A neural network-based bio-asphalt aging risk assessment method

Abstract

The application discloses a biological asphalt aging risk assessment method based on a neural network, relates to the field of road material testing technology, and comprises the following steps: acquiring asphalt gene index sets of to-be-tested biological asphalts and rheological performance parameters at different aging stages; constructing a multi-scale fusion neural network model and outputting rheological performance prediction values fused with multi-scale gene information; training the multi-scale fusion neural network model by using the asphalt gene index sets and the rheological performance parameters, and screening out key regulation target points whose influence on aging rheological performance exceeds a preset threshold; performing targeted performance verification on the biological asphalts, and outputting aging risk assessment results. The application effectively solves the problems of difficult heterogenous data fusion and low credibility of 'black box' prediction; improves the anti-aging target point screening from experience trial and error to rational navigation, and outputs a visual target point regulation priority navigation diagram; and the accuracy, timeliness and engineering practicability of the assessment are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of road material testing technology, and in particular to a method for assessing the aging risk of bio-asphalt based on neural networks. Background Technology

[0002] With increasing global emphasis on sustainable development and environmental protection, bio-asphalt, as a green pavement material that can partially or completely replace traditional petroleum asphalt, has broad application prospects. However, during its service life, bio-asphalt is highly susceptible to aging due to the combined effects of multiple climatic factors such as ultraviolet radiation, heat and oxygen, and moisture, leading to deterioration of its rheological properties and a significant shortening of pavement life. Accurately assessing and improving the anti-aging performance of bio-asphalt is a core issue that urgently needs to be addressed to promote its engineering application. Current methods for assessing the aging of asphalt materials mainly focus on the detection of macroscopic rheological performance indicators, such as measuring the complex shear modulus and phase angle using a dynamic shear rheometer (DSR) or measuring... While creep stiffness and creep rate directly reflect the macroscopic performance degradation of materials, their understanding of aging mechanisms has significant limitations. Specifically: First, macroscopic indicators are the combined result of various microscopic and molecular-level changes, failing to reveal the root causes of aging and thus hindering targeted material modification. Second, when significant, detectable changes in macroscopic rheological properties occur, irreversible deep damage to the material's internal microstructure has often already occurred, exhibiting a clear "detection lag." To overcome the limitations of macroscopic assessment, researchers have begun to focus on microscopic-level analysis. However, most of these studies remain at the level of qualitative or semi-quantitative isolated indicator analysis, failing to establish a comprehensive analysis model encompassing "molecules / functional groups → micromorphology →..." On the one hand, cross-scale, quantitative dynamic correlation models between "macro-rheology" and "macro-rheology" provide a means to explore the molecular association behavior of asphaltene. However, existing technologies usually treat it as an independent explanatory tool and fail to organically integrate it with actual micro-morphology and macro-rheological test data. This disconnect between testing and simulation, and between macro and micro, results in a lack of systematicness and precision in the evaluation of the anti-aging performance of bio-asphalt. In particular, it is impossible to accurately screen out the "key regulatory targets" that play a decisive role in anti-aging performance from the complex coupling of multiple factors.

[0003] However, current common solutions have many drawbacks, including: existing technologies have single and lagging evaluation indicators, relying heavily on macroscopic rheological indicators and failing to reveal the root causes of aging at the microscopic level; at the same time, multi-scale test data such as microscopic morphology, functional group evolution and molecular association structure are fragmented, lacking a cross-scale quantitative correlation model of "molecule → micro → macro"; existing neural network models have simple structures and cannot effectively integrate heterogeneous multi-scale features such as images, spectra, and discrete parameters, and the model training lacks physical constraints, resulting in insufficient reliability and extrapolation ability of prediction results; the screening of anti-aging targets has long relied on empirical trial and error, lacking systematic sensitivity quantification methods, and failing to form a complete closed loop from microscopic indicator detection to targeted performance verification, resulting in poor evaluation accuracy and blind modification guidance. Summary of the Invention

[0004] The purpose of this section is to outline some aspects of embodiments of the present invention and to briefly describe some preferred embodiments. Simplifications or omissions may be made in this section, as well as in the abstract and title of this application, to avoid obscuring the purpose of these documents; however, such simplifications or omissions should not be construed as limiting the scope of the invention.

[0005] In view of the problems existing in the current neural network-based bioasphalt aging risk assessment method, this invention is proposed.

[0006] Therefore, the purpose of this invention is to provide a neural network-based method for assessing the aging risk of bio-asphalt. This method addresses the shortcomings of existing technologies, such as the reliance on single and outdated assessment indicators, dependence on macroscopic rheological indicators that fail to reveal the root causes of aging at the microscopic level, fragmented multi-scale test data (microscopic morphology, functional group evolution, and molecular association structure), lack of a cross-scale quantitative correlation model ("molecular → microscopic → macroscopic"), simple existing neural network models that cannot effectively integrate heterogeneous multi-scale features such as images, spectra, and discrete parameters, and the lack of physical constraints in model training, resulting in insufficient reliability and extrapolation ability of prediction results. Furthermore, the long-term reliance on empirical trial and error in anti-aging target screening, the lack of systematic sensitivity quantification methods, and the failure to form a complete closed loop from microscopic indicator detection to targeted performance verification have led to poor assessment accuracy and blind modification guidance.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, embodiments of the present invention provide a method for assessing the aging risk of bio-asphalt based on neural networks, comprising: acquiring a set of asphalt gene indicators and rheological performance parameters of the bio-asphalt to be tested at different aging stages; constructing a multi-scale fusion neural network model and outputting rheological performance prediction values ​​that integrate multi-scale gene information; training the multi-scale fusion neural network model using the asphalt gene indicator set and rheological performance parameters, and performing sensitivity analysis based on the trained multi-scale fusion neural network model to screen out key regulatory targets that affect the aging rheological performance beyond a preset threshold; and verifying the targeted performance of the bio-asphalt based on the screened key regulatory targets and outputting the aging risk assessment results.

[0008] As a preferred embodiment of the neural network-based bioasphalt aging risk assessment method of the present invention, the asphalt gene index set includes microscopic morphology parameters, functional group parameters, and association structure parameters extracted from asphalt. The construction of the multi-scale fusion neural network model specifically includes: constructing a multi-scale fusion neural network model with parallel sub-networks, wherein the parallel sub-networks include a morphology feature extraction sub-network, a functional group feature extraction sub-network, and an association feature extraction sub-network; inputting the feature vectors output by each of the three sub-networks into a self-attention mechanism layer to dynamically learn the interaction weights between morphology features, functional group features, and association features; and fusing the output of the self-attention mechanism layer through a fully connected layer to obtain the rheological performance prediction value of the fused multi-scale gene information.

[0009] As a preferred embodiment of the neural network-based bio-asphalt aging risk assessment method of the present invention, the step of obtaining the asphalt gene index set of the bio-asphalt to be tested specifically includes: performing image segmentation on the "honeycomb-like" micro-morphological image of the asphalt and extracting individual honeycomb cells; performing morphological analysis on the individual honeycomb cells, calculating the equivalent round particle size of the honeycomb cell, the standard deviation of the local curvature of the honeycomb wall boundary as a measure of flatness, and the maximum vertical distance between the fitted planes of multiple honeycomb cell center points as a measure of height, as the micro-morphological parameters; performing Fourier transform on the micro-morphological image to obtain its amplitude spectrum to capture the periodic degradation characteristics of the honeycomb structure; inputting the amplitude spectrum into the morphological feature extraction subnetwork, and extracting the frequency domain morphological feature vector through a multilayer perceptron.

[0010] As a preferred embodiment of the neural network-based bio-asphalt aging risk assessment method of the present invention, the step of obtaining the asphalt gene index set of the bio-asphalt to be tested specifically includes: performing second derivative processing on the Fourier transform infrared spectrum of asphalt; performing Gaussian-Lorentz mixture function deconvolution on the characteristic absorption peaks of carbonyl and sulfoxide groups to separate the peak areas of free and associated functional groups; calculating the peak area ratio of free to associated functional groups as the functional group parameter; and inputting the spectral data after the second derivative processing into the functional group feature extraction subnetwork to extract local chemical feature vectors through a one-dimensional convolutional neural network.

[0011] As a preferred embodiment of the neural network-based bio-asphalt aging risk assessment method of the present invention, the step of obtaining the asphalt gene index set of the bio-asphalt to be tested specifically includes: constructing an asphaltene dimer model in a molecular simulation environment; performing tensile molecular dynamics simulation on the asphaltene dimer until the fracture limit state; capturing the centroid distance and the angle between the molecular principal axes as the association angle in the fracture limit state, calculating the maximum slope of the binding energy-displacement curve during the fracture process as the binding energy parameter, and using it as the association structure parameter; inputting the association structure parameter into the association feature extraction subnetwork, and extracting the structural feature vector through a fully connected network.

[0012] As a preferred embodiment of the neural network-based bio-asphalt aging risk assessment method of the present invention, the acquisition of rheological performance parameters at different aging stages specifically includes: applying sequential or simultaneous ultraviolet light aging, thermo-oxidative aging, and water aging to the bio-asphalt under test to constitute a full-climate coupled simulated aging; interrupting the simulated aging at different preset aging stages and removing the aged sample; obtaining the complex shear modulus and phase angle of the aged sample using a dynamic shear rheometer; and obtaining the creep stiffness and creep rate of the aged sample using a bending beam rheometer to obtain the rheological performance parameters.

[0013] As a preferred embodiment of the neural network-based bioasphalt aging risk assessment method of the present invention, the training of the multi-scale fusion neural network model specifically includes: training using a composite loss function, wherein the composite loss function is composed of a weighted sum of a rheological performance prediction error term and a physical consistency penalty function term; when the predicted complex shear modulus decreases with increasing aging or the phase angle decreases with increasing aging, the physical consistency penalty function is activated, and a residual sum of squares positively correlated with the violation amplitude is assigned; the sensitivity analysis based on the trained multi-scale fusion neural network model is used to screen out key regulatory targets that have an impact on aging rheological performance exceeding a preset threshold, specifically including: using a feature attribution method based on Shapley values ​​to quantify the marginal contribution of each asphalt gene index to aging rheological loss; selecting the top N indices by contribution as the key regulatory targets; generating a target regulation priority navigation map with target controllability and target regulation effect as coordinate dimensions, used to output the key regulatory targets.

[0014] Secondly, to further address the aforementioned technical problems, the present invention provides a neural network-based bio-asphalt aging risk assessment system, comprising: a data acquisition module for acquiring the asphalt gene index set and rheological performance parameters of the bio-asphalt to be tested at different aging stages; a model construction module for constructing a multi-scale fusion neural network model and outputting rheological performance prediction values ​​fused with multi-scale gene information; a model training module for training the multi-scale fusion neural network model using the asphalt gene index set and rheological performance parameters, and performing sensitivity analysis based on the trained multi-scale fusion neural network model to screen out key regulatory targets whose impact on aging rheological performance exceeds a preset threshold; and a performance verification module for performing targeted performance verification on the bio-asphalt based on the screened key regulatory targets and outputting aging risk assessment results.

[0015] Thirdly, embodiments of the present invention provide a computer device, including a memory and a processor, wherein the memory stores a computer program, and the computer program, when executed by the processor, implements any step of the neural network-based bioasphalt aging risk assessment method described in the first aspect of the present invention.

[0016] Fourthly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the neural network-based bioasphalt aging risk assessment method as described in the first aspect of the present invention.

[0017] The beneficial effects of this invention are as follows: By constructing a multi-scale "asphalt gene" index system covering the microscopic morphology, functional groups, and molecular association structure of asphalt, this invention fundamentally overcomes the shortcomings of existing technologies that rely on macroscopic rheological indicators, leading to detection lag and ambiguity in mechanism understanding. It employs parallel sub-networks and a self-attention mechanism to construct a multi-scale fusion neural network, and combines this with a physical consistency penalty function to constrain the model training, effectively solving the problems of difficult heterogeneous data fusion and low reliability of "black box" predictions. Furthermore, it introduces a feature attribution method based on Shapley values ​​to systematically quantify the marginal contribution of each gene index to aging rheological loss, elevating the screening of anti-aging targets from empirical trial and error to rational guidance, and outputting a visualized target regulation priority navigation map. Finally, it establishes a complete closed-loop process from microscopic index detection, multi-scale correlation analysis, key target screening to targeted performance verification, providing a full-chain systematic solution for the anti-aging modification of bio-asphalt, significantly improving the accuracy, timeliness, and engineering practicality of the evaluation. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a flowchart illustrating the implementation of the present invention in Example 1. Detailed Implementation

[0019] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0020] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0021] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0022] Example 1 Reference Figure 1This is the first embodiment of the present invention, which provides a method for assessing the aging risk of bio-asphalt based on neural networks, including the following steps: S1: Obtain the asphalt gene index set and rheological performance parameters of the bio-asphalt to be tested at different aging stages.

[0023] Preferably, the asphalt gene index set includes microscopic morphology parameters, functional group parameters, and association structure parameters extracted from asphalt.

[0024] Furthermore, the asphalt gene index set of the bio-asphalt to be tested is obtained, specifically including: Image segmentation was performed on the "honeycomb-like" micromorphological image of asphalt to extract individual honeycomb cells; Morphological analysis was performed on a single honeycomb cell unit. The equivalent round particle size of the honeycomb cell and the standard deviation of the local curvature of the honeycomb wall boundary were calculated as measures of flatness, and the maximum vertical distance between the fitted planes of multiple honeycomb cell center points was calculated as a measure of height, serving as micromorphological parameters. Fourier transform is performed on the microscopic morphology image to obtain its amplitude spectrum in order to capture the periodic degradation characteristics of the honeycomb structure; The amplitude spectrum is input into the topography feature extraction subnetwork, and the frequency domain topography feature vector is extracted by a multilayer perceptron.

[0025] Furthermore, the asphalt gene index set of the bio-asphalt to be tested is obtained, specifically including: The Fourier transform infrared spectrum of asphalt was processed by second derivative. The characteristic absorption peaks of carbonyl and sulfoxide groups were deconvolved using a Gaussian-Lorentz mixture function to separate the peak areas of free and associated functional groups. Calculate the peak area ratio of free state to associated state functional groups as functional group parameters; The spectral data processed by the second derivative is input into the functional group feature extraction subnetwork, and local chemical feature vectors are extracted by a one-dimensional convolutional neural network.

[0026] Specifically, the asphalt gene index set of the bio-asphalt to be tested is obtained, including: In a molecular simulation environment, a model of asphaltene dimer is constructed; Tensile molecular dynamics simulations were performed on asphaltene dimers until the fracture limit state was reached; In the fracture limiting state, the distance between the centroids and the angle between the molecular principal axes are used as the association angle, and the maximum slope of the binding energy-displacement curve during the fracture process is calculated as the binding energy parameter, which is used as the association structure parameter. The association structure parameters are input into the association feature extraction subnetwork, and the structure feature vector is extracted through a fully connected network.

[0027] Furthermore, rheological performance parameters at different aging stages are obtained, specifically including: The bio-asphalt to be tested is subjected to sequential or simultaneous ultraviolet light aging, thermo-oxidative aging and water aging to form a full-climate coupled aging simulation. The simulated aging process was interrupted at different preset aging stages, and the aged samples were removed. The complex shear modulus and phase angle of the aged samples were obtained using a dynamic shear rheometer. The creep stiffness and creep rate of the aged samples were obtained by using a bending beam rheometer to obtain rheological performance parameters.

[0028] Preferably, this step incorporates the quantitative geometric parameters of the microstructure of asphalt, the fine states of the free and associated states of functional groups, and the mechanical parameters of the intermolecular associated structures into the "asphalt gene index set." This fundamentally breaks through the limitations of existing technologies that rely solely on macroscopic rheological indicators, leading to detection lag and vague understanding of mechanisms. It enables early and accurate identification of the root causes of aging from three parallel dimensions: physical structure degradation, chemical composition evolution, and intermolecular force attenuation. This provides a systematic input feature with profound physical implications for subsequent modeling.

[0029] For example, taking a biological bitumen sample as an example, firstly, an environmental scanning electron microscope was used to obtain a "honeycomb-like" microscopic morphological image of its asphalt. After segmenting the image, individual honeycomb cells were extracted. Morphological analysis calculated that the equivalent spherical particle size of the honeycomb cell was 2.3 μm, the flatness characterized by the standard deviation of the local curvature of the honeycomb wall boundary was 0.15, and the height characterized by the maximum vertical distance between the fitted planes of multiple honeycomb cell center points was 0.8 μm. Simultaneously, Fourier transform infrared spectroscopy was performed on the asphalt of the sample. After second derivative processing of the obtained spectrum, the carbonyl group (approximately 1700 cm⁻¹) was analyzed. -1 ) and sulfoxide (approximately 1030 cm) -1 The characteristic absorption peaks of the sample were subjected to Gaussian-Lorentz deconvolution to separate the peak areas of the free and associated states. The calculated peak area ratio of the carbonyl free state to the associated state was 0.35, and the peak area ratio of the sulfoxide free state to the associated state was 0.28. Furthermore, an asphaltenes dimer model of this bio-asphalt was constructed in the Materials Studio molecular simulation environment, and tensile molecular dynamics simulations were performed up to the fracture limit state, capturing a centroid distance of 1.85 nm, a molecular principal axis angle of 32°, and a maximum slope of 0.78 nN / nm for the binding energy-displacement curve. For aging testing, the sample was simultaneously subjected to ultraviolet radiation (intensity 0.68 W / m²). 2The aging process involved thermo-oxidative aging at 60℃ and immersion in distilled water at 60℃, forming a full-climate coupled aging simulation. Samples were taken at 0h, 50h, 100h, and 200h, respectively. The complex shear modulus and phase angle at each stage were measured using a dynamic shear rheometer, and the creep stiffness and creep rate were measured using a bending beam rheometer, forming a complete set of asphalt gene indexes and rheological performance parameters.

[0030] S2: Construct a multi-scale fusion neural network model and output the rheological performance prediction value that integrates multi-scale gene information.

[0031] Preferably, a multi-scale fusion neural network model is constructed, specifically including: A multi-scale fusion neural network model with parallel sub-networks is constructed. The parallel sub-networks include a morphology feature extraction sub-network, a functional group feature extraction sub-network, and an association feature extraction sub-network. The feature vectors output by each of the three sub-networks are input into a self-attention mechanism layer to dynamically learn the interaction weights between morphological features, functional group features, and association features; The output of the self-attention mechanism layer is fused through a fully connected layer to obtain rheological performance prediction values ​​that incorporate multi-scale gene information.

[0032] Specifically, the shape feature extraction subnetwork is a pre-trained visual geometric group network (VGG). Its deep feature maps are Fourier transformed and then input into a multilayer perceptron to transform the local texture information in the image domain into global periodic degradation features in the frequency domain.

[0033] Furthermore, the functional group feature extraction subnetwork is a one-dimensional convolutional neural network (1D-CNN) containing two convolutional layers and one pooling layer, used to automatically extract local chemical features of functional group absorption peaks from second-order derivative spectral data; the association feature extraction subnetwork is a fully connected network containing two hidden layers, used to map discrete molecular simulation parameters into high-dimensional structural feature vectors.

[0034] Preferably, this step constructs a parallel sub-network architecture oriented towards the characteristics of heterogeneous data and introduces a self-attention mechanism to dynamically learn the interaction weights between microscopic morphology, functional groups, and associated structural features. This effectively solves the technical problem that existing simple neural networks cannot effectively fuse multi-scale heterogeneous features such as images, spectra, and discrete simulation parameters. This enables the model to autonomously capture the synergistic aging effect between microscopic structural degradation and chemical oxidation from the data, significantly improving the accuracy of rheological performance prediction and the ability to express complex nonlinear relationships.

[0035] For example, a multi-scale fusion neural network model is constructed within the PyTorch framework. The morphology feature extraction sub-network uses a pre-trained VGG-16 network. Its deep feature maps are Fourier transformed to extract the amplitude spectrum, which is then input into a three-layer multilayer perceptron, outputting a 128-dimensional frequency domain morphology feature vector. The functional group feature extraction sub-network uses a one-dimensional convolutional neural network containing two convolutional layers (kernel sizes of 3 and 5, respectively) and a max-pooling layer to extract a 64-dimensional local chemical feature vector from the second-derivative spectral data. The association feature extraction sub-network uses… Using a fully connected network with two hidden layers (32 and 16 neurons respectively), the three simulated parameters of centroid distance, association angle, and binding energy are mapped to 32-dimensional structural feature vectors. The 128-dimensional, 64-dimensional, and 32-dimensional feature vectors output by the above three sub-networks are concatenated and input into the self-attention mechanism layer. This layer dynamically learns the interaction weights between the three types of features by calculating the Query, Key, and Value matrices. Finally, the fully connected layer outputs four-dimensional rheological performance prediction values ​​of complex shear modulus, phase angle, creep stiffness, and creep rate at different aging stages.

[0036] S3: A multi-scale fusion neural network model is trained using asphalt gene index set and rheological performance parameters. Based on the trained multi-scale fusion neural network model, sensitivity analysis is performed to screen out key regulatory targets that have an impact on aging rheological performance exceeding a preset threshold.

[0037] Preferably, training a multi-scale fusion neural network model specifically includes: The training is performed using a composite loss function, which is a weighted sum of the rheological performance prediction error term and the physical consistency penalty function term. When the predicted complex shear modulus decreases with increasing aging or the phase angle decreases with increasing aging, the physical consistency penalty function is activated and assigned a residual sum of squares that is positively correlated with the magnitude of the violation. Sensitivity analysis was performed based on the trained multi-scale fusion neural network model to identify key regulatory targets whose impact on aging rheological properties exceeded a preset threshold. These targets included: The feature attribution method based on Shapley values ​​was used to quantify the marginal contribution of each asphalt gene index to aging rheological loss. The top N indicators in terms of contribution were selected as key regulatory targets. Generate a target control priority navigation map with target controllability and target control effect as coordinate dimensions, which is used to output key control targets.

[0038] Furthermore, the target regulation priority navigation map is a two-dimensional coordinate system. Its horizontal axis is the regulation effect score after normalization based on the marginal contribution of Shapley value, and the vertical axis is the controllability rating based on the pre-set accessibility of existing technical means. Among them, targets falling in the first quadrant are priority regulation targets, those in the second quadrant are reserve targets, those in the third quadrant are observation targets, and those in the fourth quadrant are low-cost and quick-win targets.

[0039] Furthermore, the Shapley value is calculated by taking the weighted average of each bitumen gene index across all possible combinations of indices, which reduces the model's prediction error.

[0040] Specifically, the preset threshold is the percentile of Shapley value contribution determined by statistical analysis of historical sample aging data, or it can be customized by the user according to the engineering accuracy requirements.

[0041] Furthermore, the method also includes a closed-loop iterative optimization step: after targeted modification based on key control targets, if the improvement in rheological properties of the modified sample does not reach the expected threshold, the modified test data is fed back to the multi-scale fusion neural network model for incremental learning, and sensitivity analysis and target screening are automatically re-executed to update the priority navigation map until the residual between the predicted improvement value and the measured improvement value converges within the preset range.

[0042] Preferably, this step, by embedding the physical consistency penalty function into the loss function to constrain model training, eliminates unreasonable predictive outputs that violate well-known physical laws such as "deeper aging, increased modulus, and decreased phase angle," transforming the model from an unexplainable "black box" into a "gray box" with physical common sense, significantly improving the credibility and extrapolation ability of the prediction results. At the same time, a feature attribution method based on Shapley values ​​is introduced to fairly quantify the marginal contribution of each gene index to aging rheological loss within a game theory framework, and to intuitively output a priority navigation map that integrates regulatory effects and technical accessibility, elevating target selection from empirical trial and error to rational decision-making, solving the core problem of the difficulty in accurately locating key regulatory targets under the coupling effect of multiple factors.

[0043] For example, using the aforementioned 200 sets of sample data from different aging stages, the data were divided into training, validation, and test sets in an 8:1:1 ratio. The model was trained using the Adam optimizer and an initial learning rate of 0.001. The weight α of the rheological prediction mean square error term in the composite loss function was set to 1.0, and the weight β of the physical consistency penalty function term was set to 0.1. After training for 85 rounds, the model's rheological prediction mean square error on the validation set decreased to 0.0032, and the physical consistency penalty function value converged to near zero, indicating that the model possesses both high prediction accuracy and strict adherence to physical laws. After training, the Shapley value was used to... Using an attribution method, the marginal contribution of each gene index combination to the loss of complex shear modulus due to aging was calculated. The results showed that the Shapley value of association angle had the highest contribution (0.31), followed by the free / associated carbonyl ratio (0.26) and the cell height (0.18), with the three contributing more than 75% in total. Based on this, the above three indicators were identified as key regulatory targets, and a target regulation priority navigation map was generated. The association angle fell in the first quadrant (high efficacy, high accessibility), the free / associated carbonyl ratio fell in the first quadrant, and the cell height fell in the second quadrant (high efficacy, low accessibility).

[0044] S4: Based on the selected key regulatory targets, the targeted performance of bio-asphalt is verified, and the aging risk assessment results are output.

[0045] Preferably, based on the selected key regulatory targets, the targeted performance of bio-asphalt is verified, and aging risk assessment results are output, specifically including: Based on the target regulation priority navigation map, select one or more targets with the highest priority; For the selected target, the corresponding targeted modifier is matched from the preset modifier knowledge base to perform targeted modification of the bio-asphalt to be tested; Steps S1 and S2 were repeated on the modified bio-asphalt to obtain the predicted values ​​of the modified rheological properties. The rheological properties before and after modification were compared, and the performance improvement rate was calculated to verify the effectiveness of the key control targets.

[0046] Specifically, the output of the aging risk assessment results includes: The target parameters of the modified bioasphalt sample were compared with those of the unaged baseline sample, and the deviation of each key regulatory target was calculated. Input the deviation vectors of all key regulatory targets into a preset risk level classifier; The risk level classifier outputs a quantified aging risk level, which is divided into three levels: low risk, medium risk, and high risk, each corresponding to different ranges of expected service life.

[0047] Preferably, this step achieves a technical closed loop from "diagnostic analysis" to "targeted modification verification" based on the target priority navigation map. The effectiveness of the target is verified by quantitatively comparing the material properties before and after modification. Furthermore, the modification data is fed back to the model for incremental learning through a closed-loop iterative mechanism, so that the model's risk assessment and target selection capabilities continue to evolve with the number of uses. Finally, the quantitative aging risk level and expected life range are output, providing a self-improving systematic solution for the anti-aging modification of bio-asphalt.

[0048] For example, based on the priority navigation map output by S3, association angles and free / associated carbonyl ratios falling in the first quadrant are preferentially selected as regulatory targets. For association angle targets, nano-silica (nano-SiO2) is selected from the preset modifier knowledge base as a targeted modifier, with an addition amount of 3% of the bio-asphalt mass. For free / associated carbonyl ratio targets, the hindered phenolic antioxidant Irganox is selected. 1010, with a dosage of 0.5%; modified samples were prepared by simultaneously incorporating two modifiers into the original bio-asphalt. The S1 and S2 processes were repeated to obtain the predicted rheological properties at each aging stage after modification. Comparing the data before and after modification after 200h aging, the aging increment of complex shear modulus decreased by 42%, and the aging decay of phase angle decreased by 38%. The calculated comprehensive performance improvement rate was 40.2%, verifying the effectiveness of the selected target. Finally, the key target parameters of the modified sample were compared with the parameters of the unaged reference sample, and the deviation vector was calculated to be [0.12, 0.08]. The input was given to the preset risk level classifier, and the output aging risk level was "low risk", corresponding to an expected service life range of 8-10 years.

[0049] In summary, this invention fundamentally overcomes the shortcomings of existing technologies that rely on macroscopic rheological indicators, leading to detection lag and ambiguous mechanism understanding, by constructing a multi-scale "asphalt gene" indicator system covering the microscopic morphology, functional groups, and molecular association structure of asphalt. It employs parallel sub-networks and a self-attention mechanism to construct a multi-scale fusion neural network, and combines this with a physical consistency penalty function to constrain the model training, effectively solving the problems of difficult heterogeneous data fusion and low reliability of "black box" predictions. Furthermore, it introduces a feature attribution method based on Shapley values ​​to systematically quantify the marginal contribution of each gene indicator to aging rheological loss, elevating the screening of anti-aging targets from empirical trial and error to rational guidance, and outputting a visualized target regulation priority navigation map. Finally, it establishes a complete closed-loop process from microscopic indicator detection, multi-scale correlation analysis, key target screening to targeted performance verification, providing a full-chain systematic solution for the anti-aging modification of bio-asphalt, significantly improving the accuracy, timeliness, and engineering practicality of the evaluation.

[0050] Example 2, an embodiment of the present invention, provides a neural network-based bio-asphalt aging risk assessment system, comprising: a data acquisition module for acquiring the asphalt gene index set and rheological performance parameters of the bio-asphalt to be tested at different aging stages; a model construction module for constructing a multi-scale fusion neural network model and outputting rheological performance prediction values ​​fused with multi-scale gene information; a model training module for training the multi-scale fusion neural network model using the asphalt gene index set and rheological performance parameters, and performing sensitivity analysis based on the trained multi-scale fusion neural network model to screen out key regulatory targets whose impact on aging rheological performance exceeds a preset threshold; and a performance verification module for performing targeted performance verification on the bio-asphalt based on the screened key regulatory targets and outputting aging risk assessment results.

[0051] Example 3 is an embodiment of the present invention, which differs from the previous embodiment in that: If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0052] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0053] More specific examples (a non-exhaustive list) of computer-readable media include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which programs can be printed, because programs can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0054] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0055] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for assessing the aging risk of bio-asphalt based on neural networks, characterized in that: include: Obtain the asphalt gene index set and rheological performance parameters of the bio-asphalt to be tested at different aging stages; Construct a multi-scale fusion neural network model and output rheological performance prediction values ​​that integrate multi-scale gene information; The multi-scale fusion neural network model was trained using the asphalt gene index set and rheological performance parameters. Sensitivity analysis was performed based on the trained multi-scale fusion neural network model to screen out key regulatory targets that have an impact on aging rheological performance exceeding a preset threshold. Based on the selected key regulatory targets, the targeted performance of bio-asphalt was verified, and the aging risk assessment results were output.

2. The method for assessing the aging risk of bio-asphalt based on neural networks as described in claim 1, characterized in that: The asphalt gene index set includes microscopic morphology parameters, functional group parameters, and association structure parameters extracted from asphalt. The construction of the multi-scale fusion neural network model specifically includes: A multi-scale fusion neural network model with parallel sub-networks is constructed, wherein the parallel sub-networks include a morphology feature extraction sub-network, a functional group feature extraction sub-network, and an association feature extraction sub-network; The feature vectors output by each of the three sub-networks are input into a self-attention mechanism layer to dynamically learn the interaction weights between morphological features, functional group features, and association features; The output of the self-attention mechanism layer is fused through a fully connected layer to obtain the rheological performance prediction value of the fused multi-scale gene information.

3. The method for assessing the aging risk of bio-asphalt based on neural networks as described in claim 2, characterized in that: The acquisition of the asphalt gene index set of the bio-asphalt to be tested specifically includes: Image segmentation was performed on the "honeycomb-like" micromorphological image of asphalt to extract individual honeycomb cells; Morphological analysis was performed on the individual honeycomb cell unit, and the equivalent round particle size of the honeycomb cell and the standard deviation of the local curvature of the honeycomb wall boundary were calculated as a measure of flatness, and the maximum vertical distance between the fitted planes of multiple honeycomb cell center points was calculated as a measure of height, which were used as the micromorphological parameters. The Fourier transform of the microscopic morphology image is performed to obtain its amplitude spectrum in order to capture the periodic degradation characteristics of the honeycomb structure. The amplitude spectrum is input into the shape feature extraction subnetwork, and the frequency domain shape feature vector is extracted by a multilayer perceptron.

4. The method for assessing the aging risk of bio-asphalt based on neural networks as described in claim 1, characterized in that: The acquisition of the asphalt gene index set of the bio-asphalt to be tested specifically includes: The Fourier transform infrared spectrum of asphalt was processed by second derivative. The characteristic absorption peaks of carbonyl and sulfoxide groups were deconvolved using a Gaussian-Lorentz mixture function to separate the peak areas of free and associated functional groups. Calculate the peak area ratio of the free state to the associated state functional groups, and use it as the parameter of the functional group; The spectral data processed by the second derivative is input into the functional group feature extraction subnetwork, and local chemical feature vectors are extracted through a one-dimensional convolutional neural network.

5. The method for assessing the aging risk of bio-asphalt based on neural networks as described in claim 1, characterized in that: The acquisition of the asphalt gene index set of the bio-asphalt to be tested specifically includes: In a molecular simulation environment, a model of asphaltene dimer is constructed; Tensile molecular dynamics simulations were performed on the asphaltene dimer until the fracture limiting state was reached; Under the fracture limiting state, the distance between the centroids and the angle between the molecular principal axes are used as the association angle, and the maximum slope of the binding energy-displacement curve during the fracture process is calculated as the binding energy parameter, which is used as the association structure parameter. The association structure parameters are input into the association feature extraction subnetwork, and the structure feature vector is extracted through a fully connected network.

6. The method for assessing the aging risk of bio-asphalt based on neural networks as described in claim 1, characterized in that: The acquisition of rheological performance parameters at different aging stages specifically includes: The bio-asphalt to be tested is subjected to sequential or simultaneous ultraviolet light aging, thermo-oxidative aging and water aging to form a full-climate coupled aging simulation. The simulated aging process is interrupted at different preset aging stages, and the aged samples are removed. The complex shear modulus and phase angle of the aged sample were obtained using a dynamic shear rheometer. The creep stiffness and creep rate of the aged specimen were obtained using a bending beam rheometer to obtain the rheological performance parameters.

7. The method for assessing the aging risk of bio-asphalt based on neural networks as described in claim 1, characterized in that: Training the multi-scale fusion neural network model specifically includes: Training is performed using a composite loss function, which is a weighted sum of a rheological performance prediction error term and a physical consistency penalty function term. When the predicted complex shear modulus decreases with increasing aging or the phase angle decreases with increasing aging, the physical consistency penalty function is activated, and a residual sum of squares that is positively correlated with the magnitude of the violation is assigned. The sensitivity analysis based on the trained multi-scale fusion neural network model identifies key regulatory targets whose impact on aging rheological properties exceeds a preset threshold, specifically including: The feature attribution method based on Shapley values ​​was used to quantify the marginal contribution of each asphalt gene index to aging rheological loss. The top N indicators in terms of contribution were selected as the key regulatory targets. Generate a target control priority navigation map with target controllability and target control effect as coordinate dimensions, which is used to output the key control targets.

8. A neural network-based bio-asphalt aging risk assessment system, based on the neural network-based bio-asphalt aging risk assessment method according to any one of claims 1 to 7, characterized in that: include, The data acquisition module is used to acquire the asphalt gene index set and rheological performance parameters of the bio-asphalt under test at different aging stages. The model building module is used to construct a multi-scale fusion neural network model and output rheological performance prediction values ​​that fuse multi-scale gene information. The model training module is used to train a multi-scale fusion neural network model using asphalt gene index set and rheological performance parameters, and to perform sensitivity analysis based on the trained multi-scale fusion neural network model to screen out key regulatory targets that have an impact on aging rheological performance exceeding a preset threshold. The performance verification module is used to perform targeted performance verification on bio-asphalt based on the selected key regulatory targets and output aging risk assessment results.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the neural network-based bio-asphalt aging risk assessment method according to any one of claims 1 to 7.

10. A 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 steps of the neural network-based bio-asphalt aging risk assessment method according to any one of claims 1 to 7.

Images