Intelligent design method and system for heat-conducting asphalt material based on large model
By employing a large-scale intelligent design method, combined with data-driven approaches and dedicated model training, the problems of long design cycles and blind parameter optimization for directional heat-conducting asphalt materials have been solved. This approach enables efficient material adaptation and performance consistency across different scenarios, thereby improving design efficiency and applicability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-19
- Publication Date
- 2026-03-27
AI Technical Summary
Existing directional thermal asphalt materials suffer from long design cycles, blind parameter optimization, and poor performance controllability, making it difficult to achieve accurate mapping and batch consistency, which restricts their engineering application efficiency and large-scale promotion.
A large-model-based intelligent design approach is adopted. Through data acquisition, preprocessing, model training, performance verification and iterative optimization, a dedicated model of the Transformer architecture is constructed. By combining supervised learning and reinforcement learning, accurate component ratios and process parameters are output to achieve the directional orientation and efficient mixing of materials.
It achieves precise matching of thermally conductive asphalt material formulation and process parameters, improves design efficiency, ensures the consistency and adaptability of material performance in different scenarios, and solves the problems of long cycle and poor controllability in traditional design.
Smart Images
Figure CN121744894A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of road engineering materials and functional asphalt materials, in particular to an intelligent design method and system for heat-conducting asphalt materials based on a large model. BACKGROUND
[0002] As a functional road engineering material with high-efficiency heat conduction performance, the directional heat-conducting asphalt material is widely used in road structure layers under high-temperature environments, airport pavements, bridge pavement, tunnel pavement, etc. Its core function is to quickly conduct and release the heat absorbed by the pavement, relieve the rutting, fatigue cracking, thermal aging and other diseases caused by high-temperature accumulation of traditional asphalt pavement, and ensure the long-term stability and service life of the road structure. It is one of the key materials to adapt to the high requirements of modern traffic engineering on pavement performance.
[0003] The existing design method needs to adjust the component content and process parameters through multiple repeated tests, which not only leads to a long design cycle and high research and development cost, but also lacks precise theoretical support and data basis for parameter optimization, making it difficult to establish a precise mapping relationship between components, processes and performance. Ultimately, the performance controllability of the design scheme is poor, and the performance consistency of different batches of products is insufficient, which seriously restricts the engineering application efficiency and large-scale promotion of the directional heat-conducting asphalt material. SUMMARY
[0004] In view of the deficiencies of the prior art, the present application provides an intelligent design method and system for heat-conducting asphalt materials based on a large model, which solves the problem of long design cycle, blind parameter optimization and poor performance controllability caused by relying on the experience of engineers for trial preparation in the prior art.
[0005] To achieve the above purpose, the following technical solutions are adopted: An intelligent design method for heat-conducting asphalt materials based on a large model, comprising the following steps:
[0006] S1, data acquisition and preprocessing, acquiring component data, process parameter data, performance data and application scenario data of the directional heat-conducting asphalt, and obtaining a training data set after preprocessing;
[0007] S2, large model training and optimization, based on the training data set of S1, constructing and training an intelligent design large model of the heat-conducting asphalt until the model prediction error meets the preset threshold;
[0008] S3, intelligent design parameter output, inputting the target performance index, outputting the initial design parameters through the large model trained in S2, the initial design parameters including component mass ratio and preparation process parameters;
[0009] S4, performance verification, preparing a sample according to the initial design parameters of S3, testing the performance and judging whether the target index is met;
[0010] S5, iterative optimization, if the judgment result of S4 is no, the verification data is fed back to S2 to update the model, and S3-S4 is repeated until the optimal design parameter is output.
[0011] Preferably, in S1, the component data includes the type and dosage range data of the matrix asphalt, MXene, mesophase pitch fiber, steel slag powder, mineral powder and dispersant aid, wherein the dosage range of MXene is 1-5 parts, and the dosage range of mesophase pitch fiber is 2-8 parts.
[0012] Preferably, in S1, the process parameter data includes the heating temperature of the matrix asphalt, the stirring speed, the stirring time, and the action parameter of the orientation device, wherein the heating temperature range is 160-180℃, and the stirring speed range is 4000-6000rpm.
[0013] Preferably, in S2, the large model is a special model based on the Transformer architecture, and the training process combines supervised learning and reinforcement learning. The reward function of the reinforcement learning is set based on the material thermal conductivity and the structure stability.
[0014] Preferably, in S3, the component mass ratio includes the following parts by weight: matrix asphalt 100 parts, steel slag powder 10-25 parts, mineral powder 5-15 parts, and dispersant aid 0.5-2 parts.
[0015] Preferably, in S4, the performance verification includes thermal conductivity test and structure stability test, the target thermal conductivity is ≥0.8W / m・K, and the structure stability meets no obvious segregation and crack phenomenon.
[0016] An intelligent design system for thermal asphalt material based on a large model, comprising:
[0017] A data acquisition and preprocessing module for acquiring and processing thermal asphalt component, process, performance and scene data, and outputting a training data set;
[0018] A large model training module connected to the data acquisition and preprocessing module, receiving the training data set and training the intelligent design large model;
[0019] An intelligent design module connected to the large model training module, receiving target performance indicators and outputting initial design parameters;
[0020] A performance verification module connected to the intelligent design module, testing sample performance and feeding back verification results;
[0021] A parameter optimization module connected to the performance verification module and the large model training module, driving model updating and parameter optimization.
[0022] Preferably, the data acquisition and preprocessing module comprises a data cleaning unit, a standardization unit and a feature extraction unit, and the feature extraction unit is used to extract the correlation features of the component content, process parameters and heat conduction performance.
[0023] Preferably, the large model training module comprises a model construction unit, a hyperparameter adjustment unit and an error evaluation unit, and the error evaluation unit calculates the prediction error through cross-validation, and the training is stopped when the error is ≤5%.
[0024] Preferably, the intelligent design module can receive user-defined constraint conditions, including cost upper limit and raw material availability, and output design parameter schemes that meet the constraints.
[0025] The application provides a large model-based intelligent design method and system for heat-conducting asphalt materials.
[0026] 1. The intelligent design technical solution of data-driven and special large model training is adopted to realize the accurate matching of the heat-conducting asphalt material formula and process parameters, and compared with the technical solution of relying on the experience of engineers for trial in the prior art, the design cycle is long, the parameter optimization is blind, and the performance controllability is poor.
[0027] 2. The cooperative process technical solution of directional orientation process, high shear dispersion and reasonable mixing sequence is adopted to construct a complete and efficient directional heat conduction network, and compared with the technical solution of random distribution of heat conduction path and lack of process cooperation in the prior art, the heat conduction efficiency is low, and the material structure is easy to separate and crack.
[0028] 3. The customized design technical solution of scene demand, constraint condition and performance index linkage is adopted to realize the flexible adaptation of different application scenes, and compared with the technical solution of single design scheme in the prior art, it is difficult to consider performance and cost, and the adaptation of different scenes such as low-load roads, highways and airport pavements is poor. BRIEF DESCRIPTION OF DRAWINGS
[0029] Figure 1 It is a flowchart of the large model-based intelligent design method for heat-conducting asphalt materials.
[0030] Figure 2 It is a module schematic diagram of the large model-based intelligent design method for heat-conducting asphalt materials. DETAILED DESCRIPTION
[0031] The technical solutions of the present application will be described clearly and completely below in combination with the drawings of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of the present application.
[0032] Please refer to the drawings of the present application Figure 1 The embodiment of the present application provides a kind of intelligent design method of heat conduction asphalt material based on large model, comprising the following steps:
[0033] S1, data acquisition and pretreatment, the component data of directional heat conduction asphalt, process parameter data, performance data and application scenario data are collected, and training data set is obtained after pretreatment;
[0034] Among them, the component data includes the type and mixing amount range data of matrix asphalt, MXene, mesophase pitch fiber, steel slag powder, mineral powder and dispersant aid, wherein the MXene mixing amount range is 1-5 parts, and the mesophase pitch fiber mixing amount range is 2-8 parts.
[0035] Process parameter data includes matrix asphalt heating temperature, stirring speed, stirring time and orientation device action parameter, wherein the heating temperature range is 160-180 DEG C, and the stirring speed range is 4000-6000 rpm;
[0036] S2, based on the training data set of S1, build and train heat conduction asphalt intelligent design large model, until the model prediction error meets the preset threshold;
[0037] Among them, the large model is a special model based on Transformer architecture, and the training process combines supervised learning and reinforcement learning, and the reward function of reinforcement learning is set based on material thermal conductivity and structure stability meeting condition;
[0038] S3, input target performance index, output initial design parameter through the large model trained in S2, and the initial design parameter includes component mass ratio and preparation process parameter;
[0039] Among them, the component mass ratio includes the following weight parts: matrix asphalt 100 parts, steel slag powder 10-25 parts, mineral powder 5-15 parts and dispersant aid 0.5-2 parts;
[0040] S4, according to the initial design parameter of S3, sample is prepared, performance is tested, and whether it meets the target index is judged;
[0041] Among them, performance verification includes thermal conductivity test and structure stability test, target thermal conductivity is greater than or equal to 0.8 W / m·K, and structure stability meets no obvious segregation and crack phenomenon;
[0042] S5, if the result of S4 is no, the verification data is fed back to S2 to update the model, and S3-S4 are repeated until the optimal design parameter is output.
[0043] Please refer to the attached Figure 2 A large model-based intelligent design system for heat-conducting asphalt materials, comprising:
[0044] A data acquisition and preprocessing module for acquiring and processing heat-conducting asphalt component, process, performance and scene data, and outputting a training data set;
[0045] A large model training module connected to the data acquisition and preprocessing module, receiving the training data set and training the intelligent design large model;
[0046] An intelligent design module connected to the large model training module, receiving target performance indicators and outputting initial design parameters;
[0047] A performance verification module connected to the intelligent design module, testing sample performance and feeding back verification results;
[0048] A parameter optimization module connected to the performance verification module and the large model training module, driving model updating and parameter optimization.
[0049] The data acquisition and preprocessing module includes a data cleaning unit, a standardization unit and a feature extraction unit, and the feature extraction unit is used to extract the correlation features of component content, process parameters and heat-conducting performance.
[0050] The large model training module includes a model construction unit, a hyperparameter adjustment unit and an error evaluation unit, and the error evaluation unit calculates the prediction error through cross-validation, and stops training when the error is ≤5%.
[0051] The intelligent design module can receive user-defined constraints, including cost upper limit, raw material availability, and output design parameter schemes that meet the constraints.
[0052] The following will be described in conjunction with specific embodiments:
[0053] Example 1
[0054] S1, data acquisition and preprocessing: select 70# road petroleum asphalt, MXene 1 part, mesophase pitch fiber 2 parts, steel slag powder 10 parts, mineral powder 5 parts, polyolefin amide dispersant 0.5 parts, process parameters are heating temperature 160℃, stirring speed 4000rpm, dispersion time 20min, magnetic field orientation strength 0.3T, performance data is the historical heat conductivity of the same formula 0.78-0.85W / m・K, application scene is suburban low load road, high temperature peak value is 37℃, annual traffic volume is ≤5000 vehicles / day, data is cleaned, standardized and low content and basic performance feature extraction to form a training data set;
[0055] S2, Large model training and optimization: based on the Transformer architecture, input the training data set, combine supervised learning and reinforcement learning, the reward function is set to the thermal conductivity ≥0.8 W / m·K and no segregation crack reward 1.0, the reward of not meeting the standard is-0.6, the hyperparameter adjustment is learning rate 0.001, batch_size 32, after 5-fold cross-validation, the model prediction error is 4.9%≤5%, stop training;
[0056] S3, Intelligent design parameter output: input target index thermal conductivity ≥0.8 W / m·K, no obvious segregation crack and constraint condition cost upper limit 800 yuan / ton, conventional raw materials, the model outputs the composition according to the above ratio, the process is heating 160℃, stirring 4000rpm, dispersing 20min, magnetic field 0.3T, powder mixing 15min;
[0057] S4, Performance verification: prepare samples according to the parameters, test the thermal conductivity 0.82 W / m·K, no segregation crack after 72h at room temperature, meet the scene demand;
[0058] S5, Iterative optimization: verification meets the standard, output the above parameters as the optimal scheme.
[0059] Example 2
[0060] S1, Data collection and preprocessing: select 90# road petroleum asphalt, MXene 3 parts, mesophase pitch fiber 5 parts, steel slag powder 18 parts, mineral powder 10 parts, non-ionic surfactant 1.2 parts, process parameters are heating temperature 170℃, stirring speed 5000rpm, dispersing time 25min, electric field orientation strength 1.5kV / cm, performance data is the thermal conductivity of historical similar formula 0.9-1.1 W / m·K, application scene is the main road of the highway, high temperature peak 43℃, annual traffic volume ≥30,000 vehicles / day, extract the medium mixing amount and balanced performance correlation characteristics to form a training data set;
[0061] S2, Large model training and optimization: use the same Transformer architecture, adjust the reward function to thermal conductivity ≥0.9 W / m·K and no defect reward 1.3, 0.8-0.9 W / m·K and no defect reward 0.7, not meeting the standard reward-0.8, hyperparameter optimization is learning rate 0.0008, batch_size 64, stop training when the model prediction error is 3.1%≤5%;
[0062] S3, intelligent design parameter output: input target index thermal conductivity ≥ 0.9 W / m·K, no segregation crack and constraint condition cost upper limit 1000 yuan / ton, raw material supply cycle ≤ 3 days, model output composition according to the above ratio, process for heating 170℃, stirring 5000rpm, dispersion 25min, electric field 1.5kV / cm, powder mixing 15min;
[0063] S4, performance verification: test thermal conductivity 1.03 W / m·K, no defect after 100h constant temperature cycle test 45℃ / 20℃ alternation, optimal balance of performance and cost is achieved;
[0064] S5, iterative optimization: verification up to standard, output the above parameters as the best solution, which adapts to most of the highway construction needs in China.
[0065] Example 3
[0066] S1, data acquisition and pretreatment: select SBS modified asphalt, MXene 5 parts, mesophase pitch fiber 8 parts, steel slag powder 25 parts, mineral powder 15 parts, and compound dispersing agent 2 parts, process parameters are heating temperature 180℃, stirring speed 6000rpm, dispersion time 30min, composite orientation magnetic field 0.6T and electric field 3kV / cm, performance data is historical high dosage formula thermal conductivity 1.1-1.3 W / m·K, application scene is airport pavement, high temperature peak 46℃, withstands airplane take-off and landing impact load, extracts high dosage and high thermal conductivity correlation characteristics, forms training data set;
[0067] S2, large model training and optimization: using Transformer architecture, reward function is set to reward 1.6 for thermal conductivity ≥ 1.1 W / m·K and no defect, and reward-1.0 for not meeting the standard, hyperparameter adjustment is learning rate 0.0006, batch_size 128, model prediction error 2.4%≤5% stops training;
[0068] S3, intelligent design parameter output: input target index thermal conductivity ≥ 1.1 W / m·K, no segregation crack and constraint condition cost upper limit 1500 yuan / ton, customizable procurement of raw materials, model output composition according to the above ratio, process for heating 180℃, stirring 6000rpm, dispersion 30min, composite orientation, powder mixing 15min;
[0069] S4, performance verification: test thermal conductivity 1.25 W / m·K, no defect after 150h high temperature aging test 50℃ constant temperature, optimal thermal conductivity performance;
[0070] S5, iterative optimization: verification up to standard, output the above parameters as the best solution, which adapts to high heat dissipation and high load demand scenarios.
[0071] Comparative Example 1
[0072] Different from example 2, the electric field orientation process was cancelled, and the rest of the process parameters were consistent with example 2.
[0073] Comparative example 2
[0074] Different from example 2, only supervised learning was used to train the model, and reinforcement learning was cancelled, and the rest of the process parameters were consistent with example 2.
[0075] Comparative example 3
[0076] Different from example 2, the ordinary stirring process was used instead of high shear dispersion, and the rest of the process parameters were consistent with example 2.
[0077] Comparative example 4
[0078] Different from example 2, the mixing order of powder and fiber was reversed, and the rest of the process parameters were consistent with example 2.
[0079] Table 1, performance test data table
[0080]
[0081] This method can accurately adapt to different scene needs: example 1 suburban low load road low dosage formula and basic process, thermal conductivity 0.82W / m・K, meet the low cost demand;
[0082] Example 2, the best solution of the main road of the highway, medium dosage and optimized process, thermal conductivity 1.03W / m・K, realize the optimal balance of performance and cost;
[0083] Example 3, airport pavement high dosage and high strength process, thermal conductivity 1.25W / m・K, adapt to high rigorous scene, three of them are once standard, design efficiency and stability outstanding.
[0084] The core process and model training method is the key to performance: cancel the electric field orientation, replace high shear dispersion with ordinary stirring, reverse the mixing order, or cancel reinforcement learning and only use supervised learning, all of which result in thermal conductivity not reaching the target 0.76-0.92W / m・K, and problems such as segregation and cracking, which confirms the necessity of directional orientation, high shear dispersion, reasonable mixing order, and the synergy of supervised and reinforcement joint training.
[0085] This intelligent design method solves the pain points of traditional design relying on experience and time-consuming optimization, and can flexibly adapt to multiple scenes, with high efficiency and practicality, providing a reliable intelligent solution for the engineering application of directional heat conduction asphalt
[0086] Table 2, detection standard and detection method
[0087]
[0088] While embodiments of the application have been shown and described, it is to be understood that the application is not limited to the details of the embodiments described, since numerous changes, modifications, substitutions and variations can be made thereto without departing from the spirit and scope of the application as defined by the appended claims and their equivalents.
Claims
1. A smart design method for thermally conductive asphalt materials based on a large model, characterized in that, Includes the following steps: S1. Data Acquisition and Preprocessing: Collect composition data, process parameter data, performance data, and application scenario data of directional heat-conducting asphalt, and obtain the training dataset through preprocessing. S2. Large model training and optimization: Based on the training dataset of S1, construct and train a large model for intelligent design of thermally conductive asphalt until the model prediction error meets the preset threshold. S3. Intelligent design parameter output: Input the target performance index, and output the initial design parameters of the large model trained by S2. The initial design parameters include the component mass ratio and preparation process parameters. S4. Performance verification: Prepare samples based on the initial design parameters in S3, test the performance, and determine whether the target indicators are met. S5. Iterative optimization: If the judgment result of S4 is negative, the verification data is fed back to S2 to update the model. S3-S4 are repeated until the optimal design parameters are output.
2. The intelligent design method for thermally conductive asphalt materials based on a large model according to claim 1, characterized in that: In S1, the composition data includes the types and dosage ranges of base asphalt, MXene, mesophase asphalt fiber, steel slag powder, mineral powder and dispersant, wherein the dosage range of MXene is 1-5 parts and the dosage range of mesophase asphalt fiber is 2-8 parts.
3. The intelligent design method for thermally conductive asphalt materials based on a large model according to claim 1, characterized in that: In S1, the process parameter data includes the base asphalt heating temperature, stirring speed, stirring time, and orientation device operation parameters, wherein the heating temperature range is 160-180℃, and the stirring speed range is 4000-6000rpm.
4. The intelligent design method for thermally conductive asphalt materials based on a large model according to claim 1, characterized in that: In S2, the large model is a dedicated model based on the Transformer architecture. The training process combines supervised learning and reinforcement learning. The reward function of reinforcement learning is set based on the material's thermal conductivity and the achievement of structural stability standards.
5. The intelligent design method for thermally conductive asphalt materials based on a large model according to claim 1, characterized in that: In S3, the component mass ratio includes the following parts by weight: 100 parts of base asphalt, 10-25 parts of steel slag powder, 5-15 parts of mineral powder, and 0.5-2 parts of dispersant.
6. The intelligent design method for thermally conductive asphalt materials based on a large model according to claim 1, characterized in that: In S4, the performance verification includes thermal conductivity testing and structural stability testing. The target thermal conductivity is ≥0.8W / m・K, and the structural stability meets the requirement of no obvious segregation or cracking.
7. A smart design system for thermally conductive asphalt materials based on a large model, characterized in that, include: The data acquisition and preprocessing module is used to collect and process data on the composition, process, performance, and scenario of thermally conductive asphalt, and output a training dataset. The large model training module connects to the data acquisition and preprocessing module, receives the training dataset, and trains the intelligent design large model. The intelligent design module connects to the large model training module, receives target performance indicators, and outputs initial design parameters. The performance verification module connects to the intelligent design module, tests the sample performance, and provides feedback on the verification results. The parameter optimization module connects the performance verification module and the large model training module, driving model updates and parameter optimization.
8. The intelligent design system for thermally conductive asphalt materials based on a large model according to claim 7, characterized in that: The data acquisition and preprocessing module includes a data cleaning unit, a standardization unit, and a feature extraction unit. The feature extraction unit is used to extract the correlation features between component dosage, process parameters, and thermal conductivity.
9. The intelligent design system for thermally conductive asphalt materials based on a large model according to claim 7, characterized in that: The large model training module includes a model building unit, a hyperparameter tuning unit, and an error evaluation unit. The error evaluation unit calculates the prediction error through cross-validation, and training stops when the error is ≤5%.
10. The intelligent design system for thermally conductive asphalt materials based on a large model according to claim 7, characterized in that: The intelligent design module can receive user-defined constraints, including cost limits and raw material availability, and output design parameter schemes that meet the constraints.