Road spatiotemporal compaction quality inspection method and system based on acoustic-solid coupling

Abstract

The present invention is a road spatiotemporal compaction quality inspection method and system based on acoustic-solid coupling, including the following contents: using an acoustic pressure sensor to contactlessly receive surface wave signals leaking in the air at compaction sampling points; constructing a VMD-LSTM denoising module for denoising and reconstructing the surface wave signal to obtain a reconstructed signal; generating a two-dimensional time-frequency graph from the reconstructed signal, and extracting features from the two-dimensional time-frequency graph using a classification module to obtain time-frequency features; obtaining compaction influencing factors and corresponding compaction degrees under different stage categories and different structural layer categories, and constructing a road spatiotemporal fusion data set; training an XGBoost machine learning algorithm with a road spatiotemporal fusion data set; combining the VMD-LSTM denoising module, the classification module, and the trained XGBoost machine learning algorithm to form a fusion intelligent model, and using the fusion intelligent model to quickly complete compaction quality detection of any structural layer at any stage. The present invention realizes full-thickness and full-life cycle road compaction quality detection.

Description

Technical Field

[0001] The present invention relates to the technical field of non-destructive testing, and particularly to a method and system for inspecting the spatio-temporal compaction quality of roads based on acoustic-solid coupling. Background Art

[0002] With the rapid development of infrastructure construction in China, the total mileage of roads has been increasing year by year. The detection of the compaction quality of road structural layers has become particularly important, which is directly related to the service quality and service life of highways. Traditional compaction quality detection methods, such as the sand replacement method and the core drilling method, are not only time-consuming and laborious, but also have a certain degree of destructiveness to the road structure. In the context of the construction of intelligent transportation infrastructure, a road quality detection method combining stress waves and intelligent models has emerged, aiming to improve the deficiencies of traditional methods and achieve fast and non-destructive detection.

[0003] Currently, on-site road compaction quality detection mainly uses a dynamic signal collector combined with an acceleration sensor or a multi-channel collector to collect surface wave signals excited artificially, and then conducts quality evaluation. However, the surface waves generated by these methods propagate in the structural layer and are captured by directly contacting sensors, which requires the sensors to be arranged on the surface of the road structure. Although non-destructive detection is achieved, the flexibility of sensor arrangement still needs to be improved. Existing research shows that ground-penetrating radar can be used for road compaction quality detection, but the cost of ground-penetrating radar is relatively high. As an economical and convenient sensor, a sound pressure sensor can receive surface wave signals leaking into the air, which meets the requirements of flexible and fast collection of surface wave signals. In addition, the existing intelligent models for road compaction established only target specific structural layers and specific stages, and it is urgent to provide a road compaction quality detection method using sound pressure sensors and applicable to the full-thickness full-life cycle. Summary of the Invention

[0004] Aiming at the deficiencies of the existing technology, the technical problem to be solved by the present invention is: to provide a method and system for inspecting the spatio-temporal compaction quality of roads based on acoustic-solid coupling. This method uses the data obtained by a non-contact sensor - a sound pressure sensor to establish a fusion intelligent model applicable to the compaction quality detection of different structural layers and different stages of roads, and realizes the compaction quality detection of the full-thickness full-life cycle of roads.

[0005] The technical solution adopted by the present invention to solve the above technical problem is:

[0006] In the first aspect, the present invention provides a method for inspecting the spatio-temporal compaction quality of roads based on acoustic-solid coupling, including the following steps:

[0007] Non-contact reception of surface wave signals excited by external forces and leaking into the air at the compaction sampling points by using a sound pressure sensor;

[0008] The surface wave signal is decomposed into several modes by the VMD algorithm. Multiple modes with smooth waveforms and consistent with the main frequency range of the surface wave signal are selected for reconstruction to obtain the reconstructed signal. All the modes obtained by decomposing the surface wave signal through the VMD algorithm and the corresponding reconstructed signals are used to form the denoising samples, and a denoising data set is constructed.

[0009] The input of the LSTM network is all the modes obtained by decomposition using the VMD algorithm, and the output is the reconstructed signal. The LSTM network is trained using the denoising data set to obtain a trained LSTM network. The VMD algorithm and the trained LSTM network are connected to form a VMD-LSTM denoising module.

[0010] The degree of compaction of the compaction sampling points is obtained. Several compaction degree intervals are divided according to the compaction degree distribution. Each compaction degree interval is used as a compaction degree category. The two-dimensional time-frequency diagram of the reconstructed signal obtained by processing through the VMD-LSTM denoising module is generated. The two-dimensional time-frequency diagram generated by the reconstructed signal is classified according to the compaction degree interval to which its corresponding compaction degree belongs. The two-dimensional time-frequency diagram and its corresponding compaction degree category are used to form classification samples, and a time-frequency diagram classification data set is constructed.

[0011] The classification network is trained using the time-frequency diagram classification data set to obtain a trained classification network. The output layer of the classification network consists of a fully connected layer and a Softmax function. After removing the Softmax function from the trained classification network, it is used as a classification module to extract features from the two-dimensional time-frequency diagram of the reconstructed signal to obtain time-frequency features. The input of the classification module is the two-dimensional time-frequency diagram of the reconstructed signal, and the output is the feature vector output by the fully connected layer.

[0012] The compaction influencing factors and the corresponding compaction degrees under different stage categories and different structural layer categories are obtained to construct a road spatio-temporal fusion data set. The different stage categories are the construction stage, the maintenance stage, and the acceptance and opening stage. The different structural layer categories are the road base layer, the cement stabilized macadam layer, and the asphalt surface layer. The compaction influencing factors are water content, soil particle gradation, surface wave velocity, time-frequency features, aggregate gradation, cement content, bulk specific gravity, Marshall stability, flow value, asphalt content, theoretical maximum density of the surface layer, and porosity of the surface layer. When a certain compaction influencing factor is missing, it is set to 0.

[0013] Among them, the surface wave velocity is determined using the reconstructed signal; the time-frequency features are obtained by extracting features from the two-dimensional time-frequency diagram of the reconstructed signal using the classification module.

[0014] The XGBoost machine learning algorithm is trained using the road spatio-temporal fusion data set to obtain a trained XGBoost machine learning algorithm for compaction degree detection.

[0015] Combine the VMD-LSTM denoising module, classification module with the trained XGBoost machine learning algorithm to form a fusion intelligent model;

[0016] During on-site use, input data of compaction influencing factors other than surface wave velocity and time-frequency characteristics, collect on-site surface wave signals, and then use the fusion intelligent model to quickly complete the detection of compaction quality at any stage of any structural layer.

[0017] Furthermore, the classification network is implemented using networks such as ViT, VGG, ResNet, MobileNet, or Swin Transformer.

[0018] Furthermore, the XGBoost machine learning algorithm is optimized using an intelligent optimization algorithm before training, and the intelligent optimization algorithm is at least one of PSO algorithm, GWO algorithm, GA algorithm, etc.

[0019] In a second aspect, the present invention provides a road spatio-temporal compaction quality inspection system based on acoustic-solid coupling, adopting the road spatio-temporal compaction quality inspection method based on acoustic-solid coupling, including a mobile vehicle, a dynamic signal collector installed on the mobile vehicle, and a control terminal. A pair of sound pressure sensors are installed at the rear of the mobile vehicle through a bracket and two sleeves, and the two sleeves are symmetrically installed on the bracket. The sound pressure sensors are electrically connected to the dynamic signal collector, and the dynamic signal collector communicates with the control terminal;

[0020] A pair of sound pressure sensors are arranged at both ends of the compaction sampling point and do not contact the compaction sampling point, and are used to receive surface wave signals excited by external forces and leaking into the air;

[0021] A wind shield structure is arranged around a pair of sound pressure sensors;

[0022] The control terminal obtains the surface wave signals collected by the sound pressure sensors through the dynamic signal collector;

[0023] Load the fusion intelligent model in the control terminal. During on-site use, input data of compaction influencing factors other than surface wave velocity and time-frequency characteristics, collect on-site surface wave signals, and then use the fusion intelligent model to quickly complete the detection of compaction quality at any stage of any structural layer.

[0024] Furthermore, the road spatio-temporal compaction quality inspection system based on acoustic-solid coupling further includes a rubber hammer.

[0025] Furthermore, the road spatio-temporal compaction quality inspection system based on acoustic-solid coupling further includes a cloud, and the cloud is used to store denoising data sets, time-frequency map classification data sets, and road spatio-temporal fusion data sets. The cloud communicates with the control terminal.

[0026] Furthermore, during sampling, the vertical distance between the sound pressure sensor and the compaction sampling point is controlled within 20 cm.

[0027] Furthermore, the two sleeves are movably connected to the bracket. The sleeves can move up and down relative to the bracket, and the wind shield structure is fixed on the two sleeves and can move up and down following the sleeves.

[0028] Furthermore, the two sleeves are electrically connected to the lifting unit, and the lifting unit is electrically connected to the control terminal. The sleeves can be automatically lifted and lowered under the control of the lifting unit.

[0029] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0030] Based on the acoustic-solid coupling principle, the present invention uses a sound pressure sensor to collect surface waves that are propagated through the road structure and leaked into the air, and applies it to the on-site road compaction quality detection. As a non-contact sensor, the sound pressure sensor is used to receive the surface wave signals leaked in the air. The sound pressure sensor can flexibly capture and is economical.

[0031] The system of the present invention is provided with a wind shield structure, and uses the dual guarantee of the VMD-LSTM denoising module and the wind shield structure to reduce the influence of the air environment on the surface wave signals, and improves the accuracy and reliability of signal denoising.

[0032] The method of the present invention constructs a road spatio-temporal fusion data set, trains a machine learning algorithm with this data set, and then obtains a fusion intelligent model, which can be used to realize the detection of the compaction quality of the full-depth full-life cycle road, including: the road base layer, the cement stabilized macadam layer and the asphalt surface layer, under different operating states of the full-life cycle, including: the construction stage, the maintenance stage and the acceptance and opening stage. When the present invention is applied to a new scenario, it can quickly, non-destructively and continuously detect the surface wave signals of the road base layer, the cement stabilized macadam layer and the asphalt surface layer, saving manpower and material resources and meeting the actual engineering requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is a schematic flow chart of an embodiment of the road spatio-temporal compaction quality inspection method based on acoustic-solid coupling of the present invention.

[0034] Figure 2 It is a schematic flow chart of the VMD-LSTM denoising module.

[0035] Figure 3 It is a schematic flow chart of the classification module.

[0036] Figure 4 It is a schematic structural diagram of an embodiment of the road spatio-temporal compaction quality inspection system based on acoustic-solid coupling of the present invention.

[0037] In the figure, 1 is the road structure layer, 2 is the mobile vehicle, 3 is the dynamic signal collector, 4 is the control terminal, 5 is the sound pressure sensor, 6 is the wind shield structure, 7 is the bracket, and 8 is the sleeve. Specific embodiments

[0038] The present invention will be further explained below in conjunction with embodiments and the accompanying drawings, but the protection scope of the present application is not limited thereby.

[0039] The road spatio-temporal compaction quality inspection system based on acoustic-solid coupling of the present invention (see Figure 4 ) includes a mobile vehicle 2, a dynamic signal collector 3 and a control terminal 4 installed on the mobile vehicle. A pair of sound pressure sensors are installed at the tail of the mobile vehicle through a bracket 7 and two sleeves 8. The two sleeves 8 are symmetrically installed on the bracket 7, and the bracket 7 is fixed to the mobile vehicle. The sound pressure sensor 5 is electrically connected to the dynamic signal collector, and the dynamic signal collector communicates with the control terminal 4;

[0040] A pair of sound pressure sensors are arranged at both ends of the compaction sampling point and do not contact the compaction sampling point, and are used to receive the surface wave signals excited by humans and leaking into the air.

[0041] A wind shield structure is arranged around the pair of sound pressure sensors. The wind shield structure is used to isolate the sound pressure sensors from the external air and reduce the influence of air on the signals.

[0042] The control terminal obtains the surface wave signals collected by the sound pressure sensors through the dynamic signal collector.

[0043] A fusion intelligent model is loaded in the control terminal. During on-site use, data of compaction influencing factors other than the surface wave velocity and time-frequency characteristics are input, and the surface wave signals on-site are collected, and then the compaction quality of any structural layer at any stage is quickly detected by using the fusion intelligent model.

[0044] During the collection process, the sound pressure sensors, as non-contact sensors, are placed at both ends of the compaction sampling point. There are two sound pressure sensors in total, which are used to receive the surface wave signals excited by humans and leaking into the air. The sound pressure sensors are connected to the dynamic signal collector and are controlled by the control terminal for collection and storage.

[0045] The wind shield structure in the present invention can adopt a wind shield or a wind shield board, etc., to isolate the two sound pressure sensors from the external air to reduce the influence of air on the signals. The specific implementation structure can be realized according to the existing technology. The lower end of the wind shield structure should be lower than the lowest installation point of the sound pressure sensors. When the lower end of the wind shield structure contacts the road structure layer 1, the influence of air on the signals can be minimized.

[0046] Figure 2It is a schematic flow diagram of the VMD-LSTM denoising module. In the present invention, the VMD algorithm and the LSTM network are combined to further denoise the surface wave signal. During signal processing, the VMD algorithm is used to decompose the surface wave signal into several modes, and appropriate modes are manually selected. The selected appropriate modes are combined and reconstructed to obtain a reconstructed signal, which is then saved. The "appropriate" mentioned here can be understood as follows: the decomposed modes are smooth in waveform and similar in shape to the original signal (i.e., the surface wave signal before decomposition); in terms of frequency, they are close to the main frequency range of the original signal, then the mode is considered an appropriate mode. That is to say, the mode with a smooth waveform and consistent with the main frequency range of the surface wave signal is selected as the appropriate mode. Here, "consistent" can be understood as close. This process can be manually selected and is considered consistent within the range of visual error.

[0047] In the early stage of training, surface wave signals at the compaction sampling points under different stage categories and different structural layer categories are obtained. All the modes obtained by decomposing the surface wave signal through the VMD algorithm and the corresponding reconstructed signals form denoising samples, and a large number of denoising samples constitute a denoising dataset.

[0048] All the modes obtained by decomposing the VMD algorithm are used as the input of the LSTM network, and the reconstructed signal is used as the target output. The LSTM network is trained using the denoising dataset, and the VMD algorithm is connected to the trained LSTM network to form a VMD-LSTM denoising module, enabling the VMD-LSTM denoising module to automatically realize mode selection and signal reconstruction, thus completing the automatic denoising process of the surface wave signal. The VMD-LSTM denoising module effectively reduces residual noise. The processed surface wave signal is smoother and has less noise, meeting the requirements of practical engineering.

[0049] The VMD algorithm can divide the surface wave signal into several intrinsic mode functions, and each intrinsic mode function contains local signal features of different scales of the original signal. The process of the VMD algorithm is as follows: First, the original signal is decomposed using VMD, and then the fast Fourier transform is performed on each mode. By observing the time-domain curve and frequency response of the mode, the mode with less noise is selected for signal reconstruction.

[0050] Figure 3It is a schematic flowchart of the classification module. The compaction degree and surface wave signals at the compaction sampling points under different stage categories and different structural layer categories are obtained. The compaction degree intervals are divided according to the compaction degree distribution. Each compaction degree interval is used as a compaction degree category. The reconstructed signals obtained by processing the surface wave signals through the VMD-LSTM denoising module are used to generate two-dimensional time-frequency diagrams. The two-dimensional time-frequency diagrams generated by the reconstructed signals are classified according to the compaction degree intervals to which their corresponding compaction degrees belong, that is, each two-dimensional time-frequency diagram is labeled with a compaction degree category classification label. The two-dimensional time-frequency diagrams and their corresponding classification labels are combined to form classification samples, and a large number of classification samples constitute the time-frequency diagram classification data set.

[0051] The classification network is trained using the time-frequency diagram classification data set. The output layer of the classification network consists of a fully connected layer and a Softmax function. After removing the Softmax function from the trained classification network, it is used as the classification module to extract features from the two-dimensional time-frequency diagrams and obtain time-frequency features. The input of the classification module is the two-dimensional time-frequency diagram of the reconstructed signal, and the output is the feature vector output by the fully connected layer, that is, the time-frequency features.

[0052] The road spatio-temporal compaction quality inspection method based on acoustic-solid coupling of the present invention selects a sound pressure sensor to collect the surface wave signals leaking into the air based on the acoustic-solid coupling principle. The acoustic-solid coupling principle is as follows: A rubber hammer is used to strike the ground on one side of the compaction sampling point artificially. The rubber hammer excites stress waves outside the sound pressure sensor. Among them, the shear wave and the longitudinal wave will propagate along the depth direction, and the surface wave will propagate along the solid surface of the road base course, cement stabilized macadam layer and asphalt surface course. During the propagation process, part of the surface wave will diffuse into the air medium to form a leaking surface wave. The leaking surface wave propagates in the air medium to form an acoustic-solid coupling model. The leaking surface wave is received by two sound pressure sensors placed above and the waveform is recorded by a dynamic signal acquisition instrument. The sound pressure sensor closer to the hammering side is denoted as the first sound pressure sensor, and the sound pressure sensor farther from the hammering side is denoted as the second sound pressure sensor. The first sound pressure sensor first receives the surface wave leaking into the air. The propagation time T 1 from the hammering point to the surface wave below the first sound pressure sensor propagating in a single structural layer is t 1 plus the time t 1 ′ for the surface wave leaking into the air and propagating to the first sound pressure sensor, that is, T 1 =t 1 +t 1 ′. Subsequently, the second sound pressure sensor receives the leaking surface wave. The propagation time T 2 from the hammering point to the surface wave below the second sound pressure sensor propagating in a single structural layer is t 2 plus the time t 2 ′ for the surface wave leaking into the air and propagating to the second sound pressure sensor, that is, T 2 =t2 +t 2 ′. The time difference between the propagation times of the two acoustic pressure sensors is ΔT = T 2 -T 1 . Since the two acoustic pressure sensors are placed horizontally with respect to the ground and have the same installation height, and the propagation speed of the surface wave leaking into the air is considered to be the same, it is considered that t 1 ′ = t 2 ′. Therefore, the time difference between the propagation times of the surface wave from the hammer strike point to the surface below the two acoustic pressure sensors in the corresponding structural layer is the time difference between the propagation times of the two acoustic pressure sensors, ΔT = T 2 -T 1 = t 2 -t 1 . The surface wave velocity v is the distance L between the two acoustic pressure sensors divided by the time difference ΔT, i.e., v = L / ΔT.

[0053] In the present invention, acoustic pressure sensors are used to collect surface wave signals for different structural layers at different stages, that is, surface wave signals are collected at the compaction sampling points of the roadbed layer, the cement stabilized macadam layer, and the asphalt surface layer during the construction stage, the curing stage, and the acceptance and opening to traffic stage respectively. The VMD-LSTM denoising module is used to process the surface wave signals to obtain two reconstructed signals. Based on the two reconstructed signals, the time difference is calculated, and then the surface wave velocity is calculated. The reconstructed signals processed by the VMD-LSTM denoising module are used to generate a two-dimensional time-frequency diagram, and the classification module is used to extract time-frequency features from the two-dimensional time-frequency diagram. In this embodiment, the classification module takes the VisionTransformer (ViT) classification network as the main architecture, and the classification network can also adopt VGG, ResNet, MobileNet, SwinTransformer, etc.

[0054] For the roadbed layer, the water content, soil particle gradation (such as the content in the large, medium, and small particle size ranges), surface wave velocity, time-frequency features, and compaction degree at the compaction sampling points are jointly used to form a roadbed compaction quality detection data set.

[0055] For the cement stabilized macadam layer, the water content, gravel gradation, cement content, surface wave velocity, time-frequency features, and compaction degree at the compaction sampling points are jointly used to form a cement stabilized macadam layer compaction quality detection data set.

[0056] For the asphalt surface layer, the bulk specific gravity, Marshall stability, flow value, asphalt content, theoretical maximum density of the surface layer, surface layer porosity, surface wave velocity, time-frequency features, and compaction degree detected at the compaction sampling points are jointly used to form an asphalt surface layer compaction quality detection data set.

[0057] Union the compaction influencing factors in the subgrade compaction quality detection dataset, the cement stabilized macadam layer compaction quality detection dataset, and the asphalt surface layer compaction quality detection dataset. Additionally, include the structural layer category (different layers), the stage category (construction stage, maintenance stage, and acceptance and opening to traffic stage), and the corresponding compaction degree to obtain the road spatio-temporal fusion dataset. In the road spatio-temporal fusion dataset, when sampling the roadbed layer, the values of the data of other compaction influencing factors except those of this structural layer are all zero. The same applies to the other two layers, that is, when a compaction influencing factor is missing, it is set to 0. In the subgrade compaction quality detection dataset, the cement stabilized macadam layer compaction quality detection dataset, and the asphalt surface layer compaction quality detection dataset, the data except the compaction degree are all compaction influencing factors.

[0058] During the construction of the above datasets, after each structural layer is compacted, a compaction sampling point is set every certain distance (about 10 - 20m). During later on-site use, spot checks or continuous inspections can be carried out. The area between two sound pressure sensors is the compaction sampling point. When conducting spot checks, the compaction sampling points can be set according to the specification requirements.

[0059] This invention uses the XGBoost machine learning algorithm for compaction prediction and uses intelligent optimization algorithms to optimize the XGBoost machine learning algorithm. The intelligent optimization algorithm can be at least one of the PSO algorithm, the GWO algorithm, the GA algorithm, etc., and can establish PSO - XGBoost regression prediction models, GA - XGBoost regression prediction models, GWO - XGBoost regression prediction models, etc.

[0060] Divide the road spatio - temporal fusion dataset into a training set and a validation set with a ratio of 7:3 for training and validating the above regression prediction models. Both the PSO - XGBoost regression prediction model and the GA - XGBoost regression prediction model are set with 50 particles and 3000 iterations, and the GWO - XGBoost regression prediction model is set with 50 wolves and 3000 iterations. MAPE is used as the fitness function, and 2 the three indicators of R, RMSE, and MAPE are used to evaluate the results of the regression prediction models, and the optimal regression prediction model is selected for preservation.

[0061] Connect the VMD - LSTM denoising module, the classification module, and the optimal regression prediction model in sequence to obtain the fusion intelligent model. In this embodiment, the fusion intelligent model can be denoted as VMD - LSTM - ViT - XGB. During on - site use, input the data of the compaction influencing factors except the surface wave velocity and time - frequency characteristics, and the surface wave signals collected in real - time on - site. Then, use the fusion intelligent model to quickly complete the compaction quality detection of any structural layer at any stage, and the compaction degree at which structural layer and stage can be output.

[0062] During on-site sampling of the present invention, the vertical distance between the sound pressure sensor and the compaction sampling point is within 20 cm. The sound pressure sensor should not touch the ground, and the closer it is to the ground, the better, and the distance should not be too large.

[0063] Embodiment 1

[0064] The process of the method for spatio-temporal compaction quality inspection of roads based on acoustic-solid coupling in this embodiment (see Figure 1 ) is as follows:

[0065] 1. Surface wave signal acquisition: Build a spatio-temporal compaction quality inspection system for roads based on acoustic-solid coupling, including a mobile vehicle, a dynamic signal collector installed on the mobile vehicle, and a control terminal. Install a pair of sound pressure sensors at the rear of the mobile vehicle through a bracket and two sleeves. The two sleeves are symmetrically fixed on the bracket. The sound pressure sensor is electrically connected to the dynamic signal collector, and the dynamic signal collector communicates with the control terminal. The vertical distance between the sound pressure sensor and the compaction sampling point is 15 cm. A windshield structure is arranged around the pair of sound pressure sensors.

[0066] Before acquisition, calibrate the sensitivity of the sound pressure sensor. Use a sound calibrator to excite a standard signal of 1000 Hz and 1 Pa to correct the sensitivity of the sound pressure sensor in the current environment. After completing the sensitivity calibration, collect surface wave signals at the compaction sampling points during the construction stage, maintenance stage, and acceptance and opening stage of the road base layer, cement stabilized macadam layer, and asphalt surface layer, and save them.

[0067] 2. Signal denoising: Cut the signal output by the sound pressure sensor to obtain a signal containing mutation signals (during the acquisition process, it is a stable signal when no surface wave is received, and a mutation signal is formed after receiving the surface wave) and with a length of 1000 (1000 represents the number of data points). This is the surface wave signal. Then use the VMD algorithm for modal decomposition. Set the number of decomposition modes to 10, decompose the surface wave signal, select appropriate modes for combination to obtain a reconstructed signal, which is the denoised signal. Use all the modes obtained after VMD decomposition as the input of the LSTM network, and the reconstructed signal as the target output. Train the LSTM network using the denoising data set. Connect the VMD algorithm and the trained LSTM network to form a VMD-LSTM denoising module, and save the VMD-LSTM denoising module.

[0068] 3. Construct a road spatio-temporal fusion dataset: Use the reconstructed signal to obtain the surface wave velocity and generate a two-dimensional time-frequency map, and use the ViT classification module to extract the time-frequency features of each two-dimensional time-frequency map. Correlate the surface wave velocity, time-frequency features with the material properties (such as water content, soil particle gradation, crushed stone gradation, cement content, bulk density, Marshall stability, flow value, asphalt content, theoretical maximum density of the surface layer, porosity of the surface layer, etc.) and compaction degree at the compaction sampling points respectively to form a subgrade compaction quality detection dataset, a cement stabilized macadam layer compaction quality detection dataset, and an asphalt surface layer compaction quality detection dataset. Take the union of the compaction influencing factors in the subgrade compaction quality detection dataset, the cement stabilized macadam layer compaction quality detection dataset, and the asphalt surface layer compaction quality detection dataset, and add the structural layer category (different layers) and stage category (construction stage, maintenance stage, and acceptance and opening stage) and the corresponding compaction degree to obtain a road spatio-temporal fusion dataset.

[0069] Among them, the area between two sound pressure sensors is the compaction degree detection interval, and the surface wave velocity, time-frequency features are in one-to-one correspondence with the compaction degree and the material properties at that place.

[0070] 4. Road compaction quality evaluation: Use the PSO-XGBoost regression prediction model to train and predict the road spatio-temporal fusion dataset, and divide the road spatio-temporal fusion dataset into a training set and a validation set. The ratio of the training set to the validation set is 7:3. Use the trained PSO-XGBoost regression prediction model for compaction degree detection.

[0071] Combine the VMD-LSTM denoising module, the ViT classification module with the PSO-XGBoost regression prediction model to form a VMD-LSTM-ViT-XGB fusion intelligent model. Use the fusion intelligent model for road compaction quality evaluation, input the surface wave signal and the corresponding compaction influencing factors of the subgrade layer, cement stabilized macadam layer, and asphalt surface layer, as well as the structural layer category and stage category, and finally realize fast and lightweight inspection of road spatio-temporal compaction quality. Once the fusion intelligent model is built, in actual use, there is no need to construct the time-frequency map classification dataset, denoising dataset, and road spatio-temporal fusion dataset anymore. During on-site use, only by receiving the surface wave signal excited by external force and leaking into the air in real time can compaction detection be realized.

[0072] Example 2

[0073] The road spatio-temporal compaction quality inspection system based on acoustic-solid coupling in this example further includes a cloud, which is used to store the denoising dataset, time-frequency map classification dataset, and road spatio-temporal fusion dataset. The cloud communicates with the control terminal, and the cloud can obtain on-site data in a timely manner and update the corresponding dataset in a timely manner.

[0074] Example 3

[0075] In this embodiment, a drop hammer unit is provided. The drop hammer unit can also be fixed outside the outermost sound pressure sensor through a bracket. The drop hammer unit is electrically connected to the control terminal, and the magnitude and frequency of the force exerted by the drop hammer unit can be set, enabling it to be automatically excited at fixed time intervals during on-site use to meet the purpose of automatic continuous sampling or intermittent sampling.

[0076] Embodiment 4

[0077] In this embodiment, two sleeves are movably connected to the bracket. The sleeves can move up and down relative to the bracket and can be fixed at the target height. The windshield structure is fixed on the two sleeves and can move up and down with the sleeves. During use, the sleeves can be manually lowered at the compaction sampling point until the windshield structure on the sleeves contacts the surface to be measured, and then hammering is performed to collect surface acoustic waves.

[0078] Embodiment 5

[0079] In this embodiment, the sleeve is electrically connected to the lifting unit, and the lifting unit is electrically connected to the control terminal. The sleeve can be automatically lifted and lowered under the control of the lifting unit. The lifting unit uses a motor and a chain, etc., to adjust the height of the windshield structure and the sound pressure sensor from the surface to be measured. The horizontal distance between the two sound pressure sensors and the vertical distance between the sound pressure sensor and the compaction sampling point can be set in the control terminal. The windshield structure always covers the periphery of the sound pressure sensor and moves up and down with the sound pressure sensor, enabling the windshield structure and the sound pressure sensor to achieve automatic lifting and detection on-site without affecting the movement of the mobile vehicle, realizing automated inspection.

[0080] Matters not described in the present invention are applicable to the prior art.

Claims

1. A road spatiotemporal compaction quality inspection method based on acoustic-solid coupling, characterized in that: The road spatiotemporal compaction quality inspection method based on acoustic-solid coupling includes the following contents: The acoustic pressure sensor is used to contactlessly receive the surface wave signal excited by external force and leaked into the air at the compaction sampling point; The VMD algorithm is used to decompose the surface wave signal into several modes, and multiple modes with smooth waveforms and consistent with the main frequency range of the surface wave signal are selected for reconstruction to obtain the reconstructed signal; all modes obtained by decomposing the surface wave signal by the VMD algorithm and the corresponding reconstructed signals are combined into denoising samples to construct a denoising data set; The input of the LSTM network is all the modes obtained by decomposition of the VMD algorithm, and the output is the reconstructed signal. The LSTM network is trained using the denoising data set to obtain a trained LSTM network. The VMD algorithm is connected to the trained LSTM network to form a VMD-LSTM denoising module. The compaction degree of the compaction sampling point is obtained, and several compaction degree intervals are divided according to the compaction degree distribution. Each compaction degree interval is used as a compaction degree category. The reconstructed signal obtained by the VMD-LSTM denoising module is processed to generate a two-dimensional time-frequency graph. The two-dimensional time-frequency graph generated by the reconstructed signal is classified according to the compaction degree interval to which the corresponding compaction degree belongs. The two-dimensional time-frequency graph and its corresponding compaction degree category form a classification sample, and a time-frequency graph classification data set is constructed; The classification network is trained using the time-frequency graph classification data set to obtain a trained classification network, wherein the output layer of the classification network is composed of a fully connected layer and a Softmax function, and the trained classification network is used as a classification module after removing the Softmax function to extract features from the two-dimensional time-frequency graph to obtain time-frequency features; Obtain the compaction influencing factors and corresponding compaction degrees under different stage categories and different structural layer categories, and construct a road spatiotemporal fusion dataset; the different stage categories are construction stage, maintenance stage and acceptance and opening stage, the different structural layer categories are road base layer, cement stabilized gravel layer and asphalt surface layer, and the compaction influencing factors are water content, soil particle gradation, surface wave velocity, time-frequency characteristics, gravel gradation, cement content, gross volume density, Marshall stability, flow value, asphalt content, theoretical maximum density of surface layer and surface porosity; Train the XGBoost machine learning algorithm with the road spatiotemporal fusion dataset to obtain the trained XGBoost machine learning algorithm for compaction detection; Combine the VMD-LSTM denoising module, the classification module and the trained XGBoost machine learning algorithm to form a fusion intelligent model; When used on site, data on compaction influencing factors other than surface wave velocity and time-frequency characteristics are input, and surface wave signals on site are collected. The fusion intelligent model is then used to quickly complete the detection of compaction quality at any stage of any structural layer.

2. The road time-space compaction quality inspection method based on acoustic-solid coupling according to claim 1 is characterized in that: The classification network is implemented using a ViT, VGG, ResNet, MobileNet or Swin Transformer network.

3. The road time-space compaction quality inspection method based on acoustic-solid coupling according to claim 1 is characterized in that: The XGBoost machine learning algorithm is optimized using an intelligent optimization algorithm before training, and the intelligent optimization algorithm is at least one of a PSO algorithm, a GWO algorithm, or a GA algorithm.

4. A road time-space compaction quality inspection system based on acoustic-solid coupling, characterized in that: The road time-space compaction quality inspection system based on acoustic-solid coupling adopts the road time-space compaction quality inspection method based on acoustic-solid coupling according to claim 1, including a mobile vehicle, a dynamic signal collector installed on the mobile vehicle, and a control terminal. A pair of sound pressure sensors are installed at the rear of the mobile vehicle through a bracket and two sleeves, and the two sleeves are symmetrically installed on the bracket. The sound pressure sensor is electrically connected to the dynamic signal collector, and the dynamic signal collector communicates with the control terminal; A pair of acoustic pressure sensors are arranged at both ends of the compaction sampling point and are not in contact with the compaction sampling point, and are used to receive surface wave signals excited by external force and leaked into the air; A wind shield structure is arranged around the pair of sound pressure sensors; The control terminal obtains the surface wave signal collected by the sound pressure sensor through a dynamic signal collector; Load the fusion intelligent model in the control terminal.

5. The road time-space compaction quality inspection system based on acoustic-solid coupling according to claim 4 is characterized in that: The road spatiotemporal compaction quality inspection system based on acoustic-solid coupling also includes a rubber hammer.

6. The road time-space compaction quality inspection system based on acoustic-solid coupling according to claim 4 is characterized in that: The road spatiotemporal compaction quality inspection system based on acoustic-solid coupling also includes a cloud, which is used to store a denoised data set, a time-frequency graph classification data set, and a road spatiotemporal fusion data set, and the cloud communicates with a control terminal.

7. The road time-space compaction quality inspection system based on acoustic-solid coupling according to claim 4 is characterized in that: During sampling, the vertical distance between the sound pressure sensor and the compaction sampling point was controlled within 20 cm.

8. The road time-space compaction quality inspection system based on acoustic-solid coupling according to claim 4 is characterized in that: The two sleeves are movably connected to the bracket, the sleeves can move up and down relative to the bracket, and the wind shielding structure is fixed on the two sleeves and can move up and down with the sleeves.

9. The road time-space compaction quality inspection system based on acoustic-solid coupling according to claim 8 is characterized in that: The two sleeves are electrically connected to the lifting unit, and the lifting unit is electrically connected to the control terminal. The sleeves can be automatically lifted and lowered under the control of the lifting unit.

Images