High liquid limit fly ash roadbed compaction quality intelligent dynamic regulation method

Abstract

The application discloses a kind of high liquid limit fly ash roadbed compaction quality intelligent dynamic regulation methods, the mapping relationship of high liquid limit fly ash hyperspectral characteristics and specific surface area index, total amount of hydroxyl and moisture content is established by first neural network model, second neural network model establishes optimal moisture content prediction model, and the correlation of compaction degree and dielectric constant, stiffness, moisture content is constructed by third to fifth neural network model;Through fiber probe and unmanned aerial vehicle hyperspectral imaging, the moisture content of yard and paving layer is two-stage regulated, and pre-wetting and secondary accurate regulation are carried out in combination with intelligent spraying system;In the compaction stage, data is collected in real time by multi-frequency ground penetrating radar and vibration sensor, compaction degree distribution is inverted and stratified evaluation is carried out, and rolling parameters are dynamically adjusted.The application integrates hyperspectral technology, multi-source sensing and deep learning, realizes the intelligent management of compaction quality, and significantly improves the uniformity and quality stability of fly ash roadbed construction.

Description

Technical Field

[0001] This invention relates to the field of comprehensive utilization technology of solid waste fly ash, specifically to an intelligent dynamic control method for the compaction quality of high liquid limit fly ash roadbed. Background Technology

[0002] Fly ash, as an industrial byproduct, is widely used in roadbed filling due to its lightweight and good permeability. However, high liquid limit fly ash (liquid limit w) L Fly ash with a moisture content greater than 50% presents significant technical challenges during compaction, as its engineering characteristics are greatly affected by differences in raw material sources, particle size distribution, and mineral composition. The physicochemical properties of fly ash produced in different regions (such as power plants, steel mills, and chemical plants) and with different production processes (such as wet and dry discharge) vary significantly, making it difficult to standardize construction parameters. This manifests as large fluctuations in compaction effects and complex moisture content control.

[0003] Numerous studies have shown that the compaction quality of fly ash is significantly affected by particle size distribution and moisture content.

[0004] For coarse-grained fly ash (D 50 >0.1mm), well-graded (C) u >6, C c When the content is 1-3, the particle skeleton structure is stable, the porosity is low (approximately 25% to 30%), and the optimum moisture content is relatively low (w). opt (≈12% to 15%), easily achieving a compaction degree of over 93% through vibration compaction. Meanwhile, coarse particles are sensitive to moisture content; when w deviates from w... opt At ±1.5%, the dry density ρ d The decrease can reach 0.2 g / cm3.

[0005] For fine particulate fly ash (D 50 <0.05mm), single gradation (C) u <4), large specific surface area (greater than 500 m²) 2 For fly ash particles with a moisture content of / kg, a higher moisture content is required for lubrication, thus the optimal moisture content is relatively high. Fine particles have strong water retention and slow moisture migration, requiring a longer time to reach moisture equilibrium after humidification. Simultaneously, fine-particle fly ash is prone to forming a "springy soil" phenomenon, resulting in a loose and cracked surface after compaction. Under the same compaction process as coarse-particle fly ash, the compaction degree is typically below 85%, making it difficult to meet design requirements.

[0006] In addition, the content of hydrophilic minerals also affects the optimum moisture content to some extent. For every 5% increase in montmorillonite content, the optimum moisture content increases by 1.5% to 2.0%.

[0007] For a fixed batch of fly ash, under the condition that the particle size distribution remains unchanged, the moisture content directly affects the control of compaction quality. The control of moisture content involves two dimensions: the magnitude of the actual moisture content and the degree to which the actual moisture content approaches the optimum moisture content.

[0008] Regarding the control of moisture content in fly ash roadbed, the commonly used method is to take samples when the material arrives at the site and test them using the drying method. After paving and before compaction, since the drying method requires a long time and cannot be closely integrated with on-site construction, it is usually determined by visual observation and experience whether moisture needs to be added or if the moisture content is too high. When moisture needs to be added, water is sprayed by a sprinkler truck based on experience. When the moisture content needs to be reduced, it is appropriately dried.

[0009] The existing methods for controlling moisture content have several problems, including: only a few samples are obtained upon arrival at the site, making it impossible to grasp the overall moisture content of the fly ash upon arrival and after spreading; the quality of compaction is mainly determined by the moisture content of the fly ash before compaction, and experience alone cannot comprehensively and accurately determine the true moisture content of the fly ash before compaction, thus making it impossible to control the moisture content to the target value (generally w). opt +2%); Because high liquid limit fly ash is relatively difficult to compact, it requires more compaction passes. During the compaction process, especially shallow fly ash, the moisture content will change continuously due to the influence of weather, temperature and other factors. The extent and nature of these changes cannot be accurately grasped, and therefore, it cannot be controlled according to the moisture content control value.

[0010] Currently, the on-site sand filling method is commonly used for compaction testing of fly ash roadbeds. This method involves point sampling, resulting in low testing efficiency, limited coverage, and difficulty in timely, efficient, and accurate assessment of the compaction quality of fly ash roadbeds.

[0011] In existing technologies, there are methods that use ground-penetrating radar (GPR) data to correlate with the dielectric constant of soil in order to evaluate the compaction degree of roadbeds. The physical basis for GPR assessment of compaction degree is the dielectric constant of the material. The dielectric constant is a comprehensive indicator, simultaneously influenced by density (compaction degree), moisture content, and material composition. For the same batch of material, the dielectric constant increases with increasing compaction degree, and also increases with increasing moisture content. If the dielectric constant at a certain location increases, GPR cannot directly distinguish whether it is due to increased compaction degree, increased moisture content, or both. Therefore, under constant moisture content, a relationship between GPR signals and dielectric constant can be established, and further, a relationship between compaction degree and dielectric constant can be established, thus achieving quantitative evaluation. However, fly ash has high porosity, and its moisture content may change before and during compaction due to weather, temperature, and other factors, making the evaluation of compaction degree based on dielectric constant more complex and less accurate.

[0012] In existing technologies, there are also methods that use the CMV (Compaction Value) of vibratory rollers to evaluate the compaction degree of roadbeds. The interaction between the vibratory roller wheel and the soil generates higher-order harmonics, and the ratio of the amplitude of this higher-order harmonic to the amplitude of the fundamental wave is the CMV value. Representative points are selected at the work site, and while measuring the CMV value, the actual dry density and moisture content are measured using traditional methods (such as nuclear density gauges or sand cone methods). Only by establishing a correlation curve between the CMV value and the traditional measured values ​​can the real-time displayed CMV value be interpreted as a meaningful compaction degree index. Therefore, evaluating compaction degree based on the CMV method is indirect and relative, and its accuracy is generally lower than that of the ground-penetrating radar method; the CMV method mainly reflects the overall compaction state of the compacted layer and cannot identify the differences between the surface and deep layers.

[0013] Furthermore, during the compaction of high liquid limit fly ash, the shallow layer of fly ash becomes relatively loose and densely populated with cracks due to the vibration tension exerted by the vibratory roller. This loose layer has a low dielectric constant, which significantly affects the vibration signal, leading to a substantial underestimation of the overall compaction degree of the compacted layer. This factor also makes it difficult to directly apply existing ground-penetrating radar and CMV methods to the evaluation and dynamic control of compaction quality in high liquid limit fly ash roadbeds. Summary of the Invention

[0014] The technical problem to be solved by the present invention is to provide a method that can control the moisture content of fly ash throughout the entire process, adjust the moisture content of fly ash in stages, accurately evaluate the compaction quality of fly ash subgrade across the entire cross section, and dynamically adjust the compaction strategy based on the compaction situation, in light of the existing technology.

[0015] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows:

[0016] A method for intelligent dynamic control of compaction quality of high liquid limit fly ash roadbed includes the following steps:

[0017] A first neural network model was established to establish the relationship between the hyperspectral characteristics of high liquid limit fly ash and its specific surface area index, total hydroxyl content, and moisture content.

[0018] Based on step S1, a second neural network model was established to establish the relationship between the optimal moisture content of high liquid limit fly ash and the specific surface area index and total hydroxyl content, combining X-ray diffraction test, thermogravimetric mass spectrometry test, and compaction test.

[0019] A third neural network model was established to establish the relationship between fly ash compaction degree and dielectric constant and moisture content; a fourth neural network model was established to establish the relationship between fly ash compaction degree and dielectric constant and stiffness; and a fifth neural network model was established to establish the relationship between fly ash compaction degree and dielectric constant.

[0020] Spectral data is collected by fiber optic probes in the fly ash stockpile. The specific surface area index, moisture content and optimum moisture content are inverted based on the first neural network model and the second neural network model. When the moisture content is lower than the optimum moisture content, pre-humidification is controlled by an intelligent spray system.

[0021] The pre-controlled fly ash is transported to the site for paving. Hyperspectral images of the paving layer are obtained by a drone-borne hyperspectral camera. The moisture field of the fly ash is generated based on the first neural network model, and the optimal moisture content distribution field is generated based on the second neural network model.

[0022] The difference between the moisture field and the optimal moisture content distribution field is analyzed, and secondary moisture content regulation is carried out. Humidification regulation is applied to areas where the moisture content is lower than the moisture content control value.

[0023] High-frequency and low-frequency combined ground-penetrating radar and vibration sensors are integrated on the road roller to test the dielectric constant and stiffness parameters of fly ash subgrade in real time during the compaction process. Based on the third, fourth and fifth neural network models, the compaction degree distribution in different depth ranges from 0 to 60 cm is inverted, and the compaction degree is evaluated in layers according to the depth range. The rolling strategy is dynamically adjusted according to the compaction degree.

[0024] Preferably, the first neural network is a convolutional neural network. The model is input with a preprocessed hyperspectral image of fly ash. Image features are extracted through convolutional layers, pooling layers, and fully connected layers. The specific surface area index calculated from particle size distribution data measured by a laser particle size analyzer, the total hydroxyl content data obtained by thermogravimetric mass spectrometry (TGA), and the moisture content data measured by drying are used as labels for training. The output is the specific surface area index, total hydroxyl content, and moisture content of high-liquid-limit fly ash. The convolutional and fully connected layers of the first neural network use the ReLU activation function, and the output layer uses a linear activation function.

[0025] The specific surface area index is calculated according to the following formula:

[0026]

[0027] Among them: I s p is the specific surface area index of fly ash. i D represents the mass percentage of the i-th particle size range; i Let be the representative particle size of the i-th particle size range; the particle size range is defined as 0 to d. 10 d 10 ~d 20 d 20 ~d 30 d 30 ~d 40 d 40 ~d 50 d 50 ~d60 d 60 ~d 70 d 70 ~d 80 d 80 ~d 90 d 90 ~d 100 The particle size is divided into segments, and the representative particle size for each segment is the average value of the segment; the d... 10 d 20 d 30 d 40 d 50 d 60 d 70 d 80 d 90 d 100 The values ​​represent particle sizes representing cumulative sieve mass percentages of 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, and 100%.

[0028] Preferably, in the second neural network model, the specific surface area index and total hydroxyl content of fly ash are used as inputs, and the optimal moisture content is used as the output; in the third neural network model, the dielectric constant calculated by ground-penetrating radar wave velocity and the moisture content of fly ash are used as inputs, and the compaction degree at a depth of 0 to 5 cm is used as the output; in the fourth neural network model, the dielectric constant calculated by ground-penetrating radar wave velocity and the stiffness calculated by vibration spectrum are used as inputs, and the compaction degree at a depth of 5 to 30 cm is used as the output; in the fifth neural network model, the dielectric constant calculated by ground-penetrating radar wave velocity is used as input, and the compaction degree at a depth of 30 to 60 cm is used as the output.

[0029] Preferably, the second, third, fourth, and fifth neural network models are all fully connected neural networks, with the hidden layers all using the ReLU activation function and the output layers all using the linear activation function.

[0030] Preferably, the above-mentioned hyperspectral characteristics are hyperspectral reflectance data with wavelengths from 400 to 2500 nm.

[0031] Preferably, the fiber optic probe is a reflective multimode fiber with a wavelength range of 400 to 2500 nm. The sampling method involves random sampling of fly ash from the area to be transported and utilized, and the real-time data is processed through a computing terminal.

[0032] Preferably, the imaging band of the UAV-borne hyperspectral camera is 400 to 2500 nm, and the spectral resolution is no greater than 10 nm. The obtained moisture field and optimal moisture content distribution field are superimposed on the construction GIS platform in the form of a heat map.

[0033] Preferably, the moisture content control value mentioned above is (wopt +2%), w opt To achieve the optimal moisture content, the humidification amount is dynamically allocated based on the difference between the moisture content control value and the current moisture content of the 2m×1m grid cell area.

[0034] Preferably, the above-mentioned multi-frequency ground-penetrating radar adopts a combination of high-frequency and low-frequency methods, and the frequency selection method is as follows:

[0035]

[0036] Where: f1 is the frequency of high-frequency ground-penetrating radar; f2 is the frequency of low-frequency ground-penetrating radar; c is the speed of light; h1 is the thickness of the relatively loose layer on the surface of fly ash; ε1 is the dielectric constant of the relatively loose layer on the surface of fly ash; h2 is the thickness of the normally compacted layer of fly ash, which is generally 0.3m; ε2 is the dielectric constant of the normally compacted layer of fly ash.

[0037] Preferably, the above-mentioned method for evaluating compaction degree by stratification is as follows:

[0038] High-frequency ground-penetrating radar (GPR) was used to detect cracks and loose areas within a 5cm radius of the surface layer. Risk areas were identified by the intensity of reflected signal scattering and abnormal dielectric constant values. A fly ash moisture field was generated based on a first neural network model, and the dielectric constant and moisture content at the corresponding locations were extracted. The surface compaction degree was output through a third neural network. High-frequency GPR is not suitable for testing shallow layers, and the inversion results of vibration sensors are distorted for loose surface areas. Therefore, obtaining the moisture content at the corresponding locations based on the first neural network and combining it with high-frequency GPR has higher accuracy.

[0039] Low-frequency ground-penetrating radar and vibration sensors were used to simultaneously collect wave velocity and vibration signals from the middle layer (5 to 30 cm). The dielectric constant ε2 was calculated based on the wave velocity, and the stiffness k1 of the fly ash subgrade was calculated based on the vibration signals. The compaction degree of the middle layer was output through a fourth neural network and calculated according to the following formula:

[0040]

[0041] In the formula, k1 is the equivalent stiffness of the fly ash subgrade after eliminating the influence of the relatively loose fly ash on the surface; k0 is the equivalent stiffness of the fly ash subgrade calculated based on vibration parameters; λ is the fly ash material property coefficient, calculated based on the damping ratio and vibration frequency; h1 is the thickness of the relatively loose fly ash surface layer; m e denoted as ω, where r is the mass of the eccentric block of the vibratory roller; ω is the eccentricity; ω is the excitation angular frequency; a is the measured acceleration amplitude of the vibratory wheel; η is the fly ash damping coefficient; and m is the total mass of the roller's vibration system.

[0042] The moisture content of the middle layer of fly ash (5 to 30 cm) is difficult to test, so the dielectric constant is used as the main indicator, combined with a modified vibration signal, to achieve high accuracy.

[0043] Using low-frequency ground-penetrating radar data, the compaction degree at a depth of 30 to 60 cm is output through the fifth neural network, and trend analysis is performed by combining the deep dielectric constant and the previous compaction degree data.

[0044] The fly ash layer of 30 to 60 cm has been tested and evaluated during the compaction of the previous layer. Most of it has met the requirements, and the few deficiencies in a few places have been reinforced during the second compaction. At the same time, the moisture content of the fly ash in this layer is less affected by weather and temperature and is relatively stable. Therefore, using the dielectric constant as input can achieve high accuracy.

[0045] Preferably, the above-mentioned dynamic adjustment compaction strategy is as follows:

[0046] For surface cracks of 0-5cm, reduce the amplitude by 50%, the rolling speed should not exceed 2km / h, and increase the number of static compaction passes by 1 to 2.

[0047] For the 5-30cm under-compacted middle layer, the amplitude is increased by 10% to 20%, and the rolling speed is reduced to 1.5km / h;

[0048] For abnormal areas at a depth of 30-60cm, artificial intervention is recommended;

[0049] When multiple layers have insufficient compaction, measures should be taken gradually from the deepest layers to the shallowest layers.

[0050] This invention deeply integrates hyperspectral imaging, multi-frequency ground-penetrating radar, vibration sensing, and deep learning models to construct a closed-loop system for "sensing-analysis-control" of fly ash roadbeds. Through multi-model collaboration, multi-technology integration, and dynamic control throughout the entire process, it significantly improves the compaction quality and construction efficiency of high liquid limit fly ash roadbeds, and solves the major defects of traditional methods in terms of material heterogeneity, moisture content sensitivity, and one-sidedness in compaction assessment.

[0051] The beneficial effects of this invention are also reflected in the following aspects: through a multi-stage progressive control method, the specific surface area index, total hydroxyl content, moisture content, and optimum moisture content of the material from the stockpile to the site can be accurately grasped and controlled; through the high-precision inversion of multi-level compaction, the limitations of traditional single-parameter detection can be broken through, and the full-section non-destructive testing of fly ash roadbed compaction can be realized. Attached Figure Description

[0052] Figure 1 This is the overall flowchart of the intelligent dynamic control method for moisture content and compaction degree of high liquid limit fly ash roadbed of the present invention;

[0053] Figure 2 This is a structural diagram of the first neural network model of the present invention;

[0054] Figure 3This is a structural diagram of the second neural network model of the present invention;

[0055] Figure 4 This is a structural diagram of the third neural network model of the present invention;

[0056] Figure 5 This is a structural diagram of the fourth neural network model of the present invention;

[0057] Figure 6 This is a structural diagram of the fifth neural network model of the present invention;

[0058] Figure 7 A moisture field thermogram for paving without prior moisture content pre-control.

[0059] Figure 8 The thermal map shows the optimal moisture content distribution field before moisture content pre-control was performed on site.

[0060] Figure 9 Moisture field thermogram for pre-controlling moisture content before on-site paving;

[0061] Figure 10 The optimal moisture content distribution field is shown in the thermogram for pre-controlling moisture content before on-site paving.

[0062] Figure 11 The thermal map of surface compaction distribution without pre-control of moisture content and dynamic control of the compaction process;

[0063] Figure 12 Thermographic diagram of surface compaction distribution after moisture content control and dynamic control of the compaction process;

[0064] Figure 13 The thermal diagram of the compaction degree distribution in the vertical section without moisture content control and dynamic control of the compaction process;

[0065] Figure 14 The thermal diagram shows the vertical cross-sectional compaction degree distribution after moisture content control and dynamic control of the compaction process. Detailed Implementation

[0066] The present invention will be further described below with reference to the accompanying drawings and embodiments to more clearly and completely describe the technical solution of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The following description of at least one embodiment is illustrative in nature and is in no way intended to limit the present invention or its application or use. Other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are all within the scope of protection of the present invention. Unless otherwise specifically stated, the relative positions, quantities, and dimensions of the components and steps described in these embodiments do not limit the scope of the present invention.

[0067] Example:

[0068] This embodiment uses high-liquid-limit fly ash produced by a power plant as the treatment target. Its basic characteristics are: extremely fine particles, P 0.075 >90%, P 0.045 Approximately 50%–60%, optimum moisture content (w opt The content of hydrophilic minerals (mainly montmorillonite and illite) is as high as 42% to 47%, with a high content of hydrophilic minerals.

[0069] Figure 1 This is a flowchart of an intelligent dynamic control method for compaction quality of high liquid limit fly ash roadbed according to the present invention.

[0070] Establish such as Figure 2 The first neural network model showing the relationship between the hyperspectral characteristics of high liquid limit fly ash and particle size distribution parameters, mineral composition, and moisture content is a convolutional neural network model. The input layer is a preprocessed hyperspectral image, and the output layer is the specific surface area index, total hydroxyl content, and moisture content of fly ash.

[0071] The specific surface area index is calculated according to the following formula:

[0072]

[0073] Among them: I s p is the specific surface area index of fly ash. i D represents the mass percentage of the i-th particle size range; i Let be the representative particle size of the i-th particle size range; the particle size range is defined as 0 to d. 10 d 10 ~d 20 d 20 ~d 30 d 30 ~d 40 d 40 ~d 50 d 50 ~d 60 d 60 ~d 70 d 70 ~d 80 d 80 ~d 90 d 90 ~d 100 The particle size is divided into segments, and the representative particle size for each segment is the average value of the segment; the d... 10 d 20 d 30 d 40 d 50 d 60 d 70 d80 d 90 d 100 The values ​​represent particle sizes representing cumulative sieve mass percentages of 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, and 100%.

[0074] In the first neural network model, the hyperspectral features are the hyperspectral reflectance data of fly ash. The input is a preprocessed hyperspectral image of fly ash (wavelength 400 to 2500 nm). The preprocessing methods include normalization and denoising. Normalization uses the minimum-maximum normalization method to scale the spectral reflectance data to the range [0,1]. Denoising uses Gaussian filtering. The model includes two convolutional layers, two pooling layers, and one fully connected layer. The convolutional layers use 3×3 kernels with a stride of 1000. The pooling layer uses max pooling with a window size of 2×2. Image features are extracted through the convolutional, pooling, and fully connected layers to select feature bands highly correlated with particle size distribution parameters, mineral composition, and moisture content. The convolutional and fully connected layers use the ReLU activation function, and the output layers all use linear activation functions.

[0075] During training, the specific surface area index calculated from particle size distribution data measured by a laser particle size analyzer, the total hydroxyl content data obtained by thermogravimetric mass spectrometry (TGA / MS), and the moisture content data measured by drying were used as labels. The sample set was based on 100 representative fly ash samples, with 80% used for training and 20% for validation. The number of iterations was 200, and the learning rate was 0.001. After training, the mean absolute errors (MAE) of the first neural network inverting the specific surface area index, total hydroxyl content, and moisture content were 0.04, 0.10%, and 0.5%, respectively, with a coefficient of determination R0. 2 All are above 0.95.

[0076] like Figure 3As shown, combining X-ray diffraction, thermogravimetric-mass spectrometry (TGA-MS), and compaction tests, a second neural network model was established based on the first neural network to model the relationship between the optimal moisture content of high liquid limit fly ash and its specific surface area index and total hydroxyl content. The input layer consists of the specific surface area index and total hydroxyl content of fly ash, and the output layer is the optimal moisture content. Three fully connected layers were used, with 128, 64, and 32 neurons respectively, and the activation function was Tanh. The output layer was the optimal moisture content of fly ash, and the activation function was linear. The sample set was based on X-ray diffraction, TGA-MS, and compaction test data from 100 representative fly ash samples. The specific surface area index calculated from particle size distribution data measured by a laser particle size analyzer, the total hydroxyl content obtained from TGA-MS, and the optimal moisture content obtained from the compaction test were used as labels. 80% of the data was used for training, and 20% for validation. The number of iterations was 150, and the learning rate was 0.001. After training, the mean absolute error (MAE) of the first neural network inverting the optimal moisture content was 0.5%, and the coefficient of determination R0 was [missing value]. 2 It is 0.94.

[0077] like Figure 4 As shown, a third neural network model was established to establish the relationship between fly ash compaction degree and dielectric constant and moisture content. The input layer consists of the dielectric constant calculated from ground-penetrating radar wave velocity and the fly ash moisture content; the fully connected layers are hidden layers, with three layers having 128, 64, and 32 neurons respectively, and the activation function being ReLU; the output layer is the fly ash compaction degree, with a linear activation function. The sample set is based on the dielectric constant, moisture content measured by drying method, and compaction degree data of 100 representative compacted fly ash samples, of which 80% was used for training and 20% for validation, with 200 iterations and a learning rate of 0.001. After training, the mean absolute error (MAE) of the third neural network inverting compaction degree is 0.010, and the coefficient of determination R² is 0.99.

[0078] like Figure 5 As shown, a fourth neural network model was established to establish the relationship between fly ash compaction degree and dielectric constant and stiffness. The input layer consists of the dielectric constant calculated from ground-penetrating radar wave velocity and the stiffness calculated from vibration spectrum. Three fully connected layers were used as hidden layers, with 128, 64, and 32 neurons respectively, and the activation function was ReLU. The output layer represents the fly ash compaction degree, with a linear activation function. The sample set was based on dielectric constant, stiffness, and compaction degree data from 100 representative compacted fly ash samples, with 80% used for training and 20% for validation. The number of iterations was 300, and the learning rate was 0.001. After training, the mean absolute error (MAE) of the fourth neural network inverting compaction degree was 0.023, and the coefficient of determination R² was 0.96.

[0079] like Figure 6As shown, a fifth neural network model was established to establish the relationship between fly ash compaction degree and dielectric constant. The input layer is the dielectric constant calculated from the ground-penetrating radar wave velocity; the fully connected layers are hidden layers, set to 3 layers, with ReLU, and the number of neurons is 128, 64, and 32 respectively, with ReLU activation function; the output layer is the compaction degree of fly ash, with linear activation function. The sample set is based on dielectric constant and compaction degree data of 100 representative compacted fly ash samples, of which 80% is used for training and 20% for validation, with 300 iterations and a learning rate of 0.001. After training, the mean absolute error (MAE) of the fifth neural network inverting compaction degree is 0.031, and the coefficient of determination R0 is 0.031. 2 It is 0.94.

[0080] Spectral data was collected using fiber optic probes at the fly ash storage yard. Particle size distribution parameters, moisture content, and optimum moisture content were retrieved based on a first neural network model and a second neural network model. When the moisture content was lower than the optimum moisture content, pre-humidification was controlled via an intelligent spray system. The fiber optic probes were reflective multimode fibers with wavelengths covering 400 to 2500 nm. Sampling was performed by randomly selecting fly ash from the area to be transported and utilized. Real-time data was processed through a computing terminal.

[0081] Pre-controlled fly ash was transported to the site for paving. Hyperspectral images of the paving layer were acquired using a drone-borne hyperspectral camera. A moisture field of the fly ash was generated based on a first neural network model, and an optimal moisture content distribution field was generated based on a second neural network model. The imaging band of the drone-borne hyperspectral camera was 400 to 2500 nm, with a spectral resolution of 10 nm and a spatial resolution of 5 cm. The obtained moisture field and optimal moisture content distribution field were both overlaid onto the construction GIS platform in the form of heat maps.

[0082] The difference between the moisture field and the optimum moisture content distribution field was analyzed, and secondary moisture content regulation was implemented. Humidification was applied to areas where the moisture content was below the control value. The control value for moisture content was (w... opt +2%), the humidification amount is dynamically allocated based on the difference between the moisture content control value of the 2m×1m grid cell area and the current moisture content.

[0083] Figure 7 The moisture field thermogram is shown for areas where moisture content was not pre-controlled before paving. The moisture content ranges from 40.2% to 42.4%, with some areas ranging from 36.1% to 40.7%.

[0084] Figure 8 The image shows a heat map of the distribution of moisture control values ​​before on-site paving without pre-control, ranging from 44.5% to 47.0%. The difference between the actual moisture content and the moisture control value is mostly between -4% and -6%, with some local differences reaching -7% to -8%.

[0085] Figure 9 The moisture field thermogram is for pre-control of moisture content before on-site paving. The moisture content distribution is between 43% and 47%, with only a few local areas having a moisture content distribution between 41% and 43%.

[0086] Figure 10 The image shows a heat map of the distribution of moisture control values ​​before on-site paving, ranging from 44.5% to 47.0%. The difference between the actual moisture content and the moisture control value is mostly between -1% and -2%, with only a few local areas showing a difference of -3% to -4%.

[0087] Therefore, after implementing moisture content pre-control measures, the moisture content distribution is significantly more uniform and much closer to the moisture content control value. After on-site paving, only a few areas need to undergo secondary moisture content adjustment, achieving a significant effect in moisture content control.

[0088] Subsequently, a 26t steel-drum vibratory roller was used, with a compaction width of 2.2m, an excitation force of 360 to 430kN, a nominal amplitude of 1.3 to 2.0mm, and a vibration frequency of 28 to 32Hz. A high-frequency and low-frequency combined ground-penetrating radar and vibration sensors were integrated into the roller to collect real-time data on the dielectric constant and stiffness parameters of the fly ash subgrade during compaction. Based on third, fourth, and fifth neural network models, the compaction degree distribution at different depths from 0 to 60cm was inverted. Compaction degree was evaluated in layers according to depth range, and the compaction strategy was dynamically adjusted based on the compaction degree.

[0089] The multi-frequency ground-penetrating radar uses a dual-frequency array of 400MHz and 2.5GHz.

[0090] The compaction degree stratification evaluation method is as follows: High-frequency ground-penetrating radar is used to detect cracks and loose areas within a 5cm range of the surface layer. Risk areas are determined by the scattering intensity of reflected signals and the abnormal value of dielectric constant. Based on the first neural network model, a fly ash moisture field is generated, and the dielectric constant and moisture content at the corresponding locations are extracted. The compaction degree of the surface layer from 0 to 5cm is output through the third neural network. Low-frequency ground-penetrating radar and vibration sensors are used to synchronously collect wave velocity and vibration signals from the middle layer from 5 to 30cm. The dielectric constant ε2 is calculated based on the wave velocity, and the stiffness k1 of the fly ash subgrade is calculated based on the vibration signal. The compaction degree of the middle layer is output through the fourth neural network.

[0091]

[0092] In the formula, k1 is the equivalent stiffness of the fly ash subgrade after eliminating the influence of relatively loose fly ash on the surface; k0 is the equivalent stiffness of the fly ash subgrade calculated based on vibration parameters; λ is the fly ash material property coefficient, calculated based on damping ratio and vibration frequency; k0 is the overall stiffness of the fly ash subgrade calculated based on vibration parameters; m eThe mass of the eccentric block of the vibratory roller is 320 kg; r is the eccentricity, 0.14 m; ω is the excitation angular frequency, 151 rad / s; a is the measured acceleration amplitude of the vibratory wheel, ranging from 12 to 25 m / s². 2 η is the damping coefficient of fly ash, taken as 0.20; m is the total mass of the road roller vibration system, which is 10500 kg.

[0093] Using low-frequency ground-penetrating radar data, the compaction degree at a depth of 30 to 60 cm is output through the fifth neural network, and trend analysis is performed by combining the deep dielectric constant and the previous compaction degree data.

[0094] For surface cracked areas, reduce the amplitude by 50%, the rolling speed should not exceed 2 km / h, and increase the number of static compaction passes by 2; for under-compacted areas in the middle layer, increase the amplitude by 15% and reduce the rolling speed to 1.5 km / h; for deep abnormal areas, artificial intervention is recommended; when multiple layers have substandard compaction, measures should be taken gradually from deep to shallow layers.

[0095] like Figure 11 As shown, without adjusting the moisture content of fly ash and dynamically controlling the compaction process, the surface compaction degree of fly ash roadbed ranges from 0.855 to 0.879, which is significantly different from the compaction degree control standard of 0.900.

[0096] like Figure 12 As shown, after adjusting the moisture content of fly ash and dynamically controlling the compaction process, the surface compaction degree of fly ash roadbed ranges from 0.890 to 0.899, basically reaching the compaction degree control standard of 0.900.

[0097] Figure 13 The thermal map shows the compaction degree distribution of the vertical cross section without control of fly ash moisture content and dynamic control of the compaction process. The compaction degree in the 0-5cm range is 0.875 to 0.879, the compaction degree in the 5-30cm range is 0.887 to 0.898, and the compaction degree in the 30-60cm range is 0.894 to 0.902. Overall, the compaction degree is high at the bottom and low at the top, and it is difficult to meet the compaction degree control standard of 0.900 specified in the standard.

[0098] Figure 14 The thermal map of vertical cross-section compaction degree distribution after dynamic control of fly ash moisture content and compaction process was obtained. The compaction degree in the 0-5cm range was 0.896 to 0.899, the compaction degree in the 5-30cm range was 0.909 to 0.914, and the compaction degree in the 30-60cm range was 0.917 to 0.919. Overall, the compaction degree showed a high degree at the bottom and a low degree at the top. The surface compaction degree basically met the compaction degree control standard, while the compaction degree in the bottom was significantly better than the compaction degree control standard.

[0099] Therefore, compared to before moisture content control, dynamic compaction assessment, and adjustment of the compaction strategy, the overall compaction degree of the surface plane of the fly ash subgrade after adjustment generally met the design requirement (0.900), the compaction degree within a 5cm depth of the surface layer basically met the design requirement, and the overall compaction degree of the lower layer was further improved, better meeting the design requirements. The above examples further demonstrate that high liquid limit fly ash is difficult to compact to the specified compaction degree of 0.900 using existing control methods. However, the intelligent dynamic control method for compaction quality of high liquid limit fly ash subgrade according to this invention achieves outstanding control results.

[0100] Obviously, the above embodiments are merely illustrative examples for clear explanation and not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom should still be considered within the scope of protection of this invention.

Claims

1. A method for intelligent dynamic control of compaction quality of high liquid limit fly ash roadbed, characterized in that, Includes the following steps: S1. Establish a first neural network model for the relationship between the hyperspectral characteristics of high liquid limit fly ash and specific surface area index, total hydroxyl content, and moisture content; S2. Based on step S1, and combining X-ray diffraction, thermogravimetric mass spectrometry and compaction test, establish a second neural network model relating the optimal moisture content of high liquid limit fly ash to the specific surface area index and total hydroxyl content. S3. Establish a third neural network model relating fly ash compaction degree to dielectric constant and moisture content; establish a fourth neural network model relating fly ash compaction degree to dielectric constant and stiffness; and establish a fifth neural network model relating fly ash compaction degree to dielectric constant. S4. Use fiber optic probes to collect spectral data at the fly ash stockpile. Based on the first and second neural network models, invert the specific surface area index, moisture content, and optimum moisture content. When the moisture content is lower than the optimum moisture content, use an intelligent spray system for pre-humidification control. S5. Transport the pre-controlled fly ash to the site for paving, acquire hyperspectral images of the paving layer using a drone-borne hyperspectral camera, generate a moisture field of fly ash based on the first neural network model, and generate an optimal moisture content distribution field based on the second neural network model; S6. Analyze the difference between the moisture field and the optimal moisture content distribution field, perform secondary moisture content control, and humidify areas with moisture content lower than the moisture content control value. S7. Integrate high-frequency and low-frequency combined ground-penetrating radar and vibration sensors on the road roller to test the dielectric constant and stiffness parameters of fly ash subgrade in real time during the compaction process. Based on the third, fourth and fifth neural network models, invert the compaction degree distribution in different depth ranges from 0 to 60 cm, evaluate the compaction degree in layers according to the depth range, and dynamically adjust the rolling strategy according to the compaction degree. The first neural network is a convolutional neural network. The model takes a pre-processed hyperspectral image of fly ash as input, extracts image features through convolutional layers, pooling layers, and fully connected layers, and uses the specific surface area index calculated from particle size distribution data measured by a laser particle size analyzer, the total hydroxyl content data obtained by thermogravimetric mass spectrometry, and the moisture content data measured by drying as labels for training. The output is the specific surface area index, total hydroxyl content, and moisture content of high-liquid-limit fly ash. The convolutional and fully connected layers of the first neural network use the ReLU activation function, and the output layer uses a linear activation function. The specific surface area index is calculated according to the following formula: ； Among them: I s p is the specific surface area index of fly ash. i D represents the mass percentage of the i-th particle size range; i Let be the representative particle size of the i-th particle size range; the particle size range is defined as 0~d. 10 d 10 ~ d 20 d 20 ~ d 30 d 30 ~ d 40 d 40 ~ d 50 d 50 ~d 60 d 60 ~ d 70 d 70 ~ d 80 d 80 ~ d 90 d 90 ~ d 100 The particle size is divided into segments, and the representative particle size for each segment is the average value of the segment; the d... 10 d 20 d 30 d 40 d 50 d 60 d 70 d 80 d 90 d 100 The values ​​represent particle sizes representing cumulative sieve mass percentages of 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, and 100%.

2. The intelligent dynamic control method for compaction quality of high liquid limit fly ash roadbed according to claim 1, characterized in that, In the second neural network model, the specific surface area index and total hydroxyl content of fly ash are used as inputs, and the optimal moisture content is used as the output. In the third neural network model, the dielectric constant calculated by ground-penetrating radar wave velocity and the moisture content of fly ash are used as inputs, and the compaction degree is used as the output. In the fourth neural network model, the dielectric constant calculated from the ground-penetrating radar wave velocity and the stiffness calculated from the vibration spectrum are used as inputs, and the compaction degree is used as the output. In the fifth neural network model, the dielectric constant calculated from the ground-penetrating radar wave velocity is used as input, and the compaction degree is used as output. The second, third, fourth, and fifth neural network models are all fully connected neural networks, with ReLU activation function used in the hidden layers and linear activation function used in the output layers.

3. The intelligent dynamic control method for compaction quality of high liquid limit fly ash roadbed according to claim 1, characterized in that, The hyperspectral features mentioned in step S1 are hyperspectral reflectance data with wavelengths from 400 to 2500 nm.

4. The intelligent dynamic control method for compaction quality of high liquid limit fly ash roadbed according to claim 1, characterized in that, The fiber optic probe mentioned in step S4 is a reflective multimode fiber with a wavelength range of 400 to 2500 nm. The sampling method is random sampling of fly ash in the area to be transported and utilized, and the real-time data is processed through a computing terminal.

5. The intelligent dynamic control method for compaction quality of high liquid limit fly ash roadbed according to claim 1, characterized in that, In step S5, the imaging band of the UAV-borne hyperspectral camera is 400 to 2500 nm, and the spectral resolution is no greater than 10 nm. The obtained moisture field and optimal moisture content distribution field are superimposed on the construction GIS platform in the form of heat maps.

6. The intelligent dynamic control method for compaction quality of high liquid limit fly ash roadbed according to claim 1, characterized in that, The moisture content control value mentioned in step S6 is w opt +2%, the humidification amount is dynamically allocated based on the difference between the moisture content control value of the 2m×1m grid unit area and the current moisture content.

7. The intelligent dynamic control method for compaction quality of high liquid limit fly ash roadbed according to claim 1, characterized in that, The frequency selection method for the high-frequency and low-frequency combined ground-penetrating radar mentioned in step S7 is as follows: ； ； Where: f1 is the frequency of high-frequency ground-penetrating radar; f2 is the frequency of low-frequency ground-penetrating radar; c is the speed of light; h1 is the thickness of the relatively loose layer on the surface of fly ash; ε1 is the dielectric constant of the relatively loose layer on the surface of fly ash; h2 is the thickness of the normally compacted layer of fly ash, which is generally 0.3m; ε2 is the dielectric constant of the normally compacted layer of fly ash.

8. The intelligent dynamic control method for compaction quality of high liquid limit fly ash roadbed according to claim 1, characterized in that, The aforementioned method for evaluating compaction degree by stratification is as follows: High-frequency ground-penetrating radar is used to detect cracks and loose areas within a 5cm range of the surface layer. Risk areas are determined by the intensity of reflected signal scattering and abnormal values ​​of dielectric constant. Based on the first neural network model, a fly ash moisture field is generated, and the dielectric constant and moisture content at the corresponding locations are extracted. The compaction degree of the surface layer from 0 to 5cm is output through the third neural network model. Low-frequency ground-penetrating radar and vibration sensors were used to simultaneously collect wave velocity and vibration signals from the middle layer (5 to 30 cm). The dielectric constant ε2 was calculated based on the wave velocity, and the stiffness k1 of the fly ash subgrade was calculated based on the vibration signals. The compaction degree of the middle layer was output through a fourth neural network model and calculated according to the following formula: ； ； In the formula, k1 is the equivalent stiffness of the fly ash subgrade after eliminating the influence of the relatively loose fly ash on the surface; k0 is the equivalent stiffness of the fly ash subgrade calculated based on vibration parameters; λ is the fly ash material property coefficient, calculated based on the damping ratio and vibration frequency; h1 is the thickness of the relatively loose fly ash surface layer; m e Let r be the mass of the eccentric block of the vibratory roller; r be the eccentricity. ω is the excitation angular frequency; α is the measured acceleration amplitude of the vibrating wheel; η is the damping coefficient of fly ash, and m is the total mass of the roller vibration system; Using low-frequency ground-penetrating radar data, the compaction degree at a depth of 30 to 60 cm was output through the fifth neural network model, and trend analysis was performed by combining the deep dielectric constant and the previous compaction degree data.

9. The intelligent dynamic control method for compaction quality of high liquid limit fly ash roadbed according to claim 1, characterized in that, The aforementioned dynamic adjustment compaction strategy is as follows: For surface cracks of 0-5cm, reduce the amplitude by 50%, the rolling speed should not exceed 2km / h, and increase the number of static compaction passes by 1 to 2. For the 5-30cm under-compacted middle layer, the amplitude is increased by 10% to 20%, and the compaction speed is reduced to 1.5 km / h; For abnormal areas at a depth of 30-60cm, artificial intervention is recommended; When multiple layers have insufficient compaction, measures should be taken gradually from the deepest layers to the shallowest layers.

Images