A subgrade humidity nondestructive testing method based on wireless radio frequency signal multi-parameter inversion

Abstract

The present application belongs to the technical field of road engineering subgrade nondestructive testing, and discloses a subgrade humidity nondestructive testing method based on wireless radio frequency signal multi-parameter inversion. In order to solve the problems of traditional subgrade humidity testing methods, such as destructiveness, low efficiency, discontinuous monitoring and easily disturbed testing parameters, the specific process is as follows: first, burying RFID tags and obtaining the received signal strength indication value; then, establishing a received signal strength indication value and subgrade moisture content relationship model, and extracting a characteristic parameter affected only by moisture content; then, constructing a neural network model, taking the extracted characteristic parameter as input, and using field data to train the model to invert the subgrade moisture content; finally, combining the test point distribution and subgrade structure information, obtaining the spatial distribution of the subgrade moisture content, and realizing efficient, non-contact and continuous monitoring of the subgrade humidity state.

Description

Technical Field

[0001] This invention belongs to the field of road engineering technology, and in particular relates to a non-destructive testing method for roadbed moisture based on multi-parameter inversion of wireless radio frequency signals. Background Technology

[0002] Changes in subgrade moisture directly affect the strength, stiffness, and stability of road structures. Increased moisture during the operational period leads to decreased strength, resilience modulus, and load-bearing capacity, potentially causing structural damage and ultimately impacting road service performance. Therefore, efficient and accurate measurement of subgrade moisture during the operational period is crucial for ensuring long-term road service performance.

[0003] Traditional methods for testing the moisture content of roadbeds include the drying method and the alcohol combustion method. Although these methods offer high accuracy, they are destructive and inconvenient to operate, especially for testing existing highways. Therefore, in recent years, a number of non-destructive or low-disturbance technologies have emerged, such as time-domain reflectometry, resistivity methods, and ground-penetrating radar (GPR), which are widely used for rapid on-site moisture measurement of roadbeds. These methods estimate the moisture content by determining the dielectric constant or resistivity of the soil. Time-domain reflectometry and resistivity methods require pre-installed sensors, which are limited by high equipment maintenance costs and fixed locations. While GPR enables non-destructive in-situ measurements, it also suffers from complex operation, difficult data analysis, and expensive equipment.

[0004] Passive Radio Frequency Identification (RFID) technology enables target identification and data exchange without mechanical connection or optical contact. Its Received Signal Strength Indication (RSSI) is significantly affected by the physical characteristics of the radio frequency signal propagation medium, and the equipment cost is extremely low, making it a potential high-efficiency measure for measuring humidity changes in roadbeds during operation. An RFID system mainly consists of a reader and electronic tags, which exchange energy and data via spatial coupling elements in the form of radio frequency signals. As an efficient means of automatic identification and data acquisition, RFID technology has been widely used in manufacturing, logistics management, and public safety. However, to apply RFID technology to roadbed humidity monitoring, it is crucial to establish a theoretical-empirical model relating RSSI, roadbed humidity, roadbed soil properties, and measurement spatial parameters. Furthermore, it is necessary to address numerous issues related to the technology's operational procedures, measurement accuracy, and application limitations. Summary of the Invention

[0005] The present invention aims to provide a non-destructive detection method for roadbed humidity based on multi-parameter inversion of wireless radio frequency signals, in order to solve the problems of destructive sampling, difficult equipment maintenance, high cost, difficulty in long-term monitoring, and pollution risk in traditional methods for roadbed humidity measurement.

[0006] To solve the above-mentioned technical problems, the technical solution adopted in this invention is a non-destructive testing method for roadbed moisture based on multi-parameter inversion of wireless radio frequency signals, which is carried out according to the following steps:

[0007] S1. Install RFID tags and obtain the received signal strength indication value of the RFID test site on the road to be tested;

[0008] S2. Establish a model relating the received signal strength indication value to the subgrade moisture content, and determine the characteristic parameters that are only affected by the subgrade moisture content.

[0009] S3. Construct a neural network model between the received signal strength indication value and the subgrade moisture content;

[0010] S4. Organize the received signal strength indication values ​​collected on site into a dataset. Use the feature parameters that are only affected by the subgrade moisture content obtained by the relationship model between the received signal strength indication values ​​and the subgrade moisture content established in S2 as the input features of the neural network model between the received signal strength indication values ​​and the subgrade moisture content, train and output the subgrade moisture content at the time of collection.

[0011] S5. Based on the on-site test point installation, roadbed structure, and the moisture content obtained in step S4, obtain the spatial distribution of roadbed moisture content.

[0012] Furthermore, in S1, the minimum distance between the RFID reader and the RFID tag is 35cm; the distance between the RFID reader and the road surface is 25~40cm; the RFID tag is buried at a depth of 10~50cm in the roadbed; the RFID reader collects data from the RFID tags at each test point; the collected data includes antenna transmission power, electromagnetic wave operating frequency, and received signal strength indication value.

[0013] Furthermore, in S2:

[0014] The relationship model between Received Signal Strength Indication (RSSI) and subgrade moisture content is as follows:

[0015] (1)

[0016] Among them, P RSSI The RSSI (dBm) value is the received signal strength indication value for the reader. The signal strength (dBm) transmitted by the RFID reader. For the transmit antenna gain, For receiving antenna gain; The path loss of electromagnetic waves propagating in air; A r A is the reflection loss of electromagnetic waves when they pass through the interface between the roadbed soil and the air. s It is the path loss of electromagnetic waves propagating in the subgrade soil.

[0017] Furthermore, the path loss of the electromagnetic wave propagating in the air medium The process of determining is as follows:

[0018] (2)

[0019] in, The distance between the reader and the road surface (m); The operating frequency of electromagnetic waves is Hz.

[0020] Furthermore, the reflection loss when the electromagnetic wave passes through the interface between the roadbed soil and the air... Specifically:

[0021] (3)

[0022] Where H is a parameter (Ω) that is only affected by the subgrade moisture content, used to represent the equivalent wave impedance of the subgrade soil; Wave impedance in air (Ω):

[0023] (4)

[0024] in, is the vacuum permeability (H / m); is the relative permittivity (F / m).

[0025] Furthermore, the path loss of the electromagnetic wave propagating in the subgrade soil Specifically:

[0026] (5)

[0027] in, The distance (m) is the propagation distance of electromagnetic waves in the subgrade soil. Let be the operating frequency of the electromagnetic wave (Hz), and K and M be parameters (s / m) that are only affected by the moisture content of the subgrade. K represents the phase change of the electromagnetic wave when it propagates in the subgrade soil, and M represents the amplitude change of the electromagnetic wave when it propagates in the subgrade soil.

[0028] Furthermore, the parameters H, K, and M, which are only affected by the subgrade moisture content, are specifically as follows:

[0029] (6)

[0030] (7)

[0031] (8)

[0032] k1, k2, k3, k4, k5, k6, k7, k8, and k9 are the fitting parameters for the formula; VWC is the subgrade moisture content.

[0033] Furthermore, the neural network model between the received signal strength indication value and the subgrade moisture content in S3 is a subgrade moisture content prediction model based on residual MLP, which uses the feature parameters H, K, and M, which are only affected by the subgrade moisture content, as the model input and the subgrade moisture content as the output. The input layer receives three features: H, K, and M; the hidden layer consists of three stacked residual blocks, each containing two fully connected layers, with 32 hidden nodes per layer; the activation function of the hidden layer is the ReLU function; skip connections are introduced in each residual block; the output layer is a single neuron, using a linear activation function, which directly outputs the predicted value of the subgrade moisture content.

[0034] Furthermore, in S4, the neural network model between the received signal strength indication value and the roadbed moisture content is trained using a 5-fold cross-validation method for model selection, the optimizer is Nadam, the loss function is mean squared error, and L2 regularization is introduced.

[0035] Furthermore, 80% of the dataset mentioned in S4 is used to train the roadbed moisture content prediction model based on residual MLP neural network, and the remaining 20% ​​is used as a test set to input the model for verification and prediction, outputting the roadbed moisture content value at the time of collection.

[0036] Compared with existing technologies, the advantages of this invention are as follows: This invention does not damage the soil structure during subgrade moisture content testing, thus ensuring a comprehensive and accurate measurement of the subgrade moisture field. This invention not only maintains the original physical and chemical properties of the soil but also provides reliable data support for subsequent engineering assessments. This invention designs a simple and easy-to-use testing equipment system. Users only need to follow the operating instructions to complete the equipment installation steps to directly obtain key data for analyzing subgrade moisture content. This design greatly simplifies the operation process, enabling even personnel without professional background knowledge to complete the entire testing process efficiently and accurately, reducing technical barriers and labor costs. This invention uses low-cost RFID tags, suitable for large-scale deployment and demonstrating significant cost-effectiveness. Since these RFID tags do not rely on batteries or cables for power, they have almost no negative impact on the environment during operation, highly consistent with modern green and sustainable development concepts. This invention performs excellently in data acquisition, with high data stability and a small error range. Compared with other buried non-destructive measurement methods for subgrade moisture, this invention is less affected by external environmental factors and can provide more stable and reliable test results. This not only improves measurement accuracy but also provides a solid foundation for subsequent data analysis and decision-making. In summary, this invention achieves non-destructive roadbed moisture detection while possessing multiple advantages such as economy, practicality, and environmental friendliness. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 This is a schematic diagram of the basic RFID model;

[0039] Figure 2 This is a diagram illustrating the backscattering operation mode of the RFID system in this embodiment.

[0040] Figure 3 This is a system framework diagram of the method for measuring the moisture content of the roadbed in this embodiment;

[0041] Figure 4 This is a flowchart for calculating the subgrade moisture content in this implementation method;

[0042] Figure 5 This is a schematic diagram of the test equipment structure in this embodiment;

[0043] Figure 6This is a comparison chart showing how well different functions fit the parameter H;

[0044] Figure 7 This is a comparison chart showing how well different functions fit the parameter K;

[0045] Figure 8 This is a comparison chart of how different functions fit the parameter M.

[0046] Figure 9 This is a flowchart of a residual MLP neural network;

[0047] Figure 10 This is a comparison chart of the predicted and actual values ​​of the roadbed moisture content when the label is buried at a depth of 10cm.

[0048] Figure 11 This is a comparison chart of the predicted and actual values ​​of the roadbed moisture content when the label is buried at a depth of 30cm.

[0049] Figure 12 This is a comparison chart of the predicted and actual values ​​of the roadbed moisture content when the label is buried at a depth of 50cm. Detailed Implementation

[0050] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] Subgrade moisture content is a key factor determining the bearing capacity of the subgrade and a crucial indicator for evaluating its strength and stability. Therefore, accurate measurement of subgrade moisture content is essential for ensuring the long-term service performance of roads during operation. Current engineering practices primarily employ time-domain reflectometry and resistivity methods to measure subgrade moisture content; however, the measurement results are easily affected by external environmental interference, leading to decreased accuracy and potential damage to the pavement structure. While existing non-destructive testing technologies (such as ground-penetrating radar) can avoid damaging the pavement, their high equipment cost and complex operation limit their large-scale application.

[0052] This embodiment provides a non-destructive testing method for roadbed moisture based on multi-parameter inversion of wireless radio frequency signals, specifically including the following steps:

[0053] S1. Obtain the Received Signal Strength Indication (RSSI) value of the passive radio frequency identification (RFID) test on the road to be tested;

[0054] like Figure 1The RFID system mainly consists of a reader and electronic tags, which exchange energy and data via radio frequency signals through spatial coupling elements. This implementation plan places the host computer, RFID tags, reader, and antenna at designated locations on the road to be inspected. Based on the roadbed structure and material properties, surrounding environmental conditions, and the performance parameters of the RFID tags and reader, the burial depth of the RFID tags and the distance between adjacent tags are determined. During installation, it should be ensured that all RFID tags are on the same horizontal plane and parallel to the road surface to guarantee testing accuracy.

[0055] In this embodiment, the minimum distance between the RFID reader and the tag is set to 35cm. By comparing the signal reception quality of RFID tags at different burial depths, the optimal range for the reader's distance from the road surface is determined to be 25-40cm. Within this range, the signal reception is optimal when the RFID tag is buried at a roadbed depth of 10-50cm. Simultaneously, the dynamic frame slotted ALOHA algorithm is used to evaluate the signal reception quality of adjacent tags at different distances, determining the optimal distance range between adjacent tags to be 50-100cm.

[0056] After road construction is completed, a reader is used to collect data from the RFID tags at each test point to determine the antenna transmission power (P). t The system determines the operating frequency (f) and records the corresponding RSSI value for each location. This implementation uses an UHF RFID reader with a directional antenna of 6-12dB gain, connected to the host computer, reader, and antenna via TNC and LAN ports. The RFID tags used are anti-interference tags with a gain of 3-6dB, which offer advantages such as strong anti-interference capability, long lifespan, fast identification speed, good anti-collision capability, and high stability.

[0057] After data collection at each test point, the moisture content of soil samples from the same location was determined using the alcohol combustion method, serving as the true reference value for the subgrade moisture content at that point. Subsequently, the moisture content values ​​collected from each test point were integrated with the RSSI values ​​into a dataset, where moisture content is the dependent variable and RSSI is the independent variable. A correlation model between RSSI and moisture content was constructed, generating a calibration curve. This calibration curve allows for accurate analysis of the intrinsic relationship between subgrade moisture content and RSSI values, providing a basis for rapid prediction of subsequent moisture content.

[0058] S2. Construct a model relating the Received Signal Strength Indication (RSSI) value to the subgrade moisture content based on passive radio frequency identification (RFID) technology;

[0059] S21. Construct a relationship model for Received Signal Strength Indication (RSSI) values;

[0060] like Figure 2As shown, the RFID system uses backscatter technology for data transmission and reception. The communication link between the reader and the tag is divided into two types: a forward link and a reverse link. Therefore, the RSSI value ultimately recorded by the reader is derived from the power conversion received in the reverse link. Figure 3 As shown, based on the above working principle, the path loss of the wireless radio frequency signal in free space (A) is comprehensively considered. a ), Reflection loss at the interface between subgrade soil and air (A r ) and path loss in the subgrade soil (A s Based on three factors, the theoretical-empirical model for RSSI values ​​is established as follows:

[0061] (1)

[0062] (2)

[0063] (3)

[0064] From equations (1) to (3), we get:

[0065] (4)

[0066] Among them, P r Reader receive power (mW); P RSSI The RSSI (dBm) value is the received signal strength indication value for the reader; P t The reader's transmit power (mW); The strength of the reader's transmitted signal (dBm); G t For reader antenna gain; G r Tag antenna gain; A a The path loss of electromagnetic waves propagating in air; A r A is the reflection loss of electromagnetic waves when they pass through the interface between the roadbed soil and the air. s It is the path loss of electromagnetic waves propagating in the subgrade soil.

[0067] like Figure 3 As shown, electromagnetic wave signals gradually attenuate as the transmission distance increases when propagating in air. This attenuation mainly involves two mechanisms: one is the geometric attenuation caused by the spherical diffusion of electromagnetic waves in free space, the intensity of which is inversely proportional to the square of the propagation distance; the other is the energy loss caused by the absorption and scattering of electromagnetic waves by the air medium. To calculate the path loss (A) of electromagnetic waves propagating in the air medium... a This implementation method is based on the free-space propagation loss model. Combining the characteristics of electromagnetic wave propagation in free space, a corresponding mathematical model is constructed, as shown in formulas (5) to (7). Finally, the free-space loss (A...) is...a Convert to logarithmic form for calculation:

[0068] (5)

[0069] (6)

[0070] (7)

[0071] Among them: A a The path loss of electromagnetic waves propagating in air; d a The distance between the reader antenna and the road surface (m); λ is the signal wavelength (m); c is the speed of light in a vacuum, typically 1000 m. m / s; The operating frequency of electromagnetic waves is Hz.

[0072] like Figure 3 As shown, due to differences in parameters such as wave impedance, dielectric constant, and magnetic permeability between the two media, electromagnetic waves will be reflected at the air-soil interface when propagating between the air and the subgrade soil, resulting in energy loss. Therefore, this embodiment defines this energy loss as the reflection loss at the subgrade soil-air interface (A0). r Furthermore, as a typical lossy medium, the equivalent wave impedance of subgrade soil is a dynamic parameter significantly affected by its physical state. With increasing water content, the dielectric constant of the soil typically increases significantly, directly impacting its equivalent wave impedance. This implementation method, based on Fresnel's formula and the law of conservation of energy, establishes the following mathematical expression:

[0073] (8)

[0074] Among them: A r Z1 is the reflection loss of electromagnetic waves passing through the interface between the subgrade soil and the air; Z2 is the wave impedance in the air, which is usually about 377Ω; H represents the change in the equivalent wave impedance of the subgrade soil (Ω), which is only affected by the water content of the subgrade. In addition, the mathematical expression for the wave impedance in the air is shown in equation (9):

[0075] (9)

[0076] in: The vacuum permeability is typically 1. H / m; The relative permittivity is typically 1. F / m.

[0077] like Figure 3As shown, when electromagnetic waves propagate in the subgrade soil medium, their energy is easily absorbed by water molecules and conductive minerals in the soil, leading to gradual signal attenuation. Simultaneously, the propagation characteristics of electromagnetic waves exhibit systematic changes with distance, specifically in three aspects: first, the equivalent wavelength of the signal changes due to the complex dielectric properties of the soil; second, the signal phase undergoes a cumulative shift with propagation distance; and third, the signal amplitude attenuates exponentially with increasing propagation distance, as shown in formulas (10) to (12). Therefore, this implementation method, based on the Longley-Rice model, studies the path loss (A) of electromagnetic waves propagating in the subgrade soil. s Quantitative characterization and analysis were performed, and the established mathematical expression is as follows:

[0078] (10)

[0079] (11)

[0080] (12)

[0081] (13)

[0082] (14)

[0083] (15)

[0084] From equations (10) to (15), we get:

[0085] (16)

[0086] Among them: A s It is the path loss (dB) of electromagnetic waves propagating in the subgrade soil; d s The propagation distance of electromagnetic waves in the subgrade soil (m); λ is the signal wavelength (m); c is the speed of light in a vacuum, typically 1000 m. m / s; The electromagnetic wave operating frequency (Hz) is given; the phase shift constant B is the change in phase per unit distance when the electromagnetic wave propagates in the subgrade soil, measured in rad / m; the attenuation constant C is the attenuation of the amplitude per unit distance when the electromagnetic wave propagates in the subgrade soil, measured in Np / m; L s L represents the geometric loss caused by the spherical diffusion of electromagnetic wave energy in the soil; K L represents the energy loss caused by phase changes when electromagnetic waves propagate through the subgrade soil. MThe energy loss caused by the amplitude change when the electromagnetic wave propagates in the subgrade soil; K and M are parameters (s / m) that are only affected by the subgrade moisture content, where K represents the phase change when the electromagnetic wave propagates in the subgrade soil; and M represents the amplitude change when the electromagnetic wave propagates in the subgrade soil.

[0087] S22. Determine the parameters H, K, and M that are only affected by the subgrade moisture content;

[0088] First, based on field measurement data, the raw RSSI values ​​were preprocessed using Matlab software. The signal was smoothed and denoised using the moving average method and statistical principles, identifying and removing duplicate, outlier, and missing values ​​to obtain a stable RSSI dataset. With parameters such as transmit power, operating frequency, antenna height, and tag burial depth fixed, parameters were solved and separated according to the RSSI value theoretical-empirical model. Key parameters H, K, and M, which are only affected by the subgrade moisture content, were extracted, providing interpretable input features for the subsequent establishment of a neural network model of RSSI and subgrade moisture content.

[0089] Secondly, as shown in Table 1, a nonlinear regression analysis was performed on the relationship between parameters H, K, M and the subgrade moisture content. By comparing the fitting effects of different function forms, the sine function with the best fitting effect was selected to establish the calibration relationship, as shown in equations (17) to (19). Figures 6-8 The results show that there is a significant nonlinear relationship between parameters H, K, and M and the subgrade moisture content, further verifying that the sine function can effectively obtain nonlinear laws that cannot be described by the linear model, and its fitting effect is significantly better than that of the exponential function and the polynomial function.

[0090] (17)

[0091] (18)

[0092] (19)

[0093] Where: k1, k2, k3, k4, k5, k6, k7, k8, and k9 are the fitting parameters of the formula; H, K, and M are the parameters in the theoretical-empirical model of RSSI values ​​that are only affected by the subgrade moisture content; VWC is the subgrade moisture content; R 2 The coefficient of determination.

[0094] Table 1 Comparison of fitting effects of different models

[0095]

[0096] Finally, by substituting parameters H, K, and M into formulas (4) and (16) for inversion, the predicted accuracy R² of the subgrade moisture content was found to be only 0.50, which is an improvement compared to the linear model inversion result (R²=0.29). However, due to the high heterogeneity of the subgrade soil itself, the complexity of the electromagnetic wave propagation path, and the reflection, absorption, and scattering losses experienced by the signal at different physical interfaces (such as the air-soil interface), the RSSI value and the subgrade moisture content exhibit a highly complex nonlinear relationship. It is difficult to accurately describe this relationship using theoretical-empirical models of RSSI values ​​and traditional fixed-form analytical functions (such as polynomials, exponentials, or sine functions), and there are generally problems with low overall fitting accuracy and insufficient generalization ability. For example, under different soil types, compaction degrees, or environmental conditions, the model parameters often need to be recalibrated, resulting in a limited range of applicability and difficulty in stably responding to dynamic changes in moisture content. Therefore, this implementation introduces a neural network method based on the RSSI value and subgrade moisture content relationship model. Using H, K, and M as input features, a nonlinear mapping model driven by fusion of mechanism and data is constructed. This method not only significantly improves the prediction accuracy of subgrade moisture content and broadens its applicability, but also enables dynamic updates by combining measured data, thus supporting high-precision, widely applicable, and real-time monitoring of subgrade moisture content.

[0097] S3. Construct a neural network model based on the received signal strength indication (RSSI) value and the subgrade moisture content using passive radio frequency identification (RFID) technology;

[0098] To achieve accurate prediction of subgrade moisture content, this implementation method constructs a neural network model based on Residual Multilayer Perceptron (ResMLP) to establish a nonlinear mapping relationship between RSSI signals and subgrade moisture content. The specific implementation steps are as follows:

[0099] First, in the task of calculating subgrade moisture content, the original RSSI data is preprocessed by using a moving average filtering method to remove noise and outliers, forming a stable RSSI dataset. Then, based on the feature parameters H, K, and M extracted in step S2 that are only affected by subgrade moisture content, and using subgrade moisture content as the output, a subgrade moisture content prediction model based on residual MLP is constructed. This model exhibits good adaptability to different label burial depths and working conditions, achieving accurate and dynamic prediction of subgrade moisture content.

[0100] Secondly, to improve the learning efficiency and expressive power of the model, a residual MLP neural network with the following parameter configuration was constructed, the structure of which is as follows: Figure 9As shown in Table 2: The input layer receives three features: H, K, and M; the hidden layer consists of three stacked residual blocks, each containing two fully connected layers, with 32 hidden nodes per layer; the ReLU activation function is used in the hidden layers to enhance nonlinear expressiveness and alleviate the gradient vanishing problem; skip connections are introduced in each residual block to achieve identity mapping and promote the training stability of deep networks; the output layer is a single neuron using a linear activation function, directly outputting the predicted value of roadbed water content. During model training, 5-fold cross-validation was used for model selection, and full-batch training was employed, with 500 training iterations. The loss value continuously decreased with the number of iterations during training, indicating that the model has good convergence and generalization ability. The mean squared error (MSE) loss function was used, the Nadam optimizer was used, the initial learning rate was set to 0.02, and L2 regularization was introduced to suppress overfitting.

[0101] Table 2 Training parameters of neural network models

[0102]

[0103] Finally, to comprehensively evaluate the model performance, the complete training set was divided into training, validation, and test sets at a ratio of 80%, 10%, and 10%, respectively. The coefficient of determination (R²), root mean square error (RMSE), and mean absolute error (MAE) were selected as the core evaluation indicators, as shown in equations (20) to (22). After training, a high-performance residual MLP model was output for subsequent real-time monitoring of subgrade moisture content and generation of spatial distribution maps. This model can effectively capture the complex nonlinear relationship between RSSI and subgrade moisture content, significantly improving prediction accuracy and generalization ability, and providing solid technical support for non-destructive testing of subgrade moisture.

[0104] (20)

[0105] (twenty one)

[0106] (twenty two)

[0107] Where: a i b is the true value; i The predicted value is denoted by m; m is the number of sample points. is the average of the true values; i is the i-th sample point.

[0108] S4. The RSSI values ​​collected on-site are organized into a dataset. After preprocessing the relationship model between the Received Signal Strength Indication (RSSI) value and the subgrade moisture content, the parameters H, K, and M are obtained as input features of the neural network model, and the predicted subgrade moisture content at the time of collection is output.

[0109] S5. Based on the actual burial conditions of the test points, the subgrade structure information, and the moisture content data of each point obtained in step S4, generate a spatial distribution map of subgrade moisture content.

[0110] In this implementation plan, the calculation process for the subgrade moisture content is as follows: Figure 4 As shown. First, a host computer, an RFID reader, and an antenna are installed at predetermined locations on the selected test section. The distance d between the reader and the road surface is measured. a and the label burial depth d s And determine the initial parameters, including the transmit antenna gain G. t Receiver antenna gain G r Transmit power P t The operating frequency f and air wave impedance Z1, etc. are then used. Based on electromagnetic wave propagation theory (such as Friesian transmission model, Fresnel reflection theory, etc.), combined with the established received signal strength indication (RSSI) value and roadbed moisture content relationship model (1)~(15), the key parameters H, K, and M that are closely related to moisture content are extracted from the RSSI data obtained from the field RFID test, so as to transform the actual engineering problem into a mathematical inversion problem with clear characteristics. Next, in order to construct a roadbed moisture content prediction model, the real roadbed moisture content data measured by the traditional alcohol combustion method are used simultaneously to form a historical dataset consisting of parameters H, K, M and the real value of roadbed moisture content. Finally, 80% of the dataset is used to train the roadbed moisture content prediction model based on residual MLP neural network, and the remaining 20% ​​is used as the test set input model for verification and prediction, and the roadbed moisture content value at the time of collection is output, so as to realize the fast and accurate lossless inversion of the roadbed moisture content corresponding to any newly collected RSSI data.

[0111] In road engineering, the soil medium has a significant impact on electromagnetic wave propagation. Due to the combined effects of soil dielectric properties and environmental factors such as moisture content, electromagnetic waves experience significant attenuation during propagation. Simultaneously, the reflection effect at the air-soil interface further exacerbates energy loss, leading to significant fluctuations in Received Signal Strength Indication (RSSI) values. Especially under high humidity conditions, the soil dielectric constant increases, weakening the penetration ability of RFID signals, further increasing energy loss, and thus reducing communication efficiency and reception quality.

[0112] To address the aforementioned issues, this implementation combines the Friesian propagation model, Fresnel reflection model, and soil medium absorption model to quantitatively analyze energy loss during signal propagation. Based on the theoretical-empirical relationship between RSSI and subgrade moisture content, three key parameters H, K, and M, which are only affected by moisture content, are extracted. A high-precision residual MLP neural network is used to construct a nonlinear mapping relationship between these parameters and subgrade moisture content to improve the accuracy of moisture content inversion. To enhance signal penetration and reduce attenuation, the system is equipped with a 12dB high-gain antenna, effectively improving transmission efficiency and signal reception quality. Simultaneously, by employing anti-interference tags and optimizing their deployment, long-term, stable, and non-destructive monitoring of the subgrade condition is achieved without damaging the pavement structure, effectively avoiding environmental pollution problems caused by batteries. The system ultimately integrates an RFID reader, a high-gain antenna, and a host computer data processing platform to collect and transmit RSSI data in real time. Combined with a spatial distribution algorithm, a subgrade moisture content field distribution map is quickly generated, significantly improving detection accuracy and engineering applicability, providing reliable technical support for road condition assessment and scientific maintenance.

[0113] Experimental verification:

[0114] To verify the feasibility of this design, a system test experiment was conducted. The test site was located in Tianxin District, Changsha City, Hunan Province, and the soil type was high liquid limit silt. The main physical properties of this soil are as follows: optimum moisture content is 21%, maximum dry density is 1.65 g / cm³, liquid limit is 52%, and plastic limit is 38%. In the experiment, RFID tags were buried at depths of 10 cm, 30 cm, and 50 cm, respectively. The RFID reader, host computer, and equipment platform were integrated into a mobile test device. The initial parameters of the RFID test device were set, as shown in Table 3. The antenna was installed at the bottom of the device and kept parallel to the ground. Figure 5 As shown.

[0115] Table 3 Initial parameters of the RFID testing device

[0116]

[0117] The specific testing steps and data processing flow of this verification scheme are as follows: First, the RFID testing device is fixed at a height of 40cm above the ground to collect the Received Signal Strength Indication (RSSI) data of the tags in the soil in real time and transmit it to the host computer for recording. Second, the collected raw RSSI data is systematically organized, as shown in Tables 4 to 6. Simultaneously, the raw RSSI dataset is preprocessed using Matlab software, combining the moving average method and statistical principles to achieve signal smoothing and noise reduction. Third, based on the RSSI value theory-empirical model, the key parameters H, K, and M closely related to moisture content are derived from the preprocessed dataset. Fourth, 80% of the H, K, and M data, along with the actual subgrade moisture content measured by the traditional alcohol combustion method, are input into a high-precision residual MLP neural network for training. The remaining 20% ​​of the data is used to verify the model performance to predict the subgrade moisture content at the time of data collection. The model's evaluation results on the test set show that at a burial depth of 10cm, the model's coefficient of determination (R²) is 0.97, the root mean square error (RMSE) is 1.20, and the mean absolute error (MAE) is 0.97; at a burial depth of 30cm, R² is 0.93, RMSE is 2.13, and MAE is 1.17; and at a burial depth of 50cm, R² is 0.87, RMSE is 3.43, and MAE is 2.75.

[0118] Table 4. Raw RSSI dataset when the label burial depth is 10cm

[0119]

[0120] Table 5. Raw RSSI dataset when the label embedment depth is 30cm

[0121]

[0122] Table 6. Raw RSSI dataset when the tag burial depth is 50cm

[0123]

[0124] To objectively evaluate the measurement accuracy of this method and compare it with existing technologies, this study also uses the time-domain reflectometry (TDR) method to obtain the subgrade moisture content as a reference. Figures 10-12 As shown, under three burial depth conditions, the proposed method demonstrates high accuracy in predicting subgrade moisture content. Especially at a burial depth of 10 cm, the prediction accuracy R² of the residual MLP neural network model reaches 0.97, significantly better than the 0.80 of the TDR method, fully demonstrating the accuracy advantage of this method in subgrade moisture content prediction. In summary, the subgrade moisture content monitoring system constructed in this study performs well in terms of prediction accuracy, adaptability, and engineering practicality, and can meet the high-precision real-time monitoring requirements of subgrade moisture content in practical engineering projects.

[0125] The various embodiments in this specification are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0126] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of protection of the present invention.

Claims

1. A method for non-destructive testing of subgrade humidity based on multi-parameter inversion of wireless radio frequency signals, characterized in that, Specifically, the following steps are performed: S1, embedding an RFID tag, obtaining a received signal strength indication value of an RFID test on a road site to be detected; S2, establishing a relationship model of the received signal strength indication value and the roadbed water content, and determining a characteristic parameter affected only by the roadbed water content; the received signal strength indication value RSSI and the roadbed water content relationship model is: （1） Wherein, P RSSI is the received signal strength indicator value RSSI, dBm, of the reader; is the transmitted signal strength of the RFID reader; is the gain of the transmitting antenna, is the gain of the receiving antenna; is the path loss of the electromagnetic wave propagating in the air medium; A r is the reflection loss of the electromagnetic wave passing through the interface between the roadbed soil and the air; A s is the path loss of the electromagnetic wave propagating in the roadbed soil; S3, constructing a neural network model between the received signal strength indication value and the roadbed water content; S4, arranging the received signal strength indication value collected on site into a data set, and using the characteristic parameter affected only by the roadbed water content obtained by the relationship model of the received signal strength indication value and the roadbed water content in S2 as the input feature of the neural network model between the received signal strength indication value and the roadbed water content, training and outputting the roadbed water content at the collection time; S5, obtaining the spatial distribution of the roadbed water content according to the embedding situation of the test point, the roadbed structure and the water content obtained in step S4.

2. The method according to claim 1, wherein, In S1, the minimum distance between the RFID reader and the RFID tag is 35 cm; the distance between the RFID reader and the road surface is 25-40 cm; the RFID tag is embedded at a depth of 10-50 cm in the roadbed; the RFID reader collects data of the RFID tag of each test point; the collected data includes antenna transmission power, electromagnetic wave operating frequency and received signal strength indication value.

3. The method of claim 1, wherein, Path loss of the electromagnetic wave propagating in the air medium The determination process is: （2） wherein, is the distance between the reader and the road surface; is the electromagnetic wave operating frequency.

4. The method of claim 1, wherein, Reflection loss of the electromagnetic wave when passing through the air-soil interface of the roadbed Specifically: （3） where H is a parameter only affected by the water content of the subgrade, and is used to represent the equivalent wave impedance of the subgrade soil; for the wave impedance in air: （4） wherein is the vacuum permeability; is the relative permittivity.

5. The method according to claim 4, wherein, Path loss of the electromagnetic wave propagating in the embankment soil Specifically: （5） wherein, is the propagation distance of the electromagnetic wave in the subgrade soil, is the operating frequency of the electromagnetic wave, and K and M are parameters affected only by the water content of the subgrade, wherein K represents the phase change when the electromagnetic wave propagates in the subgrade soil, and M represents the amplitude change when the electromagnetic wave propagates in the subgrade soil.

6. The method according to claim 5, wherein, The parameters H, K and M affected only by the roadbed water content are specifically: （6） （7） （8） k1, k2, k3, k4, k5, k6, k7, k8 and k9 are formula fitting parameters; VWC is the roadbed water content.

7. The method according to any one of claims 1, 4, 5, wherein, The neural network model between the received signal strength indication value and the roadbed water content in S3 is a roadbed water content prediction model based on residual MLP, which takes the characteristic parameters H, K and M affected only by the roadbed water content as the model input and takes the roadbed water content as the output; wherein, the input layer receives three characteristics H, K and M; the hidden layer is composed of 3 residual blocks, each residual block contains 2 fully connected layers, and the number of hidden nodes in each layer is 32; the hidden layer activation function adopts ReLU function; a skip connection is introduced in each residual block; the output layer is a single neuron, which adopts a linear activation function and directly outputs the predicted value of the roadbed water content.

8. The method of claim 1, wherein, The training of the neural network model between the received signal strength indication value and the roadbed water content in S4 adopts 5-fold cross-validation method for model selection, the optimizer adopts Nadam, the loss function adopts mean square error, and L2 regularization is introduced.

9. The method of claim 1, wherein, 80% of the data set in S4 is used to train the roadbed water content prediction model based on residual MLP neural network, and the remaining 20% is input into the model as a test set for verification and prediction, and the roadbed water content value at the collection time is output.

Images