Compaction quality detection method and device based on multi-domain analysis and artificial neural network

Abstract

The present invention discloses a compaction quality detection method and device based on multi-domain analysis and artificial neural network. The method obtains acceleration signals in the compaction process through sensors installed on the roller, and screens out multiple feature information according to the time domain characteristics of the acceleration signal, the frequency domain characteristics of the power spectrum and the correlation with the comprehensive compaction quality. The screened feature information is used as the training set of the artificial neural network, and the classification and regression neural network are trained to judge the compaction quality of the road surface and evaluate the compaction quality of the road surface. Experiments show that compared with the traditional compaction quality detection method, the proposed method can accurately judge the compaction quality of the road surface without damaging the road surface.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent compaction, and in particular relates to a compaction quality detection method and device based on multi-domain analysis and artificial neural network. Background Art

[0002] The compaction quality of roads is related to the service life of roads and driving safety. The accurate detection of compaction quality has always been a concern of on-site engineers. The traditional compaction quality detection method is to first take samples from the compacted road surface, and then conduct relevant tests and calculations on the samples in the laboratory to obtain the compaction degree of the road material samples. This method will cause damage to the road and cannot achieve real-time monitoring. The current intelligent compaction measurement method generally measures the compaction quality through the compaction meter value and the mechanical drive power of the roller, but these methods only consider the characteristics within the amplitude spectrum range and are easily affected by factors such as the construction environment and roller parameters. It is difficult to accurately evaluate the compaction quality. Summary of the invention

[0003] The purpose of the present invention is to provide a compaction quality detection method and device based on multi-domain analysis and artificial neural network, which can monitor the soil compaction quality in real time to determine the soil compaction quality and obtain the opportunity to change the roller parameters.

[0004] To achieve the above object, the present invention provides a compaction quality detection method based on multi-domain analysis and artificial neural network, comprising the following steps:

[0005] S1. Collect vertical acceleration signals at the vibrating steel wheel of the roller during different compaction processes on the test section, and measure the soil compaction degree of the test section pavement; classify the soil into three compaction qualities: under-compacted, optimally compacted, and over-compacted according to the comprehensive compaction degree of the soil;

[0006] S2, calculating the time domain characteristics of the vertical acceleration signal and the frequency domain characteristics of the power spectrum density diagram, calculating the correlation between the characteristics of the vertical acceleration signal and the compaction quality, and screening out multiple characteristics of the vertical acceleration signal according to the correlation between the characteristics of the vertical acceleration signal and the compaction quality;

[0007] S3, labeling the multiple features with the compaction quality type labels they belong to, using the multiple features as input variables of the artificial neural network, and the compaction quality labels as output variables, to train the artificial neural network model;

[0008] S4. Collect and record the vertical acceleration signals at the vibrating steel wheel of the roller on the remaining road sections, and use the trained artificial neural network model and multiple features of the vertical acceleration signal to evaluate the compaction quality of the tested road section.

[0009] Furthermore, in S1, the ring knife method is used to measure the comprehensive compaction of the soil on the road surface of the test section.

[0010] Furthermore, in S1, the measurement process of the comprehensive compaction degree of the soil is: use the ring knife method to take multiple samples of the surface, 20 cm underground and 40 cm underground soil after each compaction, calculate the average compaction degree of the multiple samples as the final compaction degree of the soil at different depths, and use the average value of the final compaction degree of the soil at different depths as the comprehensive compaction degree of the soil.

[0011] Furthermore, in S2, the correlation between the characteristics of the vertical acceleration signal and the compaction quality is calculated by variance analysis.

[0012] Furthermore, in S2, the features of the screened vertical acceleration signal include: a first peak value, a bandwidth power, a standard deviation, a first peak frequency, a time domain peak value, a fifth peak value, a second peak value, and a fourth peak value.

[0013] Furthermore, in S3, K-fold cross validation is used when training the artificial neural network.

[0014] Furthermore, in S3, the artificial neural network loss function is:

[0015]

[0016] Among them, loss is the loss value, N is the number of observations, C is the number of categories, and the categories are under-compaction, over-compaction, and optimal compaction. ni is the actual category of the sample, Y ni To predict the category, the weight parameters and bias parameters of the fully connected layer in the artificial neural network are updated by minimizing the loss function.

[0017] An intelligent compaction quality evaluation device based on multi-domain analysis and artificial neural network includes an acquisition module and a processing module which are electrically connected; the acquisition module is used to acquire the vertical acceleration signal of the tested road surface and transmit it to the processing module; the processing module is used to judge the compaction quality of the tested road surface according to the received vertical acceleration signal of the tested road surface.

[0018] Compared with the prior art, the present invention has at least the following beneficial technical effects:

[0019] The present invention discloses a compaction quality detection method based on multi-domain analysis and artificial neural network. The method obtains the acceleration signal at the vibrating steel wheel generated by the vertical rebound energy of the soil during the compaction process through a sensor installed on the roller. According to the time domain characteristics of the acceleration signal, the frequency domain characteristics of the power spectrum and the correlation with the comprehensive compaction quality, the energy and other signal feature information with the greatest correlation with the compaction quality are screened out, and the screened feature information is used as the training set of the artificial neural network model. The training classification and artificial neural network are used to judge the compaction quality of the road surface and evaluate the compaction quality of the road surface without being affected by the double jump of the roller. Compared with the traditional compaction quality detection method, the proposed method can accurately judge the compaction quality of the road surface without damaging the road surface.

[0020] When calculating the acceleration signal characteristics, the present invention comprehensively considers the time domain characteristics and the frequency domain characteristics of the power spectrum density diagram, and uses an artificial neural network to establish a nonlinear model to determine the compaction quality. Compared with the method of only considering the amplitude spectrum characteristics, the present invention has a higher accuracy.

[0021] Furthermore, the ring knife method was used to take multiple samples of the surface, 20 cm underground and 40 cm underground soil after each compaction, and the average compaction degree of the multiple samples was calculated as the final compaction degree of the soil at different depths. The average value of the final compaction degree of the soil at different depths was used as the comprehensive compaction degree of the soil to improve the accuracy of the comprehensive compaction degree of the entire soil.

[0022] Furthermore, the present invention performs variance analysis on the features of the acceleration signal, calculates the correlation between the features and the compaction quality, removes invalid features, reduces data dimensions, improves algorithm performance, and facilitates real-time monitoring.

[0023] Furthermore, the present invention adopts K-fold cross validation during artificial neural network training to avoid overfitting, so that the artificial neural network model can find objective laws instead of simply retrieving samples. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 A flow chart of the quality evaluation method provided by the present invention;

[0025] Figure 2 This is a diagram of the road roller construction and signal collection process;

[0026] Figure 3 It is the acceleration signal segmentation process diagram;

[0027] Figure 4 This is the result graph of correlation calculation;

[0028] Figure 5 It is the structure diagram of artificial neural network;

[0029] Figure 6 This is the compaction quality prediction result diagram;

[0030] Figure 7 A schematic diagram of the module structure of the compaction quality evaluation device provided by the present invention;

[0031] Figure 8 A schematic diagram of the structure of a computer device provided by the present invention.

[0032] In the attached figure: 1. vibrating steel wheel, 2. acceleration sensor, 3. bracket, 4. cab, 5. rear wheel. DETAILED DESCRIPTION

[0033] In order to make the purpose and technical solution of the present invention clearer and easier to understand, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. The specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0034] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first" and "second" are used only for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, "multiple" means two or more. In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0035] Example 1

[0036] Reference Figure 1The present invention discloses a compaction quality detection method based on multi-domain analysis and artificial neural network, which adopts soil vertical rebound energy to reflect soil compaction quality, and soil rebound energy is observed by acceleration sensor. Firstly, the vertical acceleration signal at the vibrating steel wheel of the roller is recorded on the test section, and the soil compaction degree is measured destructively on the test section by the ring knife method. According to the soil compaction degree, the soil is divided into three compaction qualities: under-compacted, optimally compacted and over-compacted. Then, the time domain characteristics of the vertical acceleration signal and the frequency domain characteristics of the power spectrum density diagram are calculated, and 8 features are screened out according to the correlation between the characteristics of the vertical acceleration signal and the compaction quality through variance analysis. Using artificial neural network, a nonlinear model between the 8 features and the compaction quality is established. Finally, the vertical acceleration signal at the vibrating steel wheel of the roller on the remaining sections is collected and recorded, and the compaction quality of the remaining sections is evaluated using the nonlinear model and the 8 features of the acceleration signal.

[0037] Reference Figure 2 The road roller includes a vibrating steel wheel 1, a cab 4 and a rear wheel 5.

[0038] The specific steps include:

[0039] Step 1: Install the acceleration sensor 2 on the bracket 3 of the vibrating steel wheel 1 of the roller to collect and record the acceleration sensor data, such as Figure 2 The roller compacts the graded soil of the test section 12 times, and records the vertical acceleration signal during the compaction process at intervals. For example, the vertical acceleration signals during the 2nd, 4th, 6th, 8th, 10th, and 12th compaction processes can be recorded, and the vertical acceleration signals during the 1st, 3rd, 5th, 7th, 9th, and 11th compaction processes can also be recorded. The acceleration signals of the sampling points of the acceleration sensor are extracted each time the roller is in stable operation, and the vertical acceleration signals are cut into segments. Each segment contains an acceleration signal of 0.3s and 8 vibration cycles. The signal cutting process is shown in FIG. Figure 3 As shown in the figure. 100 signal segments are obtained in each compaction process, and 6×100 signal segments are obtained in total in six compaction processes. Each segment contains acceleration signals of 600 sampling points. The test section and the tested section are different sections of the road of the same project.

[0040] Step 2: Calculate the time domain features of the acceleration signal and the frequency domain features of the power spectrum density diagram, and perform variance analysis using formula (1). Select eight features based on the correlation between the features and the compaction quality.

[0041] The characteristics of the acceleration signal include time domain characteristics and frequency domain characteristics:

[0042] Frequency domain features include: the first to fifth peaks (PeakAmp) and peak frequencies (Freq) in the power spectrum, and bandwidth power (Bandpower)

[0043] The time domain features include: signal-to-noise ratio (SNR), signal-to-distortion and distortion ratio (SINAD), total harmonic distortion (THD), standard deviation (std), crest factor (CrestFactor), impulse factor (ImpulseFactor), kurtosis (Kurtosis), skewness (Skewness), mean (Mean), shape factor (ShapeFactor), clearance factor (ClearanceFactor), and peak value (PeakValue).

[0044] The formula for the correlation between the variance analysis calculation characteristics and compaction quality is as follows:

[0045]

[0046] Among them, SSE is the residual sum of squares, SSR is the regression sum of squares, k is the total number of categories, in this embodiment k is 3 (a total of 3 categories), N is the total number of observations, in this embodiment N is 600, SSR / k-1 is the difference within the category, SSE / Nk is the difference between categories. F is the ratio of the difference within the category to the difference between the categories, and F is the evaluation of the correlation. The larger the F value, the higher the correlation. The expressions of SSR and SSE are shown in Formula 2:

[0047]

[0048]

[0049]

[0050]

[0051] Among them, y ij is the feature value, i represents the number of samples of a certain feature, in this embodiment, i = 1, 2, ..., 100; j is the number of categories, in this embodiment, j = 1, 2, 36; n j is the number of samples in the jth class, and in this embodiment, nj is 100. is the average value of all data of a certain feature, is the average value of the jth type of data in a certain feature data. SST is the total sum of squares of deviations.

[0052] The time domain and frequency domain characteristics of 600 acceleration signal segments were calculated, and the correlation between the features and the compaction quality was calculated by variance analysis using formula (1). The correlations were sorted by size, and the top 8 features with the highest scores were used as the input data set of the artificial neural network. The 8 features selected based on the correlation are: the first peak, bandwidth power, standard deviation, first peak frequency, time domain peak, fifth peak, second peak, and fourth peak. The correlation calculation results are shown in Figure 2. Figure 4 And as shown in Table 1.

[0053] Table 1 Eigenvalue scoring

[0054]

[0055] Step 3: Use artificial neural network to establish a nonlinear model between features and compaction quality, and use the nonlinear model to evaluate compaction quality. Input the data set consisting of the selected 8 features into the artificial neural network, and the output of the artificial neural network is three types of compaction quality: undercompacted, optimal compacted and overcompacted, and train the artificial neural network.

[0056] The process of determining the compaction quality is as follows: the surface, 20 cm underground and 40 cm underground soil are sampled 5 times by using the ring knife method after each compaction, and the average compaction degree of the 5 samples is calculated as the final compaction degree of the soil at different depths, and the average final compaction degree of the soil at different depths is used as the comprehensive compaction degree of the soil. The compaction process with the largest comprehensive compaction degree is defined as the best compaction, the compaction quality of the compaction times less than the best compaction is defined as under-compaction, and the compaction quality of the compaction times greater than the best compaction is defined as over-compaction. The compaction degrees during the 2nd, 4th, 6th, 8th, 10th and 12th compaction processes are shown in Table 2. According to the compaction degree measurement results, the compaction quality of 2 and 4 times of compaction is defined as under-compaction, that is, the labels of the 8 features corresponding to the 2nd and 4th acceleration signals are under-compacted; the compaction quality of 6 times of compaction is defined as optimal compaction, that is, the labels of the 8 features corresponding to the 6th acceleration signal are optimal compaction; the compaction quality of 8 and 12 times of compaction is defined as over-compaction, that is, the labels of the 8 features corresponding to the 8th and 12th acceleration signals are over-compacted.

[0057] Table 2 Soil compaction results

[0058]

[0059] The structure of the artificial neural network consists of an input layer, a first fully connected layer, an activation layer, a second fully connected layer, a Softmax layer, and a category output layer, such as Figure 5 As shown in the figure, the first fully connected layer contains 25 neurons, the activation function of the activation layer is ReLU, and the second fully connected layer contains 3 neurons.

[0060] The artificial neural network loss function expression is:

[0061]

[0062] Among them, loss is the loss value, N is the number of observations, and C is the number of categories, which are under-compaction, over-compaction, and optimal compaction. ni is the actual category of the sample, Yni To predict the category, the weight parameters and bias parameters of the fully connected layer in the artificial neural network are updated by minimizing the loss function.

[0063] The maximum number of iterations of the artificial neural network is 1000. The training of the artificial neural network takes 8 signal features as input and compaction quality (undercompacted, optimal compacted, overcompacted) as output. During the training process, the artificial neural network achieves parameter adjustment by reducing the loss value in formula 3.

[0064] During the training process, K-fold cross validation is used for the test section and other sections, where K = 5. The specific process is:

[0065] The 600 signal segments were randomly divided into 5 parts. 4 of them were used as simulated test section data to train the artificial neural network. One of the 5 parts was selected as simulated data for other sections to verify the accuracy of the method. The accuracy rate was the average accuracy rate of the above K-fold cross-validation process repeated 5 times.

[0066] The prediction results of the nonlinear model are as follows Figure 6 As shown in the figure, the prediction results are expressed in the form of confusion matrix, with an overall accuracy of 97.8%. Among the 200 undercompacted soils, 197 were correctly identified and 3 were incorrectly identified as overcompacted soils, with an accuracy of 98.5%; among the 100 optimally compacted soils, 98 were correctly identified and 2 were incorrectly identified as overcompacted soils, with an accuracy of 98.0%; among the 300 overcompacted soils, 292 were correctly identified, 6 were incorrectly identified as undercompacted soils, and 2 were incorrectly identified as optimally compacted soils, with an accuracy of 97.3%, while the existing prediction methods can only achieve an accuracy of about 60%.

[0067] Step 4: Collect the vertical acceleration sensor signal of the tested section, input the vertical acceleration sensor signal of the tested section into the trained artificial neural network, obtain the compaction quality of the tested section, and realize non-destructive detection of the compaction quality of other sections.

[0068] Example 2

[0069] Reference Figure 7 The present invention provides a road surface compaction quality evaluation device, comprising an electrically connected acquisition module and a processing module; the acquisition module is used to collect the vertical acceleration signal at the vibrating steel wheel of the roller on the tested road surface and transmit it to the processing module; the processing module is used to judge the compaction quality of the tested road section according to the received vertical acceleration signal.

[0070] Example 3

[0071] The present invention provides a computer device, such as Figure 8As shown, it includes an electrically connected memory and a processor, wherein the memory stores a calculation program that can be run on the processor, and when the processor executes the calculation program, the steps of the above-mentioned road section compaction quality method are implemented. Figure 1 Alternatively, the processor implements the functions of each module in the above-mentioned device embodiments when executing the computer program.

[0072] The computer program may be divided into one or more modules / units, and the one or more modules / units are stored in the memory and executed by the processor to accomplish the present invention.

[0073] The road surface compaction quality evaluation device / terminal device may be a computing device such as a desktop computer, a notebook, a palm computer, a cloud server, etc. The road surface compaction quality evaluation device / terminal device may include, but is not limited to, a processor and a memory.

[0074] The processor can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0075] The memory can be used to store the computer program and / or module, and the processor implements various functions of the pavement compaction quality evaluation device / terminal equipment by running or executing the computer program and / or module stored in the memory and calling the data stored in the memory.

[0076] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0077] Example 4

[0078] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium that is executable by a processor. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0079] The above contents are only for explaining the technical idea of ​​the present invention and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the claims of the present invention.

Claims

1. A compaction quality detection method based on multi-domain analysis and artificial neural network, characterized in that: The following steps are involved: S1. Collect vertical acceleration signals at the vibrating steel wheel of the roller during different compaction processes on the test section, and measure the soil compaction degree of the test section pavement; classify the soil into three compaction qualities: under-compacted, optimally compacted, and over-compacted according to the comprehensive compaction degree of the soil; S2. Calculate the time domain features of the vertical acceleration signal and the frequency domain features of the power spectrum density diagram, calculate the correlation between the features of the vertical acceleration signal and the compaction quality, and screen out multiple features of the vertical acceleration signal according to the correlation between the features of the vertical acceleration signal and the compaction quality; the features of the vertical acceleration signal include time domain features and frequency domain features; In S2, the correlation between the characteristics of the vertical acceleration signal and the compaction quality is calculated by variance analysis; The features of the screened vertical acceleration signal include: first peak, bandwidth power, standard deviation, first peak frequency, time domain peak, fifth peak, second peak and fourth peak; S3, labeling the multiple features with the compaction quality type labels they belong to, using the multiple features as input variables of the artificial neural network, and the compaction quality labels as output variables, to train the artificial neural network model; S4. Collect and record the vertical acceleration signals at the vibrating steel wheel of the roller on the remaining road sections, and use the trained artificial neural network model and multiple features of the vertical acceleration signal to evaluate the compaction quality of the tested road section.

2. The compaction quality detection method based on multi-domain analysis and artificial neural network according to claim 1 is characterized in that: In S1, the comprehensive soil compaction degree of the road surface of the test section is measured by the ring knife method.

3. The compaction quality detection method based on multi-domain analysis and artificial neural network according to claim 1 is characterized in that: In S1, the measurement process of the comprehensive compaction degree of the soil is: use the ring knife method to take multiple samples of the surface, 20 cm underground and 40 cm underground soil after each compaction, calculate the average compaction degree of the multiple samples as the final compaction degree of the soil at different depths, and use the average value of the final compaction degree of the soil at different depths as the comprehensive compaction degree of the soil.

4. The compaction quality detection method based on multi-domain analysis and artificial neural network according to claim 1 is characterized in that: In S3, K-fold cross validation is used when training the artificial neural network.

5. The compaction quality detection method based on multi-domain analysis and artificial neural network according to claim 1 is characterized in that: In S3, the artificial neural network loss function is: in, is the loss value, N is the number of observations, C is the number of categories, which are under-compacted, over-compacted, and optimally compacted. is the actual category of the sample, To predict the category, the weight parameters and bias parameters of the fully connected layer in the artificial neural network are updated by minimizing the loss function.

6. An intelligent compaction quality evaluation device based on multi-domain analysis and artificial neural network, used to implement the method of claim 1, characterized in that: It includes an electrically connected acquisition module and a processing module; the acquisition module is used to acquire the vertical acceleration signal of the tested road surface and transmit it to the processing module; the processing module is used to determine the compaction quality of the tested road surface based on the received vertical acceleration signal of the tested road surface.

Images