Asphalt pavement compactness detection method based on partial least square regression

Abstract

The invention provides an asphalt pavement compactness detection method based on partial least squares regression, which comprises the following steps: acquiring a rolling acceleration signal of a vibratory roller to obtain an equifrequency component, and calculating a corresponding compactness value CMV; temperature data T of the asphalt pavement are collected, and the actual asphalt pavement temperature is obtained through calculation; positioning data of the vibratory roller are collected, and the rolling speed V of the roller is calculated; the method comprises the following steps: performing interpolation on a CMV calculation abnormal point caused by the change of the working state of a road roller, and estimating an actual CMV value of the CMV abnormal point by calculating an average value of a plurality of known CMV values near a target point by adopting a moving average method based on the integrity of an asphalt pavement; and carrying out partial least square regression on the equifrequency component, CMV, T and V variables. The method is a nondestructive testing method, comprehensive and accurate compaction degree evaluation is provided, the compaction state of the asphalt pavement is sensed in advance, and compaction degree detection errors caused by change of the working state of a road roller are eliminated.

Description

Technical Field

[0001] The invention provides a method for detecting the compaction degree of an asphalt pavement based on partial least squares regression, belonging to a method for detecting the compaction degree of an asphalt pavement. Background Art

[0002] Compaction is the final process of asphalt pavement construction and a key process for asphalt mixture molding and functional quality. The full compaction of asphalt mixture can ensure the strength of the pavement, reduce deformation, and thus ensure the working performance of the pavement. At present, during the compaction construction of asphalt pavement, the compaction quality is mostly controlled by the core drilling method after completion, which is difficult to display the change information of the pavement compaction quality in real time and accurately. Due to the unevenness of asphalt mixture, the specified number of rolling passes alone cannot ensure that the target compaction degree is achieved. Therefore, compaction quality problems such as under-compaction and over-compaction may occur, causing early diseases and affecting the performance of asphalt pavement. The current technical specifications for highway asphalt pavement construction in my country require that the length of the rolling section, the temperature of the asphalt mixture, the number of rollers and the rolling speed should be controlled during the compaction construction of asphalt pavement, and the quality of asphalt pavement compaction should be detected by sampling and coring. However, the sampling method is difficult to fully reflect the compaction quality, and the accuracy of the results will also be affected by the number of samples sampled. If the number of samples sampled is small, the accuracy of the results is poor. If the number of samples sampled is large, it will affect the structure of the pavement.

[0003] At present, the compaction detection methods commonly used in asphalt pavement construction sites include traditional methods such as sampling method, sand filling method, ring knife method, non-nuclear density meter and nuclear density meter method.

[0004] The sampling method will drill the core after the road surface is compacted, which will cause local damage to the road surface and reduce the strength and integrity of the road surface. The sampling method cannot fully detect the entire road surface, and can only characterize the overall road surface compaction by the compaction of a few local points. Such results are not accurate.

[0005] Although the nuclear density meter method can achieve non-destructive testing, it is radioactive and can cause harm to testers, and the instrument itself is expensive and not suitable for widespread engineering construction.

[0006] Non-nuclear density meters eliminate the impact of radiation, but they lack accuracy when penetrating into the underlying layers of the road surface for testing, and still cannot provide comprehensive and real-time compaction testing of the road surface.

[0007] In addition, abnormal values ​​will appear in the FFT calculation process of the traditional method, which seriously deviates from the actual value range. The reason for the abnormality is that the vibration conditions have changed.

[0008] In addition, most non-destructive testing technologies for the compaction of asphalt pavements remain at the stage of processing only single variables such as acceleration, temperature or speed, and have not yet comprehensively considered multiple data such as vibration acceleration, rolling speed, and pavement temperature. Therefore, it is impossible to effectively associate these related data with the compaction of asphalt pavements. Summary of the invention

[0009] The purpose of the present invention is to provide a method for detecting the compaction degree of asphalt pavement based on partial least squares regression to solve the problems raised in the above background technology. The method comprises the following steps:

[0010] The asphalt pavement compaction detection method based on partial least squares regression includes the following steps:

[0011] S1. Device Setup

[0012] Including: vibration acceleration sensor, infrared temperature sensor, GPS positioning module;

[0013] The vibration acceleration sensor is equipped with a magnetic adsorption base, which is used to be adsorbed on the center line of the front and rear steel wheel axles and remain relatively still with the vehicle body.

[0014] The infrared temperature sensor is installed on the side of the vehicle body and is perpendicular to the ground.

[0015] The GPS positioning module is installed on the roof of the steel wheel roller through a suction cup, and the antenna is parallel to the forward direction of the vehicle body.

[0016] S2. Data Collection

[0017] The rolling acceleration signal of the vibratory roller is collected, and then the signal is analyzed by FFT to obtain the frequency spectrum information of the acceleration signal; A 0.5Ω , A Ω , A 1.5Ω , A 2Ω , A 2.5Ω , which are the 0.5 times, fundamental frequency, 1.5 times, 2 times, 2.5 times frequency components of the acceleration signal, and the corresponding CMV values ​​are calculated.

[0018] The temperature data T of the asphalt pavement is collected, compensated by software, and the relationship between the data collected by the infrared temperature sensor and the actual temperature data is compared to perform curve fitting to obtain the actual asphalt pavement temperature.

[0019] The positioning data of the vibratory roller is collected, some noise is filtered out, the coordinate system is converted, the longitude and latitude data are converted into UTM coordinates, and the rolling speed V of the roller is calculated.

[0020] S3, interpolating the missing or unavailable CMV data points, using the moving average method to estimate the CMV value of the unknown point by calculating the average of multiple known CMV values ​​near the target point;

[0021] Specifically, the basic steps of the moving average method are as follows:

[0022] (1) Select the window size: For each point to be interpolated, first determine a window size that includes several known data points near the target point. The window size is the number of points involved in the calculation;

[0023] (2) Calculate the moving average: Add the CMV values ​​of all known points in the window and divide it by the number of points to obtain the moving average of the points in the window, which is the estimated CMV value of the point to be interpolated.

[0024] Its formula is:

[0025]

[0026] Among them, SMA t is the moving average at time t, N is the number of points in the window, CMV t-i is the CMV value of each point in the window.

[0027] (3) Apply interpolation: Use the calculated SMA value as the CMV value of the target point to fill in the missing data.

[0028] S4, against A 0.5Ω , A Ω , A 1.5Ω , A 2Ω , A 2.5Ω , CMV, T, and V variables were subjected to partial least squares regression.

[0029] Partial least squares regression (PLS) is a regression method for processing multivariate data, especially for high-dimensional input data and multicollinearity. The basic idea of ​​PLS ​​is to establish a regression model by extracting the common principal components of the input variables and output variables. The PLS model aims to find the optimal linear relationship between the input variables and the output variables, thereby improving the prediction accuracy and model stability.

[0030] The specific calculation steps of partial least squares regression are as follows:

[0031] 1) Set the independent variable X = (X1, X2, ..., X m ) and dependent variable Y=(Y1,Y2,...,Y p ) are standardized respectively, and the standardized data matrix after n observations is shown in the following formula (2):

[0032]

[0033] 2) Extract the first pair of components of the independent variable and the dependent variable, denoted as T1 and S1. They are the linear combinations of the independent variable X and the dependent variable Y, as shown in the following equations (2) and (3):

[0034]

[0035] T1 and S1 should extract as much variation information of the independent variable and the dependent variable as possible, and their correlation should be maximized. The score vectors of T1 and S1 are calculated using the standardized matrices X0 and Y0, respectively, and are recorded as t1 and s1, as shown in the following equations (4) and (5):

[0036]

[0037] The above two conditions are transformed into a conditional extreme value problem, namely maxCov(t1,s1), and the constraints are satisfied as shown in the following formula (6):

[0038]

[0039] The extreme value problem is transformed as shown in the following formula (7):

[0040]

[0041] By using the Lagrange multiplier method, we can get formula (8):

[0042]

[0043] Taking partial derivatives of w1 and v1 respectively, we can get equations (9) and (10):

[0044]

[0045] Combining equations (9) and (10), we can get:

[0046] λ=θ (11)

[0047]

[0048] Formula (12) can be regarded as a matrix The characteristic equation of M is λ. 2 , the eigenvector is w1, and the formula (12) is combined with the formula (9) to obtain:

[0049]

[0050] 3) Set the independent variable X = (X1, X2, ..., X m ) and dependent variable Y=(Y1,Y2,...,Y p) and t1 respectively to get:

[0051]

[0052] Y0=t1r1 T +Y1 (15)

[0053] Among them, X1 and Y1 are residual matrices, and p1 and r1 are estimated by the least squares method:

[0054]

[0055] 4) Replace X0 and Y0 in 1) with the residual matrices X1 and Y1, repeat 2) and 3), and remember:

[0056]

[0057] Then the residual matrix is:

[0058]

[0059] If the absolute value of the elements in the residual matrix is ​​approximately 0, stop extracting the components.

[0060] 5) Assume that a total of l components are extracted, and use these l components to regress the dependent variable Y, as shown in the following formula (22):

[0061]

[0062] Finally, the regression expression of the characteristic variable X and the dependent variable Y can be written as:

[0063]

[0064] The present invention adopts a non-destructive testing method. By collecting data such as vibration acceleration, rolling speed and asphalt surface temperature during the rolling process, the partial least squares regression (PLS) model is used to predict the compaction of the road surface. This method has the advantages of non-destructive testing, comprehensive coverage and high efficiency and real-time. Compactness information can be obtained without physically destroying the road surface, maintaining the integrity and structural strength of the road surface; during the rolling process, this method can collect a wide range of road surface data in real time, provide a comprehensive and accurate compaction evaluation, and effectively reduce the errors caused by insufficient local sampling.

[0065] The present invention first determines a window size that includes several known data points near the target point. The window size is the number of points involved in the calculation. The CMV values ​​of all known points in the window are added and divided by the number of points to obtain the moving average of the points in the window, thereby estimating the CMV value of the point to be interpolated. This method can effectively handle missing data points and further improve the integrity of CMV data and the accuracy of analysis.

[0066] The present invention collects data such as vibration acceleration, rolling speed, and asphalt surface temperature during the rolling process, and uses a partial least squares regression (PLS) model to predict the pavement compaction. Taking into account a variety of data such as vibration acceleration, rolling speed, and pavement temperature, these relevant data are effectively associated with the compaction of the asphalt pavement. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 is a flow chart of the present invention;

[0068] Figure 2 It is a comparison chart of the predicted values ​​and the actual values ​​of the training set and the test set of the embodiment. DETAILED DESCRIPTION

[0069] The present invention provides a method for detecting the compaction degree of asphalt pavement based on partial least squares regression, such as Figure 1 As shown, the following steps are included:

[0070] S1. Device Setup

[0071] The vibration acceleration sensor uses the Ruifen Technology AKF392 three-axis accelerometer, which has a measurement range of ±40g, a measurement accuracy of 1mg, and a maximum sampling frequency of 1000Hz. The vibration acceleration sensor is equipped with a magnetic adsorption base, which can be directly adsorbed on the center line of the front and rear steel wheel shafts and remain relatively still with the vehicle body.

[0072] The infrared temperature sensor is the domestic SA50R online infrared thermometer made by SAAO, which has a temperature measurement range of 0℃ to 500℃, a temperature measurement accuracy of ±1.5% of the measured value, a response time of 300ms, a response wavelength of 8-14 microns, a distance coefficient of 20:1 (D:S), and a signal output of RS485. The infrared temperature sensor is installed on the side of the vehicle body and is perpendicular to the ground.

[0073] The GPS positioning module uses the Hi-Target Sky3 positioning and orientation GNSS receiver, and is paired with the Qianxun RTK virtual base station service to achieve vehicle positioning, with a plane positioning accuracy of 1cm. The GNSS antenna is installed on the top of the steel wheel roller through a suction cup, and the antenna is parallel to the forward direction of the vehicle body.

[0074] S2. Data Collection

[0075] The rolling acceleration signal of the vibratory roller is collected, and then the signal is analyzed by FFT to obtain the frequency spectrum information of the acceleration signal; A 0.5Ω , A Ω , A 1.5Ω , A 2Ω , A 2.5Ω, which are the 0.5 times, fundamental frequency, 1.5 times, 2 times, 2.5 times frequency components of the acceleration signal, and the corresponding CMV values ​​are calculated.

[0076] The temperature data T of the asphalt pavement is collected, compensated by software, and the relationship between the data collected by the infrared temperature sensor and the actual temperature data is compared to perform curve fitting to obtain the actual asphalt pavement temperature.

[0077] The positioning data of the vibratory roller is collected, some noise is filtered out, and then the coordinate system is converted to convert the longitude and latitude data into UTM coordinates, so as to calculate the rolling speed V of the roller.

[0078] S3. In the present invention, in order to interpolate missing or unavailable CMV (CompactionMeterValue) data points, a moving average (SMA) method is used. This method estimates the CMV value of an unknown point by calculating the average of multiple known CMV values ​​near the target point. This method can effectively eliminate short-term fluctuations in the data when the data is incomplete, obtain smooth and reasonable estimation results, and further improve the integrity of CMV data and the accuracy of analysis.

[0079] For the point to be interpolated, first determine a window size that includes several known data points near the target point. The window size is the number of points involved in the calculation.

[0080] Specifically, the basic steps of the moving average method are as follows:

[0081] (1) Selecting the window size: For each point to be interpolated, first determine a window size that includes several known data points near the target point. The window size is the number of points involved in the calculation, which is usually determined based on the distribution and characteristics of the data to balance the smoothing effect and data accuracy.

[0082] (2) Calculate the moving average: Add the CMV values ​​of all known points in the window and divide it by the number of points to obtain the moving average of the points in the window, which is the estimated CMV value of the point to be interpolated.

[0083] Its formula is:

[0084]

[0085] Among them, SMA t is the moving average at time t, N is the number of points in the window, CMV t-i is the CMV value of each point in the window.

[0086] (3) Apply interpolation: Use the calculated SMA value as the CMV value of the target point to fill in the missing data.

[0087] When interpolating one of the points, the window size is selected as 2. Calculate the two known CMV values ​​near the point, for example, CMV1 = 17.196, CMV2 = 14.68, then the estimated CMV value of the point is 15.93. This method can effectively handle missing data points and further improve the integrity of CMV data and the accuracy of analysis.

[0088] S4, against A 0.5Ω , A Ω , A 1.5Ω , A 2Ω , A 2.5Ω , CMV, T, and V variables were subjected to partial least squares regression.

[0089] This embodiment collects data such as vibration acceleration, rolling speed, and asphalt surface temperature during the rolling process, and uses a partial least squares regression (PLS) model to predict the road surface compaction degree.

[0090] First, load the independent and dependent variable data from the data file and divide the data set into a training set and a test set, where the test set accounts for 30% of the total data. Next, standardize the independent variables in the training set, and the standardized data will be used to train and evaluate the regression model.

[0091]

[0092] Among them, μ X is the mean of the independent variable, σ X is the standard deviation of the independent variable.

[0093] When selecting the number of principal components, we first determine a range of principal component numbers and use 10-fold cross validation to evaluate the performance of the model under each number of principal components. The predictive ability of the model is measured by calculating the mean square error (MSE), and the number of principal components with the smallest mean square error is selected as the optimal number of principal components.

[0094] The PLS regression model is trained using the selected number of principal components, and the regression coefficients and related statistics are calculated. Then, the test set or new data is subjected to the same standardization process, and the regression coefficients obtained from the training are used for prediction, and the performance indicators of the prediction results such as the root mean square error (RMSE) and the coefficient of determination (R 2 ). Through these steps, the predictive ability of the model can be evaluated and its validity can be confirmed.

[0095]

[0096] When there are fewer principal components, the goodness of fit of partial least squares regression is low, and the RMSE and MAPE values ​​are large, indicating that the fitting effect is poor. As the number of principal components increases, the goodness of fit of the regression model gradually improves and the error gradually decreases. When the number of principal components increases to 4, the goodness of fit no longer increases significantly with the increase in the number of principal components. When the number of principal components increases to more than 6, the differences in goodness of fit, MAPE and RMSE between the training set and the test set become insignificant. Therefore, the number of principal components is selected as 5, and the predicted values ​​of the training set and the test set are plotted for comparison with the actual values.

[0097] The data set was randomly divided into a training set and a test set in a ratio of 7:3. The training set was used for model training, and the test set was used for model verification. When the number of principal components was 5, the goodness of fit was around 0.85, and the fitting effect was good. Table 1 lists the PLS model performance parameters of the training set and the test set when the number of principal components was 5. The comparison between the predicted values ​​and the actual values ​​of the training set and the test set is shown in Figure 2 .

[0098] Table 1 Performance parameters of PLS ​​model for training set and test set

[0099] <![CDATA[R 2 ]]> RSME MAPE Training set 0.8618 1.52634 1.2971％ Test Set 0.8562 1.6111 1.4445％

[0100] After the partial least squares regression (PLS) model is established, it can be used to predict new data. The specific steps are as follows:

[0101] 1. Data standardization: Standardize the newly collected independent variable data (such as acceleration signal spectrum characteristics, temperature, speed, etc.) using the same standardization method as the model training phase to ensure data consistency.

[0102]

[0103] Among them, μ X is the mean of the independent variable, σ X is the standard deviation of the independent variable.

[0104] 2. Predict compaction: Using the regression coefficient β obtained in the training phase, the standardized independent variable data is input into the PLS model for calculation to obtain the predicted compaction value of the asphalt pavement.

[0105] Y=X·β

[0106] in:

[0107] X is the standardized independent variable matrix. X=[A 0.5Ω A Ω A 1.5Ω A 2Ω A 2.5Ω CMV TV];

[0108] β is the regression coefficient matrix obtained by the PLS regression model. β = [0.9199 0.0023 -0.0013 0.0042-0.0028 0.0019 0.0015 -0.0332 -0.0105] T ;

[0109] The original variable regression equation is:

[0110] Y=0.919+0.099X1-0.032X2+0.046X3-0.223X4+0.5X5+0.003X6-0.003X7-0.047X8

[0111] Table 2 Predicted compaction and measured compaction

[0112] Experiment No. The compaction degree obtained by this method Measured compaction Absolute value of compaction error / % 1 84.561 84.321 -0.42 2 89.611 87.251 1.98 3 91.983 90.692 1.42 4 94.629 93.854 0.83 5 95.580 95.927 -0.36

[0113] To verify the effectiveness of this method, experiments were conducted on five randomly selected asphalt pavements. Table 2 lists the compaction predicted by this method and the measured compaction, where the measured compaction is measured by a non-nuclear density meter.

[0114] The compaction degree of traditional asphalt pavement is measured by the core drilling method, where two core samples are drilled every 100m. This is a destructive detection method that cannot detect poor compaction quality. The existing method of using roller construction status parameters to detect compaction only uses vibration data, ignoring two important factors: asphalt mixture temperature and rolling speed. This method, based on practical applications, solves the problem of FFT anomalies in vibration data through interpolation, greatly improving the application effect. The partial least squares regression method is fast and accurate, and is an effective method for calculating compaction degree through roller construction status parameters.

Claims

1. A method for detecting compaction of asphalt pavement based on partial least squares regression, characterized in that: The following steps are involved: S1. Device settings, including: vibration acceleration sensor, infrared temperature sensor, GPS positioning module; The vibration acceleration sensor is equipped with a magnetic adsorption base, which is used to be adsorbed on the center line of the front and rear steel wheel shafts and remain relatively still with the vehicle body; The infrared temperature sensor is installed on the side of the vehicle body and is perpendicular to the ground; The GPS positioning module is installed on the roof of the steel wheel roller through a suction cup, and the antenna is parallel to the forward direction of the vehicle body; S2. Data Collection Collect the rolling acceleration signal of the vibratory roller, obtain the equal frequency component, and calculate the corresponding compaction value CMV value; Collect the temperature data T of the asphalt pavement and calculate the actual asphalt pavement temperature; Collect the positioning data of the vibratory roller and calculate the rolling speed V of the roller; S3, interpolating the missing or unavailable CMV data points, using the moving average method to estimate the CMV value of the unknown point by calculating the average of multiple known CMV values ​​near the target point; S4. Perform partial least squares regression on equal frequency components, CMV, T, and V variables.

2. The asphalt pavement compaction detection method based on partial least squares regression according to claim 1 is characterized in that: S2 is specifically: The rolling acceleration signal of the vibratory roller is collected, and then the signal is analyzed by FFT to obtain the frequency spectrum information of the acceleration signal; A 0.5Ω , A Ω , A 1.5Ω , A 2Ω , A 2.5Ω , are the frequency components of the acceleration signal such as 0.5 times, fundamental frequency, 1.5 times, 2 times, and 2.5 times, and the corresponding CMV values ​​are calculated; Collect the temperature data T of the asphalt pavement, compensate it through software, compare the relationship between the data collected by the infrared temperature sensor and the actual temperature data, perform curve fitting, and obtain the actual asphalt pavement temperature; The positioning data of the vibratory roller is collected, some noise is filtered out, the coordinate system is converted, the longitude and latitude data are converted into UTM coordinates, and the rolling speed V of the roller is calculated.

3. The asphalt pavement compaction detection method based on partial least squares regression according to claim 1 is characterized in that: In S3, the moving average method comprises the following steps: (1) Selecting the window size: For each point to be interpolated, first determine a window size that includes several known data points near the target point; the window size is the number of points involved in the calculation; (2) Calculate the moving average: add the CMV values ​​of all known points in the window and divide by the number of points to obtain the moving average of the points in the window, which is the estimated CMV value of the point to be interpolated; Its formula is: Among them, SMA t is the moving average at time t, N is the number of points in the window, CMV t-i is the CMV value of each point in the window; (3) Apply interpolation: Use the calculated SMA value as the CMV value of the target point to fill in the missing data.

4. The asphalt pavement compaction detection method based on partial least squares regression according to claim 1 is characterized in that: The specific calculation steps of partial least squares regression in S4 are as follows: 1) Set the independent variable X = (X1, X2, ..., X m ) and dependent variable Y=(Y1,Y2,...,Y p ) are standardized respectively, and the standardized data matrix after n observations is shown in the following formula (2): 2) Extract the first pair of components of the independent variable and the dependent variable, denoted as T1 and S1; they are the linear combinations of the independent variable X and the dependent variable Y, as shown in the following equations (2) and (3): Among them, T1 and S1 extract the variation information of the independent variable and the dependent variable, and their correlation should be maximized; the score vectors of T1 and S1 are calculated using the standardized matrices X0 and Y0, respectively, and are recorded as t1 and s1, as shown in the following equations (4) and (5): The above two conditions are transformed into a conditional extreme value problem, namely maxCov(t1,s1), and the constraints are satisfied as shown in the following formula (6): The extreme value problem is transformed as shown in the following formula (7): By using the Lagrange multiplier method, we can get formula (8): Taking partial derivatives of w1 and v1 respectively, we can get equations (9) and (10): Combining equations (9) and (10), we get: λ=θ (11) Formula (12) is regarded as a matrix The characteristic equation of M is λ. 2 , the eigenvector is w1, and the formula (12) is combined with the formula (9) to obtain: 3) Set the independent variable X = (X1, X2, ..., X m ) and dependent variable Y=(Y1,Y2,...,Y p ) and t1 respectively to get: Among them, X1 and Y1 are residual matrices, and p1 and r1 are estimated by the least squares method: 4) Replace X0 and Y0 in 1) with the residual matrices X1 and Y1, repeat 2) and 3), and remember: Then the residual matrix is: If the absolute value of the elements in the residual matrix is ​​approximately 0, stop extracting components; 5) Let l components be extracted in total, and use these l components to regress with the dependent variable Y, as shown in the following formula (22): Finally, the regression expression of the characteristic variable X and the dependent variable Y can be written as:

Images