Digital control preparation method for concrete bonding materials

Abstract

This application relates to a digitally controlled preparation method for a high-solid-waste, low-carbon, and high-durability concrete bonding material. It solves the technical problems of existing UHPC bonding materials, such as difficulty in balancing high performance and low carbon emissions, low solid waste resource utilization, and a lack of precise digital control methods, leading to high carbon emissions, unstable durability, and insufficient adaptability to industrial production capacity. The method includes: first, integrating data to construct a dedicated database for ultra-high-performance concrete; establishing a raw material-process-performance mapping relationship through multi-scale simulation; then, optimizing target parameters through machine learning; pre-processing raw materials; verifying the iterative model through high-throughput experiments; determining the optimal process parameters; and importing them into the system for mixing, molding, and curing. This application has the following effects: achieving high solid waste resource utilization and low-carbon emission reduction through digital control, simultaneously ensuring high strength and high durability of the material, and precisely adapting to the core application requirements of prefabricated structures.

Description

Technical Field

[0001] This invention relates to the field of building materials technology, and in particular to a digitally controlled preparation method for a high-solid-waste, low-carbon, and high-durability concrete bonding material. Background Technology

[0002] With the rapid development of modern bridge engineering and urban modular construction, prefabricated structures place dual core requirements on the performance of connection materials: on the one hand, they need to possess high strength and high durability to adapt to complex service environments such as high-chloride and salty coastal areas; on the other hand, they need to meet the needs of low carbon emissions and efficient resource utilization. Ultra-high performance concrete (UHPC) has become the preferred connection material due to its dense microstructure and excellent mechanical properties. However, existing technologies have not yet formed a mature solution that balances high performance and low carbon emissions, and it is urgent to break through the bottleneck through innovative technologies.

[0003] In current related technologies, UHPC bonding materials are mostly constructed with high cement content to form a cementitious system. They are prepared through traditional empirical proportioning design and conventional processes. Industrial solid waste is mostly added in a low proportion as an auxiliary admixture, and no systematic optimization and utilization scheme has been formed.

[0004] The core problem with existing technologies is that they fail to effectively balance the dual goals of high performance and low carbon emissions. On the one hand, they rely on high cement usage to ensure strength, resulting in high carbon emissions from the cementitious system. On the other hand, they lack precise design and control methods, resulting in low utilization of solid waste resources and difficulty in ensuring the long-term durability of materials in complex environments, thus failing to fully meet the actual application needs of prefabricated structures. Summary of the Invention

[0005] In order to achieve high solid waste resource utilization and low carbon emission reduction through digital intelligent control, while ensuring high strength and high durability of materials and accurately adapting to the core application requirements of prefabricated structures, this application provides a digital intelligent control preparation method for high solid waste, low carbon and high durability concrete connecting materials.

[0006] This application provides a digitally controlled preparation method for a high-solid-waste, low-carbon, and high-durability concrete bonding material, employing the following technical solution:

[0007] A digitally controlled preparation method for a high-solid-waste, low-carbon, and high-durability concrete bonding material includes:

[0008] Acquire and integrate basic physicochemical parameters, historical process parameters and performance data of raw materials to construct a database of special materials for ultra-high performance concrete;

[0009] Based on a database of special materials for ultra-high performance concrete, a fundamental mapping relationship between raw materials, processes, and performance is constructed through multi-scale computational simulation, and parameter boundaries are defined. Then, a machine learning model is used to perform nonlinear fitting and optimization on this mapping relationship and boundaries to screen raw material ratios and key process parameters that meet the preset goals of high solid waste content, low carbon emissions, and high durability.

[0010] Based on the above screening results, the raw material pretreatment parameters were determined, and industrial solid waste was activated, fibers were surface modified, and aggregates were combined according to gradation.

[0011] Based on the pre-treated raw materials, according to the key process parameters and preset experimental schemes determined by the screening, multiple sets of parallel high-throughput experiments were carried out using automated equipment. Process and performance data were collected through preset monitoring equipment and fed back to the database of special materials for ultra-high performance concrete. The aforementioned machine learning model was iteratively calibrated using this as a new sample until the deviation between the model's predicted value and the experimental measured value reached the preset deviation threshold, thereby determining the optimal process parameters suitable for industrial production.

[0012] By importing the optimal process parameters into the production control system and executing mixing, molding, and curing according to the preset production process, a high-solid-waste, low-carbon, and high-durability concrete bonding material is prepared. Attached Figure Description

[0013] Figure 1 This is a schematic flowchart of a digital control preparation method for a high-solid-waste, low-carbon, and high-durability concrete connecting material according to an embodiment of this application. Detailed Implementation

[0014] The present application will be further described in detail below with reference to the accompanying drawings.

[0015] Reference Figure 1 This application discloses a digitally controlled preparation method for a high-solid-waste, low-carbon, and high-durability concrete bonding material, comprising:

[0016] Step S100: Obtain and integrate basic physical and chemical parameters of raw materials, historical process parameters and performance data to construct a database of special materials for ultra-high performance concrete.

[0017] The specific process is described as follows: First, the basic physicochemical parameters of the raw materials are obtained from the laboratory. These parameters are obtained through precise instrumental testing to ensure the accuracy and reliability of the data. For example, X-ray fluorescence spectrometry (XRF) is used to determine the chemical composition of the raw materials, and a laser particle size analyzer is used to determine the particle size distribution of the slag powder. Next, the company's past production process parameters and performance test data are integrated. This data reflects the actual performance of the materials under different process conditions. For example, process parameters such as stirring time, curing temperature, and vibration frequency are usually stored in the company's production management system. Through data mining technology, this scattered data can be integrated and cleaned to ensure data consistency and integrity. At the same time, performance data such as compressive strength, tensile strength, and durability are also obtained through laboratory testing, such as measuring compressive strength using a pressure testing machine and measuring chloride ion penetration resistance using electrochemical methods. Finally, these data are integrated into a unified database. Through data standardization processing, the data format from different sources is ensured to be consistent, facilitating subsequent analysis and use.

[0018] Step S200: Based on the database of special materials for ultra-high performance concrete, the basic mapping relationship between raw materials, processes and performance is constructed through multi-scale calculation simulation and the parameter boundaries are defined. Then, the mapping relationship and boundaries are nonlinearly fitted and optimized using a machine learning model to screen the raw material ratio and key process parameters that meet the preset goals of high solid waste content, low carbon emissions and high durability.

[0019] Among them, the construction of the basic mapping relationship between raw materials, process and performance and the delineation of parameter boundaries through multi-scale calculation simulation can refer to steps S210 to S240, or steps S2A0 to S2D0. The nonlinear fitting and optimization of the mapping relationship and boundary by machine learning model can refer to steps SA00 to SE00, which will not be elaborated here.

[0020] This step integrates microscopic thermodynamic / kinetic models, mesoscopic particle packing models, and macroscopic finite element models to construct a cross-scale digital materials performance simulation system. Based on first principles of physicochemistry, this system transforms raw material properties and process parameters into computable inputs, and the simulation outputs predicted results ranging from hydration products and microstructure to macroscopic mechanical and durability properties. The resulting fundamental mapping relationship between raw materials, processes, and properties, along with its corresponding multi-scale parameter boundaries, not only provides a physically feasible design space but also serves as crucial domain knowledge embedded in subsequent machine learning models, ensuring that data-driven optimization does not deviate from the fundamental laws of materials science.

[0021] Step S300: Based on the above screening results, determine the raw material pretreatment parameters, and then perform activation of industrial solid waste, surface modification of fibers, and aggregate gradation combination accordingly.

[0022] Among them, raw material pretreatment parameters refer to specific treatment conditions set to optimize raw material performance, such as grinding fineness, type and concentration of activator, and fiber surface modification methods. These parameters are obtained through machine learning model screening in step S200 to ensure optimal performance in subsequent preparation processes. Activation: Enhancing the reactivity of industrial solid waste (such as slag powder and fly ash) through physical or chemical methods, enabling it to participate more effectively in hydration reactions in concrete. Common methods include ultrafine grinding and the addition of chemical activators. Fiber surface modification: Improving the properties of the fiber surface through chemical or physical methods, enhancing the bonding performance between the fiber and the concrete matrix, and improving the fiber dispersibility and toughening effect in concrete. Common methods include silane coupling agent treatment. Aggregate gradation combination: Optimizing the aggregate particle size distribution according to the particle packing model to achieve optimal bulk density and porosity, improving the density and strength of concrete.

[0023] The specific process is as follows: 1. Data Input and Preliminary Screening: The raw material ratios and key process parameters screened by the machine learning model in step S200 are input into the pretreatment parameter analysis system. Based on historical data and experimental results, combined with preset performance targets (such as high solid waste content, low carbon emissions, and high durability), the system performs preliminary screening of the input parameters. For example, based on the model-recommended parameters such as slag powder content ≥40%, fly ash content 10%-20%, steel fiber content 1.5%-2.5%, and polypropylene fiber content 0.05%-0.15%, combined with the actual performance of these parameters in historical data, the system preliminarily determines the range of pretreatment parameters.

[0024] 2. Intelligent analysis and optimization:

[0025] Grinding fineness optimization: The particle size distribution of slag powder and fly ash was measured using a laser particle size analyzer, combined with a preset specific surface area target (e.g., 450 m²). 2 The system optimizes grinding process parameters using intelligent algorithms (such as genetic algorithms). Based on particle size distribution data, the system automatically adjusts the ball mill's rotation speed and grinding time to ensure that the specific surface area after grinding reaches the target value.

[0026] Chemical activator optimization: The effect of different concentrations of sodium sulfate on the activity of slag powder was tested experimentally. Based on the experimental data and a preset activity target (such as a 30% increase in early strength), the system uses a machine learning model (such as random forest) to predict the optimal activator concentration. For example, the system predicts that adding 3% sodium sulfate can effectively activate the activity of slag powder and promote its early hydration reaction.

[0027] Fiber surface modification optimization: Silane coupling agents are used to modify the surface of steel fibers. Based on preset bonding performance targets (e.g., increasing the bond strength between the fiber and the matrix by 50%), the system experimentally verifies the effect of different concentrations of silane coupling agents on fiber surface modification. The system automatically adjusts the concentration of the silane coupling agent and the treatment time to ensure the formation of a uniform modified layer on the fiber surface.

[0028] Aggregate gradation optimization: Based on the Andreasen & Andersen particle packing model, the aggregate gradation is optimized. The system adjusts the ratio of quartz sand and tailings powder using intelligent algorithms (such as particle swarm optimization) based on preset bulk density and porosity targets (e.g., porosity ≤ 5%). The system automatically calculates and adjusts the aggregate mixing ratio to ensure optimal bulk density and porosity.

[0029] 3. Parameter Verification and Adjustment: Experimentally verify the initially determined pretreatment parameters. For example, test the specific surface area of ​​the ground slag powder to ensure it reaches 450 m². 2 / kg; hydration reaction experiments were conducted on slag powder with added activator to verify whether its early strength increased by 30%; bond strength tests were conducted on surface-modified steel fibers to ensure that their bond strength with the matrix increased by 50%; compaction and porosity tests were conducted on aggregates with optimized gradation to ensure that the porosity was ≤5%. Based on the experimental results, the pretreatment parameters were fine-tuned.

[0030] Step S400: Based on the pretreated raw materials, according to the key process parameters and preset experimental scheme determined by the screening, multiple sets of parallel high-throughput experiments are carried out using automated equipment. Process and performance data are collected through preset monitoring equipment and fed back to the database of special materials for ultra-high performance concrete. The aforementioned machine learning model is iteratively calibrated using this as a new sample until the deviation between the model's predicted value and the experimental measured value reaches a preset deviation threshold, thereby determining the optimal process parameters suitable for industrial production.

[0031] The specific process is as follows:

[0032] 1. High-throughput experimental verification:

[0033] Experimental preparation: Based on the raw materials pretreated in step S300, and following the key process parameters and preset experimental scheme determined in step S200, multiple sets of experimental samples were prepared. The experimental scheme included different raw material ratios, stirring times, curing temperatures, vibration frequencies, and other process parameters.

[0034] Application of automated equipment: Automated mixing, molding, and curing equipment enable parallel operation of multiple sets of experiments. For example, automated mixing equipment can precisely control the mixing process according to preset mixing times and sequences; molding equipment can ensure consistent molding conditions for each sample; and curing equipment can perform curing according to preset temperature and humidity curves.

[0035] Experimental execution: Multiple sets of experiments are conducted simultaneously under the control of automated equipment to ensure the accuracy and repeatability of experimental conditions. For example, 10 sets of ultra-high performance concrete (UHPC) samples with different mix proportions are prepared simultaneously, and the process parameters such as mixing time and curing temperature for each set of samples are adjusted according to a preset plan.

[0036] 2. Data Acquisition and Feedback: 2.1. Monitoring Equipment Deployment: During the experiment, pre-set monitoring equipment is deployed to collect process and performance data in real time. For example, a rheometer is used to monitor the rheological parameters of the slurry, a hydration heat and temperature measurement system is used to monitor the hydration exothermic curve, a pressure testing machine is used to determine the compressive strength, and an electrochemical method is used to determine the chloride ion penetration resistance. 2.2. Data Acquisition: Through automated monitoring equipment, rheological parameters, hydration exothermic curves, compressive strength, tensile strength, durability, and other data for each sample are collected in real time. These data will be used as experimental measured values ​​for subsequent model calibration. 2.3. Data Feedback: The collected process and performance data are fed back to the ultra-high performance concrete special material database. These data serve as new samples for iterative calibration of the machine learning model in step S200.

[0037] 3. Model Iteration and Calibration: 3.1 Data Integration and Preprocessing: Integrate the data collected from high-throughput experiments into the materials database, perform data cleaning and standardization to ensure data consistency and usability. 3.2 Model Training and Calibration: Retrain and calibrate the machine learning model using the newly added experimental data. For example, using a random forest model, use the newly added experimental data as the training set to optimize model parameters and improve the model's prediction accuracy. 3.3 Comparison of Predicted and Measured Values: Evaluate the model's accuracy by comparing the model's predicted values ​​with the experimental measured values ​​using a preset deviation threshold. For example, set the deviation threshold for compressive strength to ±5 MPa. If the deviation between the model's predicted values ​​and the experimental measured values ​​exceeds this threshold, it indicates that the model needs further optimization.

[0038] 4. Determining Optimal Process Parameters: 4.1 Parameter Optimization: Based on the comparison between model predictions and experimentally measured values, optimize the process parameters. For example, if the model-predicted compressive strength is lower than the experimentally measured value, it may be necessary to adjust parameters such as stirring time or curing temperature. 4.2 Experimental Verification: Conduct further experimental verification of the optimized process parameters to ensure their feasibility and stability in actual production. For example, conduct experiments on the adjusted stirring time and curing temperature to verify whether they can achieve the expected compressive strength and durability. 4.3 Final Determination: Through multiple experimental verifications and model calibration, determine the optimal process parameters suitable for industrial production.

[0039] In step S500, the optimal process parameters are imported into the production control system, and the mixing, molding, and curing processes are executed according to the preset production flow to prepare a high-solid-waste, low-carbon, and high-durability concrete bonding material.

[0040] The process is described as follows: 1. Importing optimal process parameters: Import the optimal process parameters determined in step S400 into the production control system (MES). These parameters include stirring time, stirring sequence, vibration frequency, curing temperature, curing humidity, etc.

[0041] 2. Production Process Execution: 2.1 Mixing Process: Based on the optimal process parameters in the MES system, the automated mixing equipment mixes materials according to a preset time and sequence. For example, dry materials (such as cement, slag powder, fly ash, etc.) are first added to the mixer and dry-mixed for 30 seconds, then water and admixtures are added and wet-mixed for another 3 minutes to ensure thorough and uniform mixing of the raw materials. 2.2 Molding Process: The molding equipment vibrates and molds the concrete according to a preset vibration frequency and time. For example, a high-frequency vibrating table is used, with a vibration frequency set to 50Hz and a vibration time of 30 seconds to ensure dense molding of the concrete and reduce air bubbles and porosity. 2.3 Curing Process: The curing equipment cures the concrete according to a preset temperature and humidity curve. For example, the molded concrete is placed in a curing chamber with a curing temperature set to 20°C, a relative humidity of 95%, and a curing time of 28 days. During the curing process, sensors monitor the temperature and humidity in real time to ensure the stability and consistency of the curing conditions.

[0042] The fundamental mapping relationship between raw materials, processes, and performance is constructed and parameter boundaries are defined through multi-scale computational simulation, including:

[0043] Step S210, Microscale Simulation: Using preset thermodynamic calculation software, based on the principle of minimizing Gibbs free energy, the stable hydration product phase of the multi-component cementation system is predicted by setting different temperature conditions, with the goal of avoiding the unfavorable phase of calcium hydroxide crystals, and the type, content and microstructure parameters of the hydration products are output.

[0044] Thermodynamic calculation software, such as GEMS or FactSage, is based on the Gibbs free energy minimization principle and can predict the stable phase composition of multi-component systems under different conditions. The Gibbs free energy minimization principle states that the Gibbs free energy reaches its minimum value when the system reaches equilibrium. This is the fundamental principle of thermodynamic calculations used to predict the stable phase of a system. Hydration product phases refer to compounds with specific chemical compositions and microstructures formed during the hydration process of cement, such as calcium silicate hydrate (CSH) and calcium aluminate hydrate (CAH). Microstructure parameters describe the microstructural characteristics of hydration products, such as specific surface area and porosity.

[0045] The process is described as follows: 1. Precise characterization of system parameters and setting of temperature conditions: The chemical composition of the cementitious system was determined using X-ray fluorescence spectrometry (XRF) and inductively coupled plasma optical emission spectrometry (ICP-OES), including the mass fraction of cement clinker mineral phases (C3S, C2S, C3A, C4AF), the glass content and activity index of slag powder, and the total content of SiO2+Al2O3+Fe2O3 in fly ash. Reaction conditions were set using piecewise functions to simulate the actual curing process (e.g., 3 days of curing at 20°C → 4 days of curing at 40°C → standard curing at 20°C to 28 days). The temperature dependence of the hydration process was simulated by setting different temperature conditions (10°C, 20°C, 40°C). Reaction kinetic parameters were obtained by fitting the hydration exothermic rate curves at different temperatures using the Arrhenius equation (ln(k)=ln(A)-Ea / RT). The measurement errors of the pre-exponential factor A and activation energy Ea were controlled within ±5%. 2. Thermodynamic Equilibrium Calculation and Kinetic Correction: To bridge the thermodynamic equilibrium final state with the kinetic energy barrier of the actual hydration process, a Gibbs free energy database for the CaO-SiO2-Al2O3-Fe2O3-MgO-SO3-H2O multi-component system was constructed in GEMS software. The number of independent variables was calculated using the Gibbs phase rule F=C-P+2. Element conservation was constrained by the mass balance equation and charge balance equation. The nonlinear equation system was solved using the Newton-Raphson iterative algorithm (convergence criterion: residual <10). -6The system predicts the stable phase composition (e.g., CSH, CAH, CH, AFt) and its equilibrium content. In practical applications, the curing age (1d, 3d, 7d, 28d) needs to be considered. The equilibrium content is converted into a time function by introducing the Johnson-Mer-Avramy (JMA) phase transformation kinetic equation, where the nucleus growth index n and rate constant k are calibrated by isothermal calorimetry. 3. Avoidance of unfavorable phases and parameter verification: Set the CH volume fraction threshold to <5%. When the predicted CH content exceeds the limit, an inverse optimization algorithm based on gradient descent (using the variance of CH content and target threshold as the loss function) is automatically triggered. This is achieved by increasing the slag powder content to 55-70% (preferably 55-65%) or fly ash content to 15-28% (preferably 22-28%), simultaneously reducing the cement content, and introducing an alkaline activator (Na2SO4, etc.) to adjust the liquid phase pH to 12.5-13.0, thereby reducing the CH generation to <5%. The final output is the quantitative results of hydration products at each age (e.g., CSH phase content of 58-62% at 28 days). The simulated values ​​are verified using XRD-Rietveld refinement, mercury intrusion porosimetry (MIP), and BET methods to ensure that the deviation between the simulated and measured values ​​is <10%. The verified parameters are then used as input constraints for mesoscale simulations.

[0046] Step S220, Mesoscale Simulation: Based on the hydration product data output at the microscale, it is incorporated into the particle system as ultrafine powder. Combining the basic physicochemical parameters of aggregates and powders, the packing density of the system is calculated based on the particle size distribution through a preset particle packing model. The gradation design is simulated and optimized in a digital manner, and the target porosity is virtually calculated to determine the optimal packing scheme and output the corresponding mesoscale structure parameters.

[0047] The specific process is as follows: 1. Input data and particle packing model construction: The particle size distribution of aggregates and powders (e.g., quartz sand 0.1-2mm, tailings powder 0.01-0.1mm) is measured using a laser particle size analyzer. This is combined with particle density measured by a densitometer, and the volume percentage of hydration products (e.g., CSH gel 60%) and pore distribution characteristics output at the microscale. The Andreasen & Andersen particle packing model is selected to construct a three-dimensional mesoscopic model of the aggregate-fiber-cement system, and hydration products are incorporated as ultrafine powder components into the particle size distribution system. 2. Grading optimization and mesoscopic parameter calculation: To maximize packing density (target ≥75%) and minimize porosity (target 10%-15%), optimization algorithms (e.g., genetic algorithms) are used to automatically adjust the gradation ratio. Based on the three-dimensional mesoscopic model constructed using the discrete element method (DEM), the packing density of each generation of schemes is calculated through the mass balance equation, and the microscopic pore distribution is transformed into the mesoscopic particle gap range. After iterative optimization (e.g., setting the population size to 50 and iterating 100 times), mesoscopic structural parameters such as packing density and fiber spacing variation coefficient (target ≤0.3) under optimal gradation are output. 3. Parameter verification and cross-scale transfer: Optimized mesoscopic structural parameters (e.g., packing density 74-76%, porosity 10%-12%) are output. Porosity is verified by mercury intrusion porosimetry (MIP), and particle interstitial distribution is analyzed by scanning electron microscopy (SEM-BSE) to ensure that the deviation between simulated and measured values ​​is <10%. Finally, these parameters are used as key inputs to the macroscopic-scale finite element model to achieve a quantitative mapping from mesoscopic structure to macroscopic performance.

[0048] Step S230, Macroscale Simulation: Based on the optimal packing scheme and porosity parameters at the mesoscale, a finite element model reflecting the mechanical properties of the matrix is ​​established. Using finite element analysis technology, the distribution state, orientation characteristics and stress transmission path of fibers in the matrix are simulated to quantify the enhancing effect of fibers on the macroscopic mechanical properties of the material. Then, the fiber type and dosage are optimized, and simulation data related to macroscopic properties are output.

[0049] The specific process is as follows: 1. Multi-scale data input and finite element modeling: The optimal packing scheme and porosity parameters output from the mesoscale simulation are used as the input basis for the finite element model, including packing density (75%), porosity (10%), and aggregate (quartz sand particle size distribution 0.1–2 mm, density 2.65 g / cm³). 3 ) and powder (tailings micro powder with a particle size distribution of 0.01–0.1 mm and a density of 2.80 g / cm³) 3The physicochemical parameters of the concrete are determined. In a finite element analysis platform (such as ABAQUS or ANSYS), a three-dimensional heterogeneous concrete numerical model containing aggregates, powders, and hydration products is established based on the optimal packing structure determined at the mesoscale, and the spatial distribution parameters of each component are set. 2. Numerical simulation of fiber distribution and orientation characteristics: Based on the established finite element model, fiber type and dosage parameters (e.g., 2% volumetric steel fiber and 0.1% volumetric polypropylene fiber) are input, and a random distribution algorithm (e.g., generating statistically consistent fiber positions and orientations using MATLAB) is used to simulate the fiber distribution in the matrix, ensuring its randomness and uniformity in three-dimensional space. By setting the orientation probability of different directions (e.g., considering the anisotropy caused by the pouring and vibration process), the fiber distribution under actual construction conditions is simulated, providing a realistic fiber spatial configuration for subsequent stress transfer analysis. 3. Stress transfer path simulation and quantitative analysis of reinforcement effect: External loads consistent with actual service conditions are applied to the finite element model to simulate the stress transfer path of concrete under stress. Post-processing tools (such as ABAQUS / Viewer) are used to visualize the stress distribution of the fiber-matrix system, and the suppression mechanism of fiber on matrix stress redistribution and crack propagation is analyzed. Key mechanical indicators (such as compressive strength, tensile strength, and elastic modulus) are extracted, and the performance changes before and after fiber reinforcement are compared to quantify the fiber's reinforcing contribution. For example, the percentage increase in compressive strength after fiber incorporation is calculated to assess its specific impact on macroscopic mechanical properties. 4. Fiber Parameter Optimization and Verification Output: Based on the above quantification results, with the goal of optimal macroscopic performance, the system compares the reinforcing effects of different fiber types and incorporation combinations to determine the optimal fiber parameters. The optimized fiber type, incorporation, and corresponding macroscopic performance prediction data (such as compressive strength 250 MPa, tensile strength 30 MPa, and elastic modulus 50 GPa) are output. The accuracy of the simulation results is further verified through standard mechanical experiments (such as cube compression test and splitting tensile test). If there is a significant deviation between the experimental and simulated values, the model parameters are corrected, and the finite element calculation is repeated until the results meet the preset design goals.

[0050] Step S240, Mapping Relationship Construction and Parameter Boundary Delineation: Integrate simulation results at the micro, meso, and macro scales, systematically link the logical relationships between raw material characteristics, process parameters, and performance at each scale, and combine historical data accumulated in the database of ultra-high performance concrete special materials to construct a basic mapping relationship from raw materials to performance; at the same time, based on the parameter variation range covered by the simulation at each scale, delineate multi-scale parameter boundaries that match the design objectives.

[0051] The specific process is as follows: 1. Multi-source data fusion and feature engineering: Integrating microscale data (types and contents of hydration products, specific surface area of ​​CSH gel 400-600m²). 2Simulated data at the following scales were used: / kg, mesoscale (bulk density 70%-80%), and macroscale (fiber content 1.5%-2.5%, compressive strength 200-300MPa), combined with historical process-performance data from the UHPC database. Data preprocessing employed Z-score normalization and Min-Max normalization. Highly linearly correlated features were removed using Pearson correlation coefficient analysis (|r|>0.8). Principal component analysis (PCA) reduced the feature dimensions from 45 to 8 (cumulative variance contribution ≥85%), providing high-quality input data for mapping relationship construction. 2. Machine Learning Modeling and Mapping Relationship Construction: The Gradient Boosting Decision Tree (GBDT) algorithm was selected, using raw material characteristics (12-dimensional features such as chemical composition and particle size distribution) and process parameters (6-dimensional features such as curing regime and stirring parameters) as inputs, and macroscopic performance (compressive strength, elastic modulus, etc.) as outputs. Hyperparameters were optimized using five-fold cross-validation (learning rate 0.1, tree depth 6). Model performance requirements: Test set determination coefficient R0. 2 >0.92, Root Mean Square Error (RMSE) <5MPa, ensuring high accuracy and generalization ability of the mapping relationship. SHAP values ​​are used to analyze the contribution of key parameters, clarifying core control parameters such as fiber content (contribution 18%) and water-gel ratio (contribution 22%). 3. Multi-scale parameter boundary delineation and verification: Based on multi-objective optimization theory, a systematic method is used to delineate the boundaries of key parameters: First, a multi-objective function including solid waste content, carbon emissions, and durability indicators is established. The Pareto optimal solution set is solved using the Non-Dominated Sorting Genetic Algorithm (NSGA-II). Based on the distribution characteristics of the solution set, the K-means clustering algorithm is used to identify high-performance regions. Finally, the microscale (CSH gel density ≥2.0 g / cm³) is determined by combining the physical constraints of parameters at each scale. 3 The critical boundaries were defined at the following scales: CH crystal content ≤ 5%, mesoscale (packing density ≥ 72%, porosity ≤ 12%), and macroscale (compressive strength ≥ 220 MPa, fiber spacing coefficient of variation ≤ 0.3). Subsequently, parameter combinations were generated through Latin hypercube sampling, and performance predictions were performed using a trained model. Pareto front analysis was then used to determine the feasible parameter space, and independent test sets were used to verify the validity of the boundaries. A bias warning mechanism was established to complete the closed-loop verification.

[0052] The fundamental mapping relationship between raw materials, processes, and performance is constructed and parameter boundaries are defined through multi-scale computational simulation, including:

[0053] Step S2A0, Microscale Simulation: Based on the basic physicochemical parameters of raw materials in the database of special materials for ultra-high performance concrete, the hydration reaction kinetics simulation tool is used to simulate the hydration reaction process of the cementitious system, predict the type of hydration products and quantify the density of hydrated calcium silicate gel, control the amount of calcium hydroxide crystal formation to not exceed the preset upper limit to ensure high durability, and output the key parameters of hydration product volume ratio and pore distribution.

[0054] The above process can be referred to in steps S2A1 to S2A5, and will not be repeated here.

[0055] Step S2B0, Mesoscale Simulation: Using the volume ratio of hydration products and pore distribution output at the microscale as constraints, and combining aggregate characteristic data with high solid waste content from the database of special materials for ultra-high performance concrete, the aggregate gradation is optimized and the fiber distribution is evaluated through particle packing model and volume filling model. The particle gap characteristics are quantified, and a quantitative transformation relationship between micro-hydration characteristics and mesoscale structure is established. The bulk density and fiber spacing variation coefficient are output as mesoscale structure parameters.

[0056] The above process can be referred to in steps S2B1 to S2B4, and will not be repeated here.

[0057] Step S2C0, Macroscale Simulation: Based on mesoscopic structural parameters, using finite element analysis tools and performance mapping equations, a quantitative mapping relationship between mesoscopic structure, process parameters and macroscopic performance is constructed, thereby linking macroscopic mechanical properties, durability and low-carbon emission indicators under high solid waste content, and outputting preset thresholds for macroscopic performance and low-carbon indicators.

[0058] The above process can be referred to in steps S2C1 to S2C4, and will not be repeated here.

[0059] Step S2D0, Mapping Relationship Construction and Parameter Boundary Delineation: Integrate the quantitative transformation and mapping relationships formed by the three-level simulation of micro-meso-macro, combine the historical process-performance correlation data in the database of ultra-high performance concrete special materials, construct the basic mapping relationship of raw materials-process-performance, and simultaneously delineate the parameter boundaries of each scale to adapt to the goals of high solid waste, low carbon, and high durability based on the parameter range output by each scale simulation.

[0060] The specific process is as follows:

[0061] 1. Data Integration and Feature Engineering: The data integrates the volumetric composition and pore distribution of hydration products from step S2A0, the bulk density and fiber spacing variation coefficient from step S2B0, and the macroscopic performance and low-carbon indicators from step S2C0. Combined with historical formulations and measured data from the UHPC database, a dataset covering raw material characteristics (chemical composition, activity index), process parameters (stirring time, curing temperature), and target performance (strength, durability, carbon emissions) is formed. Z-score normalization and Min-Max normalization are used. Highly linear redundant features are removed through Pearson correlation coefficient analysis, and principal component analysis is used to reduce the dimensionality to 8 dimensions (cumulative variance contribution rate ≥ 85%), providing high-quality input for mapping relationship construction. 2. Mapping Relationship Construction: A nonlinear mapping model is constructed using the Gradient Boosting Decision Tree (GBDT) algorithm. The inputs include raw material proportions and process parameters, macroscopic performance (compressive strength ≥ 220 MPa, tensile strength ≥ 20 MPa), and low-carbon indicators (carbon emissions ≤ 45 kg CO2 / m³). 3 The output is 0.1. Hyperparameters (learning rate 0.1, tree depth 6) are optimized using five-fold cross-validation to ensure the model's test set determination coefficient R0. 2 >0.92, RMSE <5MPa. SHAP value analysis was used to clarify the water-cement ratio (contribution 22%), fiber content (contribution 18%), and solid waste activity index (contribution 15%) as core control parameters, accurately quantifying the influence weight of each factor on performance. 3. Multi-scale parameter boundary delineation: Based on high solid waste content ≥60% and carbon emissions ≤50kgCO2 / m³... 3 With a chloride ion permeability resistance of ≥1000°C as a constraint, the NSGA-II algorithm was used to solve for the Pareto optimal solution set. Combined with K-means clustering to identify high-performance regions, the following comprehensive delineation was established: microscopic boundaries (CSH gel density ≥2.0 g / cm³). 3 The following parameters are required: CH crystal content ≤ 5%, porosity 5%-15%, mesoscopic boundary (packing density ≥ 72%, fiber spacing variation coefficient ≤ 0.3, interface thickness ≤ 50 μm), and macroscopic boundary (compressive strength ≥ 220 MPa, carbon emissions ≤ 45 kg CO2 / m³). 3 4. Validation and Consolidation of Output: 100 parameter combinations are generated through Latin hypercube sampling. Performance prediction is performed using the trained GBDT model, and the validity of the boundary conditions is validated using an independent test set (30 sets of data not used in training). The deviation between predicted and actual values ​​is required to be <8%. The final output includes the basic mapping equation, a multi-scale parameter boundary table, and a model confidence assessment report (including R). 2 (RMSE, MAE and extrapolation risk warnings) provide benchmark constraints for subsequent machine learning optimization and production control.

[0062] Predicting hydration product types and quantifying the density of hydrated calcium silicate gel, controlling the amount of calcium hydroxide crystal formation to not exceed a preset upper limit to ensure high durability, and outputting key parameters such as the volume ratio of hydration products and porosity distribution:

[0063] Step S2A1: Retrieve data on the content of active components in industrial solid waste, specific surface area, and mineral composition of cementitious materials from the database of special materials for ultra-high performance concrete. Combine the differences in reactivity between industrial solid waste and cementitious materials to establish an initial parameter matrix for multi-component hydration reaction.

[0064] Among them, the content of active ingredients in industrial solid waste refers to the content of components with potential hydration activity in industrial solid waste, such as silicates and aluminates in slag powder. These data are obtained through chemical analysis, for example, using X-ray fluorescence spectrometry (XRF). Specific surface area refers to the surface area per unit mass of a material, usually expressed in meters (m²). 2 / kg represents the surface area. A larger specific surface area indicates higher material reactivity. Specific surface area is measured using a surface area meter, such as a Blaine surface area meter. Mineral composition of cementitious materials: refers to the mineral composition of cementitious materials such as cement, slag powder, and fly ash, including tricalcium silicate (C3S), dicalcium silicate (C2S), and tricalcium aluminate (C3A). These data are obtained through X-ray diffraction (XRD) analysis. Differences in reactivity: refers to the differences in reactivity of different cementitious materials in hydration reactions, usually determined experimentally, such as by measuring the heat of hydration of different materials using a hydration heat analyzer. Initial parameter matrix of hydration reaction: a matrix containing all initial conditions of the hydration reaction, including the content of active ingredients, specific surface area, mineral composition, etc., used to simulate the initial state of the hydration reaction.

[0065] The specific process is described as follows: 1. Data retrieval: Data on the content of active ingredients, specific surface area, and mineral composition of industrial solid waste and cementitious materials were extracted from the UHPC database. Outlier data with a coefficient of variation >15% were removed, and at least 30 sets of valid data were retained to ensure statistical significance. 2. Reactivity analysis: The hydration exothermic curves at 3, 7, and 28 days were determined using isothermal calorimetry. The reaction rate constant k and activation energy Ea (for slag powder: k = 0.01-0.03 s⁻¹) were obtained by fitting the Arrhenius equation. -1 Ea = 45-55 kJ / mol; cement: k = 0.08-0.12 s -13. Constructing the initial parameter matrix: An n×m dimensional parameter matrix is ​​established, with rows corresponding to material components such as cement, slag, and fly ash, and columns corresponding to parameters such as active ingredient content, specific surface area, k, and Ea, and including metadata such as material batch and test date to ensure traceability. 4. Parameter calibration and verification: The initial parameter matrix is ​​calibrated and verified using experimental data (such as the content of hydration products at different ages), and the reaction rate constant and activation energy are adjusted to ensure its accuracy and reliability. For example, experiments show that the CSH content in the hydration products of slag powder at 7 days is 60%, and the CH content is 5%; the corresponding values ​​for cement are 70% and 10%, respectively. The initial parameter matrix is ​​optimized based on these data. 5. Outputting the initial parameter matrix: The calibrated initial parameter matrix of the multi-component hydration reaction is output to provide accurate initial conditions for subsequent hydration reaction simulations, ensuring the scientific validity and practicality of the simulation results.

[0066] Step S2A2: Based on the initial parameter matrix of the hydration reaction, a hydration reaction model of the high solid waste system is constructed using a hydration reaction kinetic simulation tool. The reaction rate equation is corrected by introducing the solid waste activity activation coefficient. The hydration path at different ages is simulated. The types of hydration products, including hydrated calcium silicate and calcium hydroxide, are predicted by calculating the reaction process. The generation rate curves of each product and the characteristic parameters of the stoichiometry and microstructure of hydrated calcium silicate are output.

[0067] Among them, the hydration reaction kinetics simulation tool refers to a simulation platform based on the coupling of thermodynamic equilibrium and reaction kinetics (such as the GEMS kinetics module or Python+Cantera), which can realize the simulation of time-varying hydration process by inputting the initial parameter matrix; the solid waste activity activation coefficient α is a correction factor that takes into account the improvement of solid waste reactivity by alkaline activators; the generation rate curve describes the evolution of the content of each hydration product with age; the characteristic parameters include microstructure indicators such as CSH stoichiometry, specific surface area, and porosity.

[0068] The specific process is as follows: 1. Model Construction and Initialization: A hydration reaction model of the high solid waste system is constructed using a hydration reaction kinetics simulation tool. Input the initial hydration reaction parameter matrix output in step S2A1, including the content of active ingredients, specific surface area, mineral composition, and reaction kinetic parameters of industrial solid waste and cementitious materials. For example, input slag powder with a glass content of 85% and a specific surface area of ​​450 m². 2 / kg, reaction rate constant k=0.015s -1 The cement has a C3S content of 55% and a specific surface area of ​​350 m². 2 / kg, reaction rate constant k=0.10s -11. To construct a reaction system that conforms to the actual material properties. 2. Modify the reaction rate equation: Introduce the solid waste activity activation coefficient α to modify the classical JMA equation: Where r is the reaction rate, k is the rate constant, n is the Avramie exponent (taken as 1.5-2.0), and α is the excitation coefficient. The cumulative heat release over 72 hours was experimentally determined under different Na₂SO₄ dosages (2%, 3%, 4%), and the α value was determined by inversion (e.g., α=1.58 is optimal for 3% dosage), ensuring that the deviation between the corrected simulated heat release curve and the measured value is <5%. 3. Age Simulation and Product Prediction: The hydration reaction process at key ages of 1, 3, 7, 28, and 90 days was simulated. The adaptive step-size Runge-Kutta algorithm was used to solve the reaction kinetic equations, dynamically updating the consumption and generation of each component. The type and content of hydration products at each age were predicted by calculating the reaction process, along with the generation rate curve. For example, the model predicts that at 7 days old, the formation rate of CSH is 0.48 mg / h, with a content of 45%, and the formation rate of CH is 0.12 mg / h, with a content of 3.5%; at 28 days old, the formation rate of CSH decreases to 0.12 mg / h, with a content of 62%, and the CH content is 4.2%. 4. Output characteristic parameters and model validation: Output the stoichiometry and microstructural characteristic parameters of hydrated calcium silicate (CSH). Based on the simulation results, the Ca / Si ratio of CSH is calculated to be 1.65 ± 0.05, and the specific surface area is 520 m². 2 / kg, porosity 12%. XRD-Rietveld refinement of measured values ​​was used to correct simulation results, requiring a product content deviation of <8% at each age. Porosity was verified using mercury intrusion porosimetry; a deviation of <10% was considered convergence. If the deviation exceeded the range, the process returned to step S2A1 to adjust the initial parameter matrix, forming an iterative optimization closed loop. The final output was a JSON file containing age, product type and content, formation rate curve, and CSH microstructure parameters, providing time-varying constraints for the mesoscopic simulation in step S2B0.

[0069] Step S2A3: Based on the above characteristic parameters, the apparent density of hydrated calcium silicate gel is calculated using density function theory. The density is then corrected using measured data of similar gels from the database of special materials for ultra-high performance concrete to obtain a density value with a preset accuracy.

[0070] Density function theory (DFT) is based on quantum mechanical simulation of the electronic structure of CSH gel to calculate the theoretical density; measured data correction adjusts the model parameters by comparing the simulated values ​​with measured values ​​of the same type of gel in the database; the preset accuracy requirement is that the relative error of the corrected density value is <3%.

[0071] The specific process can be referred to as follows: 1. Feature parameter integration: Collect the feature parameters output in step S2A2, including CSH stoichiometry (Ca / Si=1.65±0.05), specific surface area (520m²), etc. 2 The parameters (g / kg) and porosity (12%) were used to form the input parameter set for the DFT calculation. The parameter range was ensured to be physically consistent with the S2A2 simulation results (CSH content 62%). 2. Application of density function theory: The theoretical density of the CSH gel was calculated using DFT. A CSH molecular model (selecting the tobermorite 14Å structure) was constructed using MaterialsStudio or VASP software. The geometry was optimized under GGA-PBE functional theory, and the total energy and volume were calculated. The initial calculation yielded a theoretical density of 2.62 g / cm³. 3 3. Correction and optimization of measured data: Compare the measured density (2.40 g / cm³) of similar CSH gels in the UHPC database. 3 The deviation was found to be 8.2%. After adjusting the porosity parameter in the DFT model (increasing it from 12% to 14.5%) and introducing surface hydroxylation correction, the apparent density was recalculated to be 2.48 g / cm³. 3 The deviation from the measured value was reduced to 3.3%, meeting the preset accuracy requirements. 4. Output density value: The output corrected apparent density value of CSH gel is 2.48±0.05g / cm³. 3 Simultaneously output correction parameters (porosity 14.5%, surface hydroxylation degree 1.2OH / nm) 2 The data is then transferred to the UHPC database to provide baseline density parameters for the volume filling calculation of the mesoscopic simulation in step S2B0.

[0072] Step S2A4: With a preset high durability target as a constraint, the upper limit of calcium hydroxide production is determined by using the correlation data between calcium hydroxide content and durability in the database of special materials for ultra-high performance concrete. The production amount is monitored in real time during the simulation. If the limit is exceeded, the amount of solid waste admixture or the activity activation coefficient is adjusted in reverse and the hydration reaction model is updated simultaneously.

[0073] The specific process can be referred to as follows: 1. Set the upper limit of CH generation driven by durability: Based on more than 200 sets of experimental data in the UHPC database, a quantitative correlation model between CH volume fraction and RCPT charge and freeze resistance level is established using random forest or support vector regression. The critical threshold is determined to be 5% through 5-fold cross-validation (when CH > 5%, chloride ion charge > 1000C and freeze resistance level < F300). This threshold is fixed in the GEMS / Cantera thermodynamic database and set as the core constraint condition for hydration simulation. At the same time, 4.5% is set as a warning value to reserve an adjustment window. 2. Monitor CH generation in real time during simulation: Define the CH phase as the key tracking target in the GEMS or Cantera platform. The reaction kinetic equation is solved using an adaptive step size (minimum 0.1d). The CH volume fraction evolution curves at 3d, 7d, 28d and 90d are dynamically extracted. When the predicted value is ≥ 4.5%, the system pushes a yellow warning to the MES interface. When it is ≥ 5%, the PID or fuzzy feedback control mechanism is automatically triggered and subsequent calculations are paused. 3. Reverse adjustment mechanism: Upon triggering, gradient descent or response surface optimization algorithms are initiated to maintain total solid waste ≥60% and carbon emissions ≤45kgCO2 / m³. 3 Under multi-objective constraints, the following parameters are adjusted iteratively: slag powder content (30%→35%~40%), fly ash ratio (15%→20%~25%), Na2SO4 concentration (3%→4%), and water-cement ratio (reduced from 0.22 to 0.18~0.20). This forces the secondary hydration reaction to consume CH to ≤4%, with each adjustment controlled within 1%~2% to ensure algorithm convergence. 4. Synchronously update the hydration reaction model: After parameter adjustment, immediately rewrite the initial reactant concentration matrix, rate constant k (corrected according to the Arrhenius equation), and excitation coefficient α (inverted from isothermal calorimetric measurements) of the kinetic model. Rerun the simulation until the CH content stabilizes within the target range. If the target is not met after three consecutive iterations, a secondary optimization strategy with a 50% reduction in adjustment step size is triggered. 5. Experimental Verification and Model Calibration Closed Loop: 100mm cubes and φ100×50mm cylindrical specimens were molded according to the final formulation. RCPT (ASTM C1202, 56-day charge), freeze-thaw cycle testing (ASTM C666, 300-cycle mass loss rate), XRD-Rietveld refinement (CH phase quantitative analysis), and MIP (porosity) tests were conducted. The deviation between simulated and measured values ​​was required to meet the following requirements: CH content < 8%, chloride ion charge < 10%, and freeze resistance level < 1. Otherwise, the error was propagated back to the thermodynamic database to correct the activity coefficient, forming a "simulation-optimization-experiment-feedback" closed loop. 6. Output and Consolidation: The final output includes the optimized formulation, a solid waste admixture-activator concentration mapping table, CH dynamic control logic, and updated k and α parameter sets. These are encapsulated as reusable knowledge modules with version numbers and stored in the standard interface of the UHPC materials database for direct use in mesoscale particle packing and macroscale finite element simulations.

[0074] In step S2A5, the volume percentage of hydration products at different ages is calculated by combining the corrected gel density and the product generation rate curves. The microstructure evolution caused by the hydration reaction is transformed into pore distribution parameters through a pore network model, forming a pore feature matrix that is compatible with the mesoscopic simulation, which serves as the underlying constraint for the mesoscopic simulation.

[0075] The specific process is as follows: 1. Calculate the volume percentage of hydration products: Combining the corrected calcium silicate hydrate (CSH) gel density from step S2A3 and the product formation rate curves from step S2A2, calculate the volume percentage of hydration products at different ages (e.g., 3 days, 7 days, 28 days). Using the law of conservation of mass and the density formula, convert the mass of each product into volume. For example, assuming that at 28 days, the formation rate of CSH is 0.5 mg / h and the formation rate of CH is 0.2 mg / h, the volume percentage of CSH is calculated to be 60% and the volume percentage of CH is 5% through integration. 2. Application of the pore network model: Using the pore network model, the microstructure evolution caused by the hydration reaction is transformed into pore distribution parameters. The pore network model simulates the filling process of hydration products in the matrix, calculating porosity and pore size distribution. For example, the model predicts that at 28 days, the porosity is 10%, and the pore size distribution is mainly concentrated in the range of 0.01-0.1 μm. 3. Forming the Pore Feature Matrix: The calculated porosity and pore size distribution parameters are integrated into a pore feature matrix. This matrix describes the pore characteristics at different ages, providing underlying constraints for mesoscopic simulation. For example, the pore feature matrix includes parameters such as porosity, pore size distribution, and pore connectivity, ensuring that the mesoscopic simulation can accurately reflect changes in microstructure. 4. Verification and Optimization: The accuracy of the pore feature matrix is ​​verified experimentally. For example, mercury intrusion porosimetry (MIP) is used to test the porosity and pore size distribution of concrete, ensuring consistency between simulation results and experimental data. Based on the experimental results, the pore network model is further optimized to ensure its predictive accuracy at different ages.

[0076] Quantitative analysis of interparticle spacing characteristics was conducted to establish a quantitative transformation relationship between microscopic hydration properties and mesoscopic structure. Bulk density and fiber spacing variation coefficient were output as mesoscopic structure parameters, including:

[0077] Step S2B1 uses the volume ratio of hydration products and pore distribution at the microscale as basic constraints, and combines the aggregate characteristic data with high solid waste content in the database of special materials for ultra-high performance concrete, to construct a three-dimensional mesoscopic model of the aggregate-fiber-cement system using the discrete element method, and transforms the microscopic pore distribution into the initial gap range between aggregate, solid waste particles and fibers at the mesoscopic level.

[0078] The specific process is as follows:

[0079] 1. Basic Constraints and Data Integration: Aggregate gradation (0.1-2mm quartz sand), fiber parameters (length 10-15mm, diameter 0.2mm, volumetric admixture 2%), and the volume percentage of hydration products at the microscale output (e.g., CSH accounts for 60% and CH accounts for 5% at 28d) and pore distribution (total porosity 10%, pore size 0.01-0.1μm) are extracted from the UHPC database and then standardized by Z-score to form a unified multidimensional constraint vector.

[0080] 2. Construction of a 3D Mesoscopic Model: A 3D mesoscopic model of the aggregate-fiber-cement system is constructed using the Discrete Element Method (DEM). The DEM simulates the interactions between particles to generate a 3D model that reflects the internal structure of actual concrete. The specific steps are as follows:

[0081] Particle generation: Based on the particle size distribution and shape of the aggregate, aggregate particles of different sizes and shapes are generated. For example, quartz sand particles with a particle size of 0.1-2 mm are generated.

[0082] Fiber distribution: Fibers are randomly distributed based on their length, diameter, and volume fraction. This ensures uniform fiber distribution within the model and avoids aggregation. For example, a random distribution algorithm can be used to evenly distribute the fibers throughout the model.

[0083] Cementitious material filling: Based on the volume percentage and pore distribution of hydration products, cementitious material is filled. Ensure the cementitious material fills the pores between the aggregate and fibers to form a continuous matrix. For example, when filling with CSH gel, ensure its volume percentage is 60% and its porosity is 10%.

[0084] 3. Determining the initial gap range:

[0085] In the constructed three-dimensional mesoscopic model, the initial gap range between aggregates, fibers, and cementitious materials is determined by simulating their distribution. The specific steps are as follows:

[0086] Gap Calculation: Calculate the initial gaps between aggregate particles, between fibers and aggregate, and between fibers. For example, the calculated average gap between aggregate particles is 0.05 mm, and the average gap between fibers and aggregate is 0.03 mm.

[0087] Pore ​​distribution transformation: Transforming the microscopic pore distribution into the initial gap range of the mesoscopic layer. For example, transforming a microscopic porosity of 10% into an initial gap range of 0.01-0.1 mm in the mesoscopic layer.

[0088] Step S2B2: Based on the above three-dimensional mesoscopic model, the gap size data between aggregate particles and between solid waste particles and fibers are extracted by three-dimensional structural scanning technology. The gap filling requirement is calculated by combining the volume ratio of micro-hydration products, and the average size and distribution uniformity characteristic parameters of the particle gaps are obtained.

[0089] The specific process is as follows:

[0090] 1. Application of 3D structural scanning technology: Based on the 3D mesoscopic model constructed in step S2B1, 3D structural scanning technology (such as X-ray CT) is used to scan the model and extract the gap size data between aggregate particles and between solid waste particles and fibers. The specific steps are as follows:

[0091] Data Acquisition: X-ray CT scans were performed on actual concrete samples to obtain detailed information about their internal structure. The scan data included the distribution of aggregate particles, fibers, and cementitious materials, as well as the dimensions of the gaps between them.

[0092] Image processing: The scanned images are processed using image segmentation techniques (such as thresholding, region growing, etc.) to distinguish different phases (aggregate, fiber, cementitious material, and pores). For example, thresholding can be used to label aggregate particles, fibers, and cementitious materials separately and extract their geometric features.

[0093] 2. Gap Size Data Extraction: Extract the gap size data between aggregate particles and between solid waste particles and fibers from the processed image. The specific steps are as follows:

[0094] Gap identification: Identifying gaps between aggregate particles, between fibers and aggregates, and between fibers. For example, the gap size between adjacent particles can be obtained by subtracting the sum of their radii from the distance between their centers.

[0095] Data statistics: The extracted gap size data are statistically analyzed, and statistical parameters such as the average gap size, standard deviation, and distribution range are calculated. For example, the average gap between aggregate particles is calculated to be 0.05 mm with a standard deviation of 0.01 mm; the average gap between fibers and aggregate is calculated to be 0.03 mm with a standard deviation of 0.005 mm.

[0096] 3. Gap Filling Requirement Calculation: Based on the volume percentage of microscopic hydration products, the gap filling requirement is calculated. The specific steps are as follows:

[0097] Volume percentage conversion: Converting the volume percentage of microscopic hydration products into the filling volume of the cementitious material. For example, assuming the volume percentage of hydration products is 60% and the porosity is 10%, the cementitious material needs to fill 60% of the total pore volume.

[0098] Filling requirement calculation: Based on the extracted gap size data and the filling volume of the cementitious material, the gap filling requirement is calculated. For example, the calculated total gap volume to be filled is 0.1 cm³. 3 Combined with the density of the cementitious material (e.g., 2.6 g / cm³) 3The required mass of cementitious material to be filled was 0.26 g / cm³. 3 .

[0099] 4. Characteristic Parameter Calculation: Calculate the average size and distribution uniformity of the interparticle gaps. The specific steps are as follows:

[0100] Average size calculation: Calculate the average size of all extracted gaps. For example, the average gap between aggregate particles is 0.05 mm, and the average gap between fibers and aggregate is 0.03 mm.

[0101] Distribution uniformity calculation: The uniformity of the gap size distribution is calculated, using standard deviation or coefficient of variation as the metric. For example, the standard deviation of the gap between aggregate particles is 0.01 mm, and the coefficient of variation is 20%; the standard deviation of the gap between fibers and aggregate is 0.005 mm, and the coefficient of variation is 16.7%.

[0102] Step S2B3: Using the volume ratio of micro-hydration products and pore distribution as independent variables and the interparticle spacing characteristic parameters as intermediate variables, the correlation equation between micro-hydration characteristics and mesoscopic structure parameters is fitted through multiple regression analysis to clarify the positive correlation between bulk density and volume ratio of hydration products, and the negative correlation between fiber spacing variation coefficient and uniformity of interparticle spacing distribution.

[0103] The specific process is as follows: 1. Data Preparation: Collect data on the volume percentage of micro-hydration products, pore distribution, and intergranular spacing characteristics. These data come from the outputs of steps S2B1 and S2B2, including the volume percentage of hydration products, porosity, inter-particle spacing size, and inter-fiber-aggregate spacing size. 2. Multiple Regression Analysis: Construct a multiple regression model using the volume percentage of micro-hydration products and pore distribution as independent variables, and intergranular spacing characteristics (such as average spacing size and distribution uniformity) as intermediate variables. Perform regression analysis using statistical analysis software (such as SPSS, R, or Python's scikit-learn library) to fit the correlation equation between micro-hydration characteristics and mesoscopic structure parameters. 3. Correlation Analysis: In the regression analysis, focus on analyzing the positive correlation between bulk density and the volume percentage of hydration products, and the negative correlation between the coefficient of variation of fiber spacing and the distribution uniformity of intergranular spacing. Quantify these correlations by calculating correlation coefficients (such as the Pearson correlation coefficient). For example, the correlation coefficient between bulk density and the volume percentage of hydration products is 0.85, indicating a significant positive correlation; the correlation coefficient between the coefficient of variation of fiber spacing and the uniformity of particle interstitial distribution is -0.78, indicating a significant negative correlation. 4. Equation Validation and Optimization: The fitted correlation equations are validated using independent datasets to evaluate their predictive power and accuracy. Based on the validation results, the model is optimized and adjusted, such as by adding or deleting independent variables or adjusting model parameters, to improve the goodness of fit and predictive power. 5. Output Correlation Equations: The final output is the correlation equation between microscopic hydration characteristics and mesoscopic structural parameters, clarifying the positive correlation between bulk density and the volume percentage of hydration products, and the negative correlation between the coefficient of variation of fiber spacing and the uniformity of particle interstitial distribution.

[0104] Step S2B4: Based on the above quantitative conversion relationship, calculate the theoretical values ​​of bulk density and fiber spacing variation coefficient. Combine the measured data of mesoscopic structure of high solid waste system in the ultra-high performance concrete special material database for error calibration. Finally, output the verified bulk density and fiber spacing variation coefficient as mesoscopic structure parameters.

[0105] The specific process is as follows: 1. Theoretical value calculation: Using the correlation equation obtained in step S2B3, and taking the volume ratio of micro-hydration products and pore distribution as inputs, calculate the theoretical values ​​of bulk density and fiber spacing variation coefficient. For example, assume the correlation equation is: Bulk density = 0.5 + 0.3 × volume ratio of hydration products - 0.1 × porosity. Fiber spacing variation coefficient = 0.2 - 0.05 × uniformity of particle spacing distribution. Substitute the specific values ​​for calculation. 2. Error calibration: Compare the calculated theoretical values ​​with the measured data of mesoscopic structures in the ultra-high performance concrete special material database to calculate the error. For example, the measured bulk density is 2.4 g / cm³. 3The theoretical calculated value is 2.35 g / cm³. 3 The error is 0.05 g / cm³. 3 The measured coefficient of variation for fiber spacing was 0.15, while the theoretically calculated value was 0.16, with an error of 0.01. 3. Model Optimization: The parameters of the correlation equation were adjusted based on the error to optimize the model. For example, by introducing correction coefficients or adjusting the parameters of the regression model, the error between the theoretical and measured values ​​was reduced, improving the model's accuracy. 4. Validation and Output: The optimized model was recalculated and compared with another set of independent measured data for validation. If the error was within acceptable limits, the validated packing density and coefficient of variation for fiber spacing were output as mesoscopic structural parameters, providing accurate input for subsequent macroscopic performance simulations.

[0106] Finite element analysis tools are used to construct a quantitative mapping relationship between mesoscopic structure, process parameters, and macroscopic performance, and preset thresholds for macroscopic performance and low-carbon indicators are output, including:

[0107] Step S2C1: Collect the mesoscopic structural parameters output at the mesoscopic scale, and combine them with the process parameters and basic performance data under high solid waste content from the ultra-high performance concrete special material database to form the input parameter set for finite element analysis.

[0108] The specific process is as follows: 1. Data collection: Collect mesoscopic structural parameters such as packing density and fiber spacing variation coefficient from the mesoscopic scale. For example, the packing density is 2.4 g / cm³. 3 The fiber spacing variation coefficient is 0.15. 2. Database Integration: Combining process parameters and basic performance data under high solid waste content from the ultra-high performance concrete special materials database, an input parameter set for finite element analysis is formed. For example, process parameters include mixing time, curing temperature, vibration frequency, etc., while basic performance data includes compressive strength, tensile strength, elastic modulus, etc.

[0109] In step S2C2, based on the above set of input parameters, a macroscopic performance correlation model is built using finite element analysis tools. The mesoscopic structural parameters are transformed into microscopic structural inputs of the model, and the process parameters are transformed into environmental and operational inputs. Mechanical performance sub-models and durability sub-models are constructed, and the conversion relationship between process parameters and carbon emissions is embedded.

[0110] The specific process is as follows: 1. Model building and tool selection: A macroscopic performance correlation model is built using finite element analysis tools (such as ABAQUS and ANSYS). Mesoscopic structural parameters (such as bulk density and fiber spacing variation coefficient) are converted into microscopic structural inputs for the model, and process parameters (such as mixing time, curing temperature, and vibration frequency) are converted into environmental and operational inputs. Specific steps are as follows: 1.1 Tool selection: ABAQUS is selected as the finite element analysis tool, which has powerful nonlinear analysis capabilities and a rich material model library. 1.2 Model building: A three-dimensional model of concrete materials is built in ABAQUS, including the distribution of aggregates, fibers, and cementitious materials. Based on the mesoscopic structural parameters, the distribution characteristics of aggregates and fibers are set, such as bulk density and fiber spacing variation coefficient. 2. Microscopic structural input setting: The mesoscopic structural parameters are converted into microscopic structural inputs for the model. For example, the bulk density of the aggregate is set to 2.4 g / cm³. 3 The fiber spacing variation coefficient is 0.15. The distribution characteristics of aggregates and fibers are defined in detail through the model's geometry and material property modules to ensure the model accurately reflects the microstructure of concrete. 3. Environment and Operation Input Settings: Process parameters are converted into environmental and operational inputs for the model. For example, the mixing time is set to 3 minutes, the curing temperature to 20°C, and the vibration frequency to 50Hz. Through the model's boundary conditions and load modules, the stress and environmental conditions of concrete under different process conditions are defined to ensure the model can simulate the process influences in actual production. 4. Sub-model Construction: Mechanical performance sub-models and durability sub-models are constructed, and the conversion relationship between process parameters and carbon emissions is embedded. Specific steps are as follows: 4.1 Mechanical Performance Sub-model: Define the mechanical performance parameters of concrete, such as compressive strength, tensile strength, and modulus of elasticity. Select appropriate constitutive relations from the material model library, such as the Drucker-Prager model, to simulate the mechanical behavior of concrete under different stress conditions. 4.2 Durability Sub-model: Defines the durability performance parameters of concrete, such as chloride ion penetration resistance and freeze-thaw resistance. By defining damage and diffusion models, it simulates the durability behavior of concrete under different environmental conditions. 4.3 Carbon Emission Conversion Relationship: Embeds the conversion relationship between process parameters and carbon emissions. By defining the environmental impact factor of the model, it calculates the carbon emissions under different process parameters. For example, the relationship between carbon emissions per cubic meter of concrete and mixing time, curing temperature, and vibration frequency is set as: Carbon emissions = 50 - 10 × mixing time + 20 × curing temperature - 5 × vibration frequency. 5. Model Validation and Optimization: Uses experimental data to validate and optimize the model. For example, the model's predicted compressive strength is validated using a standard cube compressive strength test, and the model's predicted chloride ion penetration resistance is validated using a rapid chloride ion penetration test (RCPT). Based on the validation results, the model parameters are adjusted to optimize the model's predictive ability, ensuring that the model accurately reflects the actual performance of concrete.

[0111] Step S2C3 involves conducting iterative simulations with multiple sets of variables using the model described above, adjusting the mesoscopic structure parameters and process parameters, recording the corresponding macroscopic performance and low-carbon indicators, and using response surface methodology to fit the quantitative mapping equation of mesoscopic structure parameters-process parameters-macroscopic performance-low-carbon indicators.

[0112] The specific process is as follows: 1. Iterative simulation with multiple variables: A finite element model is used to conduct iterative simulations with multiple variables, adjusting mesoscopic structural parameters (such as bulk density and fiber spacing variation coefficient) and process parameters (such as stirring time, curing temperature, and vibration frequency), and recording the corresponding macroscopic properties (such as compressive strength, tensile strength, and elastic modulus) and low-carbon indicators (such as carbon emissions). For example:

[0113] Simulation 1: Bulk density 2.4 g / cm³ 3 With a fiber spacing variation coefficient of 0.15, a stirring time of 3 minutes, a curing temperature of 20°C, and a vibration frequency of 50Hz, a compressive strength of 220MPa and a carbon emission of 45kgCO2 / m³ were obtained. 3 .

[0114] Simulation 2: Bulk density 2.5 g / cm³ 3 With a fiber spacing variation coefficient of 0.10, a stirring time of 4 minutes, a curing temperature of 22°C, and a vibration frequency of 55Hz, a compressive strength of 230MPa and a carbon emission of 40kgCO2 / m³ were obtained. 3 .

[0115] 2. Response Surface Methodology Fitting: The response surface methodology is used to fit quantitative mapping equations between mesoscopic structural parameters, process parameters, macroscopic performance, and low-carbon indicators. The specific steps are as follows: 2.1 Data Preparation: Multiple sets of simulation data are prepared to form a dataset. 2.2 Model Building: A response surface model is built using statistical analysis software (such as Design-Expert, R, or Python's scikit-learn library). 2.3 Equation Fitting: Quantitative mapping equations are obtained through fitting, for example: Compressive strength = 200 + 50 × bulk density - 30 × fiber spacing coefficient of variation + 10 × stirring time - 5 × curing temperature + 8 × vibration frequency. Carbon emissions = 50 - 10 × stirring time + 20 × curing temperature - 5 × vibration frequency.

[0116] 3. Equation Validation: Validate the fitted mapping equation using an independent dataset to ensure its predictive power and accuracy. For example, use a set of simulated data not involved in the fitting to validate the deviation between the equation's predicted values ​​and the actual values, ensuring the deviation is within an acceptable range.

[0117] In step S2C4, with high solid waste, low carbon, and high durability as constraints, and referring to the high-performance sample range in the database, the preset thresholds for macroscopic performance and low carbon indicators are determined by substituting them into the mapping equation for back calculation.

[0118] 1. Constraint Setting: High solid waste, low carbon, and high durability are set as constraints, referencing the range of high-performance samples in the ultra-high performance concrete special materials database. 2. Mapping Equation Inverse Calculation: Substitute the mapping equation fitted in step S2C3 and perform inverse calculation to determine the preset thresholds for macroscopic performance and low carbon indicators that satisfy the constraints.

[0119] The nonlinear fitting and optimization of this mapping relationship and boundary using a machine learning model includes:

[0120] Step SA00: Collect the raw material-process-performance basic mapping relationship data from the multi-scale simulation output, and combine it with the historical formulas, processes and corresponding performance data in the ultra-high performance concrete special material database to form a training dataset containing raw material proportions, process parameters, characteristic parameters at each scale and target performance.

[0121] The specific process is as follows: 1. Data integration scope: Extract the basic mapping relationship of "raw materials-process-performance" from the multi-scale simulation results, including the volume ratio of micro-hydration products, meso-bulk density, fiber spacing variation coefficient, as well as macro-compressive strength, chloride ion penetration resistance and carbon emissions; Simultaneously call the historical formula, mixing system and measured performance data in the database of special materials for ultra-high performance concrete to form an original data pool covering multiple solid waste admixtures, multiple ages and multiple environmental conditions.

[0122] 2. Feature Engineering and Cleaning:

[0123] Consistency checks were performed on the original data pool: samples missing key performance items were removed; outliers were removed using the 3σ criterion; variables with large dimensional differences, such as raw material chemical composition, specific surface area, process time, and temperature, were Z-score standardized; feature importance was evaluated using random forest, and variables with importance > 0.01 were retained. The final training dataset contained raw material proportions, process parameters, feature parameters at various scales, and target performance (strength + durability + carbon emissions), with a sample size of approximately 1.2 × 10⁻⁶. 4 These can be directly used for subsequent machine learning modeling.

[0124] Step SB00: Based on the nonlinear correlation characteristics of the training dataset, a machine learning model is constructed using the gradient boosting tree algorithm. The model takes the raw material ratio and process parameters as inputs, and the macroscopic performance and low-carbon indicators as outputs. The parameter boundaries defined by the multi-scale simulation are used as input constraints.

[0125] Specifically, regarding algorithm selection and model construction: based on the complex nonlinear characteristics in the training dataset, the gradient boosting decision tree algorithm was selected to construct the machine learning model. The input features of this model are clearly defined as: 1. Raw material proportions (such as the mass fraction of cement, slag powder, fly ash, and fiber, and aggregate gradation parameters); 2. Key process parameters (such as water-cement ratio, mixing time, and curing regime number).

[0126] The model's output objectives are clearly defined as: 1. Macroscopic performance indicators (such as 28-day compressive strength and chloride ion diffusion coefficient); 2. Low-carbon indicators (such as carbon emission equivalent per cubic meter of material).

[0127] Before model training, the physical boundaries of each parameter defined through multi-scale simulation in step S200 (e.g., solid waste content ≥60%, water-cement ratio ≤0.22) are encoded as hard constraints and directly applied to the selection of training data and boundary checks during the model prediction stage. Through grid search and cross-validation, the optimal hyperparameter set for the model was determined to be: learning rate 0.05, maximum tree depth 8, and subsample ratio 0.8. Validated on the test set, the model's prediction determination coefficient (R²) for the core performance indicators was [value missing]. 2 All are greater than 0.90.

[0128] Step SC00: Divide the above training dataset into a training set and a validation set according to a preset ratio, input the constructed machine learning model into the training set and perform iterative training. Monitor the prediction error in real time through the validation set, and optimize the model hyperparameters using a grid search method until the error meets the preset accuracy requirements.

[0129] The specific process is as follows: 1. Dataset Partitioning: Divide the training dataset into a training set and a validation set according to a preset ratio. Typically, 60% of the data is used as the training set, 20% as the validation set, and the remaining 20% ​​as the test set. This partitioning ensures the model has sufficient data to learn features and provides an independent validation set for hyperparameter tuning and to prevent overfitting. 2. Model Training: Input the training set into the constructed machine learning model for iterative training. During training, the model continuously adjusts its parameters using optimization algorithms (such as gradient descent) to minimize prediction error. For example, when training a Gradient Boosting Tree (GBDT) model, the model gradually improves prediction accuracy through iterative optimization. 3. Validation Set Monitoring: During training, use the validation set to monitor the model's prediction error in real time. This helps to promptly detect whether the model is overfitting or underfitting. For example, if the error on the validation set starts to increase while the error on the training set continues to decrease, this may indicate that the model is starting to overfit. 4. Grid Search for Hyperparameter Optimization: Use a grid search method to search for the optimal combination of hyperparameters within a predefined hyperparameter space. For example, for the GBDT model, hyperparameters such as the learning rate, tree depth, and subsampling ratio can be adjusted. The performance of different hyperparameter combinations is evaluated on the validation set, and the combination that minimizes the prediction error is selected as the optimal hyperparameters. 5. Model Optimization and Validation: The model is retrained using the optimized hyperparameters and finally evaluated on the test set to ensure the model's generalization ability and prediction accuracy. For example, the model's performance is validated by calculating metrics such as mean squared error (MSE) and mean absolute error (MAE) on the test set. If the model's performance on the test set meets the preset accuracy requirements, the model is considered to have completed training and can be used in practical applications.

[0130] Step SD00: Based on the trained machine learning model, the core target values ​​of high solid waste content, low carbon emissions, and high durability are imported. Under parameter boundary constraints, the genetic algorithm drives the model to carry out multi-objective optimization and outputs a set of candidate solutions for a preset number of raw material ratios and their key process parameters.

[0131] The specific process is as follows: 1. Target value setting and boundary constraints: High solid waste content, low carbon emissions, and high durability are used as core target values ​​and imported into the trained machine learning model. For example, the target for solid waste content is set at 60%, and the target for carbon emissions is 50 kg CO2 / m³. 3The target compressive strength is 220 MPa, and the target chloride ion permeability resistance is 1000°C. Simultaneously, the parameter boundaries defined by multi-scale simulation are used as input constraints to ensure the model does not exceed these boundaries during the optimization process. 2. Genetic Algorithm Initialization: Initialize the genetic algorithm (GA) by setting parameters such as population size, crossover probability, and mutation probability. For example, the population size is set to 100, the crossover probability to 0.8, and the mutation probability to 0.01. Each individual in the population represents a combination of raw material ratios and process parameters. 3. Fitness Function Definition: Define the fitness function to evaluate the quality of each individual. The fitness function combines macroscopic performance and low-carbon indicators, for example: Fitness = w1 × Compressive Strength + w2 × Chloride Ion Permeability Resistance + w3 × Carbon Emissions. Where w1, w2, and w3 are weighting coefficients, set according to actual needs. 4. Genetic Algorithm Iterative Optimization: Utilize the genetic algorithm to drive the model to perform multi-objective optimization. Through selection, crossover, and mutation operations, the population is continuously optimized to improve the fitness value. For example, in each iteration, the individual with the highest fitness is selected for crossover and mutation to generate a new population. This iteration is repeated until a termination condition is met (such as the maximum number of iterations or fitness convergence). 5. Candidate Solution Output: Output a preset number of candidate solutions for raw material proportions and key process parameters. For example, output one set of candidate solutions, where each solution includes raw material proportions (such as the volume fraction of slag powder, fly ash, cement, aggregate, and fiber) and process parameters (such as mixing time, curing temperature, and vibration frequency). These solutions meet the goals of high solid waste, low carbon, and high durability while possessing high fitness values.

[0132] Step SE00: For the candidate solutions output above, filter parameter combinations whose deviations from the performance prediction values ​​and target values ​​are within a preset deviation threshold, verify them by combining historical measured data from the database of special materials for ultra-high performance concrete, retain solutions with errors within a preset range, and determine the comprehensive optimal parameter combination.

[0133] The specific process is as follows: 1. Screening candidate solutions: For the candidate solutions output in step SD00, calculate the deviation between the predicted performance value and the target value for each solution. For example, for a target compressive strength of 220 MPa, if a solution predicts a value of 210 MPa, the deviation is -10 MPa. Set a deviation threshold, such as ±5%, and screen solutions with deviations within this range. 2. Verification using historical data: Verify the screened solutions by combining historical measured data from the database of special materials for ultra-high performance concrete. Compare the actual performance under similar mix proportions and process conditions in the historical data to verify the reliability of the predicted values. 3. Retaining effective solutions: Retain solutions with errors within the preset range to form a preliminary list of optimal parameter combinations. For example, if the key indicators such as compressive strength and carbon emissions of a solution are all within the threshold, then retain that solution. 4. Comprehensive evaluation to determine the optimal combination: Conduct a comprehensive evaluation of the retained solutions, considering multi-objective balance, and determine the final optimal parameter combination. For example, select the solution with the lowest carbon emissions while meeting all performance requirements as the optimal solution.

[0134] A digitally controlled preparation method for a high-solid-waste, low-carbon, and high-durability concrete bonding material also includes:

[0135] In step S600, during the mixing, molding, and curing processes, sensors pre-positioned at each key node of the mixing, molding, and curing processes are used to collect key data and simultaneously upload it to the production control system.

[0136] Specifically, sensors are deployed at key production stages such as mixing, molding, and curing to collect critical process data (such as temperature, humidity, and pressure) and material performance data (such as early strength) in real time, and the data is synchronously uploaded to the production control system. For example, temperature and humidity sensors are installed in the curing room to monitor and upload curing environment data in real time.

[0137] In step S700, the uploaded real-time data is dynamically compared with the preset digital twin model. If the parameter deviation exceeds the preset range, the corresponding process parameters are automatically fine-tuned, and an early warning is issued in real time when the deviation exceeds the preset threshold, so as to realize adaptive control of the production process.

[0138] For details of the process, please refer to steps S710 to S750, which will not be elaborated here.

[0139] Step S800: After curing, the finished product is tested for macroscopic mechanical properties, durability, and low carbon index, and the test data is compared with the preset target values ​​for compliance determination.

[0140] Specifically, after the maintenance is completed, the finished products are tested for macroscopic mechanical properties, durability performance and low-carbon indicators. The test items include compressive strength, chloride ion penetration resistance, carbon emissions, etc. The test data is judged for compliance with the preset target values. For example, if the target value of compressive strength is 220 MPa and the test value is 215 MPa, and the deviation is within the allowable range (±5 MPa), the judgment is passed.

[0141] Step S900, if the judgment fails, immediately identify and isolate the non-conforming products, trace the reasons in combination with the production data and handle them specifically, record the whole process data and feedback it to the database; if the judgment passes, assign a unique digital ID to the factory-out connectors, associate the raw material batches, ratios, processes and test data, and form a traceability file.

[0142] 1. Handling of non-conforming products:

[0143] Immediate identification and isolation: If the test result of the finished product fails to pass the verification, the system immediately identifies the non-conforming products and isolates them. For example, if the test value of compressive strength is lower than the target value, the system automatically marks the products of this batch as non-conforming and moves them to the isolation area.

[0144] Traceability and handling: Trace the reasons in combination with the production data, such as deviations in certain process parameters or raw material quality problems. After specific handling, record the whole process data and feedback it to the database for optimizing subsequent production.

[0145] 2. Traceability file for qualified products:

[0146] Unique digital ID: If the verification passes, assign a unique digital ID to the factory-out connectors, and associate the raw material batches, ratios, processes and test data.

[0147] Form a traceability file: Integrate the whole process data from raw materials to finished products to form a complete traceability file, ensure the traceability of product quality, and provide a guarantee for users' consistency verification.

[0148] It should be further pointed out that the model system constructed in this invention is an intelligent system with continuous evolution capabilities. New data obtained from high-throughput experimental verification (step S400), process data recorded in production adaptive control (step S700), and long-term performance data fed back from connector service monitoring (as described in subsequent steps in the specification) are all synchronously fed back to the ultra-high performance concrete special material database. These new data samples from the real world are periodically used to calibrate the parameters of the multi-scale physicochemical simulation model and to incrementally train the machine learning prediction and optimization model. Through this closed-loop data feedback and model iterative optimization mechanism across the entire "design-production-service" chain, the system can continuously adapt to real-world changes such as raw material batch fluctuations and aging process equipment, making the optimization results of material proportions and process parameters increasingly accurate and robust, thereby achieving continuous self-improvement and performance enhancement of the preparation method.

[0149] The uploaded real-time data is dynamically compared with the preset digital twin model. If the parameter deviation exceeds the preset range, the corresponding process parameters are automatically fine-tuned, including:

[0150] Step S710: Extract the material temperature and viscosity during the stirring stage, the vibration frequency and molding pressure during the molding stage, and the ambient temperature, humidity and steam flow rate during the curing stage from the uploaded real-time data, and compare them with the standard parameter ranges corresponding to each stage in the digital twin model to identify single disturbance or concurrent disturbance scenarios.

[0151] A digital twin model is a dynamic mirror image of a physical production line in the information space. Its core consists of three parts: 1. A "state-performance" mapping sub-model, which is an experimentally verified rapid prediction model from the optimal process parameters finally determined in step S400 to product performance; 2. An equipment response and control sub-model, describing the response characteristics of key equipment such as mixers, molding machines, and curing kilns to control commands; 3. A multi-objective control decision sub-model, embedding adjustment strategies (such as PID control logic and priority rules) based on historical data and expert rules. This model synchronizes data in real time with the production line's sensors and control system through industrial protocols such as OPCUA. During dynamic comparison, the model not only determines whether parameters exceed limits but also simulates and calculates the deviation of the final product performance under the current disturbance, whether no adjustment is made or different adjustment strategies are adopted, thereby generating the optimal control command that simultaneously ensures stable quality and minimizes energy consumption.

[0152] 1. Real-time data extraction: Extract key process parameters from the uploaded real-time data, including: 1.1, stirring stage: material temperature and viscosity; 1.2, molding stage: vibration frequency and molding pressure; 1.3, curing stage: ambient temperature and humidity and steam flow rate.

[0153] 2. Data Comparison: The extracted real-time data is compared with the standard parameter ranges corresponding to each stage in the digital twin model to identify single disturbances or concurrent multiple disturbances: 2.1 Single Disturbance Identification: For example, if only the material temperature exceeds the standard range in the real-time data (standard range is 20-25°C, real-time temperature is 26°C), it is identified as a single disturbance. 2.2 Concurrent Multiple Disturbance Identification: For example, if the material temperature is 26°C in the real-time data, and the molding pressure exceeds the standard range (standard range is 50-60MPa, real-time pressure is 65MPa), it is identified as a concurrent multiple disturbance scenario.

[0154] 3. Deviation Identification: Determines whether real-time data exceeds the allowable range by setting a deviation threshold. For example: 3.1 Material Temperature: Standard range 20-25°C, deviation threshold ±2°C. If the real-time temperature is 26°C, it exceeds the deviation threshold. 3.2 Molding Pressure: Standard range 50-60MPa, deviation threshold ±5MPa. If the real-time pressure is 65MPa, it exceeds the deviation threshold.

[0155] 4. Scene Recognition: Based on the comparison results, the system identifies the types of disturbances in the current production process, providing a basis for subsequent graded control decisions. For example, if both material temperature and molding pressure exceed the standard range simultaneously, the system identifies this as a multi-disturbance concurrent scenario, requiring comprehensive evaluation and handling.

[0156] Step S720: Execute hierarchical control decisions based on the identification results: For a single disturbance, directly call historical process-performance correlation data and model sensitivity analysis results to lock in the core process parameters; for multiple concurrent disturbances, use the analytic hierarchy process (AHP) to construct a multi-scale priority evaluation model, using the microstructure parameters, mesostructure parameters, and macroscopic performance parameters output from the aforementioned multi-scale calculation simulation as performance influence criteria, and combine historical production data from the ultra-high performance concrete special material database to quantify the degree of influence of each disturbance on the multi-scale performance criteria, calculate the comprehensive influence weight of each disturbance parameter, and determine the adjustment priority order.

[0157] 1. Single disturbance handling:

[0158] Access historical data: Directly access historical process-performance correlation data to find which process parameter adjustments effectively resolved the problem under similar single disturbance conditions.

[0159] Model sensitivity analysis: Based on the results of the model sensitivity analysis, identify the core process parameters that have the greatest impact on performance. For example, if the material temperature exceeds the standard range, historical data shows that adjusting the stirring time or cooling system parameters can effectively control the temperature, and the model shows that temperature has a significant impact on early strength, then the stirring time and cooling system parameters are identified as the core adjustment targets.

[0160] 2. Handling multiple disturbances concurrently:

[0161] Evaluation Model Construction: A multi-scale prioritization evaluation model was constructed using the Analytic Hierarchy Process (AHP). Microstructural parameters (such as hydration product density), mesostructural parameters (such as aggregate distribution uniformity), and macroscopic performance parameters (such as compressive strength) were used as performance influence criteria.

[0162] Quantifying the degree of impact: Combining historical production data from the database of special materials for ultra-high performance concrete, the degree of impact of various disturbances (such as abnormal material temperature and abnormal molding pressure) on the above performance criteria is quantified.

[0163] Weight Calculation and Ranking: The combined impact weight of each disturbance parameter is calculated using the AHP model to determine the priority order for adjustment. For example, if the calculated impact weight of abnormal material temperature on macroscopic compressive strength is 0.4, abnormal molding pressure on ...

[0164] Step S730: Based on the output of the hierarchical control decision, formulate a parameter adjustment plan: allocate adjustment actions sequentially based on priority order, prioritize the adjustment of high-priority parameters, and then adjust the secondary parameters one by one.

[0165] The process is described below:

[0166] 1. Based on the adjustment priority order determined in step S720, formulate a detailed parameter adjustment plan. For example, if the priority order is material temperature, molding pressure, and ambient humidity, the adjustment plan is as follows:

[0167] Step 1: Adjust the material temperature by adjusting the cooling system or heating device of the mixer to control the temperature within the standard range (e.g., 20-25°C).

[0168] Step 2: Adjust the molding pressure. Adjust the pressure to the standard range (e.g., 50-60MPa) by adjusting the pressure control system of the molding equipment.

[0169] Step 3: Adjust the ambient humidity by adjusting the humidification equipment in the curing room to control the humidity within the standard range (e.g., 90-95%).

[0170] 2. Serialization of Allocation and Adjustment Actions: Allocation and adjustment actions are serialized according to priority, ensuring that high-priority parameters are adjusted first, followed by lower-priority parameters. For example:

[0171] Priority 1: Material temperature adjustment, to be performed by the mixer operator.

[0172] Priority 2: Molding pressure adjustment, performed by the molding equipment operator.

[0173] Priority 3: Adjustment of ambient humidity, to be performed by the curing room operator.

[0174] 3. Specific implementation of the adjustment actions:

[0175] Material temperature adjustment: If the real-time temperature is 26°C, which is 20-25°C outside the standard range, the temperature can be gradually reduced to the standard range by increasing the power of the cooling system or reducing the power of the heating device.

[0176] Molding pressure adjustment: If the real-time pressure is 65MPa, which exceeds the standard range of 50-60MPa, the pressure will be gradually reduced to the standard range by reducing the output of the pressure controller.

[0177] Environmental humidity adjustment: If the real-time humidity is 85%, which is lower than the standard range of 90-95%, the humidity can be gradually increased to the standard range by increasing the operating power of the humidifier.

[0178] 4. Record the adjustment plan: Record the established parameter adjustment plan in the production control system, including the parameters to be adjusted, the adjustment sequence, and the target values. For example, the recorded adjustment plan is: "Adjust the material temperature to 20-25°C, adjust the molding pressure to 50-60MPa, and adjust the ambient humidity to 90-95%."

[0179] Step S740: Calculate the fine-tuning range of each core process parameter by combining the built-in parameter response equation of the digital twin model with the degree of deviation and priority order; convert the calculated fine-tuning range into equipment control commands and send them to the corresponding process equipment for execution.

[0180] Fine-tuning range calculation: Using the parameter response equations built into the digital twin model, combined with the degree of deviation and priority order, the specific fine-tuning range of each core process parameter is calculated. For example, if the real-time material temperature is 26°C, exceeding the standard range (20-25°C), the model calculates that the cooling system needs to increase its power by 15% to reduce the temperature; if the molding pressure is 65MPa, exceeding the standard range (50-60MPa), the model calculates that the pressure controller output needs to be reduced by 20% to reduce the pressure.

[0181] Equipment control command issuance and execution: The calculated fine-tuning range is converted into specific equipment control commands and issued to the corresponding process equipment for execution. For example, the system sends a command to the cooling system of the mixer to increase the cooling water flow or increase the cooling fan speed; it sends a command to the pressure control system of the molding equipment to adjust the output pressure of the hydraulic pump; and it sends a command to the humidification equipment in the curing room to increase the operating power of the humidifier or the spray frequency. After receiving the command, the equipment executes the adjustment action. The system collects the adjusted process data in real time and feeds it back to the production control system for comparison to ensure that the parameters return to the preset range. If the deviation is still not eliminated, the system repeats the adjustment process until the parameters stabilize within the standard range.

[0182] In step S750, the system collects the adjusted process data in real time and compares it with the model again. If the deviation returns to the preset range, the process data is recorded. If it still exceeds the range, the adjustment process is repeated.

[0183] After passing the conformity assessment, the intelligent control method for preparing a high-solid-waste, low-carbon, and high-durability concrete bonding material further includes:

[0184] Step 1: For qualified connectors, before leaving the factory, sensing elements are precisely embedded in key structural stress areas to ensure tight fit and coordinated deformation with the concrete substrate. Specifically, before leaving the factory, fiber optic gratings or piezoelectric ceramic sensing elements are precisely embedded in key structural stress areas of the connectors (such as the section with the maximum bending moment and the shear concentration area). Positioning molds are used to ensure that the elements are tied to the reinforcing steel frame and poured synchronously with the concrete. After vibration, the sensing elements are made to fit tightly with the substrate and deform in sync, avoiding voids and interface slippage.

[0185] Step 2: Calibrate and adjust the monitoring accuracy of the sensing element, associate it with the unique digital ID of the connector, and enter it into the UHPC database to form an integrated traceability file for production, testing, and sensing. Specifically: After calibrating and adjusting the monitoring accuracy of the sensing element to ±2με / ±0.1°C, associate it with the unique digital ID of the connector via RFID or QR code, and enter it together with the production batch, process parameters, and performance test data into the UHPC database to form an integrated traceability file for production, testing, and sensing.

[0186] Step 3: Establish a real-time monitoring system for the service status of connectors through network access. Specifically, establish a real-time monitoring system for the service status of connectors through NB-IoT or 5G network access to achieve encrypted data transmission and remote access.

[0187] Step 4: During the service life of the connector, the monitoring system collects service status data such as stress, strain, temperature, and vibration frequency at a preset frequency and uploads it to the intelligent operation and maintenance platform in real time. Specifically, during service life, stress, strain, temperature, and vibration frequency data are automatically collected at a preset frequency (e.g., once / hour) and uploaded to the intelligent operation and maintenance platform in real time with encryption.

[0188] Step 5: The intelligent operation and maintenance platform combines the integrated traceability archive to perform real-time analysis of the collected data, determine the health status of the connectors, and implement graded early warning for anomalies. Specifically, the intelligent operation and maintenance platform combines the integrated traceability archive with preset health thresholds to perform real-time analysis of the collected stress, strain, temperature, and vibration data, and uses edge computing to determine the health status of the connectors and trigger three-level anomaly warnings: green, yellow, and red.

[0189] Step 6: Generate targeted preventative maintenance plans based on the analysis results. Specifically, based on the warning level, a targeted preventative maintenance plan is automatically generated, specifying the maintenance cycle, reinforcement measures, and spare parts list to ensure that the situation is addressed before it deteriorates further.

[0190] Step 7 involves transmitting the service data back to the UHPC database, iteratively calibrating the multi-scale computational simulation model and machine learning model, and continuously optimizing the raw material ratio and key process parameters. Specifically, the service data, after being anonymized, is transmitted back to the UHPC database. A transfer learning strategy is used to iteratively calibrate the micro / meso / macro multi-scale computational simulation model and machine learning model, continuously optimizing the raw material ratio and key process parameters.

[0191] The embodiments described in this specific implementation are preferred embodiments of this application and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A method for intelligently controlling the preparation of a concrete bonding material, characterized in that, include: Acquire and integrate basic physical and chemical parameters, historical process parameters and performance data of raw materials to construct a database of concrete-specific materials; Based on a database of concrete-specific materials, a fundamental mapping relationship between raw materials, processes, and performance is constructed through multi-scale computational simulation, and parameter boundaries are defined. Then, a machine learning model is used to perform nonlinear fitting and optimization on the fundamental mapping relationship and boundaries to screen raw material ratios and key process parameters that meet preset solid waste content, preset carbon emissions, and preset durability targets. Based on the screening results, the raw material pretreatment parameters are determined, and the industrial solid waste is activated, the fiber surface is modified, and the aggregate is combined according to the gradation. Based on the pre-treated raw materials, multiple sets of parallel high-throughput experiments were conducted using automated equipment according to key process parameters and preset experimental schemes. Process and performance data were collected through preset monitoring equipment and fed back to the concrete-specific material database. This data was used as new samples to iteratively calibrate the aforementioned machine learning model until the deviation between the model's predicted value and the experimental measured value reached a preset deviation threshold, thereby determining the optimal process parameters suitable for industrial production. The optimal process parameters are imported into the production control system, and the mixing, molding and curing processes are executed according to the preset production flow to prepare concrete bonding materials. During the mixing, molding, and curing processes, sensors pre-positioned at key nodes of mixing, molding, and curing are used to collect key data and upload it synchronously to the production control system. Extract material temperature and viscosity during the mixing stage, vibration frequency and molding pressure during the molding stage, and ambient temperature, humidity and steam flow during the curing stage from the uploaded real-time data. Compare these data with the standard parameter ranges corresponding to each stage in the digital twin model to identify single disturbance or concurrent disturbance scenarios. For a single disturbance: directly call upon historical process-performance correlation data and model sensitivity analysis results to pinpoint the core process parameters; For concurrent multi-disturbance scenarios: A multi-scale priority evaluation model is constructed using the analytic hierarchy process (AHP). The microstructure parameters, mesostructure parameters, and macroscopic performance parameters output from the aforementioned multi-scale calculation simulation are used as performance influence criteria. Combined with historical production data from the concrete special materials database, the influence of each disturbance on the multi-scale performance criteria is quantified, the comprehensive influence weight of each disturbance parameter is calculated, and the adjustment priority order is determined. Adjustment actions are assigned sequentially based on priority, with higher priority parameters adjusted first, followed by lower priority parameters. By using the built-in parameter response equations of the digital twin model, and combining the degree of deviation with the priority order, the fine-tuning range of each core process parameter is calculated; the calculated fine-tuning range is then converted into equipment control commands and sent to the corresponding process equipment for execution.

2. The method for intelligent control and preparation of a concrete bonding material according to claim 1, characterized in that, The fundamental mapping relationship between raw materials, processes, and performance is constructed and parameter boundaries are defined through multi-scale computational simulation, including: Microscale simulation: Using pre-set thermodynamic calculation software, based on the principle of minimizing Gibbs free energy, the stable hydration product phase of the multi-component cementation system is predicted by setting different temperature conditions, with the goal of avoiding the unfavorable phase of calcium hydroxide crystals, and the type, content and microstructure parameters of the hydration products are output. Mesoscale simulation: Based on hydration product data, hydration products are incorporated into the particle system as ultrafine powders. Combining the basic physicochemical parameters of aggregates and powders, the packing density of the system is calculated based on the particle size distribution through a preset particle packing model. The gradation design is simulated and optimized in a digital way, the target porosity is virtually calculated, the optimal packing scheme is determined, and the corresponding mesoscale structure parameters are output. Macroscale simulation: Based on the optimal packing scheme and porosity parameters at the mesoscale, a finite element model reflecting the mechanical properties of the matrix is ​​established. Using finite element analysis technology, the distribution state, orientation characteristics and stress transmission path of fibers in the matrix are simulated, the enhancement effect of fibers on the macroscopic mechanical properties of the material is quantified, and the fiber type and dosage are optimized, and simulation data related to macroscopic properties are output. Construction of basic mapping relationships and delineation of parameter boundaries: Integrating simulation results at the micro, meso, and macro scales, the system connects the logical relationships between raw material characteristics, process parameters, and performance at each scale. Combining historical data accumulated in the concrete special material database, a basic mapping relationship from raw materials to performance is constructed. At the same time, based on the range of parameter changes covered by simulations at each scale, multi-scale parameter boundaries that match the design objectives are delineated in a coordinated manner.

3. The method for intelligent control and preparation of a concrete bonding material according to claim 1, characterized in that, The fundamental mapping relationship between raw materials, processes, and performance is constructed and parameter boundaries are defined through multi-scale computational simulation, including: Microscale simulation: Based on the basic physicochemical parameters of raw materials in the concrete special material database, the hydration reaction kinetics simulation tool is used to simulate the hydration reaction process of the cementitious system, predict the type of hydration products and quantify the density of hydrated calcium silicate gel, control the amount of calcium hydroxide crystal formation to not exceed the preset upper limit to ensure high durability, and output key parameters of hydration product volume ratio and pore distribution. Mesoscale simulation: Using the volume ratio of hydration products and pore distribution output at the microscale as constraints, and combining aggregate characteristic data under high solid waste content in the concrete special material database, the aggregate gradation is optimized and the fiber distribution is evaluated through particle packing model and volume filling model. The particle gap characteristics are quantified, and a quantitative transformation relationship between micro-hydration characteristics and meso-structure is established. The bulk density and fiber spacing variation coefficient are output as meso-structure parameters. Macroscale simulation: Based on mesoscopic structural parameters, using finite element analysis tools and performance mapping equations, a quantitative mapping relationship between mesoscopic structure, process parameters and macroscopic performance is constructed, thereby linking macroscopic mechanical properties, durability and low carbon emission indicators under high solid waste content, and outputting preset thresholds for macroscopic performance and low carbon indicators. Construction of basic mapping relationships and delineation of parameter boundaries: Integrating the quantitative transformation and mapping relationships formed by the three-level simulation of micro-meso-macro, and combining the historical process-performance correlation data in the concrete special material database, we construct the basic mapping relationship of raw materials-process-performance. Simultaneously, based on the parameter range output by each scale simulation, we comprehensively delineate the parameter boundaries of each scale that are suitable for the goals of high solid waste, low carbon, and high durability.

4. The method for intelligent control and preparation of a concrete bonding material according to claim 3, characterized in that, Predicting hydration product types and quantifying the density of hydrated calcium silicate gel, controlling the amount of calcium hydroxide crystal formation to not exceed a preset upper limit to ensure high durability, and outputting key parameters such as the volume ratio of hydration products and porosity distribution: Data on the content of active components in industrial solid waste, specific surface area, and mineral composition of cementitious materials were retrieved from the concrete-specific materials database. Based on the difference in reactivity between industrial solid waste and cementitious materials, an initial parameter matrix for multi-component hydration reaction was established. Based on the initial parameter matrix of the hydration reaction, a hydration reaction model of the high solid waste system is constructed using a hydration reaction kinetic simulation tool. The reaction rate equation is modified by introducing the solid waste activity activation coefficient. The hydration path at different ages is simulated. The types of hydration products, including hydrated calcium silicate and calcium hydroxide, are predicted by calculating the reaction process. The generation rate curves of each product and the characteristic parameters of the stoichiometry and microstructure of hydrated calcium silicate are output. Based on the above characteristic parameters, the apparent density of hydrated calcium silicate gel is calculated using density function theory. The density is then corrected by combining measured data of similar gels from the concrete-specific materials database to obtain a density value with a preset accuracy. With a preset high durability target as a constraint, the upper limit of calcium hydroxide production is determined by using the correlation data between calcium hydroxide content and durability in the concrete special material database. The production amount is monitored in real time during the simulation. When the limit is exceeded, the amount of solid waste admixture or the activity activation coefficient is adjusted in reverse and the hydration reaction model is updated simultaneously. By combining the corrected gel density and the generation rate curves of each product, the volume ratio of hydration products at different ages is calculated. The microstructure evolution caused by hydration reaction is transformed into pore distribution parameters through a pore network model, forming a pore feature matrix that is compatible with mesoscopic simulation, which serves as the underlying constraint for mesoscopic simulation.

5. The method for intelligent control and preparation of a concrete bonding material according to claim 3, characterized in that, Quantitative analysis of interparticle spacing characteristics was conducted to establish a quantitative transformation relationship between microscopic hydration properties and mesoscopic structure. Bulk density and fiber spacing variation coefficient were output as mesoscopic structure parameters, including: Based on the volume ratio of hydration products and pore distribution at the microscale, and combined with aggregate characteristic data under high solid waste content in the concrete special material database, a three-dimensional mesoscopic model of aggregate-fiber-cement system was constructed using the discrete element method, transforming the micropore distribution into the initial gap range between aggregate, solid waste particles and fiber at the mesoscopic level. Based on the above three-dimensional mesoscopic model, the gap size data between aggregate particles and between solid waste particles and fibers are extracted by three-dimensional structural scanning technology. The gap filling requirement is calculated by combining the volume ratio of micro-hydration products, and the average size and distribution uniformity characteristic parameters of particle gaps are obtained. Using the volume ratio of micro-hydration products and pore distribution as independent variables and interparticle spacing characteristic parameters as intermediate variables, the correlation equation between micro-hydration characteristics and mesoscopic structure parameters was fitted by multiple regression analysis. The positive correlation between bulk density and volume ratio of hydration products and the negative correlation between fiber spacing variation coefficient and interparticle spacing uniformity were clarified. Based on the above quantitative transformation relationship, the theoretical values ​​of bulk density and fiber spacing variation coefficient are calculated. The error is calibrated by combining the measured data of the mesoscopic structure of the high solid waste system in the concrete special material database. Finally, the verified bulk density and fiber spacing variation coefficient are output as mesoscopic structure parameters.

6. The method for intelligent control and preparation of a concrete bonding material according to claim 2 or 3, characterized in that, Using machine learning models to perform nonlinear fitting and optimization of basic mapping relationships and boundaries includes: Collect the raw material-process-performance basic mapping relationship data from multi-scale simulation outputs, and combine it with historical formulas, processes and corresponding performance data from the concrete special materials database to form a training dataset that includes raw material proportions, process parameters, characteristic parameters at each scale and target performance. Based on the nonlinear correlation features of the training dataset, a machine learning model is constructed using the gradient boosting tree algorithm. The model takes the raw material ratio and process parameters as inputs, and the macroscopic performance and low-carbon indicators as outputs. The parameter boundaries defined by multi-scale simulation are used as input constraints. The training dataset is divided into training and validation sets according to a preset ratio. The dataset is then input into the constructed machine learning model for iterative training. The prediction error is monitored in real time using the validation set. The model hyperparameters are optimized using a grid search method until the error meets the preset accuracy requirements. Based on the trained machine learning model, the core target values ​​of high solid waste content, low carbon emissions, and high durability are imported. Under the parameter boundary constraints, the genetic algorithm drives the model to carry out multi-objective optimization and outputs a set of candidate solutions for a preset number of raw material ratios and their key process parameters. For the candidate solutions output above, parameter combinations whose deviations from the performance prediction values ​​and target values ​​are within a preset deviation threshold are selected. These combinations are then verified using historical measured data from the concrete special materials database. Solutions with errors within a preset range are retained, and the optimal combination of parameters is determined.

7. The method for intelligent control and preparation of a concrete bonding material according to claim 1, characterized in that, It also includes the following steps: The uploaded real-time data is dynamically compared with the preset digital twin model. If the parameter deviation exceeds the preset range, the corresponding process parameters are automatically fine-tuned, and an early warning is issued in time when the deviation exceeds the preset threshold, so as to realize the adaptive control of the production process. After curing, the finished product is tested for macroscopic mechanical properties, durability and low carbon index, and the test data is judged to be consistent with the preset target values. If the product fails the test, it is immediately identified and isolated. The cause is traced through production data and targeted measures are taken. The entire process data is recorded and fed back to the database. If the product passes the test, a unique digital ID is assigned to each connector, which is linked to the raw material batch, proportion, process and testing data to form a traceability file.

Images