Method for measuring tensile strength of asphalt concrete pavement

Abstract

The invention provides a method for measuring the tensile strength of an asphalt concrete pavement, which comprises the following steps of: preparing a standard sample, making an optical mark, carrying out constant-temperature and constant-humidity multi-stage loading, and synchronously acquiring images, force, temperature and humidity; a four-dimensional space-time deformation field is constructed by using a three-dimensional digital image correlation method, loading force is combined to invert mesoscopic parameters by using a cohesion model, finally, an elastic-plastic constitutive finite element is embedded to simulate tensile failure, and the macroscopic tensile strength is calculated. Scientific basis can be provided for pavement design and maintenance, and the accuracy and reliability of pavement performance evaluation are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of road engineering material performance testing, and more particularly, to a tensile strength measurement method for asphalt concrete pavement. BACKGROUND

[0002] In road engineering construction and maintenance, the tensile strength of asphalt concrete pavement is one of the key indicators for evaluating pavement performance. Traditional tensile strength measurement methods mainly rely on physical tests, such as uniaxial tensile test or flexural tensile test. Although these methods can provide certain strength data, they have some limitations. For example, physical tests usually require destructive sampling, which is not only time-consuming and labor-intensive, but also difficult to reflect the true performance of actual pavement under complex environmental conditions. In addition, traditional methods are difficult to capture the damage and crack propagation process of materials at the micro level, which has an important influence on the long-term performance of the pavement.

[0003] In the implementation process of the embodiments of the present application, there are at least the following problems or defects in the prior art: the traditional method cannot comprehensively and real-time monitor the micro deformation and damage evolution of asphalt concrete during loading, it is difficult to accurately reflect the macro tensile strength of asphalt concrete in actual service environment, and it is not accurate enough for material performance evaluation under complex environmental conditions. SUMMARY

[0004] The present application provides a tensile strength measurement method for asphalt concrete pavement, comprising: S1, preparing a standardized asphalt concrete pavement sample, and preparing an optical measurement mark on the surface of the asphalt concrete pavement sample; S2, under a predetermined environmental condition, performing multi-level progressive loading on the asphalt concrete pavement sample, and synchronously collecting multi-modal data during the loading process; the multi-modal data includes image sequence, loading force data and environmental temperature and humidity data; S3, processing the multi-modal data to construct a four-dimensional spatiotemporal deformation field of the asphalt concrete pavement sample during the loading process; S4, based on the four-dimensional spatiotemporal deformation field and the loading force data, inversely calculating the mesoscopic mechanical parameters of the asphalt concrete pavement sample through a cohesive force model; S5, calculating the macro tensile strength of the asphalt concrete pavement sample according to the mesoscopic mechanical parameters and the material constitutive relation.

[0005] Further, the step S1 comprises: S1.1, using a core drill to drill a cylindrical sample from a service asphalt concrete pavement, and precisely milling two end faces of the cylindrical sample to ensure parallelism; S1.2, polishing the side surface of the cylindrical sample to form a smooth observation surface; S1.3, spraying a high-contrast random speckle pattern on the observation surface, the random speckle pattern having a predetermined particle size range; S1.4, placing the prepared asphalt concrete pavement sample in a constant temperature and humidity environment for a predetermined time.

[0006] Further, the step S2 comprises: S2.1, installing the asphalt concrete pavement sample on a mechanical testing machine in an environmental chamber; S2.2, performing the multi-stage progressive loading, each loading stage comprising: applying a compression load at a constant rate until a preset force threshold or displacement threshold is reached, and then keeping the load constant for a certain holding time; the force threshold is set in proportion to the total number of loading stages and the estimated maximum loading force; S2.3, using at least two high-speed cameras to synchronously collect a stereoscopic image sequence of the surface of the asphalt concrete pavement sample during the entire loading process; S2.4, recording the loading force data in real time through the sensors of the mechanical testing machine; S2.5, recording the environmental temperature and humidity data in real time through the sensors arranged in the environmental chamber.

[0007] Further, the step S3 comprises: S3.1, performing image preprocessing on the image sequence, the image preprocessing comprising flat field correction, filtering and contrast enhancement; S3.2, processing the stereoscopic image sequence using a three-dimensional digital image correlation algorithm to calculate a three-dimensional displacement field of the surface of the asphalt concrete pavement sample; the three-dimensional digital image correlation algorithm uses a zero-mean normalized cross-correlation function as a matching criterion, and the function expression is:

[0008] wherein, is a zero-mean normalized cross-correlation coefficient, is a reference sub-region image gray matrix, is a reference sub-region image gray mean value, is a deformed sub-region image gray matrix, is a deformed sub-region image gray mean value, is a coordinate according to displacement and deformation hypothesis; S3.3, performing spatial and time series analysis on the three-dimensional displacement field to calculate a full-field strain tensor, the four-dimensional spatiotemporal deformation field being composed of the three-dimensional displacement field and the full-field strain tensor.

[0009] Further, step S4 includes: S4.1 Identify the microcrack initiation location and propagation path of the asphalt concrete pavement sample from the four-dimensional spatiotemporal deformation field; S4.2 Establish a finite element model with the same geometric dimensions as the asphalt concrete pavement sample, and embed cohesive elements in the finite element model on the potential fracture path; S4.3 Using the loading force data and the displacement history of key points as boundary conditions, drive the finite element model to perform inverse analysis, and iteratively adjust the traction-separation law parameters of the cohesive unit through optimization algorithm until the model output matches the measured four-dimensional spatiotemporal deformation field within the preset tolerance. S4.4. The optimized converged traction-separation law parameters are determined as the micromechanical parameters.

[0010] Furthermore, the traction-separation law is a bilinear constitutive model, and the micromechanical parameters include: cohesive strength. Critical fracture energy and initial stiffness K; The formula for the traction-separation law is: When separation displacement At that time, traction force ; When separation displacement At that time, traction force ; in, To achieve the required cohesive strength, the separation displacement This represents the separation displacement at the point of complete destruction.

[0011] Further, step S5 includes: S5.1 Construct the macroscopic material constitutive relation of the asphalt concrete pavement material, wherein the macroscopic material constitutive relation is an elastoplastic constitutive model that can reflect the material damage and softening behavior; S5.2. The micromechanical parameters obtained by inversion calculation from the cohesive force model are embedded as key input parameters into the macroscopic material constitutive relation; S5.3 Establish a finite element model representing the target pavement structure, and assign the macroscopic material constitutive relation embedded with the micromechanical parameters to the asphalt concrete material in the finite element model; S5.4 Simulate a standard tensile test or bending tensile test on the finite element model, and calculate the virtual failure load of the asphalt concrete pavement sample under uniaxial tensile stress or bending tensile stress. S5.5. The macroscopic tensile strength is calculated based on the virtual failure load and the standard geometric dimensions of the asphalt concrete pavement sample.

[0012] Furthermore, in step S2, a dynamic loading strategy is implemented, specifically including: S8.1 During the load holding period of each loading level, the latest acquired image sequence is processed in real time to quickly calculate the current full-field strain distribution; S8.2 Based on the calculated full-field strain distribution, identify local strain concentration areas in real time and calculate the local strain concentration index; S8.3. Compare the local strain concentration index with a number of preset thresholds; S8.4. Based on the comparison results, dynamically adjust the loading parameters for the next loading stage: when the local strain concentration index exceeds the first threshold but is lower than the second threshold, automatically reduce the constant rate; when the local strain concentration index exceeds the second threshold, pause loading and enter the load maintenance stage.

[0013] Furthermore, the method also includes a model building and updating step S6 performed after obtaining the macroscopic tensile strength: S6.1 Data Collection: Collect test data from multiple asphalt concrete pavement samples to form a training dataset; the test data includes the macroscopic tensile strength of each sample, as well as the corresponding environmental temperature and humidity data, the microscopic mechanical parameters, and the original material mix information of the asphalt concrete pavement; S6.2 Model Architecture Establishment: Establish a machine learning prediction model with ambient temperature, ambient humidity, asphalt content, aggregate gradation parameters and material age as input variables and predicted tensile strength as output variable; the machine learning prediction model adopts a gradient boosting decision tree architecture; S6.3 Model Training: The gradient boosting decision tree architecture is trained using the training dataset. The splitting rules of all nodes and the output values ​​of leaf nodes are determined by optimizing the loss function, and a trained service performance prediction model is generated. S6.4 Model Deployment and Update: Deploy the service performance prediction model in the road maintenance management system; after testing new road samples using the method and obtaining new test data, input the new test data into the service performance prediction model to update the model parameters in an incremental learning manner.

[0014] Furthermore, the mathematical expression of the service performance prediction model is based on the decision rule set of the gradient boosting decision tree, and its final output is the weighted sum of the outputs of all decision trees, expressed as:

[0015] in, The predicted service tensile strength, where N is the total number of decision trees. Let be the weight of the i-th decision tree. The output of the i-th decision tree for the input feature vector X includes ambient temperature, ambient humidity, asphalt content, aggregate gradation parameters, and material age.

[0016] The embodiments of the present invention have at least the following beneficial effects: 1. By using multi-level progressive loading and multi-modal data synchronous acquisition, combined with three-dimensional digital image correlation algorithms to construct a four-dimensional spatiotemporal deformation field, it is possible to comprehensively and in real time monitor the micro-deformation and damage evolution of asphalt concrete samples during the loading process. This solves the problem that traditional methods are difficult to capture the micro-damage process of materials, improves the accuracy and reliability of tensile strength measurement, and provides more accurate data support for pavement performance evaluation.

[0017] 2. The cohesive model is used to invert the microscopic mechanical parameters and embed them into the macroscopic material constitutive relation, realizing the correlation between mechanical properties from microscopic to macroscopic. This solves the problem that traditional methods cannot accurately reflect the macroscopic tensile strength of materials under complex stress states, and improves the ability to predict the performance of asphalt concrete pavement under actual service conditions.

[0018] 3. A dynamic loading strategy is introduced, which dynamically adjusts the loading parameters based on the strain distribution monitored in real time, avoiding test errors caused by improper loading rate. At the same time, the test data is analyzed and updated by combining machine learning prediction models, which solves the problem of insufficient adaptability of traditional methods to different environmental conditions and material ratios, and enhances the versatility and practicality of the pavement tensile strength measurement method. Attached Figure Description

[0019] The above and other objects, features, and advantages of exemplary embodiments of the present invention will become readily apparent from the following detailed description taken in conjunction with the accompanying drawings. Several embodiments of the invention are illustrated in the drawings by way of example and not limitation, wherein: Figure 1 This is a flowchart illustrating a method for measuring the tensile strength of asphalt concrete pavement according to an embodiment of the present invention. Detailed Implementation

[0020] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0021] Traditional methods for assessing the tensile strength of asphalt concrete pavements rely on destructive sampling in physical testing, resulting in insufficient sample representativeness and an inability to fully characterize the heterogeneous properties of actual pavement structures. Furthermore, the lack of simultaneous multimodal data acquisition during testing makes it difficult to capture the dynamic response of microcrack initiation and propagation within the material under multi-level loading conditions. The impact of environmental parameter fluctuations on the material's mechanical behavior is not quantified, leading to systematic biases in macroscopic strength predictions under complex temperature and humidity coupling effects.

[0022] For example, in highway maintenance projects, when using traditional uniaxial tensile tests to evaluate asphalt concrete pavements that have been in service for more than five years, the spatiotemporal evolution of the three-dimensional displacement field on the sample surface under compressive load cannot be monitored in real time. Therefore, the initiation location of microcracks at the aggregate-asphalt interface cannot be accurately identified during the test. Insufficient temperature and humidity control precision within the environmental chamber leads to nonlinear changes in the viscoelastic behavior of the material during loading, and the load-displacement curves collected by the testing machine cannot reflect the heterogeneity of the true strain field distribution. Consequently, the obtained macroscopic tensile strength data deviates significantly from the in-situ performance of the pavement core samples, directly affecting the accuracy of maintenance plan formulation.

[0023] If the above problems are not addressed, strength indicators obtained using traditional methods will be insufficient to reliably assess the remaining life of pavement structures, leading to misjudgments in the timing of preventative maintenance. The invisibility of the micro-damage accumulation process will mask early signs of material degradation, accelerating the development of pavement fatigue cracking. Decoupling analysis of environmental factors and mechanical response will introduce implicit errors, resulting in a lack of a unified benchmark for comparing pavement performance across different climatic regions, ultimately leading to improper allocation of maintenance resources and a significant reduction in pavement service life.

[0024] Faced with the aforementioned problems, this application first recognizes that traditional methods cannot capture the spatiotemporal evolution of the three-dimensional displacement field of materials during loading, resulting in the invisibility of the microscopic damage accumulation process. To address this, this application attempts to achieve coordinated monitoring of mechanical response and environmental parameters through multimodal data fusion technology, where synchronously acquired image sequences and loading force data can establish a correlation between macroscopic load and microscopic deformation. Further investigation reveals that relying solely on two-dimensional image analysis is insufficient to reconstruct the three-dimensional strain distribution within the material; therefore, stereoscopic vision measurement technology is introduced to improve the accuracy of displacement field reconstruction. To address the interference of environmental factors, this application proposes implementing a multi-stage loading strategy under controlled environmental conditions, observing the viscoelastic recovery behavior of the material during the load-holding phase. Finally, it is determined that a four-dimensional spatiotemporal deformation field needs to be constructed to characterize the dynamic damage process of the material, and a cross-scale correlation between microscopic mechanical parameters and macroscopic strength is established based on a cohesive force model.

[0025] like Figure 1 As shown, this application proposes a method for measuring the tensile strength of asphalt concrete pavement, comprising the following steps: S1. Prepare standardized asphalt concrete pavement samples, and prepare optical measurement marks on the surface of the asphalt concrete pavement samples; S2. Under preset environmental conditions, the asphalt concrete pavement sample is subjected to multi-level progressive loading, and multimodal data during the loading process is collected simultaneously; the multimodal data includes image sequences, loading force data, and environmental temperature and humidity data; S3. Process the multimodal data to construct a four-dimensional spatiotemporal deformation field of the asphalt concrete pavement sample during the loading process; S4. Based on the four-dimensional spatiotemporal deformation field and the loading force data, the micromechanical parameters of the asphalt concrete pavement sample are calculated by inversion using the cohesive force model. S5. Calculate the macroscopic tensile strength of the asphalt concrete pavement sample based on the micromechanical parameters and material constitutive relations.

[0026] Standardized asphalt concrete pavement samples refer to test samples prepared according to unified specifications, possessing geometric consistency and identifiable surface features. Specifically, this can be achieved by drilling cylindrical samples with a core drill followed by precision milling and surface polishing to ensure that the sample's geometric accuracy and surface quality meet optical measurement requirements, providing a benchmark for subsequent deformation field analysis. Optical measurement markers refer to surface feature patterns with high contrast and random distribution, specifically achieved by spraying a speckled material with a specific particle size range. By forming traceable texture features, this provides a displacement calculation benchmark for digital image correlation techniques.

[0027] Multi-stage progressive loading refers to a mechanical loading method that applies progressively increasing loads in stages. This can be achieved using a graded control strategy with preset force or displacement thresholds, capturing the material's response characteristics at different damage stages through a stepped loading process. Simultaneous multimodal data acquisition refers to acquiring measurement data of multiple physical quantities at the same time. This can be achieved by using a high-speed stereo vision system, mechanical sensors, and environmental sensors working in tandem, establishing a foundation for multi-factor correlation analysis by fusing image, mechanical, and environmental data.

[0028] The construction of a four-dimensional spatiotemporal deformation field refers to the continuous deformation representation combining three-dimensional spatial coordinates and the time dimension. Specifically, it can be achieved by processing stereoscopic image sequences using three-dimensional digital image correlation algorithms, revealing the internal damage evolution process of the material by calculating the displacement field and strain tensor. The cohesive model inversion calculation refers to a reverse analysis method that infers the material's microscopic parameters based on the macroscopic response. Specifically, it can be achieved by embedding cohesive elements into a finite element model and matching experimental data with optimization algorithms, improving parameter identification accuracy by establishing a quantitative relationship between macroscopic response and microscopic damage. The cross-scale correlation of material constitutive relations refers to a mathematical model connecting microscopic mechanical parameters and macroscopic mechanical properties. Specifically, it can be achieved by combining an elastoplastic constitutive model with damage evolution equations, realizing the mechanical transfer analysis from local damage to overall failure through multi-scale modeling.

[0029] This application constructs a full-chain analysis method from experimental observation to numerical inversion through multi-dimensional data fusion and cross-scale modeling techniques. By simultaneously acquiring multimodal data on mechanics, deformation, and environment, and combining four-dimensional spatiotemporal deformation field construction and mesoscopic parameter inversion techniques, it achieves dynamic tracking and quantitative characterization of the damage evolution process of asphalt concrete materials. This technical approach, which combines optical measurement, graded loading, and cohesion models, effectively solves the technical bottlenecks of traditional methods in real-time monitoring of microscopic damage and accurate assessment of the impact of environmental factors, significantly improving the scientific rigor and engineering applicability of tensile strength measurement.

[0030] In practice, standardized asphalt concrete pavement samples were first prepared, and optical measurement marks were then applied to the sample surface. These marks were used for subsequent image analysis to track surface deformation.

[0031] Under preset environmental conditions, the samples were subjected to multi-stage progressive loading. Multi-stage loading simulates the varying degrees of stress experienced by road surfaces in actual use. Simultaneously, multimodal data was acquired during the loading process, including image sequences, loading force data, and environmental temperature and humidity data. Image sequences captured surface deformation, loading force data reflected the applied stress, and environmental temperature and humidity data accounted for the influence of environmental factors on material properties.

[0032] By processing these multimodal data, a four-dimensional spatiotemporal deformation field of the sample during loading is constructed. The four-dimensional spatiotemporal deformation field contains the deformation information of the sample in three-dimensional space over time, which can comprehensively reflect the dynamic response of the material.

[0033] Based on the constructed four-dimensional spatiotemporal deformation field and loading force data, the micromechanical parameters of the sample are calculated by inversion using a cohesive force model. The cohesive force model can describe the microscopic fracture process inside the material, and the inverted micromechanical parameters reflect the mechanical properties of the material at the microscale.

[0034] Finally, based on the obtained micromechanical parameters and material constitutive relations, the macroscopic tensile strength of the sample is calculated. This step enables cross-scale prediction from microscopic parameters to macroscopic properties.

[0035] Preferably, firstly, asphalt concrete pavement test specimens are prepared using standardized molds. The specimens are cylinders with a diameter of 100 mm and a height of 150 mm. Random speckle patterns are sprayed onto the surface of the specimens as optical measurement markers, with the speckle particle size controlled within the range of 0.5-1 mm.

[0036] The specimen was placed in an environmental chamber at 25°C and 60% relative humidity, and subjected to multi-stage progressive loading using an MTS material testing machine. The loading was divided into 5 levels, with a loading rate of 0.1 mm / min for each level, until the preset force threshold or displacement threshold was reached, and then the load was kept constant for 300 seconds.

[0037] During the loading process, two high-speed cameras simultaneously acquired a sequence of three-dimensional images of the specimen surface at a frequency of 50 frames per second. Simultaneously, the loading force data was recorded by the force sensor of the testing machine, and environmental parameters were recorded by the temperature and humidity sensors inside the environmental chamber.

[0038] A three-dimensional digital image correlation algorithm was used to process the stereo image sequence to calculate the three-dimensional displacement field on the specimen surface. Combined with time dimension information, a four-dimensional spatiotemporal deformation field was constructed.

[0039] Based on the constructed four-dimensional spatiotemporal deformation field and loading force data, an inversion analysis was performed using a bilinear cohesive model. Through iterative optimization, micromechanical parameters such as cohesive strength, critical fracture energy, and initial stiffness were determined.

[0040] Finally, the obtained micromechanical parameters are substituted into the pre-established macroscopic material constitutive model, and the macroscopic tensile strength of the asphalt concrete pavement sample is calculated by simulating the standard tensile test through finite element analysis.

[0041] This application further proposes steps for preparing standardized asphalt concrete pavement samples, including: drilling cylindrical samples from in-service asphalt concrete pavement using a core drilling rig, and precision milling the two end faces of the cylindrical samples to ensure parallelism; polishing the side surfaces of the cylindrical samples to form a smooth observation surface; spraying a high-contrast random speckle pattern onto the observation surface, the random speckle pattern having a predetermined particle size range; and curing the prepared asphalt concrete pavement samples in a constant temperature and humidity environment for a predetermined time.

[0042] The core drilling rig was perpendicular to the pavement layer to ensure the axis of the cylindrical sample was aligned with the actual stress direction. Precision milling was performed using a CNC surface grinder, with the end face flatness controlled within 0.02 mm / m. Polishing was done using sandpaper with a grit of 800 mesh or higher, achieving a surface roughness Ra ≤ 0.8 μm. High-contrast random speckle patterns were formed by aerosol spraying, with a grayscale difference greater than 150 between the black base and the white speckles, and the speckle particle size controlled within the range of 0.1-0.3 mm. During the static curing stage, the ambient temperature was controlled at 20±1℃, the relative humidity at 50±5%, and the curing time was no less than 72 hours.

[0043] In some embodiments, the core drill extracts samples vertically to ensure that the internal structure of the material is consistent with its actual service condition. End-face milling employs diamond tools for multi-pass finishing, with flatness monitored in real-time by a laser interferometer to eliminate eccentric loads during sample installation. Side surface polishing is performed using a three-axis linkage polishing machine to uniformly remove surface micro-protrusions circumferentially, creating the mirror-like effect required for optical measurements. Speckle spraying uses a two-component epoxy resin as the base coating, generating non-repeating patterns using a computer-controlled random dot matrix generator to ensure the uniqueness of digital image correlation algorithm matching. The constant temperature and humidity curing chamber incorporates a temperature and humidity feedback control system, maintaining stable environmental parameters through a PID algorithm to eliminate residual stress within the sample.

[0044] Preferably, a core drilling rig is used to drill cylindrical samples from the in-service asphalt concrete pavement. The drilled cylindrical samples have a diameter of 100 mm and a height of 150 mm. The two end faces of the cylindrical samples are precision milled to ensure that the parallelism error of the end faces does not exceed 0.05 mm.

[0045] The side surfaces of the cylindrical samples were polished to create a smooth observation surface. The polishing process used progressively finer sandpaper, from 400 grit to 2000 grit, and finally polished with polishing paste to achieve a surface roughness Ra value of less than 0.2 μm.

[0046] A high-contrast random speckle pattern was sprayed onto the observation surface. Black and white acrylic paints were used, and the pattern was sprayed using a pneumatic spray gun at a pressure of 0.2 MPa. The average particle size of the speckle pattern was controlled within the range of 0.5-1 mm, and the coverage rate reached over 50%.

[0047] The prepared asphalt concrete pavement samples were placed in a constant temperature and humidity environment for static curing. The curing conditions were: temperature 20±2℃, relative humidity 60±5%, and curing time 24 hours.

[0048] This application further proposes a method for performing multi-stage progressive loading on asphalt concrete pavement samples and simultaneously acquiring multimodal data under preset environmental conditions, including the following steps: mounting the asphalt concrete pavement sample on a mechanical testing machine inside an environmental chamber; performing multi-stage progressive loading, each loading stage including applying a compressive load at a constant rate until a preset force threshold or displacement threshold is reached, followed by maintaining a constant load for a holding time, the force threshold being set proportionally according to the total number of loading stages and the estimated maximum loading force; simultaneously acquiring a sequence of stereoscopic images of the asphalt concrete pavement sample surface using at least two high-speed cameras throughout the loading process; recording the loading force data in real time using sensors on the mechanical testing machine; and recording the ambient temperature and humidity data in real time using sensors installed inside the environmental chamber.

[0049] The mechanical testing machine installed in the environmental chamber can precisely control loading conditions and environmental parameters, eliminating external interference factors. Multi-stage progressive loading employs a dual-control mode of force and displacement thresholds to avoid the overload risk caused by a single control mode. Stereo image sequence acquisition uses two high-speed cameras to form a binocular vision system, meeting the baseline distance requirements of 3D digital image-related algorithms. The controlled holding time allows for complete recording of the material stress relaxation process, providing a stable data foundation for subsequent deformation field construction.

[0050] In some embodiments, during the sample installation stage, a fixture is used to ensure that the loading axis coincides with the geometric center of the sample, avoiding measurement errors caused by eccentric loading. During loading, the force threshold for each stage is distributed in a geometric progression; for example, when there are 5 loading stages, the force threshold is applied in stages at 20%, 40%, 60%, 80%, and 100% of the maximum estimated load. A stereoscopic image sequence is synchronously recorded at a sampling frequency of no less than 1000 frames per second to ensure complete capture of the dynamic deformation process. An ambient temperature and humidity sensor continuously records data at a sampling interval of 0.5 seconds, aligned with the loading force data using a unified timestamp. The holding time is set to 10-30 seconds to allow sufficient transfer of internal stress in the material and to reach a quasi-static equilibrium state.

[0051] Preferably, the asphalt concrete pavement sample is mounted on a mechanical testing machine inside an environmental chamber. Multi-stage progressive loading is performed, with each loading stage consisting of applying a compressive load at a constant rate of 10 N / s until a preset force or displacement threshold is reached, followed by maintaining the load constant for 60 seconds. The force threshold is set proportionally based on the total number of loading stages and the estimated maximum loading force. For example, if the estimated maximum loading force is 10 kN and the total number of loading stages is 10, then the force threshold for the first stage is 1 kN, the second stage is 2 kN, and so on.

[0052] Throughout the loading process, two high-speed cameras simultaneously acquired a sequence of stereoscopic images of the asphalt concrete pavement sample surface. The high-speed cameras were set to a sampling frequency of 100 Hz and a resolution of 2048×2048 pixels. The mechanical testing machine's sensors recorded the loading force data in real time, with a sampling frequency of 1 kHz. Temperature and humidity sensors were installed inside the environmental chamber to record ambient temperature and humidity data in real time, with a sampling frequency of 1 Hz.

[0053] This application further proposes a processing method including the following steps: image preprocessing of the image sequence, including flat field correction, filtering and contrast enhancement; processing the stereo image sequence using a three-dimensional digital image correlation algorithm to calculate the three-dimensional displacement field of the asphalt concrete pavement sample surface; performing spatial and temporal analysis on the three-dimensional displacement field to calculate the full-field strain tensor, and the four-dimensional spatiotemporal deformation field is composed of the three-dimensional displacement field and the full-field strain tensor.

[0054] Among them, the three-dimensional digital image correlation algorithm uses the zero-mean normalized cross-correlation function as the matching criterion, and its function expression is: Flat-field correction improves image quality by eliminating illumination inhomogeneities; Gaussian filtering eliminates high-frequency noise; and contrast enhancement improves speckle pattern recognition through histogram equalization. Spatial analysis obtains strain distribution by calculating the displacement gradient tensor of adjacent measuring points, while temporal analysis uses the sliding window method to extract the strain rate field.

[0055] In some embodiments, the image preprocessing stage first performs flat-field correction on the original image to eliminate grayscale differences caused by uneven light source distribution. Filtering uses a 3×3 pixel window Gaussian filter with a standard deviation of 0.5 to effectively suppress random noise during image acquisition. Contrast enhancement is achieved by limiting the contrast adaptive histogram equalization algorithm, expanding the image's grayscale dynamic range to 0-255. The preprocessed image is input into a 3D digital image correlation algorithm, performing matching calculations with a 31×31 pixel sub-region size. The displacement field is iteratively solved using the Levenberg-Marquardt optimization algorithm, achieving a spatial resolution of 0.01 pixels. The 3D displacement field calculation is based on a binocular stereo vision system, converting 2D image coordinates to 3D spatial coordinates through calibration parameters, with measurement errors controlled within ±2μm. The full-field strain tensor is calculated using the central difference method to determine the displacement gradient tensor, employing the Green-Lagrange strain metric, with a temporal resolution synchronized with the image acquisition frame rate at 500Hz. The constructed four-dimensional spatiotemporal deformation field contains displacement and strain data in three-dimensional spatial coordinates and time dimension, which can accurately characterize the initiation and propagation behavior of microcracks inside the material during loading.

[0056] Preferably, the acquired image sequence is first preprocessed. Image preprocessing includes flattening correction, filtering, and contrast enhancement. Flattening correction is used to eliminate inconsistent image brightness caused by uneven illumination. Filtering uses a Gaussian filter to remove noise from the image. Contrast enhancement uses histogram equalization to improve the overall image sharpness.

[0057] Next, a three-dimensional digital image correlation algorithm is used to process the stereo image sequence and calculate the three-dimensional displacement field on the surface of the asphalt concrete pavement sample.

[0058] Finally, spatial and temporal analyses were performed on the three-dimensional displacement field to calculate the full-field strain tensor. The four-dimensional spatiotemporal deformation field consists of the three-dimensional displacement field and the full-field strain tensor. The spatial analysis used the finite difference method to calculate the displacement gradient, and the temporal analysis used the central difference method to calculate the displacement-time derivative. The full-field strain tensor was calculated through the symmetrical part of the displacement gradient.

[0059] This application further proposes a technical solution including the following steps: identifying the microcrack initiation location and propagation path of asphalt concrete pavement samples from a four-dimensional spatiotemporal deformation field; establishing a finite element model with the same geometric dimensions as the asphalt concrete pavement sample, embedding cohesive elements on the potential fracture path; using the loading force data and the displacement history of key points as boundary conditions to drive the finite element model to perform inverse analysis, and iteratively adjusting the traction-separation law parameters of the cohesive elements through optimization algorithms until the model output matches the measured four-dimensional spatiotemporal deformation field within a preset tolerance; and determining the optimized and converged traction-separation law parameters as mesomechanical parameters.

[0060] The microcrack initiation location was identified through local strain abrupt change regions in the four-dimensional spatiotemporal deformation field, and the propagation path was determined by time-series analysis of the continuous evolution trajectory of strain concentration regions. The geometric dimensions of the finite element model were modeled proportionally based on the diameter and height of the actual sample, and the potential fracture path was set according to the direction of crack initiation and propagation into the material. Cohesive elements were distributed in a banded pattern along the fracture path, with element thickness set to 1 / 10 to 1 / 5 of the average aggregate particle size. The Levenberg-Marquardt algorithm was used for optimization, with preset tolerances of displacement field error less than 0.05 mm and strain field error less than 0.3%.

[0061] In some embodiments, during the inverse analysis, the finite element model is dynamically subjected to boundary conditions using force-time curves recorded by a mechanical testing machine. The displacement history of key points is extracted from the four-dimensional spatiotemporal deformation field, containing displacement components in three orthogonal directions. During each iteration, the optimization algorithm synchronously adjusts the initial stiffness, cohesive strength, and critical fracture energy of the cohesive elements. By comparing the displacement field calculated by the model with the measured data, an error gradient vector is generated. When the magnitude of the error gradient vector is lower than a preset threshold, parameter convergence is determined. In the traction-separation law parameter optimization process, regularization is used to eliminate the influence of measurement noise on the inverse results, and sensitivity analysis is used to determine the identifiability weights of each parameter. The final obtained micromechanical parameters directly characterize the energy dissipation properties of the material during crack propagation, providing accurate input data for macroscopic strength calculations.

[0062] Preferably, the initiation location and propagation path of microcracks in asphalt concrete pavement samples are identified from the four-dimensional spatiotemporal deformation field. This can be achieved by analyzing the gradient and discontinuities of the strain field. For example, a strain gradient threshold of 0.001 / mm can be set; when a local strain gradient exceeds this threshold, the region is marked as a potential microcrack location.

[0063] A finite element model with the same geometry as the asphalt concrete pavement sample was established. Assuming the sample is a cylinder with a diameter of 100 mm and a height of 150 mm, a 20-node hexahedral element was used for meshing in the finite element model, with the element size set to 2 mm. Zero-thickness cohesive elements were embedded along potential fracture paths.

[0064] Using the applied force data and the displacement history of key points as boundary conditions, the finite element model is driven to perform inverse analysis. Specifically, the measured applied force-time curve is used as the load input to the model, and five feature points are selected on the sample surface, with their measured displacement-time curves serving as displacement constraints. The traction-separation law parameters of the cohesive elements are iteratively adjusted through an optimization algorithm until the model output matches the measured four-dimensional spatiotemporal deformation field within a preset tolerance. Here, a genetic algorithm is used for parameter optimization, with a population size of 100, 50 iterations, a crossover probability of 0.8, and a mutation probability of 0.1.

[0065] The optimized convergent traction-separation law parameters are defined as micromechanical parameters. These parameters include cohesive strength, critical fracture energy, and initial stiffness.

[0066] This application further proposes a bilinear constitutive model for the traction-separation law, with mesoscopic mechanical parameters including cohesive strength. Critical fracture energy And the initial stiffness K. The formula for the traction-separation law is: when the separation displacement... At that time, traction force When the separation displacement At that time, traction force ,in To achieve the required cohesive strength, the separation displacement This represents the separation displacement at the point of complete destruction.

[0067] The bilinear constitutive model comprises two characteristic stages: an elastic stage and a softening stage. The initial stiffness K controls the stress-displacement linearity and cohesive strength of the material during the elastic deformation stage. The critical stress value and critical fracture energy that characterize a material at its maximum load-bearing capacity. The area under the integral traction-separation curve is obtained, reflecting the energy required for crack propagation. and The difference determines the slope of the softening segment, which is negatively correlated with the material damage rate.

[0068] In some embodiments, during the finite element model inverse analysis, the bilinear constitutive model accurately describes the entire mechanical behavior of the material from elastic deformation to crack initiation, propagation, and eventual fracture through piecewise functions. The initial stiffness K is controlled within the range of 50-100 GPa / m to ensure that the calculated displacement during the elastic stage matches the measured data. Cohesive strength... Values ​​ranging from 1.5 to 3.0 MPa correspond to the interfacial bond strength of asphalt concrete materials. Critical fracture energy. The parameter is set to 200-500 J / m². This parameter is adjusted through an optimization algorithm to ensure that the simulated crack propagation path matches the measured deformation field by more than 95%. / The ratio is controlled within the range of 0.2-0.4. This ratio was determined through statistical analysis of a large amount of experimental data and can effectively balance the convergence of calculations and the accuracy of results.

[0069] Preferably, the traction-separation law adopts a bilinear constitutive model, and the micromechanical parameters include: cohesive strength. Critical fracture energy and initial stiffness K.

[0070] The formula for the traction-separation law is: when the separation displacement... At that time, traction force When the separation displacement At that time, traction force ; in, To achieve the required cohesive strength, the separation displacement This represents the separation displacement at the point of complete destruction.

[0071] Furthermore, in practical applications, the specific values ​​of these parameters can be determined through experimental measurements or numerical simulations. For example, for typical asphalt concrete pavement materials, the cohesive strength... c is likely in the range of 2-5 MPa, critical fracture energy. Possibly 100-500 J / Within the range, the initial stiffness K may be Within a certain range. These parameter values ​​will vary with factors such as material composition, environmental conditions, and loading rate.

[0072] Therefore, by employing the traction-separation law of the bilinear constitutive model, the fracture behavior of asphalt concrete pavement materials at the microscale can be described more accurately. Specifically, this model can capture the entire process of the material from initial elastic deformation to damage development and finally fracture, thus providing more reliable micromechanical parameter inputs for subsequent macroscopic strength calculations.

[0073] This application further proposes step S5, which includes: constructing a macroscopic material constitutive relation for asphalt concrete pavement material, wherein the macroscopic material constitutive relation is an elastoplastic constitutive model that can reflect the material damage and softening behavior; embedding the microscopic mechanical parameters obtained by inversion calculation from the cohesion model as key input parameters into the macroscopic material constitutive relation; establishing a finite element model representing the target pavement structure, and assigning the macroscopic material constitutive relation with embedded microscopic mechanical parameters to the asphalt concrete material in the finite element model; simulating a standard tensile test or flexural test on the finite element model, and calculating the virtual failure load of the asphalt concrete pavement sample under uniaxial tensile stress or flexural stress state; and calculating the macroscopic tensile strength based on the virtual failure load and the standard geometric dimensions of the asphalt concrete pavement sample.

[0074] The elastoplastic constitutive model employs an isotropic hardening criterion, and its yield function includes hydrostatic pressure-related terms to reflect the pressure-sensitive characteristics of the asphalt mixture. The geometric dimensions of the finite element model are parametrically modeled based on the actual pavement structure layer thickness and aggregate distribution characteristics. An adaptive mesh refinement strategy is used, achieving local refinement in potentially high stress gradient regions. The virtual failure load is determined by monitoring the load value corresponding to the maximum principal stress in the finite element model reaching the critical value of the material's tensile strength. The critical value is jointly calibrated by the cohesive strength and fracture energy in the micromechanical parameters.

[0075] In some embodiments, when constructing the macroscopic material constitutive relation, based on the theory of continuous damage mechanics, the cohesive strength and critical fracture energy in the microscopic mechanical parameters are converted into key coefficients in the macroscopic damage variable evolution equation. By embedding the inverted microscopic parameters into the constitutive model, a unified description of cross-scale mechanical behavior is achieved. When establishing the finite element model, three-dimensional solid elements are used to simulate the asphalt concrete layer, and interface elements are used to characterize the interlayer bonding state. During the simulation of the standard tensile test, the bottom degree of freedom of the model is constrained, and displacement control loading is applied at the top, and the element stress state and damage accumulation are calculated in real time. When the damage accumulation reaches the critical threshold, the material is determined to have undergone macroscopic failure, and the recorded load peak at this time is the virtual failure load. Based on the ratio of the virtual failure load to the nominal cross-sectional area of ​​the sample, the final macroscopic tensile strength value is output. This method replaces physical testing with numerical simulation, and while preserving the integrity of the sample, it can simulate the material response under various stress states, significantly improving the engineering applicability of the test results.

[0076] Preferably, a macroscopic material constitutive relation for asphalt concrete pavement material is constructed. This macroscopic material constitutive relation adopts a modified Drucker-Prager model, which can reflect the damage softening behavior of the material. The yield function expression of the modified Drucker-Prager model is: F=√(J2)+αI1-k=0 Where J2 is the second invariant of stress deviator, I1 is the first invariant of stress tensor, and α and k are material parameters.

[0077] The micromechanical parameters obtained from the inversion calculation of the cohesive model are embedded into the macroscopic material constitutive relation. Specifically, the cohesive strength σc, critical fracture energy Gc, and initial stiffness K are used as input parameters, and the values ​​of α and k in the modified Drucker-Prager model are determined by numerical algorithm.

[0078] A finite element model representing the target pavement structure was established. This model uses three-dimensional solid elements with a mesh size of 5 mm. An asphalt concrete material region was defined within the model, and macroscopic material constitutive relations embedding micromechanical parameters were assigned to this region.

[0079] A standard tensile test was simulated on a finite element model. Displacement-controlled loading was set with a loading rate of 0.1 mm / min. The virtual failure load of the asphalt concrete pavement sample under uniaxial tensile stress was calculated through nonlinear finite element analysis.

[0080] The macroscopic tensile strength was calculated based on the virtual failure load and the standard geometric dimensions of the asphalt concrete pavement sample. The calculation formula is as follows:

[0081] in, Where P is the macroscopic tensile strength, P is the virtual failure load, and r is the radius of the cylindrical specimen.

[0082] This application further proposes to process the latest acquired image sequence in real time during the hold period of each loading stage, and quickly calculate the current full-field strain distribution; based on the calculated full-field strain distribution, identify local strain concentration areas in real time, and calculate the local strain concentration index; compare the local strain concentration index with multiple preset thresholds; and dynamically adjust the loading parameters of the next loading stage according to the comparison results: when the local strain concentration index exceeds the first threshold but is lower than the second threshold, the constant rate is automatically reduced; when the local strain concentration index exceeds the second threshold, loading is paused and the hold stage is entered.

[0083] The real-time image sequence processing adopts a parallel computing architecture, completing image preprocessing and three-dimensional displacement field calculation within the first 30 seconds of the load holding phase. The local strain concentration index is defined as the ratio of the maximum principal strain to the average principal strain of the entire field, with a preset first threshold of 3.0 and a second threshold of 5.0. The loading rate adjustment is achieved through a closed-loop control system, and the rate adjustment range changes linearly according to the proportion of the index exceeding the threshold. For example, when the index is between the first and second thresholds, the rate reduction is 20%-50% of the original rate.

[0084] In some embodiments, an image processing thread is synchronously started during the load holding phase. GPU acceleration is used to filter the image sequence and perform 3D digital image correlation calculations to generate a full-field strain distribution map. Based on the strain distribution map, a region segmentation algorithm is used to identify regions where the strain value exceeds the average value, and their geometric center coordinates are extracted as monitoring points. By calculating the strain gradient change rate at the monitoring points, it is determined whether the local strain concentration has reached a preset threshold. When the strain concentration at the monitoring point exceeds the first threshold, the control module sends a command to the mechanical testing machine to adjust the constant rate of the next loading stage from 0.5 mm / min to 0.3 mm / min; if it exceeds the second threshold, loading is immediately stopped and the current load is maintained for 120 seconds. This process, through a real-time feedback mechanism, avoids premature local damage propagation, ensures the continuity of the four-dimensional spatiotemporal deformation field data, and provides more complete crack evolution data for cohesion model inversion.

[0085] As a preferred embodiment, the solution of this application is implemented as follows: During the multi-stage loading process of the mechanical testing machine, when entering the hold-load stage, the image processing thread is immediately started. The latest acquired stereoscopic image is used to perform full-field strain calculation through a digital image correlation algorithm. The image processing employs GPU parallel acceleration technology, with the processing time for each frame controlled within 50ms. When a local strain exceeding 0.15% is detected on the observed surface, covering more than 10% of the area, the loading rate adjustment mechanism is automatically triggered. Specifically, if the strain concentration index is in the range of 0.8-1.2, the compressive load application rate is reduced from 2mm / min to 1mm / min; when the index exceeds 1.5, the actuator movement is immediately stopped and the current hold-load time is extended to 1.5 times the original plan. This dynamic adjustment process achieves closed-loop control of the actuator displacement through a PID controller, with the control cycle synchronized with the image acquisition frequency at 100Hz.

[0086] This application further proposes a dynamic loading strategy, which includes: processing the latest image sequence in real time and calculating the full-field strain distribution during the load retention period; identifying local strain concentration areas based on the full-field strain distribution and calculating the concentration index; comparing the concentration index with a preset threshold; and dynamically adjusting the loading parameters for the next stage based on the comparison results.

[0087] The image processing employs a parallel computing architecture to achieve millisecond-level latency, ensuring the real-time nature of strain data. The local strain concentration index is obtained by calculating the ratio of the maximum principal strain to the overall average strain, with a first threshold of 2.5 and a second threshold of 3.2. The loading rate adjustment uses a PID control algorithm; when the index exceeds the first threshold, the rate decays exponentially; when the second threshold is triggered, the hydraulic servo system cuts off the loading power within 50ms. The strain monitoring area is divided into 5mm × 5mm grid cells, with each cell independently calculating its strain gradient.

[0088] In some embodiments, an image processing thread is initiated during the holding phase to input the acquired stereo images into a GPU cluster for three-dimensional digital image correlation calculations, generating a full-field strain distribution map. The image processing thread and the loading control thread exchange data via shared memory, updating the strain field data every 200ms. The strain concentration region identification module traverses all mesh cells, marking regions where the principal strain exceeds 1.8 times the full-field average as regions of interest. The loading parameter adjustment module dynamically corrects the loading curve based on real-time strain data, automatically extending the next holding time by 15% when a 20% increase in the number of regions of interest is detected. This dynamic control process forms a closed-loop feedback, reducing the loading rate to 10% of the initial value when the specimen is nearing failure, ensuring the capture of complete deformation data during the stable crack propagation stage, while avoiding data loss due to sudden specimen fracture.

[0089] As a preferred embodiment, the specific implementation of this application is as follows: After completing the macroscopic tensile strength calculation of the asphalt concrete pavement samples, model construction and updating operations are performed. First, asphalt concrete pavement samples with different service years are collected from multiple road engineering sites. The ambient temperature data for each sample is recorded as a range of -20℃ to 60℃ using a temperature sensor; the ambient humidity data is recorded as a range of 30%RH to 95%RH using a humidity sensor; the asphalt content is measured to be in the range of 4.5%-6.2% through chemical analysis; the aggregate gradation parameters are obtained through sieve analysis to determine the proportion of each particle size range; and the material age is determined to be 1-15 years based on construction records. The above parameters, along with the measured tensile strength of the corresponding samples, are stored in a database to form a training set containing 200 sets of data.

[0090] A gradient boosting decision tree model was established, with ambient temperature, humidity, asphalt content, aggregate gradation parameters, and material age as input layers, and the output layer being the predicted tensile strength value. During model training, mean squared error was used as the loss function, and a greedy algorithm was employed to generate decision tree nodes layer by layer. The maximum depth of each tree was set to 5 layers, the learning rate was set to 0.1, and the number of iterations was 1000. The trained model was deployed to a cloud server and connected to the road maintenance management system via an API interface.

[0091] When new test data is added, the environmental data, material parameters, and measured strength values ​​of the new samples are uploaded to the server in real time. The model employs an online learning mechanism, using the new data as an incremental training set. While maintaining the original tree structure, parameters are updated by adjusting the weight coefficients of the leaf nodes. The system automatically retrains the model every quarter to ensure continuous optimization of prediction accuracy.

[0092] This application further proposes a method for constructing and updating a service performance prediction model based on gradient boosting decision trees, including data collection, model architecture establishment, model training, deployment and updating steps.

[0093] The data collection step integrates macroscopic tensile strength, environmental temperature and humidity, microscopic mechanical parameters, material mix proportions, and age data from multiple samples to form a training dataset covering the coupling relationships of multiple factors. The model architecture establishment step uses a gradient boosting decision tree as its basic architecture, with input variables including temperature, humidity, asphalt content, aggregate gradation parameters, and material age, and the output variable being the predicted tensile strength. The model training step optimizes the loss function to determine the splitting rules and leaf node output values ​​of the decision tree, generating a predictive model with nonlinear fitting capabilities. The deployment and update step integrates the trained model into the road maintenance management system and uses incremental learning to continuously integrate new test data to update the model parameters.

[0094] In some embodiments, during the data collection phase, the material composition information for each sample includes the percentage of asphalt mass and aggregate particle size distribution parameters. Environmental temperature and humidity data are taken from environmental chamber sensor records during the loading test. Material age is measured in days from sample preparation completion to the test date. In the model architecture, the gradient boosting decision tree achieves high-precision prediction through the weighted superposition of multiple weak classifiers. The optimal splitting condition for each decision tree's splitting node is determined based on the information gain ratio of the input variables. During training, the loss function is defined as the mean square error between the predicted intensity and the measured intensity, and the decision tree weights are iteratively adjusted using the gradient descent algorithm. During deployment, the model is embedded into the maintenance management system via an application programming interface (API). After a new sample is tested, the system automatically extracts its feature vector and triggers the incremental learning module, using an online learning algorithm to update the parameters of the existing decision tree set, enabling the model to continuously adapt to the effects of material aging or changes in environmental conditions. The mathematical expression of this method calculates the final predicted intensity by weighted summation of the outputs of all decision trees, ensuring that the nonlinear interactions of different tree models on multidimensional features are effectively captured.

[0095] As a preferred embodiment, the specific implementation of this application is as follows: When establishing a machine learning prediction model, firstly, ambient temperature, ambient humidity, asphalt content, aggregate gradation parameters, and material age data are extracted from the tested asphalt concrete pavement samples to form a training dataset containing 300 samples. The input feature vector of each sample consists of data from a 25℃ constant temperature chamber recorded by a temperature sensor, 60% RH data collected by a humidity sensor, an asphalt content of 6.2%, a maximum particle size of 13.2mm in the aggregate gradation parameters, and a material age of 28 days. A gradient boosting decision tree model is constructed using the XGBoost framework, with 150 decision trees and a maximum depth of 6 layers. During the model training phase, the learning rate is optimized to 0.05 through cross-validation, and the information gain threshold for feature splitting is determined. During the model deployment phase, the trained model parameters are encapsulated as an API interface and integrated into the pavement maintenance system. When new test data includes samples with an ambient temperature of 30℃ and an asphalt content of 5.8%, the system automatically triggers an incremental learning mechanism and updates the leaf node weights using an online gradient descent algorithm.

[0096] Through the above technical solutions, this application achieves deep fusion and dynamic modeling of multi-source heterogeneous data, effectively solving the problem that traditional experimental methods struggle to quantify the coupled influence of environmental conditions and material proportions. By enhancing the multi-level feature combination capabilities of the gradient-enhanced decision tree, the nonlinear relationship between asphalt content fluctuations and aggregate gradation parameters on macroscopic strength is accurately captured, significantly improving the reliability of tensile strength prediction under complex service environments. The introduction of an incremental learning mechanism enables the model to continuously adapt to performance evolution caused by material aging, providing a real-time updated evaluation benchmark for pavement maintenance decisions.

[0097] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for measuring the tensile strength of asphalt concrete pavement, characterized in that, Includes the following steps: S1. Prepare standardized asphalt concrete pavement samples, and prepare optical measurement marks on the surface of the asphalt concrete pavement samples; S2. Under preset environmental conditions, the asphalt concrete pavement sample is subjected to multi-level progressive loading, and multimodal data during the loading process is collected simultaneously; the multimodal data includes image sequences, loading force data, and environmental temperature and humidity data; S3. Process the multimodal data to construct a four-dimensional spatiotemporal deformation field of the asphalt concrete pavement sample during the loading process; S4. Based on the four-dimensional spatiotemporal deformation field and the loading force data, the micromechanical parameters of the asphalt concrete pavement sample are calculated by inversion using the cohesive force model. S5. Calculate the macroscopic tensile strength of the asphalt concrete pavement sample based on the micromechanical parameters and material constitutive relations.

2. The method for measuring the tensile strength of asphalt concrete pavement as described in claim 1, characterized in that, Step S1 includes: S1.

1. Use a core drilling rig to drill cylindrical samples from the in-service asphalt concrete pavement, and perform precision milling on the two end faces of the cylindrical samples to ensure parallelism. S1.2 Polish the side surface of the cylindrical sample to form a smooth observation surface; S1.3 Spray a high-contrast random speckle pattern onto the observation surface, the random speckle pattern having a predetermined particle size range; S1.

4. The prepared asphalt concrete pavement sample is left to stand and cure in a constant temperature and humidity environment for a predetermined time.

3. The method for measuring the tensile strength of asphalt concrete pavement as described in claim 1, characterized in that, Step S2 includes: S2.1 The asphalt concrete pavement sample is installed on the mechanical testing machine inside the environmental chamber; S2.2 Execute the multi-stage progressive loading, each loading stage including: applying a compressive load at a constant rate until a preset force threshold or displacement threshold is reached, and then maintaining the load constant for a period of time; the force threshold is set proportionally according to the total number of loading stages and the estimated maximum loading force. S2.3 During the entire loading process, at least two high-speed cameras are used to simultaneously acquire a sequence of stereoscopic images of the surface of the asphalt concrete pavement sample; S2.4 The loading force data is recorded in real time by the sensors of the mechanical testing machine; S2.

5. The ambient temperature and humidity data are recorded in real time by sensors installed in the environmental chamber.

4. The method for measuring the tensile strength of asphalt concrete pavement as described in claim 1, characterized in that, Step S3 includes: S3.1 Perform image preprocessing on the image sequence, the image preprocessing including flat field correction, filtering and contrast enhancement; S3.

2. The stereo image sequence is processed using a three-dimensional digital image correlation algorithm to calculate the three-dimensional displacement field of the asphalt concrete pavement sample surface. The three-dimensional digital image correlation algorithm uses a zero-mean normalized cross-correlation function as the matching criterion, and its function expression is: in, The zero-mean normalized cross-correlation coefficient For the reference sub-region image grayscale matrix, The grayscale mean of the reference sub-region image, The grayscale matrix of the deformed sub-region image. The mean gray level of the deformed sub-region image. The coordinates are based on the assumptions of displacement and deformation; S3.3 Perform spatial and temporal analysis on the three-dimensional displacement field to calculate the full-field strain tensor. The four-dimensional spatiotemporal deformation field is composed of the three-dimensional displacement field and the full-field strain tensor.

5. The method for measuring the tensile strength of asphalt concrete pavement as described in claim 1, characterized in that, Step S4 includes: S4.1 Identify the microcrack initiation location and propagation path of the asphalt concrete pavement sample from the four-dimensional spatiotemporal deformation field; S4.2 Establish a finite element model with the same geometric dimensions as the asphalt concrete pavement sample, and embed cohesive elements in the finite element model on the potential fracture path; S4.3 Using the loading force data and the displacement history of key points as boundary conditions, drive the finite element model to perform inverse analysis, and iteratively adjust the traction-separation law parameters of the cohesive unit through optimization algorithm until the model output matches the measured four-dimensional spatiotemporal deformation field within the preset tolerance. S4.

4. The optimized converged traction-separation law parameters are determined as the micromechanical parameters.

6. The method for measuring the tensile strength of asphalt concrete pavement as described in claim 5, characterized in that, The traction-separation law is a bilinear constitutive model, and the micromechanical parameters include: cohesive strength. Critical fracture energy and initial stiffness K; The formula for the traction-separation law is: When separation displacement At that time, traction force ; When separation displacement At that time, traction force ; in, To achieve the required cohesive strength, the separation displacement This represents the separation displacement at the point of complete destruction.

7. The method for measuring the tensile strength of asphalt concrete pavement as described in claim 1, characterized in that, Step S5 includes: S5.1 Construct the macroscopic material constitutive relation of the asphalt concrete pavement material, wherein the macroscopic material constitutive relation is an elastoplastic constitutive model that can reflect the material damage and softening behavior; S5.

2. The micromechanical parameters obtained by inversion calculation from the cohesive force model are embedded as key input parameters into the macroscopic material constitutive relation; S5.3 Establish a finite element model representing the target pavement structure, and assign the macroscopic material constitutive relation embedded with the micromechanical parameters to the asphalt concrete material in the finite element model; S5.4 Simulate a standard tensile test or bending tensile test on the finite element model, and calculate the virtual failure load of the asphalt concrete pavement sample under uniaxial tensile stress or bending tensile stress. S5.

5. The macroscopic tensile strength is calculated based on the virtual failure load and the standard geometric dimensions of the asphalt concrete pavement sample.

8. The method for measuring the tensile strength of asphalt concrete pavement as described in claim 1, characterized in that, In step S2, a dynamic loading strategy is implemented, specifically including: S8.1 During the load holding period of each loading level, the latest acquired image sequence is processed in real time to quickly calculate the current full-field strain distribution; S8.2 Based on the calculated full-field strain distribution, identify local strain concentration areas in real time and calculate the local strain concentration index; S8.

3. Compare the local strain concentration index with a number of preset thresholds; S8.

4. Based on the comparison results, dynamically adjust the loading parameters for the next loading stage: when the local strain concentration index exceeds the first threshold but is lower than the second threshold, automatically reduce the constant rate; when the local strain concentration index exceeds the second threshold, pause loading and enter the load maintenance stage.

9. The method for measuring the tensile strength of asphalt concrete pavement as described in any one of claims 1 to 8, characterized in that, The method further includes a model building and updating step S6 performed after obtaining the macroscopic tensile strength: S6.1 Data Collection: Collect test data from multiple asphalt concrete pavement samples to form a training dataset; the test data includes the macroscopic tensile strength of each sample, as well as the corresponding environmental temperature and humidity data, the microscopic mechanical parameters, and the original material mix information of the asphalt concrete pavement; S6.2 Model Architecture Establishment: Establish a machine learning prediction model with ambient temperature, ambient humidity, asphalt content, aggregate gradation parameters and material age as input variables and predicted tensile strength as output variable; the machine learning prediction model adopts a gradient boosting decision tree architecture; S6.3 Model Training: The gradient boosting decision tree architecture is trained using the training dataset. The splitting rules of all nodes and the output values ​​of leaf nodes are determined by optimizing the loss function, and a trained service performance prediction model is generated. S6.4 Model Deployment and Update: Deploy the service performance prediction model in the road maintenance management system; after testing new road samples using the method and obtaining new test data, input the new test data into the service performance prediction model to update the model parameters in an incremental learning manner.

10. The method for measuring the tensile strength of asphalt concrete pavement as described in claim 9, characterized in that, The mathematical expression of the service performance prediction model is based on the decision rule set of gradient boosting decision trees, and its final output is the weighted sum of the outputs of all decision trees, expressed as: in, The predicted service tensile strength, where N is the total number of decision trees. Let be the weight of the i-th decision tree. The output of the i-th decision tree for the input feature vector X includes ambient temperature, ambient humidity, asphalt content, aggregate gradation parameters, and material age.

Images