matrix feature coefficients which can more accurately represent the compaction state, then the system matches the working condition interval according to the temperature and the base type, selects corresponding compactness calculation logic in combination with the proportion of effective image blocks, fuses the initial matrix feature coefficients and the vibration acceleration to obtain the compactness, finally, the system generates a dynamic compactness interval through calculation of working condition drift compensation, compares the compactness with the interval to make an abnormality judgment, and realizes intelligent and self-adaptive monitoring of the compaction quality.

Description

Technical Field

[0001] This invention relates to the field of road construction quality monitoring technology, and in particular to an artificial intelligence-based asphalt pavement compaction monitoring system. Background Technology

[0002] The compaction degree of asphalt pavement is a key quality indicator that determines the service life, load-bearing capacity and driving comfort of the pavement. Insufficient compaction will lead to loose pavement and potholes; excessive compaction may damage the aggregate structure and also affect the pavement performance. Therefore, it is crucial to monitor the compaction degree in real time and accurately during construction.

[0003] To address the aforementioned issues, some intelligent monitoring and control systems have emerged in the prior art. For example, Chinese Patent CN120669702A discloses an unmanned asphalt pavement compaction system and intelligent control construction method for urban renewal. This solution integrates multiple sensors, such as infrared temperature, ambient temperature, steel wheel temperature, compaction degree, and lidar, onto the unmanned road roller, achieving multi-source data perception at the construction site. Its core idea is to collect standard data under ideal working conditions in advance on a test section, establishing a standard state matrix for temperature and a standard state vector for compaction degree. During actual construction, the system compares the real-time collected data with these preset standard states and dynamically adjusts the road roller's travel speed, excitation force, and compaction path based on the differences between the two using a simple linear formula.

[0004] However, this technical solution still has some shortcomings: First, its dynamic adjustment logic relies on a preset standard state and uses a simple linear formula for parameter correction. This control method based on a single ideal model has poor adaptability when facing complex and ever-changing actual working conditions, making it difficult to accurately match the nonlinear compaction law, which may lead to insufficient control accuracy. Second, this solution directly uses the raw data collected by the sensors for control calculations, lacking effective feature extraction and optimization processing of the data. This makes the system more sensitive to interference such as sensor noise and sudden environmental changes, affecting the stability and reliability of the monitoring results. Finally, the core of its quality control is to make the actual state approach the preset standard state, but it does not establish an adaptive compaction quality evaluation standard. When the working conditions undergo systematic drift, the fixed standard state may no longer be applicable, leading to misjudgment of the compaction quality. Therefore, the existing technology still needs further improvement in order to obtain an asphalt pavement compaction monitoring solution with higher accuracy, stronger adaptability, and stronger robustness under complex working conditions. Summary of the Invention

[0005] The technical problems solved by this invention are: First, existing technologies based on a single ideal model have poor adaptability to complex and ever-changing actual working conditions, making it difficult to accurately match the nonlinear compaction law, which may lead to insufficient control accuracy. Second, existing technologies lack effective feature extraction and optimization processing of data, making the system more sensitive to interference such as sensor noise and sudden environmental changes, affecting the stability and reliability of monitoring results. Finally, the core of existing quality control technology is to make the actual state approach the preset standard state, but it has not established an adaptive compaction quality evaluation standard. When the working conditions drift systematically, the fixed standard state may no longer be applicable, leading to misjudgment of compaction quality.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: an artificial intelligence-based asphalt pavement compaction monitoring system, comprising a data acquisition module, a feature extraction module, a compaction calculation module, and a judgment module;

[0007] The acquisition module is used to collect operating condition data under the current operating conditions;

[0008] The feature extraction module is used to extract an initial feature matrix from the road surface image of the working condition data, and to obtain the initial matrix feature coefficients of the optimized road surface image by optimizing the initial feature matrix with a distance constraint term.

[0009] The compaction degree calculation module is used to match the working condition interval based on the working condition data, and select the corresponding compaction degree calculation logic according to the proportion of effective image blocks to obtain the compaction degree under the current working condition.

[0010] The determination module is used to generate a dynamic compaction degree range based on road surface temperature, vibration acceleration and initial matrix characteristic coefficients through multi-parameter distance weighted fusion and normalization processing, and to make anomaly determination based on the comparison result between the compaction degree and the range.

[0011] Preferably, the operating condition data includes road surface images, road surface temperature, vibration acceleration, base course type, and base course type code;

[0012] The base layer types include semi-rigid and flexible. The base layer type code corresponding to the semi-rigid base layer is 1, and the base layer type code corresponding to the flexible base layer is 0.

[0013] Preferably, extracting the initial feature matrix specifically includes:

[0014] The road surface image is uniformly divided into K 3×3 non-overlapping image blocks, where K is a perfect square. The gray-level entropy of each image block is obtained. If the gray-level entropy is greater than or equal to the effective threshold, the image block is marked as a valid image block.

[0015] The (2D)²PCA method is used to reduce the dimensionality of each valid image block to obtain an M×M dimension reduced image block, where M is a natural number less than K. The covariance matrix of all reduced image blocks is obtained, and the covariance matrix is ​​positive definite to obtain the initial feature matrix of the current working condition.

[0016] Preferably, obtaining the initial matrix feature coefficients of the optimized road surface image specifically includes: calculating the bulldozer distance between the initial feature matrix of the current working condition and the initial feature matrix of historical similar working conditions, constructing an objective function and using the bulldozer distance as a constraint term to optimize the initial feature matrix and obtain the initial matrix feature coefficients;

[0017] Obtain the initial feature matrix corresponding to historical similar working conditions, and calculate the bulldozer distance between the distribution of the initial feature matrix of the current working condition and the distribution of the initial feature matrix of each historical similar working condition. The mathematical expression for the bulldozer distance is:

[0018] ;

[0019] in, The distance between the kth valid image block and the ith historically similar work condition bulldozer;

[0020] The distribution of the initial feature matrix for the current operating condition;

[0021] The distribution of the initial feature matrix for historically similar working conditions;

[0022] Indicates the infimum, Represents the joint probability distribution;

[0023] for and The joint distribution set;

[0024] Indicates in joint distribution The expected value of the following;

[0025] It is a single initial feature matrix sample drawn from the current distribution. It is a single initial feature matrix sample extracted from the historical distribution;

[0026] The historical similar working conditions include: constructing a current working condition data vector based on the road surface temperature, vibration acceleration, and base course type code in the working condition data; using the K-nearest neighbor algorithm to search the historical working condition database; and selecting several historical data samples that are closest to the current working condition data vector in Euclidean distance as historical similar working conditions.

[0027] Preferably, the mathematical expression of the objective function is:

[0028] ;

[0029] in, Used to collaboratively optimize the current working condition and similar historical working conditions, and obtain characteristic coefficients that characterize the current compaction state;

[0030] It is a matrix composed of the initial feature matrices of all valid image blocks in the current working condition;

[0031] It is a matrix composed of the initial feature matrices of all valid image patches in historically similar working conditions;

[0032] The historical feature coefficients are the least squares solutions of the historical feature matrix to the texture grayscale vectors in historically similar working conditions.

[0033] The texture grayscale vector for the current working condition is obtained by: expanding the 3×3 grayscale values ​​of all valid image blocks into a 9-dimensional vector in row order, and aggregating the mean of the 9-dimensional vectors of all valid image blocks to obtain the texture grayscale vector for the current working condition. The dimension of the texture grayscale vector is M.

[0034] The censored likelihood term is the matching probability between the current feature coefficients and the mean of historical feature coefficients obtained based on the Gaussian kernel function.

[0035] The bulldozer distance constraint term is the average distance between all valid image blocks under the current working condition and the bulldozer distance under similar historical working conditions.

[0036] Given the current feature coefficients, iteratively solve the objective function until convergence to obtain the current feature coefficients for each valid image block;

[0037] The initial matrix feature coefficients are the average of the current feature coefficients of all valid image blocks.

[0038] Preferably, the censored likelihood term includes: extracting the feature coefficients of all valid image blocks in historical similar working conditions, using the mean of the historical feature coefficients as the benchmark value of the normal feature coefficients, substituting the feature coefficients of each valid image block in the current working condition into the Gaussian kernel function to obtain the likelihood value, and summing the logarithmic likelihood values ​​of all valid image blocks in the current working condition to obtain the censored likelihood term.

[0039] Preferably, the working condition intervals are divided based on the pavement temperature and base course type of the current working condition data, including a first working condition interval, a second working condition interval, a third working condition interval, and a fourth working condition interval, with the following division rules:

[0040] If the temperature is less than or equal to the temperature threshold and the base layer type is semi-rigid, then it is classified as the first working condition range;

[0041] If the temperature is greater than the temperature threshold and the base layer type is semi-rigid, then it is classified as the second working condition range.

[0042] If the temperature is less than or equal to the temperature threshold and the base layer type is flexible, then it is classified as the third working condition range;

[0043] If the temperature is greater than the temperature threshold and the base layer type is flexible, then it is classified as the fourth working condition zone.

[0044] The compaction calculation logic specifically includes: calculating the proportion of the number of effective image blocks to the total number of blocks in the road surface image as the proportion of effective image blocks;

[0045] If the proportion of the effective image block is greater than or equal to the effective threshold, the final compaction degree is obtained directly based on the matched working condition interval and the compaction degree function corresponding to the working condition interval.

[0046] If the proportion of the effective image blocks is less than the effective threshold, the compaction degree calculated under the current working condition is combined with the average compaction degree of the previous N collected working conditions to obtain the final compaction degree, where N is a natural number greater than 2.

[0047] The compaction function includes a first working condition interval compaction function, a second working condition interval compaction function, a third working condition interval compaction function, and a fourth working condition interval compaction function.

[0048] Preferably, the compaction function of the first working condition interval, the compaction function of the second working condition interval, the compaction function of the third working condition interval, and the compaction function of the fourth working condition interval are constructed based on the eigencoefficients of the initial matrix and the vibration acceleration;

[0049] The compaction function of the first working condition interval and the compaction function of the third working condition interval are linear fitting functions. The initial matrix characteristic coefficients and vibration acceleration of similar historical working conditions are used as independent variables, and the actual compaction of similar historical working conditions is used as dependent variable. The compaction function is obtained by linear fitting of the first term using the least squares method.

[0050] The compaction function of the second working condition interval and the compaction function of the fourth working condition interval are quadratic polynomial fitting functions. The initial matrix characteristic coefficients and vibration acceleration of similar historical working conditions are used as independent variables, and the actual compaction of similar historical working conditions is used as dependent variable. The compaction function is obtained by fitting a quadratic polynomial using the least squares method.

[0051] Preferably, generating the dynamic compaction interval specifically includes:

[0052] Based on the current working condition data of road surface temperature, vibration acceleration and initial matrix eigenvalues, the normalized distances of road surface temperature, vibration acceleration and initial matrix eigenvalues ​​with those of similar historical working conditions are calculated as the third distance;

[0053] The third distance is weighted and fused based on the variance weights of road surface temperature, vibration acceleration, and initial matrix characteristic coefficients in historical working conditions to obtain the fourth distance;

[0054] Based on the mean and standard deviation of the fourth distance under historical operating conditions, the fourth distance under the current operating condition is normalized, and the drift compensation factor is obtained through the linear rectification function.

[0055] Based on the preset benchmark interval and the drift compensation factor, the benchmark interval is scaled proportionally to generate a dynamic compaction interval.

[0056] Preferably, the anomaly determination specifically includes:

[0057] If the final compaction degree under the current working conditions is within the dynamic compaction degree range, it is considered a normal situation.

[0058] If the final compaction degree under the current working conditions is not within the dynamic compaction degree range, it is judged as an abnormal situation.

[0059] The beneficial effects of this invention are: This invention introduces historical similar working condition data for collaborative optimization. By constructing an objective function that includes distance constraints, the system can dynamically calibrate and optimize the feature matrix extracted from the image, effectively overcoming the working condition drift problem caused by working condition fluctuations such as temperature and vibration. This makes the feature coefficients of the final initial matrix more accurately and stably reflect the real compaction state, greatly improving the accuracy and anti-interference ability of the monitoring results.

[0060] Secondly, by calculating the working condition drift compensation factor, the present invention dynamically scales the preset benchmark compaction range to generate a dynamic compaction range that can respond to changes in working conditions in real time. This adaptive judgment mechanism makes the abnormal standard more scientific and reasonable, which not only avoids misjudging qualified results under harsh working conditions, but also ensures timely detection of quality problems under ideal working conditions, significantly improving the intelligence level and practicality of the monitoring system.

[0061] Finally, the system integrates multi-dimensional data such as road surface images, temperature, vibration acceleration, and base course type, comprehensively considering key factors affecting compaction quality. Through effective image block proportion filtering and layered calculation logic, the system can intelligently integrate historical data when image quality is poor, ensuring the continuity and reliability of calculation results. This multi-modal data fusion and fault-tolerant processing mechanism ensures the stable operation of the system in complex construction sites, providing a more comprehensive and reliable quality assurance for the construction process.

[0062] In summary, this invention, by deeply integrating artificial intelligence algorithms with compaction processes, constructs a closed-loop monitoring system capable of autonomous learning, dynamic optimization, and intelligent judgment. This not only greatly improves the accuracy and efficiency of asphalt pavement compaction monitoring but also provides strong technical support for realizing digital and refined management of highway construction. Attached Figure Description

[0063] Figure 1 This is a basic flowchart of an artificial intelligence-based asphalt pavement compaction monitoring system provided in one embodiment of the present invention. Detailed Implementation

[0064] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0065] Reference Figure 1 As an embodiment of the present invention, an artificial intelligence-based asphalt pavement compaction monitoring system includes a data acquisition module, a feature extraction module, a compaction calculation module, and a judgment module.

[0066] The data acquisition module is used to collect operating condition data under the current operating conditions;

[0067] The feature extraction module is used to extract the initial feature matrix from the road surface image of the working condition data. By optimizing the initial feature matrix with distance constraint terms, the initial matrix feature coefficients of the optimized road surface image are obtained.

[0068] The compaction calculation module is used to match the working condition range based on the working condition data, and select the corresponding compaction calculation logic according to the proportion of effective image blocks to obtain the compaction under the current working condition.

[0069] The judgment module is used to generate a dynamic compaction degree range based on road surface temperature, vibration acceleration and initial matrix characteristic coefficients through multi-parameter distance weighted fusion and normalization processing, and to make anomaly judgments based on the comparison results between the compaction degree and the range.

[0070] Through the synergistic effect of the feature extraction module and the compaction degree calculation module, the traditional static detection is upgraded to dynamic adaptive monitoring. By introducing distance constraint terms for optimization and dynamic compaction degree intervals, the system can overcome the interference caused by fluctuations in on-site working conditions. This not only improves the accuracy of compaction degree calculation but also enables intelligent adjustment of anomaly judgment criteria.

[0071] Operating data includes pavement images, pavement temperature, vibration acceleration, base course type, and base course type code;

[0072] The road surface images were captured by industrial cameras;

[0073] Road surface temperature was collected by an infrared thermometer;

[0074] Vibration acceleration is collected by a vibration acceleration sensor;

[0075] The base layer types include semi-rigid and flexible. The base layer type code corresponding to the semi-rigid base layer is 1, and the base layer type code corresponding to the flexible base layer is 0.

[0076] By simultaneously acquiring road surface images, road surface temperature, vibration acceleration, and base course type, the system obtains comprehensive information affecting compaction quality. This multi-dimensional data acquisition lays a solid and reliable data foundation for subsequent feature optimization and accurate calculation, avoiding the one-sidedness and limitations brought about by a single data source.

[0077] Extracting the initial feature matrix specifically includes:

[0078] The road image is uniformly divided into K 3×3 non-overlapping image blocks, where K is a perfect square. The gray-level entropy of each image block is obtained. If the gray-level entropy is greater than or equal to the effective threshold, the image block is marked as a valid image block.

[0079] The (2D)²PCA method is used to reduce the dimensionality of each effective image block to obtain an M×M dimension reduced image block, where M is a natural number less than K. The covariance matrix of all reduced image blocks is obtained, and the covariance matrix is ​​positive definite to obtain the initial feature matrix of the current working condition.

[0080] By filtering effective image blocks using grayscale entropy, invalid areas such as uneven lighting and dirt can be intelligently filtered out, allowing subsequent analysis to focus on key texture information, thus improving computational efficiency and anti-interference capability. Secondly, by using (2D)²PCA to reduce the dimensionality and positively definite covariance matrix, the key spatial structural features of the image are preserved, and the mathematical stability of subsequent matrix operations is ensured, thus providing a guarantee for obtaining a high-quality feature matrix.

[0081] In one specific embodiment of the present invention, the system acquires a 90x90 pixel road surface image, divides it evenly into 900 3×3 non-overlapping image blocks, calculates the grayscale value of 9 pixels in one image block, and its grayscale entropy is 0.75, which is greater than the set effective threshold of 0.6. Therefore, this block is marked as a valid image block. For all valid image blocks, the (2D)²PCA method is used to reduce their dimensionality from 3x3 to 2x2 reduced image blocks. Then, the system calculates the covariance matrix of all these 2x2 reduced image blocks and performs positive definite processing to finally obtain a 2x2 initial feature matrix of the current working condition.

[0082] Obtain the initial feature matrix corresponding to historical similar working conditions, and calculate the bulldozer distance between the distribution of the initial feature matrix of the current working condition and the distribution of the initial feature matrix of each historical similar working condition. The mathematical expression for the bulldozer distance is:

[0083] ;

[0084] in, The distance between the kth valid image block and the ith historically similar work condition bulldozer;

[0085] The distribution of the initial feature matrix for the current operating condition;

[0086] The distribution of the initial feature matrix for historically similar working conditions;

[0087] Indicates the infimum, Represents the joint probability distribution;

[0088] for and The joint distribution set;

[0089] Indicates in joint distribution The expected value of the following;

[0090] It is a single initial feature matrix sample drawn from the current distribution. It is a single initial feature matrix sample extracted from the historical distribution;

[0091] Historical similar working conditions include: constructing a current working condition data vector based on pavement temperature, vibration acceleration, and base course type code in the working condition data; using the K-nearest neighbor algorithm to search the historical working condition database; and selecting several historical data samples that are closest to the current working condition data vector in Euclidean distance as historical similar working conditions; at the same time, historical similar working conditions also need to be in the same working condition interval as the current working condition.

[0092] By introducing historical similar working conditions and bulldozer distance constraints, the system can utilize prior knowledge from massive historical data to guide and correct feature extraction for the current working condition. This collaborative optimization mechanism ensures that the feature coefficients of the final initial matrix not only reflect the current road surface state but also incorporate the optimal representation under similar working conditions, greatly enhancing the accuracy and robustness of the features in expressing the compaction state.

[0093] In a specific embodiment of the present invention, based on the current working condition data, the system finds 5 historically similar working conditions in the historical database using the K-nearest neighbor algorithm. The system calculates the bulldozer distance between the initial feature matrix of the current working condition and the initial feature matrix of the historical working condition. Then, the system constructs an objective function and solves iteratively to finally obtain the current feature coefficients of the current working condition.

[0094] The mathematical expression for the objective function is:

[0095] ;

[0096] in, Used to collaboratively optimize the current working condition and similar historical working conditions, and obtain characteristic coefficients that characterize the current compaction state;

[0097] It is a matrix composed of the initial feature matrices of all valid image blocks in the current working condition;

[0098] It is a matrix composed of the initial feature matrices of all valid image patches in historically similar working conditions;

[0099] The historical feature coefficients are the least squares solutions of the historical feature matrix to the texture grayscale vectors in historically similar working conditions.

[0100] The texture grayscale vector for the current working condition is formed by: expanding the 3×3 grayscale values ​​of all valid image blocks into a 9-dimensional vector in row order, and aggregating the mean of the 9-dimensional vectors of all valid image blocks to obtain the texture grayscale vector for the current working condition. The dimension of the texture grayscale vector is M.

[0101] The censored likelihood term is the matching probability between the current feature coefficients and the mean of historical feature coefficients obtained based on the Gaussian kernel function.

[0102] The censored likelihood term includes: extracting the feature coefficients of all valid image blocks in historical similar working conditions, using the mean of the historical feature coefficients as the benchmark value of the normal feature coefficients, substituting the feature coefficients of each valid image block in the current working condition into the Gaussian kernel function to obtain the likelihood value, and summing the logarithmic likelihood values ​​of all valid image blocks in the current working condition to obtain the censored likelihood term.

[0103] The bulldozer distance constraint term is the average distance between all valid image blocks under the current working condition and the bulldozer distances under similar historical working conditions.

[0104] Given the current feature coefficients, iteratively solve the objective function until convergence to obtain the current feature coefficients for each valid image patch;

[0105] The initial matrix eigenvalues ​​are the average of the current eigenvalues ​​of all valid image patches.

[0106] By aggregating the mean values ​​of the feature coefficients of all valid image blocks, the system can generate a single, comprehensive quantitative index to characterize the macroscopic compaction state of the entire monitored area. This method effectively smooths out the influence of local noise and minor anomalies, making the final feature coefficients more stable and reliable, and facilitating subsequent unified calculations and comparisons.

[0107] The working condition intervals are divided based on the pavement temperature and base course type of the current working condition data, including the first working condition interval, the second working condition interval, the third working condition interval, and the fourth working condition interval. The division rules are as follows:

[0108] If the temperature is less than or equal to the temperature threshold and the base layer type is semi-rigid, then it is classified as the first working condition range;

[0109] If the temperature is greater than the temperature threshold and the base layer type is semi-rigid, then it is classified as the second working condition range.

[0110] If the temperature is less than or equal to the temperature threshold and the base layer type is flexible, then it is classified as the third working condition range;

[0111] If the temperature exceeds the temperature threshold and the substrate type is flexible, it is classified as the fourth working condition range.

[0112] By clearly defining the rules for dividing the working condition intervals, layered and refined management was achieved. The intervals were divided based on two key physical parameters: pavement temperature and base course type. This reflects the significant differences in asphalt compaction patterns under different materials and temperature conditions. This layered processing mechanism provides a premise for subsequent use of different compaction degree calculations that are more in line with physical reality, and is an important link in improving calculation accuracy.

[0113] In one specific embodiment of the present invention, the system presets a temperature threshold of 120°C. When the road roller is working on a semi-rigid base course and the measured road surface temperature is 110°C, the system automatically divides the current working condition into the first working condition interval. If the roller is working on the same base course but the temperature rises to 130°C, it is divided into the second working condition interval. If the roller is working on a flexible base course and the temperature is 110°C, it is divided into the third working condition interval. If the roller is working on a flexible base course and the temperature is 130°C, it is divided into the fourth working condition interval.

[0114] The compaction calculation logic specifically includes: calculating the proportion of effective image blocks to the total number of blocks in the road surface image as the proportion of effective image blocks;

[0115] If the proportion of valid image blocks is greater than or equal to the valid threshold, the final compaction degree is obtained directly based on the matched working condition interval and the compaction degree function corresponding to the working condition interval.

[0116] If the proportion of valid image blocks is less than the valid threshold, the compaction degree calculated under the current working condition is combined with the average compaction degree of the previous N collected working conditions to obtain the final compaction degree, where N is a natural number greater than 2.

[0117] The compaction function includes the compaction function for the first working condition interval, the compaction function for the second working condition interval, the compaction function for the third working condition interval, and the compaction function for the fourth working condition interval.

[0118] By introducing the proportion of effective image blocks as a criterion, the system can adaptively handle situations with poor image quality. When there is insufficient effective information, the system does not simply abandon or give incorrect results, but intelligently integrates historical data for correction, ensuring the continuity and reliability of compaction calculation results and avoiding monitoring interruptions or misjudgments caused by instantaneous data acquisition problems.

[0119] In one specific embodiment of the present invention, a road surface image is divided into 900 blocks, of which 650 are marked as valid, and the percentage of valid image blocks is 72.2%; the system sets the valid threshold to 75%; since 72.2% < 75%, the system initiates the fusion logic, calculates the compaction degree of the current working condition as 96.0%, and retrieves the average compaction degree of the previous 3 historical working conditions as 96.5%, and fuses the two by weighted average to obtain the final compaction degree of 96.3%. If the percentage of valid blocks is 80%, the system will directly calculate the final compaction degree according to the matched working condition interval function without performing fusion;

[0120] In this embodiment, the system collects the current working condition data every 1 meter it advances. When the fusion logic is activated, it retrieves the final compaction value of the working condition 3 meters ahead of the current working condition.

[0121] The compaction functions for the first, second, third, and fourth working conditions are constructed based on the eigencoefficients of the initial matrix and the vibration acceleration.

[0122] The compaction function of the first working condition interval and the compaction function of the third working condition interval are linear fitting functions. The initial matrix characteristic coefficients and vibration acceleration of similar historical working conditions are used as independent variables, and the actual compaction of similar historical working conditions is used as dependent variable. The compaction function is obtained by linear fitting of the first term using the least squares method.

[0123] The compaction function for the second working condition interval and the compaction function for the fourth working condition interval are quadratic polynomial fitting functions. The initial matrix characteristic coefficients and vibration acceleration of similar historical working conditions are used as independent variables, and the actual compaction of similar historical working conditions is used as the dependent variable. The compaction function is obtained by fitting a quadratic polynomial using the least squares method.

[0124] The mathematical expressions for the first, second, third, and fourth working conditions are as follows:

[0125] ;

[0126] ;

[0127] ;

[0128] ;

[0129] in, For the first operating condition, This is the second operating condition. This is the third working condition. This is the fourth operating condition. ~ For coefficients, The first characteristic coefficient, It represents vibration acceleration.

[0130] Linear and quadratic polynomial functions were used for low-temperature and high-temperature conditions, respectively, to accurately simulate the nonlinear characteristics of viscosity change and compaction response of asphalt mixtures at different temperatures. This differentiated approach makes the compaction degree calculation more scientific and accurate, which is superior to the traditional method of using a single fixed model.

[0131] In a specific embodiment of the present invention, historical working condition data containing four working conditions are input, a fitting function is obtained, and the current working condition belongs to the second working condition interval. The system adopts the corresponding quadratic polynomial fitting function, and the initial matrix characteristic coefficients and vibration acceleration are substituted to obtain the compaction degree.

[0132] The generation of dynamic compaction intervals specifically includes:

[0133] Based on the current working condition data of road surface temperature, vibration acceleration and initial matrix eigenvalues, the normalized distances of road surface temperature, vibration acceleration and initial matrix eigenvalues ​​with those of similar historical working conditions are calculated as the third distance;

[0134] The third distance is obtained by weighting and fusing the variance weights of road surface temperature, vibration acceleration and the characteristic coefficients of the initial matrix in historical working conditions.

[0135] Based on the mean and standard deviation of the fourth distance under historical operating conditions, the fourth distance under the current operating condition is normalized, and the drift compensation factor is obtained through the linear rectification function.

[0136] Based on the preset benchmark interval and drift compensation factor, the benchmark interval is scaled proportionally to generate a dynamic compaction interval.

[0137] By comprehensively considering the drift of temperature, vibration, and characteristic coefficients, and using historical variance for weighting, the system can keenly perceive subtle changes in working conditions. The resulting dynamic compaction range can respond to working condition drift in real time, changing the anomaly judgment criteria from fixed to floating. This ensures both the sensitivity of monitoring and enhances the system's environmental adaptability.

[0138] In a specific embodiment of the present invention, the mathematical expression for the normalized third distance between the current working condition and historically similar working conditions is:

[0139] ;

[0140] ;

[0141] in, For road surface temperature, For vibration acceleration, The mean of the eigenvalues ​​of the initial matrix. and These are the mean and standard deviation of the corresponding parameters in the historical operating condition database, respectively;

[0142] Based on the variance weights of each parameter in historical operating conditions, the third distance is weighted and fused to obtain the fourth distance. Based on historical data, this distance is normalized and then the drift compensation factor is obtained through the ReLU function. The drift compensation factor is a number greater than or equal to 0, and the preset benchmark interval is [95%, 98%]. Therefore, the generated dynamic compaction interval is […]. 98% ].

[0143] Anomaly detection specifically includes:

[0144] If the final compaction degree under the current working conditions is within the dynamic compaction degree range, it is considered a normal situation.

[0145] If the final compaction degree under the current working conditions is not within the dynamic compaction degree range, it is judged as an abnormal situation.

[0146] This invention introduces historical similar working condition data for collaborative optimization. By constructing an objective function that includes a distance constraint term, the system can dynamically calibrate and optimize the feature matrix extracted from the image, effectively overcoming the working condition drift problem caused by fluctuations in working conditions such as temperature and vibration. This makes the feature coefficients of the final initial matrix more accurately and stably reflect the real compaction state, greatly improving the accuracy and anti-interference ability of the monitoring results.

[0147] Secondly, by calculating the working condition drift compensation factor, the present invention dynamically scales the preset benchmark compaction range to generate a dynamic compaction range that can respond to changes in working conditions in real time. This adaptive judgment mechanism makes the abnormal standard more scientific and reasonable, which not only avoids misjudging qualified results under harsh working conditions, but also ensures timely detection of quality problems under ideal working conditions, significantly improving the intelligence level and practicality of the monitoring system.

[0148] Finally, the system integrates multi-dimensional data such as road surface images, temperature, vibration acceleration, and base course type, comprehensively considering key factors affecting compaction quality. Through effective image block proportion filtering and layered calculation logic, the system can intelligently integrate historical data when image quality is poor, ensuring the continuity and reliability of calculation results. This multi-modal data fusion and fault-tolerant processing mechanism ensures the stable operation of the system in complex construction sites, providing a more comprehensive and reliable quality assurance for the construction process.

[0149] In summary, this invention, by deeply integrating artificial intelligence algorithms with compaction processes, constructs a closed-loop monitoring system capable of autonomous learning, dynamic optimization, and intelligent judgment. This not only greatly improves the accuracy and efficiency of asphalt pavement compaction monitoring but also provides strong technical support for realizing digital and refined management of highway construction.

[0150] Those skilled in the art will understand that embodiments of the present invention can be provided as systems, system products, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0151] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An artificial intelligence-based asphalt pavement compaction monitoring system, characterized in that, It includes a data acquisition module, a feature extraction module, a compaction degree calculation module, and a judgment module; The acquisition module is used to collect operating condition data under the current operating conditions; The feature extraction module is used to extract an initial feature matrix from the road surface image of the working condition data, and to obtain the initial matrix feature coefficients of the optimized road surface image by optimizing the initial feature matrix with a distance constraint term. The compaction degree calculation module is used to match the working condition interval based on the working condition data, and select the corresponding compaction degree calculation logic according to the proportion of effective image blocks to obtain the compaction degree under the current working condition. The determination module is used to generate a dynamic compaction degree range based on road surface temperature, vibration acceleration and initial matrix characteristic coefficients through multi-parameter distance weighted fusion and normalization processing, and to make anomaly determination based on the comparison result of compaction degree with the range. The specific steps for obtaining the initial matrix feature coefficients of the optimized road surface image include: calculating the bulldozer distance between the initial feature matrix of the current working condition and the initial feature matrix of similar historical working conditions, constructing an objective function and using the bulldozer distance as a constraint term to optimize the initial feature matrix and obtain the initial matrix feature coefficients. Obtain the initial feature matrix corresponding to historical similar working conditions, and calculate the bulldozer distance between the distribution of the initial feature matrix of the current working condition and the distribution of the initial feature matrix of each historical similar working condition. The mathematical expression for the bulldozer distance is: ; in, The distance between the kth valid image block and the ith historical similar working condition bulldozer; The distribution of the initial feature matrix for the current operating condition; The distribution of the initial feature matrix for historically similar working conditions; Indicates the infimum, Represents the joint probability distribution; for and The joint distribution set; Indicates in joint distribution The expected value of the following; It is a single initial feature matrix sample drawn from the current distribution. It is a single initial feature matrix sample extracted from the historical distribution; The historical similar working conditions include: constructing a current working condition data vector based on the road surface temperature, vibration acceleration and base course type code in the working condition data; using the K-nearest neighbor algorithm to search in the historical working condition database; and selecting several historical data samples that are closest to the current working condition data vector in Euclidean distance as historical similar working conditions. The mathematical expression for the objective function is: ； in, Used to collaboratively optimize the current working condition and similar historical working conditions, and obtain characteristic coefficients that characterize the current compaction state; It is a matrix composed of the initial feature matrices of all valid image blocks in the current working condition; It is a matrix composed of the initial feature matrices of all valid image patches in historically similar working conditions; The historical feature coefficients are the least squares solutions of the historical feature matrix to the texture grayscale vectors in historically similar working conditions. The texture grayscale vector for the current working condition is obtained by: expanding the 3×3 grayscale values ​​of all valid image blocks into a 9-dimensional vector in row order, and aggregating the mean of the 9-dimensional vectors of all valid image blocks to obtain the texture grayscale vector for the current working condition. The dimension of the texture grayscale vector is M. The censored likelihood term is the matching probability between the current feature coefficients and the mean of historical feature coefficients obtained based on the Gaussian kernel function. The bulldozer distance constraint term is the average distance between all valid image blocks under the current working condition and the bulldozer distance under similar historical working conditions. Given the current feature coefficients, iteratively solve the objective function until convergence to obtain the current feature coefficients for each valid image block; The initial matrix feature coefficients are the average of the current feature coefficients of all valid image blocks.

2. The system according to claim 1, characterized in that: The operating data includes road surface images, road surface temperature, vibration acceleration, base course type, and base course type code; The base layer types include semi-rigid and flexible. The base layer type code corresponding to the semi-rigid base layer is 1, and the base layer type code corresponding to the flexible base layer is 0.

3. The system according to claim 2, characterized in that, Extracting the initial feature matrix specifically includes: The road surface image is uniformly divided into K 3×3 non-overlapping image blocks, where K is a perfect square. The gray-level entropy of each image block is obtained. If the gray-level entropy is greater than or equal to the effective threshold, the image block is marked as a valid image block. The (2D)²PCA method is used to reduce the dimensionality of each valid image block to obtain an M×M dimension reduced image block, where M is a natural number less than K. The covariance matrix of all reduced image blocks is obtained, and the covariance matrix is ​​positive definite to obtain the initial feature matrix of the current working condition.

4. The system according to claim 1, characterized in that, The censored likelihood term includes: extracting the feature coefficients of all valid image blocks in historical similar working conditions, using the mean of the historical feature coefficients as the benchmark value of the normal feature coefficients, substituting the feature coefficients of each valid image block in the current working condition into the Gaussian kernel function to obtain the likelihood value, and summing the logarithmic likelihood values ​​of all valid image blocks in the current working condition to obtain the censored likelihood term.

5. The system according to claim 1, characterized in that, The working condition intervals are divided based on the pavement temperature and base course type of the current working condition data, including the first working condition interval, the second working condition interval, the third working condition interval, and the fourth working condition interval. The division rules are as follows: If the temperature is less than or equal to the temperature threshold and the base layer type is semi-rigid, then it is classified as the first working condition range; If the temperature is greater than the temperature threshold and the base layer type is semi-rigid, then it is classified as the second working condition range. If the temperature is less than or equal to the temperature threshold and the base layer type is flexible, then it is classified as the third working condition range; If the temperature is greater than the temperature threshold and the base layer type is flexible, then it is classified as the fourth working condition zone. The compaction calculation logic specifically includes: calculating the proportion of the number of effective image blocks to the total number of blocks in the road surface image as the proportion of effective image blocks; If the proportion of the effective image block is greater than or equal to the effective threshold, the final compaction degree is obtained directly based on the compaction degree function corresponding to the matched working condition interval. If the proportion of the effective image blocks is less than the effective threshold, the compaction degree calculated under the current working condition is combined with the average compaction degree of the previous N collected working conditions to obtain the final compaction degree, where N is a natural number greater than 2. The compaction function includes a first working condition interval compaction function, a second working condition interval compaction function, a third working condition interval compaction function, and a fourth working condition interval compaction function.

6. The system according to claim 5, characterized in that, The compaction function for the first working condition interval, the compaction function for the second working condition interval, the compaction function for the third working condition interval, and the compaction function for the fourth working condition interval are constructed based on the eigencoefficients of the initial matrix and the vibration acceleration. The compaction function of the first working condition interval and the compaction function of the third working condition interval are linear fitting functions. The initial matrix characteristic coefficients and vibration acceleration of similar historical working conditions are used as independent variables, and the actual compaction of similar historical working conditions is used as dependent variable. The compaction function is obtained by linear fitting of the first term using the least squares method. The compaction function of the second working condition interval and the compaction function of the fourth working condition interval are quadratic polynomial fitting functions. The initial matrix characteristic coefficients and vibration acceleration of similar historical working conditions are used as independent variables, and the actual compaction of similar historical working conditions is used as dependent variable. The compaction function is obtained by fitting a quadratic polynomial using the least squares method.

7. The system according to claim 6, characterized in that, Generating the dynamic compaction interval specifically includes: Based on the current working condition data of road surface temperature, vibration acceleration and initial matrix eigenvalues, the normalized distances of road surface temperature, vibration acceleration and initial matrix eigenvalues ​​with those of similar historical working conditions are calculated as the third distance; The third distance is weighted and fused based on the variance weights of road surface temperature, vibration acceleration, and initial matrix characteristic coefficients in historical working conditions to obtain the fourth distance; Based on the mean and standard deviation of the fourth distance under historical operating conditions, the fourth distance under the current operating condition is normalized, and the drift compensation factor is obtained through the linear rectification function. Based on the preset benchmark interval and the drift compensation factor, the benchmark interval is scaled proportionally to generate a dynamic compaction interval.

8. The system according to claim 7, characterized in that, The anomaly determination specifically includes: If the final compaction degree under the current working conditions is within the dynamic compaction degree range, it is considered a normal situation. If the final compaction degree under the current working conditions is not within the dynamic compaction degree range, it is judged as an abnormal situation.

Images