A method and apparatus for aggregate formulation optimization based on multiphase heterogeneous interface modeling

By using multiphase heterogeneous interface modeling and deep neural network proxy model, the geometric morphology and interface distribution characteristics of RAP aggregates are accurately restored, solving the problem of low efficiency and low precision in aggregate formulation optimization, realizing efficient and accurate optimization of aggregate formulation, and improving the road performance of the mixture.

CN123120103APending Publication Date: 2026-11-17WUHAN UNIV OF TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611403874.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-10
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing aggregate formulation optimization methods fail to accurately reproduce the true geometric morphology and spatial distribution characteristics of multiphase heterogeneous interfaces of RAP aggregates, resulting in limited improvement of the road performance of the mixture. Furthermore, traditional methods are inefficient and lack precision.

Method used

By modeling multiphase heterogeneous interfaces and combining them with deep neural network surrogate models, we can perform aggregate geometry analysis, aging simulation, overlap collision simulation, and epoxy regeneration simulation to construct an accurate multiphase heterogeneous interface numerical model. We can then perform mechanical property calibration and finite element damage and cracking simulation to ultimately achieve efficient optimization of aggregate formulation.

Benefits of technology

It accurately reproduces the true geometric morphology and spatial distribution characteristics of multiphase heterogeneous interfaces of RAP aggregates, significantly improving the efficiency and accuracy of obtaining the optimal aggregate formula and enhancing the road performance of the mixture.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123120103A_ABST
    Figure CN123120103A_ABST
Patent Text Reader

Abstract

This application proposes an aggregate formulation optimization method and apparatus based on multiphase heterogeneous interface modeling, relating to the field of asphalt recycling technology. The method includes: obtaining a RAP aggregate geometric model through aging simulation based on a new aggregate geometric model; obtaining a mixed distribution geometric model of RAP and new aggregates through overlap and collision simulation based on the new aggregate geometric model, the RAP aggregate geometric model, and aggregate gradation; obtaining a multiphase heterogeneous interface numerical model through epoxy regeneration simulation based on the mixed distribution geometric model of RAP and new aggregates; obtaining a sample aggregate formulation and sample mechanical properties through finite element damage and cracking simulation based on the multiphase heterogeneous interface numerical model; obtaining a final deep neural network surrogate model by iteratively optimizing an initial deep neural network surrogate model using the sample aggregate formulation and sample mechanical properties; and obtaining the optimal aggregate formulation through gradient backpropagation-based iterative optimization of the aggregate formulation based on the target mechanical properties and the final deep neural network surrogate model.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of asphalt recycling technology, and in particular to a method and apparatus for optimizing aggregate formulation based on multiphase heterogeneous interface modeling. Background Technology

[0002] Epoxy recycled asphalt mixtures, as a green road material, achieve efficient recycling of old asphalt and aggregates through the crushing, screening, and recycling of recycled asphalt mixtures (RAP). To further improve the road performance of recycled mixtures, incorporating epoxy resin modifiers into the recycled asphalt can significantly improve its bonding and anti-aging properties, making epoxy recycled asphalt mixtures promising for applications in special scenarios such as heavy traffic and long longitudinal slopes. However, epoxy recycled asphalt mixtures are complex composite materials composed of multiple components such as RAP, new aggregates, and epoxy-modified asphalt binders. They contain numerous heterogeneous interfaces, and the mechanical properties and spatial distribution characteristics of these interfaces directly determine the overall performance and damage evolution of the mixture. Therefore, establishing a numerical model that can realistically reflect the microstructure and heterogeneous interface characteristics of epoxy recycled asphalt mixtures, and using this model to efficiently optimize aggregate formulations, is of great significance for improving the road performance of the mixture.

[0003] Existing aggregate formulation optimization methods do not consider the distribution characteristics of aged asphalt coated on the RAP surface, resulting in the inability to restore the true geometric morphology and spatial distribution characteristics of multiphase heterogeneous interfaces of RAP aggregates. Traditional methods for obtaining sample data through physical tests are inefficient and have low sample data accuracy, affecting the predictive ability of the prediction model. Traditional aggregate formulation optimization methods based on genetic algorithms have high trial-and-error costs and slow convergence speed, resulting in low efficiency and accuracy in obtaining the optimal aggregate formulation. Summary of the Invention

[0004] To address the aforementioned problems, in a first aspect, the present invention provides an aggregate formulation optimization method based on multiphase heterogeneous interface modeling, comprising: The particle geometry of the sample aggregate is analyzed to obtain the statistical distribution characteristics of the aggregate morphology parameters; a parameterized generation framework is obtained, and a new aggregate geometry model is constructed based on the statistical distribution characteristics of the aggregate morphology parameters and the parameterized generation framework. Aging simulation was performed based on the new aggregate geometric model to obtain the RAP aggregate geometric model; aggregate gradation was obtained, and overlap and collision simulation was performed based on the new aggregate geometric model, the RAP aggregate geometric model, and the aggregate gradation to obtain the mixed distribution geometric model of RAP aggregate and new aggregate; epoxy regeneration simulation was performed based on the mixed distribution geometric model of RAP aggregate and new aggregate to obtain the multiphase heterogeneous interface numerical model. The mechanical properties of the multiphase heterogeneous interface numerical model are calibrated to obtain the basic numerical model; based on the basic numerical model, finite element damage and cracking simulation is performed to obtain the sample aggregate formulation and sample mechanical properties. An initial deep neural network surrogate model is constructed. The initial deep neural network surrogate model is iteratively optimized by the sample aggregate formulation and sample mechanical properties to obtain the final deep neural network surrogate model. The target mechanical properties are obtained. Based on the target mechanical properties and the final deep neural network surrogate model, the aggregate formulation is iteratively optimized by gradient backpropagation to obtain the optimal aggregate formulation.

[0005] In one embodiment, the process of obtaining the RAP aggregate geometry model includes: The new aggregate geometric model is used as the geometric profile of the core aggregate to obtain the particle size range and preset basic performance parameters of the sample aggregate. The preset basic performance parameters include RAP total asphalt content, ineffective asphalt content, specific surface area, old asphalt density and service life. The effective asphalt film coating thickness under different particle size ranges is calculated based on the preset basic performance parameters. Using the equidistant offset algorithm, the outer normal direction of the geometric contour of the core aggregate is extended outward according to the effective asphalt film coating thickness to generate a multi-layered concentric layered contour. Using the geometric contour of the core aggregate and the concentric layered contour as boundaries, multiple initial annular regions arranged from the inside out are defined; the initial annular regions are assigned aging degree attributes that decrease from high to low according to the spatial order from the inside out, thus obtaining annular regions with aging degree attributes. By combining the annular region of aging degree attribute with the geometric contour of the core aggregate, the RAP aggregate geometric model is obtained.

[0006] In one embodiment, the process of obtaining the geometric model of the mixed distribution of RAP aggregate and new aggregate includes: The new aggregate geometry model and the RAP aggregate geometry model together constitute the aggregate to be fed, and the aggregate gradation is converted into the target probability of each particle size range. The quantity of aggregates is calculated based on the target probability. The modeling area is divided into multiple sub-areas of equal size based on the quantity of aggregates. A placement position is randomly generated in each sub-area. The aggregates to be placed are placed one by one into the placement position corresponding to the current sub-area to obtain the currently placed aggregates. If the minimum distance between the currently deployed aggregate and any successfully deployed aggregate is less than the preset distance constraint or exceeds the area boundary, then the deployment position of the aggregate to be deployed in the current sub-area will be regenerated; otherwise, the currently deployed aggregate will be regarded as the new successfully deployed aggregate. The aggregates to be added are sorted from largest to smallest according to their maximum particle size, and then added one by one to the addition position until all the aggregates to be added are successfully added or the preset maximum number of additions is reached, thus obtaining a geometric model of the mixed distribution of RAP aggregates and new aggregates.

[0007] In one embodiment, the process of obtaining the numerical model of the multiphase heterogeneous interface includes: Boolean operations are used to extract the blank areas outside the RAP aggregate geometric model and the new aggregate geometric model within the mixed distribution geometric model of RAP aggregate and new aggregate, and the blank areas are defined as epoxy asphalt mortar areas. Based on the spatial relationship of the components, the internal whole of the epoxy asphalt mastic body, the mastic-RAP coating layer interface, and the mastic-new aggregate interface are determined in the epoxy asphalt mastic region. Zero-thickness cohesive units were embedded into the entire internal domain of the epoxy asphalt mastic body, the mastic-RAP coating layer interface, and the mastic-new aggregate interface to obtain a multiphase heterogeneous interface numerical model.

[0008] In one embodiment, the process of obtaining the sample aggregate formulation and sample mechanical properties includes: Through cohesion and adhesion tests, the performance parameters of the zero-thickness cohesion units of the epoxy asphalt mortar matrix, the mortar-RAP coating layer interface, and the mortar-new aggregate interface in the multiphase heterogeneous interface numerical model were calibrated to obtain the basic numerical model. The sample aggregate formulation is generated based on the basic numerical model. The sample aggregate formulation includes gradation distribution parameters, aggregate morphology parameters, interfacial cohesion strength and adhesion strength. Finite element damage and cracking simulations were performed on the sample aggregate formulation under various working conditions to obtain the mechanical properties of the sample, including peak load, fracture energy, crack initiation location, crack propagation length, and damage variable evolution curves.

[0009] In one embodiment, the initial deep neural network proxy model includes a linear embedding layer, a multi-head self-attention network, a data-driven output network, and a physical information output network. The process of obtaining the final deep neural network proxy model includes: The sample aggregate formulation is mapped into a high-dimensional feature vector through a linear embedding layer; Multi-head self-attention networks use a scaled dot product attention mechanism to assign weights to high-dimensional feature vectors to obtain deep feature tensors. The deep feature tensor is used as input data to drive the output network to predict performance and obtain the predicted mechanical properties. The deep feature tensor is input into the physical information output network to extract physical data, thereby obtaining the damage variable field, stress field and strain energy release rate field. The data-driven loss function is calculated based on the predicted mechanical properties and the sample mechanical properties, and the physical constraint loss function is calculated based on the damage variable field, stress field, and strain energy release rate field. The data-driven loss function and the physical constraint loss function are weighted and summed to obtain the composite loss function. During the warm-up training phase, the weights of the physical constraint loss function are set to 0, and the initial deep neural network surrogate model is iteratively optimized using only the data-driven loss function. After running a preset number of training rounds, the first-stage network is obtained. During the joint training phase, the weights of the physical constraint loss function are gradually increased from 0 to a preset target value. The composite loss function is then used to iteratively optimize the first-stage network until the composite loss function stops decreasing in multiple iterations, thus obtaining the final deep neural network surrogate model.

[0010] In one embodiment, the process of obtaining the optimal aggregate formulation includes: Based on the final deep neural network surrogate model, a differentiable forward mapping function is constructed. Driven by the target mechanical properties and with the aggregate formulation as the optimization object, the inverse design loss function is constructed through the differentiable forward mapping function. The gradient of the inverse design loss function with respect to the aggregate formulation is calculated using the chain rule. The aggregate formulation is then iteratively updated along the descent direction of the gradient using a gradient descent optimizer until the inverse design loss function converges to the preset tolerance, at which point the iteration terminates and the optimal aggregate formulation is obtained.

[0011] Secondly, the present invention provides an aggregate formulation optimization device based on multiphase heterogeneous interface modeling, used to implement the aggregate formulation optimization method based on multiphase heterogeneous interface modeling, the device comprising: The new aggregate geometric model acquisition module is used to perform particle geometry morphology analysis on sample aggregates, obtain statistical distribution characteristics of aggregate morphology parameters, acquire a parameterized generation framework, and construct a new aggregate geometric model based on the statistical distribution characteristics of aggregate morphology parameters and the parameterized generation framework. The multiphase heterogeneous interface numerical model acquisition module is used to perform aging simulation based on the new aggregate geometric model to obtain the RAP aggregate geometric model; obtain the aggregate gradation; and perform overlap and collision simulation based on the new aggregate geometric model, the RAP aggregate geometric model, and the aggregate gradation to obtain the mixed distribution geometric model of RAP aggregate and new aggregate; and perform epoxy regeneration simulation based on the mixed distribution geometric model of RAP aggregate and new aggregate to obtain the multiphase heterogeneous interface numerical model. The training sample acquisition module is used to calibrate the mechanical properties of the multiphase heterogeneous interface numerical model to obtain the basic numerical model; based on the basic numerical model, finite element damage and cracking simulation is performed to obtain the sample aggregate formula and sample mechanical properties. The optimal aggregate formulation acquisition module is used to construct an initial deep neural network surrogate model. The initial deep neural network surrogate model is iteratively optimized by using sample aggregate formulations and sample mechanical properties to obtain a final deep neural network surrogate model. The target mechanical properties are obtained, and the aggregate formulation is iteratively optimized based on gradient backpropagation according to the target mechanical properties and the final deep neural network surrogate model to obtain the optimal aggregate formulation.

[0012] Thirdly, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the aggregate formulation optimization method based on multiphase heterogeneous interface modeling.

[0013] Fourthly, the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the aggregate formulation optimization method based on multiphase heterogeneous interface modeling.

[0014] The present invention has the following beneficial effects: 1. The geometric model of the new aggregate is sequentially subjected to aging simulation, overlapping collision simulation, and epoxy regeneration simulation to obtain the geometric model of RAP aggregate, the geometric model of the mixed distribution of RAP aggregate and new aggregate, and the numerical model of the multiphase heterogeneous interface, respectively. This accurately restores the true geometric morphology of RAP aggregate and the spatial distribution characteristics of the multiphase heterogeneous interface. By calibrating the mechanical properties of the multiphase heterogeneous interface numerical model and simulating damage and cracking using finite element methods, accurate sample aggregate formulations and their corresponding mechanical properties are efficiently generated. The initial deep neural network surrogate model is iteratively optimized using the sample aggregate formulations and mechanical properties, enabling the final deep neural network surrogate model to have accurate mechanical property prediction capabilities. Through iterative optimization of the aggregate formulation based on gradient backpropagation, the aggregate formulation is directly optimized end-to-end with the target mechanical properties as the driving force. The global optimal solution is searched efficiently in the continuous parameter space, which greatly improves the efficiency and accuracy of obtaining the optimal aggregate formulation.

[0015] 2. Using the new aggregate geometric model as the core aggregate geometric contour, the annular region is assigned an aging degree attribute that decreases progressively from high to low according to the spatial order from the inside out, generating a RAP aggregate geometric model that accurately restores the real gradient distribution characteristics of aged asphalt. Based on the aggregate quantity, the modeling area is divided into multiple sub-regions of equal size, and placement positions are randomly generated. Overlap and collision detection is performed through preset spacing constraints and region boundary constraints. After all placements are successful, a geometric model of mixed distribution of RAP aggregate and new aggregate is generated, which greatly improves the realism of aggregate spatial distribution. Three types of interfaces are extracted through Boolean operations, and zero-thickness cohesive units are embedded into the three types of interfaces to generate a multiphase heterogeneous interface numerical model, providing reliable numerical model support for the whole process simulation of interface cracking failure. The performance parameters of the zero-thickness cohesive units of the three types of interfaces are calibrated through cohesion and adhesion tests, and a sample aggregate formulation and sample mechanical properties covering multiple interface systems and working conditions are constructed, which improves the optimization effect of subsequent deep neural network surrogate models.

[0016] 3. The deep neural network surrogate model introduces a dual-branch architecture of data-driven output network and physical information output network. While fitting sample data, it incorporates prior physical knowledge of fracture mechanics and damage evolution, significantly improving prediction accuracy and generalization ability. During training, a two-stage collaborative strategy is adopted. In the warm-up stage, the weight of the physical constraint loss function is set to 0, and in the joint training stage, the weight of the physical constraint loss function is gradually increased from 0 to the preset target value, effectively ensuring training stability and final model reliability. In the aggregate formulation iterative optimization stage based on gradient backpropagation, a gradient descent optimizer is used to iteratively update the aggregate formulation along the gradient descent direction until convergence. This can directly solve for the optimal formulation in the continuous parameter space, reducing optimization time while improving the accuracy of obtaining the optimal aggregate formulation. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a method according to an embodiment of the present invention; Figure 2 This is a structural diagram of the device according to an embodiment of the present invention. Detailed Implementation

[0019] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0020] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to and includes any or all possible combinations of one or more of the listed items.

[0021] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0022] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0023] Reference Figure 1 The present invention provides an aggregate formulation optimization method based on multiphase heterogeneous interface modeling, including steps S100-S400.

[0024] S100 performs particle geometry analysis on the sample aggregates to obtain the statistical distribution characteristics of aggregate morphology parameters; obtains a parameterized generation framework, and constructs a new aggregate geometry model based on the statistical distribution characteristics of aggregate morphology parameters and the parameterized generation framework.

[0025] In some embodiments, the old and new aggregates of epoxy recycled asphalt mixtures are used as scanning objects, and computed tomography (CT) technology is used to scan the aggregates with a scanning resolution of 50-100 μm / pixel to obtain a sequence of three-dimensional cross-sectional images of each aggregate particle.

[0026] Image processing techniques are used to extract the contour boundaries of each aggregate particle, and the equivalent particle size and geometric morphology parameters of each aggregate particle are calculated based on these contour boundaries. Geometric morphology parameters include roundness, angularity index, and aspect ratio. Roundness... The calculation formula is:

[0027] in, The projected area of ​​the aggregate particle outline, in mm. 2 ; Perimeter of the aggregate particle outline, in mm. Roundness The value ranges from 0 to 1. The closer the value is to 1, the closer the shape of the aggregate is to a circle.

[0028] angularity index The calculation formula is: in, For the aggregate particle profile The interior angles of each vertex, in degrees (°); The total number of vertices. Angularity index. The larger the value, the more angular the aggregate.

[0029] The aggregate particles are divided into size ranges according to their equivalent particle size. The geometric morphology parameters of each aggregate particle are statistically analyzed according to the size range to obtain the distribution range and probability density function of the geometric parameters of each size range. The statistical distribution characteristics of the aggregate morphology parameters divided by size range are output.

[0030] Obtain a custom modeling region, mesh the custom modeling region, and select the meshing scale. m, with an approximate unit size of 1mm, generates a mesh for the custom modeling region.

[0031] The finite element software was further developed using Python. Based on the generated mesh, the node and element information of the component was recorded, and the total area of ​​the component was calculated. The geometric boundary of the component and the node number on the boundary were extracted from the node and element information. The mesh information of the custom modeling area and the node and element data structure used to describe the geometric boundary of the aggregate were output, and a parametric generation framework for generating the aggregate model was constructed.

[0032] Using the statistical distribution characteristics of aggregate morphology parameters categorized by particle size as input, a parameterized generation framework is invoked, and a vertex coordinate generation algorithm based on polar coordinates is employed to generate vertex coordinates. The vertex generation algorithm controls the random offset of each vertex by the radius fluctuation ratio and the angle fluctuation ratio, and controls the number and distribution density of vertices by the number of aggregate edges and the number of aggregate corner points.

[0033] In polar coordinates, with radius and angle Two parameters determine any point The position in a two-dimensional plane. The vertex coordinates are generated as follows:

[0034] in, This represents the basic radius of the aggregate, in mm. For any point The random radius, in mm; This represents the radius fluctuation ratio, with a value ranging from 0 to 1. The larger the value, the more irregular the generated aggregate profile. This represents the angle fluctuation ratio, with a value ranging from 0 to 0.5. The larger the value, the more irregular the generated aggregate profile. This is the number of edges of the aggregate. This refers to the number of corner points of the aggregate; Let be a random function, representing that in Generate a uniformly distributed random number within the interval.

[0035] The initial aggregate contour is formed by sequentially connecting the vertex coordinate sequence. The maximum diameter and aspect ratio of the custom aggregate are obtained. The maximum diameter and aspect ratio are input into the vertex random generation algorithm to generate the initial aggregate contour.

[0036] Calculate the roundness of the initial aggregate profile and angularity index The calculated roundness and angularity index The distribution range of geometric morphology parameters for the corresponding particle size range is compared to determine whether the morphology of the initial aggregate profile meets the similarity requirements.

[0037] If roundness Or angularity index If the distribution range of geometric morphology parameters deviates from the corresponding grain size range, adjust the radius fluctuation ratio in the vertex random generation algorithm. Angle fluctuation ratio , number of aggregate edges and the number of aggregate corner points The vertex coordinate sequence is regenerated to form an updated aggregate profile. The comparison and judgment are repeated until the roundness of the updated aggregate profile is satisfied. and angularity index The iteration stops when the similarity between the distribution range of the geometric morphology parameters of the corresponding particle size range and the distribution range of the corresponding particle size range reaches a preset similarity threshold, and the preset similarity threshold is ≥85%.

[0038] The aggregate contours that meet the preset similarity threshold are output as new aggregate geometric models, and the vertex coordinates, boundary information and geometric feature parameters of the new aggregate geometric models are recorded.

[0039] S200 performs aging simulation based on the new aggregate geometric model to obtain the RAP aggregate geometric model; obtains the aggregate gradation; performs overlap and collision simulation based on the new aggregate geometric model, the RAP aggregate geometric model, and the aggregate gradation to obtain the mixed distribution geometric model of RAP aggregate and new aggregate; performs epoxy regeneration simulation based on the mixed distribution geometric model of RAP aggregate and new aggregate to obtain the multiphase heterogeneous interface numerical model.

[0040] In some embodiments, the process of obtaining the RAP aggregate geometry model includes: The new aggregate geometric model is used as the geometric profile of the core aggregate to obtain the particle size range and preset basic performance parameters of the sample aggregate. The preset basic performance parameters include RAP total asphalt content, ineffective asphalt content, specific surface area, old asphalt density and service life. The effective asphalt film coating thickness under different particle size ranges is calculated based on the preset basic performance parameters. Using the equidistant offset algorithm, the outer normal direction of the geometric contour of the core aggregate is extended outward according to the effective asphalt film coating thickness to generate a multi-layered concentric layered contour. Using the geometric contour of the core aggregate and the concentric layered contour as boundaries, multiple initial annular regions arranged from the inside out are defined; the initial annular regions are assigned aging degree attributes that decrease from high to low according to the spatial order from the inside out, thus obtaining annular regions with aging degree attributes. By combining the annular region of aging degree attribute with the geometric contour of the core aggregate, the RAP aggregate geometric model is obtained.

[0041] In some embodiments, the preset basic performance parameters include the percentage of total asphalt mass in RAP. Percentage of ineffective asphalt mass Specific surface area Density of old asphalt and service life .

[0042] in, This indicates the percentage of total asphalt mass in RAP, expressed as % . This indicates the percentage of ineffective asphalt in the RAP mixture by mass, expressed in % . This indicates the specific surface area of ​​the RAP old aggregate grade after mixing, in m². 2 / g; This indicates the density of old asphalt in RAP mixtures, expressed in kg / m³. 3 ; This refers to the service life of the RAP material, in years.

[0043] Effective asphalt film coating thickness The unit is mm, and the calculation formula is as follows: in, This indicates the percentage of aggregate particles of each size that pass through the test, expressed in % . The aging influence factor has a value range of 0-1 and is used to characterize the effect of the degree of aging on the effective asphalt film coating thickness.

[0044] Aging Influencing Factors The calculation formula is: in, This is the aging rate constant, with a value ranging from 0.5 to 2. The service life of the RAP material is in years. For reference years, the value is 5 years.

[0045] Let the coordinates of any point on the geometric profile of the core aggregate be... The unit vector of the outward normal direction at that point is Then the first Coordinates of corresponding points on the concentric layered contour for: in, , The total number of layers is determined by the preset layer thickness and the effective asphalt film coating thickness. It was jointly determined that the vertical distance between the concentric layer contours of each layer is equal to the preset layer thickness.

[0046] Assign progressively decreasing aging level attributes. The specific method for assigning values ​​to the aging level attributes is as follows: Let the aging level attribute value be... ,in Let be the radial distance from the center of the geometric profile of the core aggregate, in mm. Then: in, This represents the initial aging degree attribute value of the core aggregate surface; The average radius of the geometric profile of the core aggregate, in mm; The aging attenuation coefficient has a value range of 0.1-0.5; the annular regions with completed aging degree attribute assignments are combined with the geometric contour of the core aggregate to form the RAP aggregate geometric model.

[0047] In some embodiments, the process of obtaining the geometric model of the mixed distribution of RAP aggregates and new aggregates includes: The new aggregate geometry model and the RAP aggregate geometry model together constitute the aggregate to be fed, and the aggregate gradation is converted into the target probability of each particle size range. The quantity of aggregates is calculated based on the target probability. The modeling area is divided into multiple sub-areas of equal size based on the quantity of aggregates. A placement position is randomly generated in each sub-area. The aggregates to be placed are placed one by one into the placement position corresponding to the current sub-area to obtain the currently placed aggregates. If the minimum distance between the currently deployed aggregate and any successfully deployed aggregate is less than the preset distance constraint or exceeds the area boundary, then the deployment position of the aggregate to be deployed in the current sub-area will be regenerated; otherwise, the currently deployed aggregate will be regarded as the new successfully deployed aggregate. The aggregates to be added are sorted from largest to smallest according to their maximum particle size, and then added one by one to the addition position until all the aggregates to be added are successfully added or the preset maximum number of additions is reached, thus obtaining a geometric model of the mixed distribution of RAP aggregates and new aggregates.

[0048] In some embodiments, a custom aggregate gradation is obtained, which includes the mass percentage passing through each particle size range. Based on the aggregate sphericity assumption, the three-dimensional Fuller gradation curve is converted into an equivalent gradation probability distribution of a two-dimensional cross-section within the modeling region, obtaining the target probability of each particle size range in the two-dimensional cross-section. The conversion formula is as follows:

[0049] in, This represents the maximum aggregate particle diameter, in mm. The sieve aperture size is in mm. This represents the percentage of aggregate in the total volume of concrete, expressed as % . For particles smaller than 2D cross-section The target probability of the aggregate appearing, in % %. The correction factor, with a value ranging from 0.05 to 0.15, is used to characterize the loss effect of small-diameter aggregates in a two-dimensional cross-section. The decay exponent has a value range of 1-2; It is a dynamic power exponent, which varies with... The value varies and ranges from 0.5 to 0.6.

[0050] Dynamic power index The calculation formula is: Multiply the target probability of each particle size range by the total area of ​​the modeling region to obtain the total area of ​​the aggregate for that range. Then divide this total area by the representative area of ​​a single aggregate particle for that range to obtain the aggregate quantity for that range. Based on the aggregate quantity, the modeling region is divided into multiple equal-sized sub-regions, with the number of sub-regions matching the aggregate quantity. A placement location is randomly generated within each sub-region, ensuring that all placement locations are evenly distributed throughout the entire modeling region.

[0051] Sort the aggregates to be added in descending order of their maximum particle size. Then, add the aggregates to be added one by one to the corresponding placement position in the current sub-area in descending order of particle size to obtain the aggregates that have been added so far.

[0052] Each time an aggregate is deployed, the geometric profile of the current aggregate is placed at the deployment position. All successfully deployed aggregate geometric profiles are traversed. The minimum distance between the geometric profile of the current aggregate and all successfully deployed aggregate geometric profiles is checked to see if it is greater than or equal to the preset distance constraint. The geometric profile of the current aggregate is also checked to see if it is entirely within the preset boundary of the modeling area. The preset distance constraint is set to 0.5-2mm.

[0053] If the minimum distance between the currently deployed aggregate and any successfully deployed aggregate is less than the preset distance constraint, or if the geometric contour of the currently deployed aggregate exceeds the boundary of the modeling area, then a collision is determined, the current deployment position is rejected, and a new deployment position for the aggregate to be deployed in the current sub-region is generated.

[0054] If both tests pass, the currently added aggregate will be considered as a new successfully added aggregate.

[0055] When all aggregates to be placed are successfully placed into the modeling area and all pass the overlap collision detection and boundary constraint detection, the placement will stop; if the placement of a single aggregate reaches the preset maximum number of retries and still fails to be placed, the placement will stop and an error message will be output. The preset maximum number of placements is 100-500 times.

[0056] Obtain the placement position and geometric contour of each successfully placed aggregate. Assemble the geometric contour of each aggregate into the modeling area one by one according to its corresponding placement position. Combine the spatial position and geometric contour information of all assembled aggregates and output the geometric model of the mixed distribution of RAP aggregate and new aggregate.

[0057] In some embodiments, the process of obtaining the numerical model of the multiphase heterogeneous interface includes: Boolean operations are used to extract the blank areas outside the RAP aggregate geometric model and the new aggregate geometric model within the mixed distribution geometric model of RAP aggregate and new aggregate, and the blank areas are defined as epoxy asphalt mortar areas. Based on the spatial relationship of the components, the internal whole of the epoxy asphalt mastic body, the mastic-RAP coating layer interface, and the mastic-new aggregate interface are determined in the epoxy asphalt mastic region. Zero-thickness cohesive units were embedded into the entire internal domain of the epoxy asphalt mastic body, the mastic-RAP coating layer interface, and the mastic-new aggregate interface to obtain a multiphase heterogeneous interface numerical model.

[0058] In some embodiments, the entire modeling region is assumed to be The area occupied by the geometric contour of all aggregates is Then the blank area , fill the blank area Defined as the epoxy asphalt mortar area.

[0059] Based on the spatial relationship of each component in the geometric model of RAP aggregate and new aggregate mixture distribution, in the epoxy asphalt mortar area Three types of interfaces are identified: The first type is epoxy asphalt mortar body. The entire internal domain is used for subsequent embedding of cohesive units that characterize the cracking of the adhesive matrix.

[0060] The second type is the adhesive-RAP coating layer interface. This refers to the contact boundary between the epoxy asphalt mortar region and the outermost concentric layered contour of each RAP aggregate geometric model, which is used to subsequently embed cohesive units that characterize the debonding of the mortar and the RAP coating layer interface.

[0061] The third category is the interface between the cementitious mortar and the new aggregate. This refers to the contact boundary between the epoxy asphalt mortar region and the geometric model outline of each new aggregate, which is used to subsequently embed cohesive units that characterize the debonding of the mortar and the new aggregate interface.

[0062] The zero-thickness cohesive element has a four-node zero-thickness element with an initial thickness of zero and coincident node coordinates on the upper and lower surfaces. The traction-separation constitutive relation of the zero-thickness cohesive element is described using a bilinear damage model.

[0063] in, This refers to the cohesive traction force, measured in MPa. The relative displacement between the upper and lower surfaces of the cohesive unit is expressed in mm. The displacement at the onset of damage is expressed in mm. The displacement represents the point of complete failure, expressed in mm. The value represents the interface strength, expressed in MPa.

[0064] This includes all components within the modeling area after the zero-thickness cohesive unit is embedded, including all new aggregate geometry models, all RAP aggregate geometry models, and the epoxy asphalt mortar area. The zero-thickness cohesive units distributed throughout the domain are combined to form a complete numerical model, outputting a numerical model of a multiphase heterogeneous interface.

[0065] The S300 is used to calibrate the mechanical properties of the numerical model of the multiphase heterogeneous interface to obtain the basic numerical model; based on the basic numerical model, finite element damage and cracking simulation is performed to obtain the sample aggregate formula and sample mechanical properties.

[0066] In some embodiments, the process of obtaining the sample aggregate formulation and sample mechanical properties includes: Through cohesion and adhesion tests, the performance parameters of the zero-thickness cohesion units of the epoxy asphalt mortar matrix, the mortar-RAP coating layer interface, and the mortar-new aggregate interface in the multiphase heterogeneous interface numerical model were calibrated to obtain the basic numerical model. The sample aggregate formulation is generated based on the basic numerical model. The sample aggregate formulation includes gradation distribution parameters, aggregate morphology parameters, interfacial cohesion strength and adhesion strength. Finite element damage and cracking simulations were performed on the sample aggregate formulation under various working conditions to obtain the mechanical properties of the sample, including peak load, fracture energy, crack initiation location, crack propagation length, and damage variable evolution curves.

[0067] In some embodiments, the specific steps of the cohesion test are as follows: Epoxy-modified asphalt and recycled aged asphalt from the RAP surface are uniformly mixed at a mass ratio of 1:1 to simulate the binder in the interface fusion zone. Dog-bone shaped specimens are used, ensuring the fracture surface occurs in the middle of the specimen. Tensile force is applied at a constant tensile rate of 20 mm / min using a universal testing machine at three temperature levels: 15°C, 25°C, and 35°C, until the asphalt completely fractures at the fracture surface. Based on the force-displacement relationship curve obtained from the test, the cohesion strength of the epoxy asphalt mastic is calculated. and fracture energy Cohesive strength Determined by the ratio of peak tensile force to the cross-sectional area of ​​the specimen, in MPa; fracture energy. The value is determined by dividing the area enclosed by the force-displacement curve by the cross-sectional area of ​​the specimen, with units of J / m. 2 .

[0068] The specific steps of the adhesion test are as follows: Test the adhesive-RAP coating interface separately. and the interface between the mortar and the new aggregate Adhesion tests were conducted. Epoxy asphalt adhesive was coated onto the corresponding aggregate surface. The adhesive film thickness was controlled to be 0.5-2 mm by adjusting the thickness of the gaskets, enabling multi-scale characterization of the adhesion between asphalt and aggregate. Tensile forces were applied at a constant tensile rate of 6 mm / min using a universal testing machine at three temperature levels: 15℃, 25℃, and 35℃, until the asphalt completely detached from the aggregate surface. Based on the force-displacement relationship curves obtained from the tests, the interfacial adhesion strength of the adhesive-RAP coating layer was calculated. and fracture energy and the adhesion strength of the adhesive-new aggregate interface. and fracture energy The units are both MPa and J / m 2 .

[0069] The calibrated bulk cohesive properties of the epoxy asphalt mastic were determined. , The value is assigned to the epoxy asphalt mortar region in the multiphase heterogeneous interface numerical model. The internal zero-thickness adhesive unit; the calibrated adhesive-RAP coating layer interfacial adhesion performance parameters ( , The value is assigned at the interface between the adhesive and the RAP coating layer. Zero-thickness cohesive unit; the calibrated adhesive-new aggregate interface adhesion performance parameters ( , The value is assigned to the interface between the mortar and the new aggregate. Zero-thickness cohesive units; after assignment, the basic numerical model is obtained.

[0070] Adjustable variables are generated based on the basic numerical model. These adjustable variables include the following parameters: gradation distribution parameters, including the mass passing percentage for each particle size range. The sum of the mass passing percentages for each particle size range is 100%; aggregate morphology parameters include the roundness of the aggregates at each particle size range. Distribution range and angularity index Distribution range; interfacial cohesion strength includes the cohesion strength of the epoxy asphalt mastic matrix. Interfacial adhesion strength includes the adhesion strength at the interface between the adhesive and the RAP coating layer. Adhesion strength of the adhesive-new aggregate interface .

[0071] A space-filling design method is employed to generate multiple parameter combination schemes within the domain of the controllable variable. Specifically, the Latin hypercube sampling method is used to generate schemes covering the entire domain of the controllable variable. Group parameter combination scheme, Each set of parameter combinations corresponds to a set of adjustable variable values ​​as a sample aggregate formulation.

[0072] The operating condition combinations include multiple temperature levels and various load types. The temperature levels include five levels: -10℃, 15℃, 25℃, 35℃, and 60℃. The load types include three types: uniaxial compression, semi-circular bending, and indirect tension. Each temperature level and load type is combined in pairs to form a total of 15 operating conditions.

[0073] Damage and cracking simulations of aggregate formulations for each sample were performed under various working conditions using an explicit finite element method (FEM) solver. The mechanical properties of the samples were then extracted from the simulation results. These mechanical properties include peak load. fracture energy The evolution curves of crack initiation location, crack propagation length, and damage variables.

[0074] Peak load The load-displacement curve from the simulation results is directly read, with units in kN; fracture energy. The value is determined by dividing the area enclosed by the load-displacement curve by the characteristic area of ​​the specimen, with units of J / m². 2 The crack initiation location is the spatial coordinate position corresponding to when the damage variable reaches the damage initiation threshold, in mm; the crack propagation length is the total length of the crack from the initiation location to the termination location, in mm; the damage variable evolution curve is the curve of the damage variable changing with the loading step or loading time.

[0075] The S400 constructs an initial deep neural network surrogate model. Iterative optimization of the initial deep neural network surrogate model is performed using sample aggregate formulations and sample mechanical properties to obtain a final deep neural network surrogate model. The target mechanical properties are obtained, and based on the target mechanical properties and the final deep neural network surrogate model, iterative optimization of the aggregate formulation based on gradient backpropagation is performed to obtain the optimal aggregate formulation.

[0076] In some embodiments, the initial deep neural network proxy model includes a linear embedding layer, a multi-head self-attention network, a data-driven output network, and a physical information output network. The process of obtaining the final deep neural network proxy model includes: The sample aggregate formulation is mapped into a high-dimensional feature vector through a linear embedding layer; Multi-head self-attention networks use a scaled dot product attention mechanism to assign weights to high-dimensional feature vectors to obtain deep feature tensors. The deep feature tensor is used as input data to drive the output network to predict performance and obtain the predicted mechanical properties. The deep feature tensor is input into the physical information output network to extract physical data, thereby obtaining the damage variable field, stress field and strain energy release rate field. The data-driven loss function is calculated based on the predicted mechanical properties and the sample mechanical properties, and the physical constraint loss function is calculated based on the damage variable field, stress field, and strain energy release rate field. The data-driven loss function and the physical constraint loss function are weighted and summed to obtain the composite loss function. During the warm-up training phase, the weights of the physical constraint loss function are set to 0, and the initial deep neural network surrogate model is iteratively optimized using only the data-driven loss function. After running a preset number of training rounds, the first-stage network is obtained. During the joint training phase, the weights of the physical constraint loss function are gradually increased from 0 to a preset target value. The composite loss function is then used to iteratively optimize the first-stage network until the composite loss function stops decreasing in multiple iterations, thus obtaining the final deep neural network surrogate model.

[0077] In some embodiments, the linear embedding layer is a single-layer fully connected layer used for formulating sample aggregates. Mapped to high-dimensional feature vectors Let the input sample aggregate formula be... ,in The dimension of the input feature corresponds to the total number of gradation distribution parameters, aggregate morphology parameters, interfacial cohesion strength, and adhesion strength.

[0078] Multi-head self-attention networks utilize a scaled dot product attention mechanism to process high-dimensional feature vectors. Perform weight assignment to obtain the depth feature tensor Self-attention mechanisms adaptively focus on features that significantly impact output performance by calculating correlation weights between input features. Multi-head self-attention networks are stacked. layer, The value ranges from 4 to 12. Each layer contains a multi-head self-attention sub-layer and a feedforward network sub-layer. Residual connections and layer normalization are used between layers.

[0079] The data-driven output network consists of several fully connected layers stacked together, each followed by a batch normalization layer, a ReLU activation function, and a Dropout layer. The data-driven output network is used to process the depth feature tensor... Perform performance prediction to obtain predicted mechanical properties Predicting mechanical properties Dimensions and mechanical properties of the sample The dimensions are consistent, both including peak load. fracture energy The location of crack initiation, crack propagation length, and the endpoint value of the damage variable evolution curve.

[0080] The physical information output network is also composed of stacked fully connected layers, but it does not contain dropout layers to maintain the spatial continuity and determinism of the physical field prediction. The physical information output network is used to calculate the depth feature tensor. Physical data extraction was performed to obtain the damage variable field. Stress field and strain energy release rate field .

[0081] Data-driven loss function Mean squared error loss is used: in, The number of samples in the training set; For the first Predicted mechanical properties of individual samples; For the first Mechanical properties of each sample; It is the square of the L2 norm.

[0082] Physical constraint loss function Includes three residuals: The first term is the residual of the damage evolution consistency constraint. Constraining damage variables The evolution of the loading step should satisfy the damage mechanics framework of the continuum: in, This represents the total number of time steps. For the first Damage variables at each time step; For the first The stress tensor at each time step, in MPa; For the first Spatial coordinates of each time step, in mm; This is a damage evolution function constructed based on the viscoelastic damage constitutive equation of epoxy recycled asphalt mixture.

[0083] The second term is the residual due to the energy conservation constraint. The relationship between the confined fracture energy release rate and the crack driving force: in, For the first The strain energy release rate at each time step, in J / m²; For the first Displacement at each time step, in mm; For the first The stress variation function with displacement at each time step, in MPa.

[0084] The third term is the constitutive relation constraint residual. The constrained stress-strain response should satisfy the viscoelastic constitutive equation of the epoxy asphalt material at the corresponding temperature: in, For the first The strain tensor at each time step; For the first Temperature at each time step, in °C; For the first The loading time for each time step, in seconds; It is a viscoelastic constitutive operator built based on the generalized Maxwell model.

[0085] Physical constraint loss function The weighted sum of the three residuals mentioned above: in, , , These are the weighting coefficients for the residuals of each physical constraint. The value ranges from 0.1 to 1.0. The value ranges from 0.1 to 1.0. The value ranges from 0.1 to 1.0.

[0086] Data-driven loss function and physical constraint loss function Weighted summation yields the composite loss function. : in, The weight coefficient of the physical constraint loss function is 0 during the warm-up training phase and gradually increases from 0 to the preset target value during the joint training phase. The preset target value ranges from 0.01 to 0.5.

[0087] During the warm-up training phase, the number of pre-set training epochs accounts for 10%-20% of the total training epochs, with a total of 500-2000 epochs. The optimizer used is Adam, and the learning rate is dynamically adjusted using a cosine annealing strategy, with an initial learning rate of 10. -4 -10 -3 The batch size ranges from 32 to 256.

[0088] During the joint training phase, the composite loss function value on the validation set is monitored in real time. When the composite loss function value on the validation set no longer decreases in 20-50 consecutive training rounds, early stopping is triggered, and the network parameters at the lowest composite loss function value on the validation set are saved as the parameters of the final deep neural network surrogate model.

[0089] In some embodiments, the process of obtaining the optimal aggregate formulation includes: Based on the final deep neural network surrogate model, a differentiable forward mapping function is constructed. Driven by the target mechanical properties and with the aggregate formulation as the optimization object, the inverse design loss function is constructed through the differentiable forward mapping function. The gradient of the inverse design loss function with respect to the aggregate formulation is calculated using the chain rule. The aggregate formulation is then iteratively updated along the descent direction of the gradient using a gradient descent optimizer until the inverse design loss function converges to the preset tolerance, at which point the iteration terminates and the optimal aggregate formulation is obtained.

[0090] In some embodiments, all network parameters of the final deep neural network proxy model are frozen. The final deep neural network surrogate model is viewed as a differentiable positive mapping function. .in, For aggregate formulation variables, These are predicted values ​​for mechanical properties. For the dimensions of aggregate formulation variables, This represents the dimension of the predicted mechanical properties.

[0091] Target mechanical properties Driven by aggregate formulation To optimize the object, a differentiable forward mapping function is used. Constructing a reverse design loss function : in, The preset target mechanical performance index vector includes target peak load, target fracture energy, etc. These are engineering design constraints. The constraint weight coefficient has a value ranging from 0.1 to 1.0.

[0092] Engineering design constraints Including compliance constraints of gradation and particle size range constraints Grading compliance constraints Defined as the sum of the squared penalties for the differences between the pass rate of each particle size range and the upper and lower limits of the specification, ensuring that the pass rate of each range does not exceed the allowable range of the specification. Particle size range constraint. Defined as the sum of the squared penalty terms of the boundary constraints of the maximum and minimum aggregate sizes, ensuring that the aggregate size does not exceed the preset allowable boundary value.

[0093] Calculate the reverse design loss function using the chain rule. For aggregate formulation gradient The expression is: in, For differentiable positive mapping functions for aggregate formulation The Jacobian matrix is ​​calculated using automatic differentiation techniques; This represents the gradient of the mechanical property prediction deviation with respect to the predicted mechanical property value. These are the constraint weighting coefficients; This indicates the engineering design constraints on aggregate formulation. The gradient.

[0094] The aggregate formulation is updated iteratively along the descent direction of the gradient using a gradient descent optimizer. : in, For the first The aggregate formulation variables for the next iteration; For the first The aggregate formulation variables for the next iteration; The learning rate is designed for reverse engineering and is set to a value of 0.01-0.1.

[0095] During the iteration process, the aggregate formula will be updated after each update. Each component is projected onto a preset feasible region, i.e. ,in This represents the pre-defined feasible region for aggregate formulations. Towards the feasible region The projection operator.

[0096] Set multiple sets of different initial values Parallel optimization is employed, with the number of iterations ranging from 10 to 50, to avoid getting trapped in local optima. Iteration terminates when the reverse design loss function converges to a preset tolerance or reaches the maximum number of iterations. The preset tolerance is 10. -6 -10 -4 The maximum number of iterations is between 100 and 500, and the optimal aggregate formula for epoxy recycled asphalt mixture that meets the target mechanical properties is output. .

[0097] Reference Figure 2 This invention provides an aggregate formulation optimization device 20 based on multiphase heterogeneous interface modeling, used to implement an aggregate formulation optimization method based on multiphase heterogeneous interface modeling. The device includes: The new aggregate geometric model acquisition module 21 is used to perform particle geometric morphology analysis on the sample aggregate, obtain the statistical distribution characteristics of aggregate morphology parameters, acquire the parameterized generation framework, and construct a new aggregate geometric model based on the statistical distribution characteristics of aggregate morphology parameters and the parameterized generation framework. The multiphase heterogeneous interface numerical model acquisition module 22 is used to perform aging simulation based on the new aggregate geometric model to obtain the RAP aggregate geometric model; obtain the aggregate gradation; perform overlap and collision simulation based on the new aggregate geometric model, the RAP aggregate geometric model, and the aggregate gradation to obtain the mixed distribution geometric model of RAP aggregate and new aggregate; and perform epoxy regeneration simulation based on the mixed distribution geometric model of RAP aggregate and new aggregate to obtain the multiphase heterogeneous interface numerical model. The training sample acquisition module 23 is used to calibrate the mechanical properties of the multiphase heterogeneous interface numerical model to obtain the basic numerical model; and to perform finite element damage and cracking simulation based on the basic numerical model to obtain the sample aggregate formula and sample mechanical properties. The optimal aggregate formulation acquisition module 24 is used to construct an initial deep neural network surrogate model, and to iteratively optimize the initial deep neural network surrogate model by using sample aggregate formulations and sample mechanical properties to obtain a final deep neural network surrogate model; to obtain target mechanical properties, and to iteratively optimize the aggregate formulation based on gradient backpropagation according to the target mechanical properties and the final deep neural network surrogate model to obtain the optimal aggregate formulation.

[0098] This application provides an electronic device, including a processor and a memory; the memory stores a computer program, wherein the computer program, when executed by the processor, implements the aggregate formulation optimization method based on multiphase heterogeneous interface modeling, as described above.

[0099] Specifically, the processor may include, for example, a general-purpose microprocessor, an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor may also include onboard memory for caching purposes. The processor may be a single processing unit or multiple processing units for performing different actions of the method flow according to embodiments of this application.

[0100] Memory can be any medium capable of containing, storing, transmitting, propagating, or transmitting instructions. For example, memory can include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, instruments, or propagation media. Specific examples of memory include: magnetic storage devices such as magnetic tape or hard disk drives (HDDs); optical storage devices such as optical discs (CD-ROMs); and also random access memory (RAM) or flash memory; and / or wired / wireless communication links.

[0101] This application also provides a computer-readable medium storing a computer program thereon, which, when executed by a processor, implements the aggregate formulation optimization method based on multiphase heterogeneous interface modeling, as described above. This computer-readable medium may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into that device / apparatus / system. The aforementioned computer-readable medium carries one or more programs, which, when executed, implement the method as described in the embodiments of this application.

[0102] According to embodiments of this application, a computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wired, optical fiber, radio frequency signals, etc., or any suitable combination thereof.

[0103] Those skilled in the art will understand that the features described in the various embodiments and / or claims of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments and / or claims of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application. Therefore, the scope of this application should not be limited to the above embodiments, but should be defined not only by the appended claims, but also by their equivalents. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for optimizing aggregate formulation based on multiphase heterogeneous interface modeling, characterized in that, include: The particle geometry of the sample aggregate was analyzed to obtain the statistical distribution characteristics of the aggregate morphology parameters. Obtain the parameterized generation framework, and construct a new aggregate geometric model based on the statistical distribution characteristics of aggregate morphology parameters and the parameterized generation framework; Aging simulation was performed based on the new aggregate geometric model to obtain the RAP aggregate geometric model; aggregate gradation was obtained, and overlap and collision simulation was performed based on the new aggregate geometric model, the RAP aggregate geometric model, and the aggregate gradation to obtain the mixed distribution geometric model of RAP aggregate and new aggregate; epoxy regeneration simulation was performed based on the mixed distribution geometric model of RAP aggregate and new aggregate to obtain the multiphase heterogeneous interface numerical model. Mechanical property calibration was performed on the numerical model of the multiphase heterogeneous interface to obtain the basic numerical model; Finite element damage and cracking simulation was performed based on the basic numerical model to obtain the sample aggregate formula and sample mechanical properties. An initial deep neural network surrogate model is constructed. The initial deep neural network surrogate model is iteratively optimized by the sample aggregate formulation and sample mechanical properties to obtain the final deep neural network surrogate model. The target mechanical properties are obtained. Based on the target mechanical properties and the final deep neural network surrogate model, the aggregate formulation is iteratively optimized by gradient backpropagation to obtain the optimal aggregate formulation.

2. The aggregate formulation optimization method based on multiphase heterogeneous interface modeling according to claim 1, characterized in that, The process of obtaining the RAP aggregate geometry model includes: The new aggregate geometric model is used as the geometric profile of the core aggregate to obtain the particle size range and preset basic performance parameters of the sample aggregate. The preset basic performance parameters include RAP total asphalt content, ineffective asphalt content, specific surface area, old asphalt density and service life. The effective asphalt film coating thickness under different particle size ranges is calculated based on the preset basic performance parameters. Using the equidistant offset algorithm, the outer normal direction of the geometric contour of the core aggregate is extended outward according to the effective asphalt film coating thickness to generate a multi-layered concentric layered contour. Using the geometric contour of the core aggregate and the concentric layered contour as boundaries, multiple initial annular regions arranged from the inside out are defined; the initial annular regions are assigned aging degree attributes that decrease from high to low according to the spatial order from the inside out, thus obtaining annular regions with aging degree attributes. By combining the annular region of aging degree attribute with the geometric contour of the core aggregate, the RAP aggregate geometric model is obtained.

3. The aggregate formulation optimization method based on multiphase heterogeneous interface modeling according to claim 1, characterized in that, The process of obtaining the geometric model of the mixed distribution of RAP aggregates and new aggregates includes: The new aggregate geometry model and the RAP aggregate geometry model together constitute the aggregate to be fed, and the aggregate gradation is converted into the target probability of each particle size range. The quantity of aggregates is calculated based on the target probability. The modeling area is divided into multiple sub-areas of equal size based on the quantity of aggregates. A placement position is randomly generated in each sub-area. The aggregates to be placed are placed one by one into the placement position corresponding to the current sub-area to obtain the currently placed aggregates. If the minimum distance between the currently deployed aggregate and any successfully deployed aggregate is less than the preset distance constraint or exceeds the area boundary, then the deployment position of the aggregate to be deployed in the current sub-area will be regenerated; otherwise, the currently deployed aggregate will be regarded as the new successfully deployed aggregate. The aggregates to be added are sorted from largest to smallest according to their maximum particle size, and then added one by one to the addition position until all the aggregates to be added are successfully added or the preset maximum number of additions is reached, thus obtaining a geometric model of the mixed distribution of RAP aggregates and new aggregates.

4. The aggregate formulation optimization method based on multiphase heterogeneous interface modeling according to claim 1, characterized in that, The process of obtaining the numerical model of a multiphase heterogeneous interface includes: Boolean operations are used to extract the blank areas outside the RAP aggregate geometric model and the new aggregate geometric model within the mixed distribution geometric model of RAP aggregate and new aggregate, and the blank areas are defined as epoxy asphalt mortar areas. Based on the spatial relationship of the components, the internal whole of the epoxy asphalt mastic body, the mastic-RAP coating layer interface, and the mastic-new aggregate interface are determined in the epoxy asphalt mastic area. Zero-thickness cohesive units were embedded into the entire internal domain of the epoxy asphalt mastic body, the mastic-RAP coating layer interface, and the mastic-new aggregate interface to obtain a multiphase heterogeneous interface numerical model.

5. The aggregate formulation optimization method based on multiphase heterogeneous interface modeling according to claim 1, characterized in that, The process of obtaining the sample aggregate formulation and sample mechanical properties includes: Through cohesion and adhesion tests, the performance parameters of the zero-thickness cohesion units of the epoxy asphalt mortar matrix, the mortar-RAP coating layer interface, and the mortar-new aggregate interface in the multiphase heterogeneous interface numerical model were calibrated to obtain the basic numerical model. The sample aggregate formulation is generated based on the basic numerical model. The sample aggregate formulation includes gradation distribution parameters, aggregate morphology parameters, interfacial cohesion strength and adhesion strength. Finite element damage and cracking simulations were performed on the sample aggregate formulation under various working conditions to obtain the mechanical properties of the sample, including peak load, fracture energy, crack initiation location, crack propagation length, and damage variable evolution curves.

6. The aggregate formulation optimization method based on multiphase heterogeneous interface modeling according to claim 1, characterized in that, The initial deep neural network surrogate model includes a linear embedding layer, a multi-head self-attention network, a data-driven output network, and a physical information output network. The process of obtaining the final deep neural network surrogate model includes: The sample aggregate formulation is mapped into a high-dimensional feature vector through a linear embedding layer; Multi-head self-attention networks use a scaled dot product attention mechanism to assign weights to high-dimensional feature vectors to obtain deep feature tensors. The deep feature tensor is used as input data to drive the output network to predict performance and obtain the predicted mechanical properties. The deep feature tensor is input into the physical information output network to extract physical data, thereby obtaining the damage variable field, stress field and strain energy release rate field. The data-driven loss function is calculated based on the predicted mechanical properties and the sample mechanical properties, and the physical constraint loss function is calculated based on the damage variable field, stress field, and strain energy release rate field. The data-driven loss function and the physical constraint loss function are weighted and summed to obtain the composite loss function. During the warm-up training phase, the weights of the physical constraint loss function are set to 0, and the initial deep neural network surrogate model is iteratively optimized using only the data-driven loss function. After running a preset number of training rounds, the first-stage network is obtained. During the joint training phase, the weights of the physical constraint loss function are gradually increased from 0 to a preset target value. The composite loss function is then used to iteratively optimize the first-stage network until the composite loss function stops decreasing in multiple iterations, thus obtaining the final deep neural network surrogate model.

7. The aggregate formulation optimization method based on multiphase heterogeneous interface modeling according to claim 1, characterized in that, The process of obtaining the optimal aggregate formulation includes: Based on the final deep neural network surrogate model, a differentiable forward mapping function is constructed. Driven by the target mechanical properties and with the aggregate formulation as the optimization object, the inverse design loss function is constructed through the differentiable forward mapping function. The gradient of the inverse design loss function with respect to the aggregate formulation is calculated using the chain rule. The aggregate formulation is then iteratively updated along the descent direction of the gradient using a gradient descent optimizer until the inverse design loss function converges to the preset tolerance, at which point the iteration terminates and the optimal aggregate formulation is obtained.

8. An aggregate formulation optimization device based on multiphase heterogeneous interface modeling, used to implement the aggregate formulation optimization method based on multiphase heterogeneous interface modeling as described in any one of claims 1 to 7, characterized in that, The device includes: The new aggregate geometric model acquisition module is used to perform particle geometry morphology analysis on sample aggregates, obtain statistical distribution characteristics of aggregate morphology parameters, acquire a parameterized generation framework, and construct a new aggregate geometric model based on the statistical distribution characteristics of aggregate morphology parameters and the parameterized generation framework. The multiphase heterogeneous interface numerical model acquisition module is used to perform aging simulation based on the new aggregate geometric model to obtain the RAP aggregate geometric model; obtain the aggregate gradation; and perform overlap and collision simulation based on the new aggregate geometric model, the RAP aggregate geometric model, and the aggregate gradation to obtain the mixed distribution geometric model of RAP aggregate and new aggregate; and perform epoxy regeneration simulation based on the mixed distribution geometric model of RAP aggregate and new aggregate to obtain the multiphase heterogeneous interface numerical model. The training sample acquisition module is used to calibrate the mechanical properties of the multiphase heterogeneous interface numerical model to obtain the basic numerical model; based on the basic numerical model, finite element damage and cracking simulation is performed to obtain the sample aggregate formula and sample mechanical properties. The optimal aggregate formulation acquisition module is used to construct an initial deep neural network surrogate model. The initial deep neural network surrogate model is iteratively optimized by using sample aggregate formulations and sample mechanical properties to obtain a final deep neural network surrogate model. The target mechanical properties are obtained, and the aggregate formulation is iteratively optimized based on gradient backpropagation according to the target mechanical properties and the final deep neural network surrogate model to obtain the optimal aggregate formulation.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the aggregate formulation optimization method based on multiphase heterogeneous interface modeling as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the aggregate formulation optimization method based on multiphase heterogeneous interface modeling as described in any one of claims 1 to 7.