Intelligent control method and system for high-solid-waste low-carbon high-performance grouting material

Abstract

The application relates to a digital intelligent control preparation method and system of high-solid-waste low-carbon high-performance grouting material, solves the problem that a traditional preparation technology lacks a self-controllable digital intelligent control system and has weak adaptation capability to material fluctuation, and the method comprises the following steps: obtaining XRF chemical component and XRD mineral composition data of industrial solid waste materials, and constructing a material gene database; based on the database, a performance prediction model is used to screen a proportioning scheme, working performance, strength development and shrinkage performance are predicted; the prediction result is input into a multi-objective optimization algorithm as a fitness function, and an optimal material gene combination is output; based on the optimal combination, a batching scheme is generated, a stirring process is started after presetting process parameters; data are collected through a real-time monitoring system and compared with a digital twin model; based on the comparison result, material proportioning parameters are automatically adjusted. The application has the following effects: precise design and control of high-solid-waste low-carbon high-performance grouting material are realized, material performance is improved, and carbon emission and cost are reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent manufacturing of building materials, and in particular to a numerical and intelligent control preparation method and system for high-solid-waste low-carbon high-performance grouting material. BACKGROUND

[0002] In precast concrete bridges and modular building structures, the performance of the grouting material in the field connection area directly determines the durability and safety of the overall structure. Therefore, the development of grouting material with high solid waste utilization rate, low carbon and high performance has become an urgent need for industry technology upgrading.

[0003] At present, the mainstream technology of high-performance grouting material still takes high-proportion Portland cement as the core cementitious material. Domestic related research and development focuses on single performance optimization. Although some patents try to improve strength and durability through aggregate grading or nano material modification, the core technology system still has the problem of homogeneity, and lacks a systematic solution to the resource utilization of industrial solid waste. At the same time, international advanced technology relies on special additives and complex control algorithms, which often faces multiple challenges in adapting to the complex raw material conditions and engineering needs in China, such as adaptability, stability and economy.

[0004] The core drawbacks of the prior art are concentrated in two aspects: on the one hand, high cement content leads to high carbon emissions and environmental load, and low utilization rate of industrial solid waste, which not only increases the burden of environmental governance, but also is difficult to meet the performance requirements of green building materials in the market; on the other hand, the traditional preparation technology lacks advanced numerical and intelligent control means, and the adaptability to raw material fluctuations is weak, the precision and reliability of the key performance prediction and optimization model are insufficient, and it is difficult to meet the stringent requirements of major projects on material quality consistency, long-term durability and technological self-reliance. SUMMARY

[0005] In order to realize the precise design and control of high-solid-waste low-carbon high-performance grouting material, improve the material performance, and reduce carbon emissions and costs, the present application provides a numerical and intelligent control preparation method and system for high-solid-waste low-carbon high-performance grouting material.

[0006] In the first aspect, the present application provides a numerical and intelligent control preparation method for high-solid-waste low-carbon high-performance grouting material, which adopts the following technical scheme:

[0007] A numerical and intelligent control preparation method for high-solid-waste low-carbon high-performance grouting material, comprising:

[0008] Obtaining XRF chemical composition and XRD mineral composition data of industrial solid waste raw materials, and constructing a material gene database containing key composition parameters, process parameters and performance indicators;

[0009] Based on the material gene database, a pre-trained performance prediction model is used to perform a virtual matching experiment, and a candidate matching scheme is screened through a pre-set high-throughput virtual calculation process, the working performance, strength development and shrinkage performance are predicted, and the corresponding performance prediction results are output;

[0010] The performance prediction results are input into a multi-objective optimization algorithm as a fitness function, and the optimal material gene combination is output through a reverse search under pre-set multi-objective constraints.

[0011] Based on the optimal material gene combination, an accurate batching scheme is automatically generated in combination with a pre-set batching rule library, and basic process parameters are pre-set according to material properties and target performance requirements, and then the mixing process is started, wherein the pre-set basic process parameters include mixing speed, mixing time, and feeding sequence.

[0012] Through a pre-set real-time monitoring system, the key process parameters in the mixing process are continuously collected, and the collected real-time process data are dynamically compared and analyzed with the pre-set target range in the digital twin model.

[0013] Based on the comparison and analysis results, the amount of additive or mixing water is automatically adjusted through a pre-set intelligent control algorithm to control the real-time process parameters within the target range.

[0014] In the second aspect, the application provides a digital intelligent control preparation system for high-solid-waste low-carbon high-performance grouting material, which adopts the following technical scheme:

[0015] A digital intelligent control preparation system for high-solid-waste low-carbon high-performance grouting material, comprising a memory, a processor, and a program stored in the memory and executable on the processor, which can be loaded and executed by the processor to realize the method for digital intelligent control preparation of high-solid-waste low-carbon high-performance grouting material according to the first aspect. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 It is a flowchart of the method for digital intelligent control preparation of high-solid-waste low-carbon high-performance grouting material according to an embodiment of the application. DETAILED DESCRIPTION

[0017] The application will be further described in detail below with reference to the accompanying drawings.

[0018] Reference Figure 1 The method for digital intelligent control preparation of high-solid-waste low-carbon high-performance grouting material disclosed in the application comprises:

[0019] Step S100: Obtain XRF chemical composition and XRD mineral composition data of industrial solid waste raw materials, and construct a material gene database containing key composition parameters, process parameters and performance indicators.

[0020] XRF (X-ray fluorescence spectroscopy): an analytical technique used to determine the chemical composition of materials by measuring the energy and number of X-ray photons to identify the type and content of elements in the sample. XRD (X-ray diffraction): an analytical technique used to analyze the crystal structure and mineral composition of materials by measuring the diffraction angle and intensity of X-rays in the crystal to determine the mineral composition of the material. Material gene database: a structured data set containing key composition parameters of materials (such as chemical composition, particle size distribution, etc.), process parameters (such as water-cement ratio, curing system, etc.) and performance indicators (such as rheological parameters, strength at each age, shrinkage, etc.).

[0021] The necessary process is as follows: first, obtain the chemical composition data of industrial solid waste raw materials by XRF analysis. After grinding the industrial solid waste sample to the appropriate particle size, put it into the XRF instrument for detection, which will automatically analyze and output the content of each element in the sample. For example, for slag and fly ash, XRF analysis can accurately determine the content of SiO2, Al2O3, Fe2O3, etc. major components. Then, use XRD analysis to determine the mineral composition of industrial solid waste. Put the sample into the XRD instrument, measure the diffraction of X-rays in the sample, analyze its crystal structure and mineral composition. XRD analysis can identify silicate minerals in slag, glass body in fly ash and other mineral phases.

[0022] After obtaining the above data, combine historical experimental data, literature data and high-throughput experimental data to build a material gene database. The database not only contains chemical composition and mineral composition, but also includes process parameters (such as water-cement ratio, curing system) and performance indicators (such as rheological parameters, strength at each age, shrinkage). For example, according to historical experimental data, record the compressive strength and shrinkage of materials under different water-cement ratios to form a complete data set.

[0023] Step S200, based on the material gene database, use the pre-trained performance prediction model to perform the virtual experiment, screen the candidate proportion scheme through the pre-set high-throughput virtual calculation process, predict its workability, strength development and shrinkage performance, and output the corresponding performance prediction results.

[0024] The material gene database is a structured data set containing key composition parameters of materials (such as chemical composition, particle size distribution, etc.), process parameters (such as water-cement ratio, curing system, etc.) and performance indicators (such as rheological parameters, strength at each age, shrinkage, etc.). The necessary process can refer to steps S210 to S250.

[0025] Step S300, input the performance prediction results as the fitness function into the multi-objective optimization algorithm, and output the optimal material gene combination through the reverse search under the pre-set multi-objective constraints. The necessary process can refer to steps S310 to S340.

[0026] At step S400, based on the optimal material gene combination, an accurate batching scheme is automatically generated by combining the preset batching rule library, and the basic process parameters are preset according to the material characteristics and target performance requirements, and then the stirring process is started, wherein the preset basic process parameters include stirring speed, stirring time, and feeding sequence.

[0027] The necessary process is as follows: first, according to the optimal material gene combination, the corresponding batching rule is extracted from the batching rule library. The batching rule library contains specific batching rules under different material combinations, for example, for a combination containing 80% industrial solid waste (such as slag and fly ash) and 20% sulphoaluminate cement, the rule library will provide specific batching ratios, such as slag powder 70%, fly ash 18%, and sulphoaluminate cement 12%. Then, according to the material characteristics and target performance requirements, the basic process parameters are preset. These parameters include stirring speed, stirring time, and feeding sequence, etc.

[0028] For example, for grouting material with high fluidity and high strength requirements, the stirring speed may need to be set to high speed (such as 1000 revolutions per minute), and the stirring time is set to 5 minutes to ensure uniform mixing of the materials. The feeding sequence is also very important, usually dry powder is added first, then liquid activator is added to avoid clumping and uneven mixing. After presetting the basic process parameters, the stirring process is started. The stirring process can be precisely controlled by an automated control system to ensure that each batch of material meets the expected performance. For example, a dedicated high-speed mixer is used to stir at the preset stirring speed and time to ensure the uniformity and fluidity of the material.

[0029] At step S500, the key process parameters during the stirring process are continuously collected through the preset real-time monitoring system, and the collected real-time process data are dynamically compared and analyzed with the preset target range in the digital twin model.

[0030] The necessary process is as follows: first, install and start the real-time monitoring system. The system includes various sensors, such as online rheometers, near-infrared spectroscopy (NIRS) sensors, temperature sensors, and humidity sensors, etc. These sensors can collect key process parameters such as rheological parameters (such as viscosity, yield stress), moisture content, temperature, and humidity of the slurry in real time during the mixing process. For example, using an online rheometer can monitor the viscosity and yield stress of the slurry in real time, ensuring the flowability and uniformity of the slurry. The near-infrared spectroscopy (NIRS) sensor can monitor the moisture content of the slurry in real time to prevent performance problems caused by excessive or insufficient moisture. Then, the real-time collected data is transmitted to the data processing system. The data processing system will dynamically compare and analyze the real-time data with the target range preset in the digital twin model. The digital twin model is a virtual digital model that simulates material performance and process parameters during mixing based on the optimal material gene combination and preset process parameters.

[0031] For example, the target range preset in the digital twin model may include a viscosity range of 10-15 Pa·s, a yield stress of 5-10 Pa, and a moisture content of 15%-20%. The data processing system will compare the collected data with these preset target ranges in real time and analyze whether there is a deviation. If the real-time data deviates from the preset target range, the data processing system will issue an alarm and provide adjustment suggestions. For example, if the real-time monitored viscosity is too high, the system will prompt that the amount of water-reducing agent needs to be adjusted; if the moisture content is too low, the system will suggest increasing the amount of water.

[0032] Step S600, based on the comparison and analysis results, automatically adjust the amount of admixture or mixing water through the preset intelligent control algorithm to control the real-time process parameters within the target range.

[0033] Wherein, "admixture" here mainly refers to chemical additives such as water-reducing agents, air-entraining agents, etc. for improving the performance of concrete; "mixing water" refers to process water that directly participates in the cement hydration reaction and affects the flowability and final strength of the slurry.

[0034] The necessary process is as follows: first, according to the comparison and analysis results, the intelligent control algorithm will evaluate whether the parameters in the current mixing process are within the preset target range. If a parameter deviates from the target range, the algorithm will calculate the amount of adjustment according to the preset rules. For example, if the real-time monitored slurry viscosity is higher than the target range, the algorithm will calculate the amount of water-reducing agent to be reduced or the amount of water to be increased.

[0035] Next, the intelligent control algorithm automatically adjusts the material proportioning parameters during the mixing process through the automated control system. For example, if the addition of water-reducing agent needs to be reduced, the system automatically adjusts the flow rate of the water-reducing agent pump; if the water quantity needs to be increased, the system automatically adjusts the water flow rate. These adjustments are automatically completed by the pre-set control algorithm without human intervention. The intelligent control algorithm is usually based on a feedback mechanism, such as a PID (Proportional-Integral-Derivative) controller. The PID controller calculates the deviation between the current value and the target value, and automatically adjusts the control amount according to the size and trend of the deviation. For example, if the viscosity deviation is large, the PID controller quickly adjusts the addition of water-reducing agent; if the deviation is small, the controller fine-tunes the addition to ensure the stability of the mixing process and the consistency of the performance of the final product.

[0036] Finally, the adjusted parameters are fed back to the digital twin model in real time, updating the state of the model. The digital twin model re-predicts the material performance and process parameters during the mixing process based on the new parameters, forming a closed-loop control system.

[0037] The construction of the material gene database containing key component parameters, process parameters, and performance indicators includes:

[0038] In step S110, the XRF chemical composition and XRD mineral composition data of the industrial solid waste raw material are obtained, and the process parameters and corresponding performance indicator data are simultaneously collected. The multi-source heterogeneous data is normalized and outlier removed to form a structured basic data set.

[0039] The specific process is as follows: First, the chemical composition data of the industrial solid waste raw material is obtained by XRF analysis. After grinding the industrial solid waste sample to an appropriate particle size, it is placed in the XRF instrument for detection. The instrument automatically analyzes and outputs the content of each element in the sample. For example, for slag and fly ash, XRF analysis can accurately determine the content of main components such as SiO2, Al2O3, and Fe2O3. Then, the mineral composition of the industrial solid waste is determined by XRD analysis. The sample is placed in the XRD instrument, and by measuring the diffraction of X-rays in the sample, the crystal structure and mineral composition are analyzed. XRD analysis can identify silicate minerals in slag, glass body in fly ash, and other mineral phases.

[0040] At the same time of obtaining the chemical composition and mineral composition data, the process parameters and corresponding performance indicator data are simultaneously collected. Process parameters include water-binder ratio, curing system, mixing speed, mixing time, etc., and performance indicators include rheological parameters (such as viscosity, yield stress), strength at each age (such as 3-day, 7-day, and 28-day compressive strength), shrinkage, etc. For example, during the experiment, the compressive strength and shrinkage of the material under different water-binder ratios are recorded to form a complete data set.

[0041] Then, the multi-source heterogeneous data is normalized. Since data from different sources may have different dimensions and ranges, normalization can scale these data to fall within a certain small range, such as 0 to 1. For example, viscosity data is normalized to the range of 0 to 1 for subsequent analysis and processing. Next, outlier removal processing is performed. Through statistical analysis methods (such as box plot, Z-score, etc.), outliers in the data are identified and removed. For example, if the compressive strength of a certain sample is significantly higher than that of other samples and has no reasonable explanation, it can be considered as an outlier and removed.

[0042] Finally, the processed data is organized into a structured basic data set. This data set has a unified format and structure, which is convenient for subsequent data analysis and model training. For example, chemical composition, mineral composition, process parameter and performance index data are organized into a table, each row representing a sample and each column representing a parameter or index.

[0043] Step S120, extracting key composition parameter features and process parameter features from the structured basic data set, retaining principal components with cumulative contribution rate ≥ preset contribution rate through principal component analysis algorithm, and establishing an association mapping between feature parameters and performance indicators.

[0044] Among them, the key composition parameter features: the feature parameters extracted from the chemical composition and mineral composition data that have a significant impact on material performance. Process parameter features: feature parameters extracted from process parameter data that have a significant impact on material performance. Principal component analysis (PCA): a statistical method that transforms data into a new coordinate system through linear transformation, so that the variance in the first principal component direction is maximized, representing the most important trend in the data, and the second principal component is next, and so on. Cumulative contribution rate: in principal component analysis, the sum of the variance contribution rates of the first few principal components, used to measure the representativeness of these principal components to the original data. Association mapping: a mathematical relationship between feature parameters and performance indicators, used to predict performance indicators.

[0045] The specific process is as follows: First, extract key composition parameter features and process parameter features from the structured basic data set. Key composition parameter features include chemical composition (such as SiO2, Al2O3, Fe2O3, etc.) and mineral composition (such as silicate minerals, glass body, etc.). Process parameter features include water-binder ratio, curing regime, stirring speed, stirring time, etc. For example, extract SiO2 content of slag powder, glass content of fly ash, water-binder ratio, stirring speed, etc. from the data set.

[0046] Next, principal component analysis (PCA) is performed on the extracted feature parameters. PCA is a dimensionality reduction technique that transforms data into a new coordinate system through linear transformation, maximizing the variance along the first principal component, which represents the most significant trend in the data. The second principal component has the largest variance, and so on. PCA can convert multiple related feature parameters into a few principal components, reducing data dimensionality while retaining the main information. Before performing PCA, the data needs to be standardized so that the mean of each feature parameter is 0 and the standard deviation is 1. For example, data such as the SiO2 content of slag powder and the vitreous content of fly ash are standardized. Then, the variance contribution rate of the principal components is calculated, and principal components with a cumulative contribution rate greater than or equal to a preset contribution rate are selected. For example, if the preset contribution rate is 85%, principal components with a cumulative contribution rate of 85% are selected. These principal components can represent the main information of the original data while reducing data dimensionality. Finally, a mapping between feature parameters and performance indicators is established. Mathematical relationships between principal components and performance indicators are established using regression analysis, machine learning algorithms, and other methods. For example, a linear regression model can be used to establish the relationship between principal components and 28-day compressive strength, or a neural network model can be used to establish the relationship between principal components and mobility.

[0047] Step S130: When the key components of a new batch of solid waste exceed the preset component fluctuation warning threshold, a small-batch verification experiment is automatically triggered, and the material genome database is updated in real time after the verification results meet the standards.

[0048] The necessary procedures are as follows: 1. Set key component fluctuation warning thresholds: Based on historical data and experimental experience, determine the reasonable fluctuation range for each key component. For example, for the SiO2 content of slag powder, set the fluctuation range to ±2%. These thresholds can be adjusted according to actual production needs and material performance requirements. 2. Detect key components of the new batch of solid waste: Use analytical techniques such as XRF and XRD to obtain chemical composition and mineral composition data of the new batch of solid waste. For example, if the SiO2 content of the new batch of slag powder is detected to be 48%, while the average value of historical data is 50%, with a fluctuation range of ±2%, then the SiO2 content of the new batch exceeds the warning threshold. 3. Automatically trigger small-batch verification experiments: When the key components of the new batch of solid waste exceed the preset fluctuation warning thresholds, the system automatically triggers small-batch verification experiments. The purpose of the small-batch verification experiments is to verify whether the performance of the new batch of solid waste meets the requirements. The experimental content includes, but is not limited to:

[0049] Workability test: Test the initial flowability and flowability loss rate of the grout prepared from the new batch of solid waste.

[0050] Strength test: Test the compressive strength and flexural strength of the new batch of grout prepared from solid waste.

[0051] Durability test: Test the anti-chloride ion permeability and sulfate erosion resistance of the grout prepared from the new batch of solid waste. For example, when conducting small batch verification experiments, a sample of grout prepared from a new batch of solid waste is prepared, and the initial fluidity is 300 mm, the fluidity loss rate is 10% within 60 minutes, the 28-day compressive strength is 120 MPa, and the anti-chloride ion permeability meets the design requirements.

[0052] 4. Real-time update of material gene database after verification results meet the requirements: If the verification results meet the requirements, i.e., the performance of the new batch of solid waste meets the preset performance requirements, the data of the new batch of solid waste is immediately updated to the material gene database. The update content includes the chemical composition, mineral composition, process parameters and performance indicators of the new batch of solid waste. For example, the SiO2 content, glass content, water-binder ratio, stirring speed and other data of the new batch of slag powder are updated to the database.

[0053] Step S140, based on the updated material gene database, when inputting new solid waste data from different production places, a deep neural network transfer learning algorithm based on attention mechanism is used to automatically fine-tune the model parameters to adapt to the new solid waste data from different production places. The specific process can be referred to steps S141 to S144, which will not be repeated here.

[0054] When inputting new solid waste data from different production places, a deep neural network transfer learning algorithm based on attention mechanism is used to automatically fine-tune the model parameters to adapt to the new solid waste data from different production places, which includes:

[0055] Step S141, collect the geographical information, production process parameters, environmental conditions and historical fluctuation data of the new solid waste from different production places, and use a preset multi-dimensional quantification processing method to obtain the quantified feature vector.

[0056] Among them, the multi-dimensional quantification processing method is a method for converting different types of non-numerical data into numerical feature vectors, so as to facilitate subsequent analysis and processing.

[0057] The necessary procedures are as follows: 1. Collecting geographical information of new solid waste production site, including geographical position related data such as latitude, longitude, and altitude. These information can be obtained through geographic information system (GIS) or global positioning system (GPS). For example, the latitude and longitude of the new production site are (39.9042, 116.4074), and the altitude is 43.5 meters. 2. Collecting production process parameters: production process parameters include process parameters in the production process of new solid waste, such as temperature, pressure, and additive types. These parameters can be obtained through sensors and monitoring systems in the production process. For example, the temperature in the production process is 1200°C, the pressure is 1.5 MPa, and the additive is a certain specific chemical agent. 3. Collecting environmental conditions: environmental conditions include the environmental conditions of the new production site, such as humidity, temperature, and atmospheric pressure. These conditions can be obtained through environmental monitoring equipment. For example, the humidity of the new production site is 60%, the temperature is 25°C, and the atmospheric pressure is 101.3 kPa. 4. Collecting historical fluctuation data: historical fluctuation data includes the fluctuation data of the composition and performance of new solid waste in the historical production process. These data can be obtained through historical production records and quality detection reports. For example, the SiO2 content of new solid waste in the historical production process fluctuates in the range of 45%-50%, with an average value of 47.5%. 5. Multidimensional quantification processing method: using the preset multidimensional quantification processing method, the collected geographical information, production process parameters, environmental conditions and historical fluctuation data are converted into numerical feature vectors. The specific method includes:

[0058] Geographical information quantification: latitude and longitude and altitude are directly used as numerical features. Production process parameter quantification: normalize temperature, pressure and other parameters to the range of 0-1. Environmental condition quantification: normalize humidity, temperature, atmospheric pressure and other parameters to the range of 0-1. Historical fluctuation data quantification: convert the fluctuation range and average value of composition and performance into numerical features.

[0059] Step S142, the quantized feature vector and the internal features of solid waste are fused through the gated recurrent unit network to generate a composite feature vector containing context information.

[0060] The necessary process is as follows: first, the quantitative feature vector of the new origin solid waste, the production process parameters, the environmental conditions and the historical fluctuation data is spliced with the internal feature vector of the solid waste itself, such as chemical composition and mineral composition, to form a comprehensive feature vector. This comprehensive feature vector integrates the multi-dimensional information of solid waste, providing a comprehensive data basis for subsequent fusion processing. Next, a gated recurrent unit (GRU) network is constructed and initialized. The GRU network can effectively process sequence data through the gating mechanism of the update gate and the reset gate, avoiding the problem of gradient disappearance. The update gate controls the updating degree of new information, while the reset gate controls the retention degree of old information, and the two work together to optimize the transmission and updating of information. Then, the comprehensive feature vector is input into the GRU network. The network updates the hidden state step by step through the control of the update gate and the reset gate. This process not only retains the key information of the internal features of the solid waste, but also integrates the context information of the external environment and historical fluctuations, so that the output composite feature vector can more comprehensively describe the characteristics of the solid waste. Finally, the GRU network outputs a composite feature vector containing context information. This composite feature vector integrates multi-modal information, providing richer feature representations for subsequent model fine-tuning and performance evaluation.

[0061] In step S143, a deep neural network based on an attention mechanism is used, the weights of the bottom layer general feature extraction layer are frozen, and a preset similarity matching algorithm is used to automatically select a sample subset with the closest component features in the historical database to guide the fine-tuning of the top layer origin adaptation parameter layer. The fine-tuning process uses preset fine-tuning parameters to ensure that the parameter adjustment range is controlled within a preset range.

[0062] The necessary process is as follows: first, a deep neural network model based on an attention mechanism is constructed. The model includes multiple layers, with the bottom layer being a general feature extraction layer for extracting basic features of solid waste data, and the top layer being an origin adaptation parameter layer for fine-tuning solid waste data for a specific origin. In the early stage of model training, the weights of the bottom layer general feature extraction layer are frozen to maintain its ability to learn general feature extraction on historical data. For example, the general feature extraction layer can include multiple convolutional layers and pooling layers to extract spatial features and texture information from solid waste data. Freezing the weights of these layers ensures that the model does not lose important feature extraction capabilities learned from historical data when processing new origin solid waste data. Next, a preset similarity matching algorithm is used to automatically select a sample subset with the closest component features to the new origin solid waste from the historical database. This process is achieved by calculating the similarity between the feature vector of the new origin solid waste and the feature vector of each sample in the historical database. For example, cosine similarity or Euclidean distance can be used to measure the similarity between feature vectors. Assuming that the feature vector of the new origin solid waste is v new , and the feature vector of the sample in the historical database is v histThe similarity can be represented as similarity = cos(v new ,v hist ). By calculating the similarity, the top K samples closest to the new production site solid waste characteristics are selected as the reference subset. Then, according to the selected sample subset, the top-level production site adaptation parameter layer is fine-tuned. The preset fine-tuning parameters are used during the fine-tuning process, including learning rate, iteration number, etc., to ensure that the parameter adjustment range is controlled within the preset range. For example, the learning rate can be set to 0.001, and the iteration number can be set to 100. During the fine-tuning process, the data of these sample subsets is used to update the top-level parameters, so that they can better adapt to the characteristics of new production site solid waste. In this way, the model can maintain the ability of bottom-level general feature extraction while effectively adjusting to the characteristics of new production site solid waste data, thereby improving the adaptability and prediction accuracy of the model to new production site solid waste data.

[0063] Step S144, evaluate the performance of the fine-tuned model by the preset accuracy standard, if the performance is not up to standard, trigger the incremental learning process, supplement external perturbation experimental data and re-tune, re-tune includes updating model parameters using incremental learning algorithm, and re-evaluate until the standard is met.

[0064] The necessary process is as follows: first, set the preset accuracy standard, which includes but is not limited to the prediction accuracy, recall rate, F1 score, etc. of the model. For example, the prediction accuracy of the model can be set to not less than 90%, and the recall rate can be set to not less than 85%. These standards are used to evaluate whether the performance of the fine-tuned model meets the requirements of actual application. Next, use the test data set to evaluate the fine-tuned model. The test data set should contain diversified samples of new production site solid waste to ensure the performance of the model under different conditions. During the evaluation process, calculate the performance indicators of the model, such as accuracy, recall rate, F1 score, etc., and compare them with the preset accuracy standard. If the evaluation result shows that the performance of the model does not meet the preset accuracy standard, for example, the prediction accuracy is only 88%, which is lower than the set standard of 90%, then trigger the incremental learning process. The incremental learning process includes the following steps:

[0065] Supplement external perturbation experimental data: collect additional experimental data, which should contain the performance data of new production site solid waste under different conditions, such as different temperatures, humidities, pressures, etc. These data can provide more information to help the model better adapt to the characteristics of new production site solid waste.

[0066] Updating model parameters using incremental learning algorithm: Add the supplementary external disturbance experimental data to the training data set, and retrain the model using the incremental learning algorithm. Incremental learning algorithm can gradually update the model parameters without retraining the entire model, so that it can adapt to new data. For example, online learning algorithm or small batch gradient descent algorithm can be used for incremental learning.

[0067] Re-evaluate model performance: Use the updated model to re-evaluate the performance, calculate the performance indicators, and compare with the preset accuracy standard. If the model performance still does not meet the standard, repeat the above incremental learning process until the model performance reaches the preset accuracy standard.

[0068] Based on the material gene database, the pre-trained performance prediction model is used to perform the matching experiment, and the candidate matching scheme is screened through the preset high-throughput virtual calculation process, and the working performance, strength development and shrinkage performance are predicted, including:

[0069] Step S210, retrieve the stored industrial solid waste XRF chemical composition, XRD mineral composition and particle size data from the material gene database, and generate the solid waste activity prediction results and grading optimization conclusions simultaneously through the preset solid waste comprehensive characteristic analysis algorithm; at the same time, the external input engineering actual demand is quantified into corresponding performance constraint indicators through the preset engineering demand conversion method.

[0070] The necessary process is as follows: in step S210, first, the XRF chemical composition, XRD mineral composition and particle size data of industrial solid waste are retrieved from the material gene database. These data are the basis for evaluating the activity of solid waste and optimizing the particle size distribution. For example, for a new batch of slag powder, the data stored in the database may include SiO2 content 48%, Al2O3 content 20%, Fe2O3 content 15%, and D50 particle size 10 μm, D90 particle size 30 μm, etc. These detailed data provide a rich information base for subsequent analysis. Then, the extracted data is analyzed using a preset comprehensive characteristic analysis algorithm of solid waste. This algorithm predicts the potential reaction activity of solid waste by calculating the activity index, and optimizes the particle packing density using the particle size distribution data. For example, the activity index formula may be 0.6 x SiO2 content + 0.3 x Al2O3 content + 0.1 x Fe2O3 content, and the activity index of slag powder calculated by this formula is 0.348, indicating that it has medium activity. At the same time, the Andreasen & Andersen model is used to optimize the particle size distribution to maximize the packing density, thereby improving the density and performance of the material. At the same time, the external input of engineering actual demand is quantified into specific performance constraint indicators through a preset engineering demand conversion method. For example, the engineering demand may include 28-day compressive strength ≥ 100 MPa, initial fluidity ≥ 280 mm and 60-minute fluidity loss ≤ 20%, etc. These requirements are converted into specific performance constraint indicators, such as 28-day compressive strength target value 100 MPa, allowable error range ± 5 MPa; initial fluidity target value 280 mm, allowable error range ± 10 mm, etc.

[0071] In step S220, the above activity prediction results, grading optimization conclusions and quantified performance constraint indicators are mapped into the precise range of grouting material core formula parameters through a preset multi-dimensional parameter correlation algorithm.

[0072] The necessary process is as follows:

[0073] The core of this step is to execute the "multi-dimensional parameter correlation algorithm". This algorithm is a trained optimization decision model that encapsulates the intrinsic relationship between material components, structure and performance. Its workflow is not linear superposition, but takes the activity prediction results, grading optimization state and performance constraint indicators as input variables, and performs collaborative search and optimization solution in a preset solution space.

[0074] Firstly, the algorithm takes the solid waste activity prediction result as the input of the reaction kinetics dimension. For example, the activity index 0.348 output by the analytical algorithm is matched with the "activity-maximum effective content" function preset in the algorithm. The function shows that at this activity level, the effective upper limit of the solid waste content is about 85% to achieve the target hydration degree; at the same time, combined with the strength model, the effective lower limit is about 80%. Therefore, the algorithm calculates and outputs the feasible interval of the solid waste content as 80-85%. Secondly, the algorithm reads the grading optimization conclusion as a fixed constraint of the physical structure dimension. When the conclusion shows that the current particle size distribution is optimal, the algorithm takes it as an unchangeable boundary condition, locks the particle grading parameters in subsequent optimization, and ensures the compactness of the material skeleton. Finally, the algorithm takes the quantitative performance constraint index as the boundary condition of the multi-objective optimization problem. In the face of the pair of mutually restrictive targets "high strength (≥100 MPa)" and "high fluidity (≥280 mm)", the algorithm starts the internal conflict resolution mechanism (for example, based on the Pareto frontier search or weighted penalty function method). The mechanism will solve the water-binder ratio and superplasticizer dosage combination that meets all performance constraints under the determined solid waste content interval and fixed grading. Through iterative calculation, the algorithm finally converges and outputs the precise range of key parameters, such as water-binder ratio 0.35-0.40 and superplasticizer dosage 1.5%-2.0%.

[0075] In summary, step S220 realizes the precise and automatic mapping from multi-source input information to executable formula parameters by revealing the collaborative optimization process of the "multi-dimensional parameter correlation algorithm".

[0076] Step S230 takes the precise range of the core formula parameters of the grouting material as the input variables and level boundaries of the preset orthogonal combination algorithm, and generates a set of tentative mixing ratio schemes covering the potential optimal performance space by systematically covering the parameter combinations and strengthening the key intervals through the preset orthogonal combination algorithm.

[0077] The specific process can be referred to steps S231 to S235, which will not be repeated here.

[0078] Step S240, the preset performance prediction model is used to batch predict the tentative mixing ratio scheme set, and output the prediction data of each scheme, wherein the prediction data includes workability, strength development and shrinkage performance data, and the preset performance prediction model is trained based on the material gene database with hydration kinetics equation and rheological constitutive relation as physical constraints.

[0079] The specific process is as follows: this step uses a preset performance prediction model to perform batch prediction on the set of candidate matching schemes. The model is an intelligent model trained based on a material gene database with physical constraints of hydration kinetics equations and rheological constitutive relations, and the output is prediction data of each scheme, including workability, strength development, and shrinkage performance data.

[0080] The physical kernel relied on by the performance prediction model includes hydration kinetics equations and rheological constitutive relations. Hydration kinetics equations describe the reaction rate and degree of cement and solid waste active components, and are the theoretical basis for predicting strength development at each age. For example, the extended Krstulović-Dabić model can be used to simulate and predict the strength growth curve. Rheological constitutive relations are used to characterize the flow and deformation law of the slurry, which is the key to predicting workability. For example, the modified Bingham model or Herschel-Bulkley model can be used to predict the flowability, yield stress, and viscosity changes under different proportions.

[0081] When performing batch prediction, the formula parameters of each scheme in the set of candidate matching schemes are input into the performance prediction model. The model calls its physical kernel and learned data rules to perform parallel calculation, and outputs the key performance prediction data corresponding to each scheme. For example, for a candidate proportion, the model can output a series of prediction results such as an initial flowability of 285 mm, a 60-minute flowability loss of 18%, a 3-day compressive strength of 50 MPa, a 28-day compressive strength of 120 MPa, and a final shrinkage rate of 0.035%.

[0082] The construction and training process of the performance prediction model is as follows: 1. Model architecture: a multi-layer feedforward neural network based on attention mechanism is adopted, the number of nodes in the input layer corresponds to the number of key feature parameters extracted from the material gene database (such as chemical composition, mineral composition, particle size distribution, etc.), the output layer includes workability indicators (fluidity), strength development indicators (3-day and 28-day compressive strength) and shrinkage performance indicators (shrinkage). The hidden layer is set to 3 layers, the number of neurons in each layer is 128, 64 and 32 respectively, and the activation function uses ReLU. 2. Training data: use historical experimental data in the material gene database as the training set, 80% for training and 20% for verification. Standardize the input features and normalize the output performance indicators. 3. Training process: use Adam optimizer, initial learning rate is set to 0.001, batch size is 32, training rounds (epoch) is 200. The loss function uses mean square error (MSE), and introduces physical constraints based on hydration kinetics equation as regularization term to improve the physical interpretability and extrapolation ability of the model. 4. Model verification: after training, use independent test data set to verify the prediction accuracy of the model, the average absolute error (MAE) between predicted and measured values is controlled within the preset range (such as fluidity error ≤5mm, strength error ≤5%), which is considered as model training completed.

[0083] The performance prediction model is not a general neural network, but a physical information neural network specially designed for cementitious material system. Its particularity lies in:

[0084] Engineering design of input features: In addition to chemical composition (SiO2, Al2O3, etc.) and mineral composition, the model input also includes glass content index calculated from solid waste XRD data and packing density parameter calculated based on particle size distribution. These features are engineering parameters known to those skilled in the art that are strongly related to the hydration activity and rheological properties of grouting material, ensuring the physical meaning of model input.

[0085] Embedding of physical constraints: In the loss function, not only the error term between predicted and experimental values, but also the residual term of hydration kinetics equation is added as regularization constraint. For example, the model predicted 3-day strength and the theoretical hydration degree calculated based on chemical composition must satisfy the mathematical relationship derived by the model. This forces the neural network to learn the mapping that conforms to the physical law, greatly improving the extrapolation prediction reliability when facing new solid waste composition that does not appear in the training data.

[0086] Verifiable association of output results: After training, the model reveals through feature importance analysis (such as SHAP value) that the features that contribute most to the prediction of 28-day strength are CaO / SiO2 mass ratio, D50 particle size and glass content in turn. This analysis result is consistent with the classical theory of materials science, proving that the model is not a black box.

[0087] Step S250, start the preset high-throughput virtual calculation process to perform parallel simulation verification on the prediction data, and finally form the performance prediction results including the workability, strength development and shrinkage performance corresponding to each matching scheme.

[0088] The necessary process is as follows: 1. Start the high-throughput virtual calculation process: the high-throughput virtual calculation process uses a high-performance computing cluster (such as a GPU or a distributed computing system) to perform parallel simulation on each scheme in the set of matching schemes. The prediction data of each scheme (including workability, strength development and shrinkage performance) is input into the virtual calculation model for detailed simulation verification. 2. Parallel simulation verification: in the parallel simulation process, the prediction data of each scheme is processed independently to ensure computing efficiency and resource utilization. For example, multiple GPU nodes are used to process different matching schemes simultaneously, with each node running an independent simulation task. These tasks may include: 2.1, workability simulation: simulate the flow behavior of the slurry through rheological models (such as the Bingham model) to verify the predicted initial fluidity and fluidity loss. 2.2, strength development simulation: simulate the cement hydration process through hydration kinetics models (such as the Blaine formula) to verify the predicted compressive strength at each age. 2.3, shrinkage performance simulation: simulate the shrinkage behavior of the material through physical models (such as chemical shrinkage and drying shrinkage models) to verify the predicted final shrinkage. 3. Data collection and analysis: simulation results are collected and stored in real time to form a large database containing performance data for all matching schemes. These data are analyzed and compared in detail to ensure the consistency of the prediction results of each scheme with the simulation results. For example, for a specific matching scheme, the simulation results may show an initial fluidity of 285 mm, a 60-minute fluidity loss of 18%, a 3-day compressive strength of 50 MPa, a 28-day compressive strength of 120 MPa, and a final shrinkage of 0.035%. 4. Form a precise prediction report: finally, according to the simulation verification results, a precise prediction report is formed containing the workability, strength development and shrinkage performance corresponding to each matching scheme. The report lists the prediction data and simulation verification results of each scheme in detail, and evaluates the performance of each scheme. For example, the report may show that the prediction data of a certain scheme is highly consistent with the simulation verification results, indicating that the scheme has high reliability; while the prediction data of another scheme has a large deviation from the simulation verification results, and needs to be further adjusted and optimized.

[0089] The precise range of core formulation parameters of the grouting material is taken as the input variables and level boundaries of the preset orthogonal combination algorithm, and the parameter combinations are systematically covered and key intervals are strengthened through the preset orthogonal combination algorithm, to generate a set of matching schemes covering the potential optimal performance space, including:

[0090] Step S231, according to the preset parameter level gradient rule, the horizontal gradient in the accurate range of the core formulation parameter of the grouting material is divided.

[0091] The specific process is as follows: 1. Determine the accurate range of the core formulation parameter: in step S220, we have determined the accurate range of the core formulation parameter of the grouting material by the multi-dimensional parameter association algorithm. For example, the solid waste content range is 80%-85%, the water-binder ratio range is 0.35-0.40, and the water-reducing agent content range is 1.5%-2.0%. 2. According to the preset rule, the level is divided: according to the preset parameter level gradient rule, the representative levels in the value range of each parameter are divided. For example, for the solid waste content, the level gradient can be set to 80%, 82%, 84%, and 85%; for the water-binder ratio, the level gradient can be set to 0.35, 0.37, 0.39, and 0.40; for the water-reducing agent content, the level gradient can be set to 1.5%, 1.7%, 1.8%, and 2.0%. 3. Generate level gradient combinations: combine the levels of the above parameters to form a set of tentative mixing ratio schemes for subsequent high-throughput virtual screening.

[0092] Step S232, call the preset orthogonal table matching algorithm, match the corresponding preset orthogonal table based on the number of parameters and the number of level gradients, generate a basic combination set covering the average level matching of each parameter, and realize the systematic coverage of parameter combination.

[0093] The specific process is as follows: 1. Determine the number of parameters and the number of level gradients: based on the parameter level gradient divided in step S231, count the total number of parameters and the number of levels of each parameter. For example, the solid waste content, water-binder ratio, and water-reducing agent content are 3 parameters, and each parameter has 4 levels (such as solid waste content: 80%, 82%, 84%, and 85%). 2. Match and select the corresponding orthogonal table: according to the number of parameters and the number of levels, call the preset orthogonal table matching algorithm, which automatically matches the applicable orthogonal table according to the built-in standard orthogonal table library and the principle of the least number of experiments. For example, for 3 parameters and 4 levels, L 16 (4 3)orthogonal table, indicating 16 experiments, can accommodate 3 4-level parameters. 3, generate a basic combination set: according to the structure of the selected orthogonal table, map the level number corresponding to each column in the table back to the specific level value of each parameter defined in step S231, so as to generate a set of experimental combinations by the system. Each row of the orthogonal table constitutes a specific matching ratio scheme. 4, realize the systematic coverage of parameter combination: through orthogonal table design, ensure that all levels of each parameter appear equally in the experimental combination, and that the different levels of any two parameters can achieve balanced matching. This method only needs to perform 16 times of simulation calculation (in this example), which can systematically cover the parameter space of 64 times of full experiment, greatly improving the screening efficiency while ensuring the representativeness of the results.

[0094] Step S233, identify the parameter interaction key interval which has significant influence on working property, strength development and shrinkage performance through the preset key interval judgment algorithm.

[0095] The specific process is as follows: 1, define the key interval judgment algorithm: the key interval judgment algorithm is based on statistical analysis and machine learning method, and models and analyzes the "parameter combination-performance prediction result" data set generated in step S232, to identify the influence of parameter interaction on material performance. For example, methods such as analysis of variance (ANOVA) or random forest can be used to evaluate the influence of each parameter and its interaction on working property, strength development and shrinkage performance. 2, evaluate the influence of parameter interaction: use the preset key interval judgment algorithm to quantitatively evaluate the above data set. For example, through ANOVA analysis, the significance level (p value) of each parameter and its interaction can be calculated. If the p value of a parameter or parameter interaction is less than the preset significance level (such as 0.05), it is considered that the parameter or parameter interaction has a significant influence on the material performance. 3, identify the key interval: for the identified significant interaction, further analyze its effect trend combined with the regression model or response surface model. According to the contour map or partial derivative analysis of the model, the parameter combination area with the most dramatic change in performance response (such as strength, fluidity) is located, so as to identify the key interval. For example, it is found through analysis that the interaction of solid waste content and water-binder ratio has a significant influence on 28-day compressive strength, and model analysis shows that the interaction has the strongest positive synergistic effect on strength when the solid waste content is 82%-84% and the water-binder ratio is 0.36-0.38. Therefore, this interval can be identified as the key interval. 4, generate a key interval report: record the identified key interval to form a key interval report. For example, the report may show:

[0096] Working property key interval: the interaction interval of solid waste content 82%-84% and water reducing agent content 1.7%-1.9% has a significant positive influence on initial fluidity.

[0097] The key interval of strength development: the interaction interval of solid waste content 82%-84% and water-binder ratio 0.36-0.38 has a significant positive effect on the 28-day compressive strength.

[0098] The key interval of shrinkage performance: the interaction interval of water-binder ratio 0.36-0.38 and water-reducing agent content 1.7%-1.9% has a significant effect on inhibiting the final shrinkage rate.

[0099] Step S234, for the identified key interval, the parameter level gradient number in the interval is increased according to the preset reinforcement gradient rule, and the reinforcement combination set of the key interval is generated by the orthogonal combination algorithm.

[0100] The specific operation is as follows: 1, positioning the key interval: inherit the key interval report output in step S233 to determine the parameter interaction region that needs to be explored. For example, the report indicates that the key interval is: solid waste content 82%-84%, water-binder ratio 0.36-0.38, and water-reducing agent content 1.7%-1.9%. 2, implement gradient encryption: according to the preset reinforcement gradient rule, the parameter level in the key interval is encrypted. The rule usually uses equal interval or non-equal interval based on the curvature of the response surface to increase the level number. For example:

[0101] The solid waste content in the interval of 82%-84% is encrypted as: 82.0%, 82.7%, 83.3%, and 84.0%.

[0102] The water-binder ratio in the interval of 0.36-0.38 is encrypted as: 0.360, 0.367, 0.373, and 0.380.

[0103] The water-reducing agent content in the interval of 1.7%-1.9% is encrypted as: 1.70%, 1.77%, 1.83%, and 1.90%.

[0104] 3, generate reinforcement combination set: call the preset orthogonal combination algorithm, and automatically match and construct the applicable orthogonal experimental design for the encrypted level number (in this example, three factors are 4 levels). For example, use the L 16 (4 3 ) orthogonal table, the system generates 16 groups of reinforced experimental combinations with balanced parameter levels in the key interval, so as to realize more detailed performance response surface description in a local range.

[0105] Step S235, merge the basic combination set and the reinforcement combination set, remove invalid schemes by the preset redundancy elimination algorithm, and finally generate the quasi-proportioning scheme set covering the potential optimal performance space.

[0106] The specific operation is as follows: 1, merging set: merging the basic combination set generated in step S232 and the strengthened combination set generated in step S234 to form an initial comprehensive matching scheme set. For example, the basic set contains 16 groups of orthogonal schemes, and the strengthened set contains 9 groups of encrypted schemes, and after merging, an initial set containing 25 groups of schemes is formed. 2, redundancy elimination: calling a preset redundancy elimination algorithm, the similarity of the schemes in the merged set is screened. The algorithm calculates the Euclidean distance or Manhattan distance of the parameter vectors between the schemes to judge their similarity. If the distance of two schemes is less than a preset threshold (for example, the normalized distance <0.05), it is considered as redundancy, and only one representative scheme is retained. 3, invalid scheme filtering: based on the core parameter accurate range and performance constraint boundary determined in step S220, an invalid scheme filtering algorithm is called to automatically identify and eliminate schemes whose parameter combinations exceed the feasible region or obviously violate the process feasibility. For example, invalid combinations with water-binder ratio >0.40 or solid waste content <80% are automatically removed. 4, generating the final set: after removing the redundancy and filtering, the optimized final matching scheme set is output. The set controls the overall simulation calculation scale while ensuring balanced and efficient coverage of the global design space and local key areas, providing high-quality input for subsequent high-throughput virtual calculation.

[0107] The performance prediction results are input into the multi-objective optimization algorithm as the fitness function, and the optimal material gene combination is output through reverse search under the preset multi-objective constraints.

[0108] In step S310, according to the preset fitness function construction rule, the workability, strength development, and shrinkage performance data in the performance prediction results are taken as core inputs to construct a multi-objective fitness function.

[0109] The specific operation is as follows:

[0110] 1. Select and quantify performance indicators:

[0111] Workability: taking initial fluidity and 60-minute fluidity loss rate as core indicators. The target value can be set as: initial fluidity 280 mm (±10 mm), 60-minute fluidity loss rate ≤20% (±2%).

[0112] Strength development: taking 3-day and 28-day compressive strength as core indicators. The target value can be set as: 3-day compressive strength 50 MPa, 28-day compressive strength 100 MPa (allowable error ±5 MPa).

[0113] Shrinkage performance: taking final shrinkage rate as core indicator. The target value can be set as: ≤0.04% (±0.01%).

[0114] 2. Construct fitness function:

[0115] Normalization: The actual predicted values of each performance indicator are normalized to the interval [0, 1], with 1 indicating complete achievement or better than the target, and 0 indicating far below the target.

[0116] Weighted synthesis: According to the importance of each performance to engineering application, assign a preset weight, for example, assume that the workability weight is , the strength development weight is , and the shrinkage performance weight is , then the fitness function can be expressed as:

[0117] ;

[0118] where and are the normalized workability, strength development, and shrinkage performance scores, respectively.

[0119] 3. Introduce a penalty term:

[0120] Penalty mechanism: For schemes that do not meet the constraint conditions, introduce a penalty term to reduce their fitness value. For example, if the 28-day compressive strength of a certain scheme is lower than the target value, a penalty term can be introduced:

[0121] ;

[0122] Then subtract the penalty term from the fitness function to obtain the final fitness function:

[0123] .

[0124] Step S320, input the multi-objective fitness function and the preset multi-objective constraint condition into the preset multi-objective optimization algorithm, and start the reverse search based on the preset running parameters.

[0125] The specific process is as follows: 1. Input multi-objective fitness function and constraint conditions: input the multi-objective fitness function constructed in step S310 into the multi-objective optimization algorithm. At the same time, input the preset multi-objective constraint conditions, which can include upper and lower limits of parameters, equality constraints, inequality constraints, etc. 2. Select multi-objective optimization algorithm: according to the characteristics and requirements of the problem, select a suitable multi-objective optimization algorithm, such as NSGA-II, SPEA2, IBEA, etc. 3. Set running parameters: set the running parameters of the algorithm, such as population size, maximum iteration number, crossover probability, mutation probability, etc. 4. Initialize population: randomly generate an initial population, and each individual in the population represents a set of possible solutions. 5. Evaluate fitness: calculate the fitness value of each individual in the population, i.e. the value of the multi-objective fitness function. 6. Handle constraints: for individuals that violate the constraints, adjust their fitness values according to the constraint handling mechanism (such as the penalty function method). 7. Selection operation: select excellent individuals according to the fitness value to enter the next generation population. 8. Crossover and mutation operation: perform crossover and mutation operations on the selected individuals to generate new individuals. 9. Update population: add the newly generated individuals to the population to form a new population. 10. Check stop condition: check whether the stop condition is met, such as reaching the maximum iteration number or the fitness value converging, etc. 11. Output results: when the stop condition is met, output the optimization results, such as the Pareto optimal solution set.

[0126] In step S330, during the running of the multi-objective optimization algorithm, the Pareto front is dynamically generated by the preset Pareto optimal solution screening rule, and the iteration is stopped when the preset stop condition is reached.

[0127] The specific process is as follows: 1. Initialize the Pareto front: in the initial iteration of the multi-objective optimization algorithm, evaluate the fitness value of each individual in the population. According to the fitness value, identify the non-dominated solutions (i.e. individuals that are not dominated by other individuals in all objectives), and initialize these non-dominated solutions as the Pareto front. 2. Dynamically update the Pareto front: in each iteration, evaluate the fitness value of the newly generated individuals. According to the preset Pareto optimal solution screening rule, compare the new individuals with the individuals in the current Pareto front. If the new individual is non-dominated, add it to the Pareto front and remove the existing Pareto front individuals dominated by the new individual. If the new individual is dominated, add it to the population of non-Pareto front individuals. 3. Check stop condition: after each iteration, check whether the preset stop condition is met. Common stop conditions include:

[0128] Maximum iteration number: reach the preset maximum iteration number.

[0129] Fitness convergence: the change in the fitness value of the Pareto front is less than the preset threshold.

[0130] Time limit: reach a preset running time.

[0131] Population diversity: the diversity of the population (e.g. the minimum distance between individuals) is below a preset threshold.

[0132] 4. Stop iteration: when any of the stopping conditions are met, stop the running of the multi-objective optimization algorithm.

[0133] Output the final Pareto front, which contains a series of solutions that achieve an optimal balance on multiple objectives.

[0134] The key contribution of the multi-objective optimization algorithm lies in its customized design for the grouting material proportioning optimization problem:

[0135] Constraint handling mechanism: the algorithm has a built-in process feasibility filter. When generating Pareto solutions, it automatically excludes proportions with too low water-binder ratio (e.g. <0.30) that cannot be mixed, or too high solid waste content (e.g. >90%) that leads to slow early strength development. This ensures that all candidate solutions have a basis for engineering implementation.

[0136] Search space guidance: based on historical experience in the material gene database, the algorithm preferentially samples in the historical high-performance interval of solid waste content 82%-86% and water-binder ratio 0.35-0.38. This knowledge-based guidance can quickly and reliably converge to the real high-performance Pareto front compared to completely random search.

[0137] Specific manifestation of solving technical contradictions: the final output of the Pareto front clearly shows the trade-off relationship between compressive strength and fluidity. For example, one solution represents strength 110 MPa and fluidity 260 mm, and another solution represents strength 100 MPa and fluidity 300 mm. The algorithm quantifies this trade-off curve, enabling designers to make precise and evidence-based decisions based on actual engineering requirements (such as the emphasis on strength or workability in different structural parts), which is not possible with traditional trial-and-error methods.

[0138] Step S340, from the final Pareto front, a unique optimal solution is selected by a preset optimal solution rule.

[0139] The specific process is as follows:

[0140] 1. Define the optimal solution rule: for example, use the weighted sum method to assign weights according to the importance of each objective, and calculate the weighted total score of each Pareto solution. For example, assume that the workability weight is , the strength development weight is , and the shrinkage performance weight is , then the weighted total score can be expressed as:

[0141] ;

[0142] wherein, and are the normalized workability, strength development and shrinkage performance scores, respectively.

[0143] 2. Calculate the weighted total score for each Pareto solution:

[0144] For each Pareto solution, calculate its weighted total score. = 0.4, = 0.5, = 0.1, for example, assuming there are three solutions in the Pareto frontier:

[0145] Solution 1: Workability 0.9, Strength Development 0.8, Shrinkage Performance 0.7, then the weighted total score of Solution 1 is 0.83.

[0146] Solution 2: Workability 0.8, Strength Development 0.9, Shrinkage Performance 0.6, then the weighted total score of Solution 2 is 0.83.

[0147] Solution 3: Workability 0.7, Strength Development 0.7, Shrinkage Performance 0.9, then the weighted total score is 0.72.

[0148] 3. Select the optimal solution: According to the weighted total score, select the solution with the highest score as the optimal solution. If the scores of multiple solutions are the same, further compare their performance on key performance indicators. For example, the weighted total scores of Solution 1 and Solution 2 are both 0.83, and further comparison of their strength development performance can be made. Select Solution 2 with higher strength development as the optimal solution.

[0149] Based on the optimal material gene combination, combined with the preset batching rule library, an accurate batching scheme is automatically generated, and the basic process parameters are preset according to the material properties and target performance requirements, and then the mixing process is started, including:

[0150] Step S410, call the preset intelligent batching rule library, convert each core formula parameter in the optimal material gene combination into single raw material batching amount accurate to the preset batching accuracy, and form a standardized batching scheme.

[0151] The specific process is as follows: 1. Call the intelligent batching rule library: the intelligent batching rule library is a pre-set, structured knowledge base and computing engine, and the rules are based on material characteristics, target performance, process constraints and historical production data. For example, the rule library defines how to automatically calculate and verify the actual batching amount of each material according to the solid waste mixing amount, water-binder ratio and other parameters, combined with the measured moisture content, purity and other information of raw materials. 2. Extract core formula parameters: extract core formula parameters such as solid waste mixing amount, water-binder ratio and water-reducing agent mixing amount from the optimal material gene combination. For example, the optimal material gene combination may be: solid waste mixing amount 82%, water-binder ratio 0.37, and water-reducing agent mixing amount 1.8%. 3. Calculate the batching amount of single raw material: according to the calculation logic in the rule library and the pre-set batch size, convert the core formula parameters into the specific amount of each raw material, and ensure that the results meet the pre-designed batching accuracy requirements. For example, assuming that the total cementitious material batch size is 1000kg, the solid waste amount is: 82% x 1000kg = 820kg; the cement amount is: 18% x 1000kg = 180kg; the water amount is: 0.37 x (820kg + 180kg) = 370kg; the water-reducing agent amount is: 1.8% x (820kg + 180kg) = 18kg.

[0152] 4. Form a standardized batching scheme: arrange the calculated amount of each raw material according to the pre-set format, and associate it with the measurement accuracy requirement, and output as a standardized executable batching instruction. For example:

[0153] · Solid waste: 820kg (accuracy ±0.5kg);

[0154] · Cement: 180kg (accuracy ±0.5kg);

[0155] · Water: 370kg (accuracy ±0.5kg);

[0156] · Water-reducing agent: 18kg (accuracy ±0.05kg).

[0157] Step S420, extract the material characteristic data and target performance index corresponding to the optimal material gene combination, input the pre-set material characteristic-process parameter association model, and automatically output the adapted basic process parameters.

[0158] The specific process is as follows:

[0159] 1. Extract material characteristic data and target performance index:

[0160] Material characteristic data: extract key characteristics such as main chemical components (SiO2, Al2O3, etc.), mineral composition and particle size distribution from the optimal material gene combination. For example, the chemical composition and D50 particle size of the mixed cementitious system corresponding to a solid waste mixing amount of 82% can be extracted as representative inputs.

[0161] Target performance indicators: Extract pre-set performance targets, such as 28-day compressive strength ≥ 100 MPa, initial fluidity ≥ 280 mm, etc.

[0162] 2. Input material property-process parameter correlation model:

[0163] Pre-set correlation model: This model is a machine learning model (such as neural network, gradient boosting tree, etc.) trained based on historical data in the material gene database, which can learn the complex mapping relationship between material properties, performance targets and process parameters.

[0164] Model input: Input the extracted material property data and target performance indicators into the correlation model.

[0165] 3. Automatically output adapted basic process parameters:

[0166] Model output: The correlation model automatically calculates and outputs the adapted basic process parameters based on the input data. For example, stirring speed: 1000 revolutions / minute; stirring time: 5 minutes; feeding sequence: first add solid waste and cement, then add water and water reducing agent; curing temperature: 25°C; curing time: 28 days.

[0167] 4. Verify process parameters:

[0168] Rule verification: The system automatically checks whether the output parameters are within the safe range of equipment and the feasible domain of process.

[0169] Feedback fine-tuning: If the parameters do not meet the constraints, trigger the model fine-tuning or rule backtracking mechanism to re-recommend parameters until a reasonable and executable process parameter set is output.

[0170] Step S430, start small batch pre-stirring verification process, collect real-time rheological data and homogeneity indicators in the pre-stirring process, compare with pre-set process adaptation threshold, if the deviation is beyond the threshold, dynamically correct the stirring time or feeding interval through pre-set fine-tuning algorithm, update the process parameters.

[0171] The specific process is as follows:

[0172] 1. Start small batch pre-stirring verification process:

[0173] Prepare materials: Prepare each raw material according to the standardized batching scheme generated in step S410.

[0174] Set initial process parameters: Initialize the control parameters of the stirring equipment, such as stirring speed, stirring time, feeding sequence and time interval, etc., according to the basic process parameters output in step S420.

[0175] 2. Collect real-time rheological data and homogeneity indicators:

[0176] Rheological data: The apparent viscosity and yield stress of the slurry are monitored in real time using an online rheometer.

[0177] Uniformity index: The moisture distribution uniformity and solid particle dispersion state of the slurry are monitored in real time by online monitoring equipment such as near-infrared spectroscopy (NIRS) sensors.

[0178] 3. Comparison with preset process adaptation threshold:

[0179] Preset threshold: Based on material characteristics and target performance, set the dynamic allowable range of key process quality indicators. For example, the slurry viscosity threshold range is 12-18 Pa·s, the yield stress threshold range is 5-8 Pa, and the moisture distribution uniformity deviation threshold is ≤±1.5%.

[0180] Real-time comparison: Dynamically compare the collected real-time data with the preset threshold to determine whether the process state is in the controlled interval.

[0181] 4. Dynamic correction of stirring time or feeding interval:

[0182] Deviation detection and fine-tuning trigger: If the sampling value of any indicator exceeds the threshold for multiple times (such as 3 times), the preset real-time fine-tuning algorithm is automatically triggered.

[0183] Fine-tuning execution: The fine-tuning algorithm dynamically adjusts the stirring time or the feeding interval between steps according to the direction and magnitude of the deviation. For example, if the viscosity is consistently high, the stirring time can be automatically extended by 30 seconds or the instantaneous flow of the water-reducing agent metering pump can be fine-tuned.

[0184] Parameter update and closed loop: The adjusted parameters are updated to the control system in real time and take effect immediately in the current pre-stirring process, forming an immediate closed-loop control of "monitoring-comparison-adjustment", ensuring that the process parameters quickly converge to the adaptation state.

[0185] Step S440, the pre-verified accurate batching scheme and the optimized process parameters are transmitted to the full-automatic stirring control system, and the system performs raw material metering, staged feeding, and variable-speed stirring operations according to the preset timing.

[0186] The specific process is as follows:

[0187] 1. Data transmission:

[0188] Transmission content: The accurate batching scheme generated in step S410 and the optimized process parameters in step S430 are transmitted to the full-automatic stirring control system. Transmission method: Data transmission is performed through industrial communication protocols (such as Profibus, Modbus TCP / IP, etc.) or data interfaces (such as OPCUA), ensuring the accuracy and real-time nature of the data.

[0189] 2. System initialization:

[0190] Parameter setting: The control system receives and parses the data, loads the ingredient formula, feeding timing, mixing stage parameters, etc., and completes the pre-setting of control instructions. Equipment inspection: The system performs a self-checking process to confirm that each metering scale, conveying mechanism, mixing motor, and sensor are in normal working and ready state.

[0191] 3. Raw material metering:

[0192] Accurate metering: According to the batching scheme, the system controls the metering equipment to accurately weigh each raw material. For example, 600 kg of solid waste, 200 kg of cement, 280 kg of water, and 12 kg of water-reducing agent. Real-time monitoring: During the metering process, the system collects weight signals in real time, performs closed-loop control, and ensures that the metering accuracy meets the preset tolerance (such as ±0.5%).

[0193] 4. Staged feeding:

[0194] Timing control: According to the optimized process parameters, the system controls the feeding of each raw material according to the preset timing and speed. For example, using a "dry material pre-mixing - mixing water next - adding admixture" staged feeding strategy. Process stability guarantee: The system can dynamically adjust the feeding rate based on real-time feedback to prevent clumping, dust, or splashing, ensuring a smooth and controllable feeding process.

[0195] 5. Variable speed mixing:

[0196] Mixing mode: The system performs mixing operations according to the preset mixing mode. For example, low-speed mixing (500 rpm) for 2 minutes at the initial stage, and then high-speed mixing (1000 rpm) for 3 minutes.

[0197] Real-time adjustment: The system dynamically adjusts the mixing speed and time based on real-time rheological data and homogeneity indicators to ensure optimal mixing results.

[0198] 6. Process monitoring and recording:

[0199] Real-time monitoring: The system monitors various parameters in real time during the mixing process, such as rheological data, temperature, humidity, etc., to ensure that the mixing process meets the process requirements. Data recording: The system records all data during the mixing process, including ingredient quantity, mixing time, mixing speed, etc., for subsequent quality traceability and process optimization.

[0200] After automatically adjusting the addition amount of admixture or mixing water through the preset intelligent control algorithm, it further includes maintenance monitoring and life prediction steps, specifically including:

[0201] Step S700, synchronously collect external environment parameters and component internal state data at a preset sampling frequency through a preset distributed sensing network, fuse the data based on a preset environmental sensitivity weight algorithm, the external environment parameters include real-time temperature, humidity, diurnal temperature difference, wind speed, and the component internal state data include hydration heat release rate and surface strain.

[0202] The specific operation is as follows: 1. Deploy the distributed sensing network: external environment sensors: deploy temperature sensors, humidity sensors, wind speed sensors, etc. around the component, for collecting real-time temperature, humidity, diurnal temperature difference, wind speed, etc. external environment parameters. Internal state sensors: deploy hydration heat sensors, strain sensors, etc. inside the component, for collecting hydration heat release rate, surface strain, etc. internal state data. 2. Set the sampling frequency: according to the engineering requirements and the data change rate, preset the sampling frequency. For example, temperature and humidity data are collected once every minute, wind speed data are collected once every 5 minutes, hydration heat release rate and surface strain data are collected once every 10 seconds. 3. Data collection and synchronization: through the distributed sensing network, synchronously collect external environment parameters and component internal state data at a preset sampling frequency. Use time stamp to ensure the synchronicity of the data, and transmit the collected data to the data processing system. 4. Environmental sensitivity weight algorithm: define the weight: according to the influence degree of the environmental parameters on the component performance, preset the environmental sensitivity weight. For example, temperature weight 0.4, humidity weight 0.3, wind speed weight 0.2, diurnal temperature difference weight 0.1. Data fusion: fuse the collected external environment parameters and component internal state data. For example, calculate the weighted average value: fused data = 0.4 x temperature data + 0.3 x humidity data + 0.2 x wind speed data + 0.1 x diurnal temperature difference data. Internal state data processing: normalize the hydration heat release rate and surface strain data, to ensure that the data are in the same dimension range. 5. Data fusion and output: integrate the fused external environment parameters and normalized internal state data, to form a comprehensive data set. For example, the fused data set may include the following:

[0203] Fused environmental data: 0.65; normalized hydration heat release rate: 0.72; normalized surface strain: 0.48.

[0204] Step S800, real-time access the fused data to the constructed engineering component digital twin model, call the preset improved hydration-transport-mechanical coupling algorithm, based on the fused data of the accessed model, dynamically simulate the preset short-term age strength development process of the component, real-time predict the shrinkage cracking risk and the preset long-term durability decay trend, and generate the quantitative performance evolution curve and risk warning value.

[0205] The engineering component digital twin model is a virtual model based on the finite element method and material time-varying constitutive relation, and its specific construction method is as follows:

[0206] Geometry and mesh: According to the actual component design drawing, a three-dimensional geometric model is established, and meshing is performed to ensure that the key parts are refined.

[0207] Material constitutive: The performance prediction data of the optimal material gene combination of the grouting material, such as the strength development curve, the time-varying law of the elastic modulus, and the shrinkage strain model, are input into the model as material properties. Among them, the strength development and shrinkage model are based on the output of the performance prediction model and the hydration-transport-mechanical coupling algorithm.

[0208] Boundary conditions and loads: According to the engineering design scheme, the boundary constraint conditions and actual load working conditions of the component are set.

[0209] Data interface: The model sets up a real-time data interface to receive fusion data from the distributed sensor network, such as environmental temperature and humidity, and component internal strain. These real-time data are used as dynamic inputs for model simulation to drive the model to update the internal state, such as temperature field, humidity field, and stress field.

[0210] Output and correlation: The outputs of the model include the time-varying stress / strain distribution inside the component, the predicted cracking risk index, and the long-term durability index (such as the chloride ion diffusion depth). These outputs are compared with the sensor readings and periodic detection results in the physical world to verify the model accuracy and drive the self-correction of the model parameters.

[0211] The specific process is as follows: 1. Data access digital twin model: Real-time data transmission: Through industrial communication protocols (such as MQTT, CoAP, etc.), the fusion-processed external environmental parameters and component internal state data are transmitted to the digital twin model in real time. Data synchronization: Ensure the consistency of all data in time, and use time stamp to synchronize the data. 2. Call improved hydration-transport-mechanical coupling algorithm: The improved hydration-transport-mechanical coupling algorithm is a mathematical model that comprehensively considers cement hydration reaction, water transport and mechanical property development. This algorithm can dynamically simulate the performance changes of the component under different environmental conditions. Key parameters are as follows:

[0212] Hydration reaction parameters: such as hydration rate, heat release rate, etc.

[0213] Water transport parameters: such as water vapor diffusion coefficient, capillary action, etc.

[0214] Mechanical property parameters: such as elastic modulus, compressive strength, etc.

[0215] 3. Dynamic simulation of short-term strength development:

[0216] Short-term strength simulation: Based on the fused data, simulate the strength development process of the component within a preset short-term age (e.g., 3 days, 7 days, 28 days). For example, simulate the development trend of 28-day compressive strength. Real-time update: According to real-time data, dynamically update the simulation results to ensure the accuracy and timeliness of the simulation.

[0217] 4、Real-time prediction of shrinkage cracking risk:

[0218] Shrinkage cracking prediction: By analyzing the hydration heat release rate, surface strain, and environmental temperature and humidity changes, predict the shrinkage cracking risk of the component. For example, when the surface strain exceeds the ultimate strain of the material, predict the cracking risk.

[0219] Risk warning: When the predicted cracking risk exceeds the preset threshold, generate a risk warning value. For example, the warning value can be expressed as a risk probability (e.g., 80%) or a risk level (e.g., high, medium, low).

[0220] 5、Prediction of long-term durability decay trend:

[0221] Long-term durability prediction: Based on the fused data, predict the durability decay trend of the component within a preset long-term period (e.g., 50 years, 100 years). For example, predict the rate of steel corrosion caused by chloride ion penetration. Performance evolution curve: Generate a quantitative performance evolution curve, such as the change curve of compressive strength over time, the change curve of chloride ion penetration depth over time, etc.

[0222] 6、Generation of performance evolution curve and risk warning value:

[0223] Performance evolution curve: Organize the simulation results and prediction data into curves to visually display the trend of component performance over time. Risk warning value: Quantify the predicted shrinkage cracking risk and durability decay risk into specific warning values to guide maintenance decisions.

[0224] Step S900, when the predicted shrinkage cracking risk warning value exceeds the preset threshold, the digital twin model triggers the built-in maintenance decision module, matches the adaptive scheme from the preset environment-maintenance measure rule library combined with real-time external environmental parameters, and sends control instructions to the execution system.

[0225] The specific process is as follows:

[0226] 1、Risk warning trigger:

[0227] Real-time monitoring: The digital twin model real-time monitors the predicted shrinkage cracking risk warning value. Threshold comparison: Compare the real-time predicted shrinkage cracking risk warning value with the preset threshold. For example, the preset cracking risk threshold is 80%.

[0228] 2、Trigger the maintenance decision module:

[0229] Condition met: When the predicted shrinkage cracking risk warning value exceeds the preset threshold, trigger the maintenance decision module. Module start: The maintenance decision module starts to prepare for developing a maintenance plan.

[0230] 3. Combine real-time external environmental parameters:

[0231] Data collection: Collect real-time external environmental parameters such as temperature, humidity, wind speed, etc. from the distributed sensor network. Parameter analysis: Analyze the impact of real-time external environmental parameters on maintenance measures. For example, high temperature and low humidity environment may need to increase moisture retention measures, and low temperature environment may need to increase insulation measures.

[0232] 4. Match the adaptive solution:

[0233] Rule base query: Query the adaptive maintenance solution from the preset environment-maintenance measure rule base. The rule base stores recommended maintenance measures under different environmental conditions. Solution selection: According to the real-time external environmental parameters, select the most adaptive maintenance solution. For example, if the real-time temperature is 30°C, the humidity is 40%, and the wind speed is 5m / s, it may choose to increase the spray moisture and sunshade measures.

[0234] 5. Send control instructions:

[0235] Instruction generation: According to the selected maintenance solution, generate specific control instructions. For example, the instructions may include starting the spray system, adjusting the angle of the sunshade curtain, increasing the maintenance time, etc. Instruction sending: Send the control instructions to the execution system through industrial communication protocols (such as Modbus TCP / IP, OPCUA, etc.). The execution system executes the corresponding maintenance operation according to the instructions.

[0236] 6. Record and feedback:

[0237] Record operation: Record the triggering time of the maintenance decision module, the selected maintenance solution, and the sent control instructions. Feedback optimization: Feedback the results of the maintenance operation to the digital twin model for further optimization of the maintenance decision algorithm and rule base.

[0238] Step SA00, according to the preset period, the actual maintenance process data of the component, the performance test results are returned to the digital twin model, and the deviation calculation is performed with the predicted performance evolution curve and the risk warning value. The key parameters in the coupling algorithm are corrected through the preset error back propagation algorithm, and are updated to the material gene database synchronously.

[0239] The specific operation is as follows:

[0240] 1. Data return:

[0241] Actual maintenance data: Data on the actual maintenance process of components is collected according to a preset cycle (e.g., weekly, monthly), including the implementation of maintenance measures and changes in environmental parameters. Performance test results: Actual performance data of components, such as compressive strength, fluidity, and shrinkage rate, are obtained through on-site testing and laboratory analysis.

[0242] 2. Deviation Calculation:

[0243] Comparison of Predicted and Actual Data: The actual performance data is compared with the predicted performance evolution curves in the digital twin model to calculate the deviation. For example, the difference between the actual compressive strength and the predicted compressive strength is calculated. Comparison of Risk Warning Values: The actual detected risk conditions (such as cracking) are compared with the risk warning values ​​predicted by the model to assess the accuracy of the prediction.

[0244] 3. Backpropagation algorithm:

[0245] Algorithm Application: A gradient descent-based error backpropagation algorithm is employed. Based on the deviation calculated in step 2, key parameters in the coupled algorithm are adjusted in reverse. Specifically, this includes:

[0246] Loss function construction: The bias calculated in step 2 (such as mean squared error, cross-entropy loss) is used as the loss function L.

[0247] Gradient calculation and backpropagation: The loss function L is calculated for each adjustable parameter in the coupled algorithm using automatic differentiation or pre-defined differentiation rules. (such as the hydration reaction rate constant k) h Moisture diffusion coefficient D w creep coefficient C C gradient of (etc.) .

[0248] Parameter update: Update parameters using stochastic gradient descent with momentum.

[0249] ; ;

[0250] Where η is the learning rate (preset to 10). -3 ~10 -4 ), where γ is the momentum coefficient (default value is 0.9). The momentum is updated for the parameters, where t is the number of iterations.

[0251] Iteration Termination: When the loss function L is lower than the preset threshold or the maximum number of iterations (e.g., 500 times) is reached, the optimization stops and the corrected parameter set is output.

[0252] Parameter correction: Through the above iterative optimization, the deviation between the predicted value and the actual value is gradually reduced, thereby improving the prediction accuracy of the model.

[0253] 4. Updating the Material Gene Database:

[0254] Data Synchronization: The corrected key parameters and actual performance data are synchronized and updated to the Material Gene Database. Database Optimization: Through continuous updating, the Material Gene Database is constantly enriched and improved, providing more accurate data support for subsequent material design and performance prediction.

[0255] The maintenance monitoring and life prediction steps also include a real-time damage monitoring and early warning sub-process, which specifically includes:

[0256] Step SB00, during the grouting material pouring stage, embed the preset type of distributed sensing components in the load-sensitive areas of the components. These components are co-networked with the deployed distributed sensing network, and collect strain values, sound wave propagation speed, and vibration frequency change data after the structure is stressed at a preset high-frequency sampling interval.

[0257] The specific process is as follows:

[0258] 1. Determine the load-sensitive area:

[0259] Structural Analysis: Through finite element analysis or other structural analysis methods, determine the stress concentration areas and potential damage-prone areas of the components under stress. Pre-set position: According to the analysis results, pre-set the installation position of the sensing components. For example, in the key stress parts of the bridge (such as bridge piers, beam bottoms) and the key nodes of the building (such as column feet, beam-column connections).

[0260] 2. Select distributed sensing components:

[0261] Strain sensor: Select fiber Bragg grating (FBG) sensor or strain gauge to measure the strain value of the structure under stress. Sound wave sensor: Select piezoelectric sensor or optical fiber sensor to measure the sound wave propagation speed. Vibration sensor: Select accelerometer or optical fiber sensor to measure the vibration frequency change.

[0262] 3. Pre-embed sensing components:

[0263] Installation preparation: Before pouring the grouting material, install the sensing components in the pre-set load-sensitive areas. Ensure that the installation position of the sensing components is accurate and fixed firmly. Protection measures: Protect the sensing components to prevent damage during the pouring process. For example, use protective sleeves or set up a buffer layer around the sensing components.

[0264] 4. Co-network transmission:

[0265] Network Integration: Integrate the pre-embedded sensing components with the deployed distributed sensing network, ensuring unified data transmission and management. Communication Protocol: Use industry-standard communication protocols (such as Modbus TCP / IP, OPCUA, MQTT, etc.) to ensure the stability and reliability of data transmission.

[0266] 5. High-frequency sampling data collection:

[0267] Sampling frequency setting: Set the high-frequency sampling interval according to monitoring requirements. For example, the sampling frequency of strain value and vibration frequency change data can be set to 100Hz, and the sampling frequency of sound wave propagation speed data can be set to 1kHz. Data collection: After the structure is stressed, collect strain value, sound wave propagation speed, and vibration frequency change data at the preset high-frequency sampling interval. Data transmission: Through the distributed sensing network, the collected data is transmitted to the data processing system in real time.

[0268] Step SC00, the collected real-time sensing data is connected to the engineering component digital twin model, first through the preset data preprocessing algorithm to extract the data characteristic quantity, the characteristic quantity includes strain abnormal fluctuation value, sound wave propagation attenuation coefficient and vibration modal offset; call the preset data-damage correlation algorithm, fuse and compare the characteristic quantity and the mapping relationship library, automatically identify the types of two kinds of hidden damage of micro-crack and interface void.

[0269] The specific process is as follows:

[0270] 1. Data access digital twin model:

[0271] Real-time data transmission: Through industrial communication protocols (such as MQTT, CoAP, etc.), the collected real-time sensing data (strain value, sound wave propagation speed, vibration frequency change, etc.) is transmitted to the digital twin model. Data synchronization: Ensure the consistency of all data in time, use timestamp to synchronize the data.

[0272] 2. Data preprocessing:

[0273] Filtering: Use filtering algorithms (such as low-pass filtering, high-pass filtering) to remove noise and retain valid signals.

[0274] Feature extraction:

[0275] Strain abnormal fluctuation value: Calculate the short-term and long-term average of strain value, and identify the abnormal fluctuation that exceeds the preset threshold. For example, when the difference between the short-term average strain value and the long-term average strain value exceeds the preset threshold, it is marked as abnormal fluctuation.

[0276] Sound wave propagation attenuation coefficient: Calculate the sound wave propagation attenuation coefficient by measuring the propagation time and amplitude of sound waves in the structure. For example, use the following formula: Where a is the sound wave propagation attenuation coefficient, d is the propagation distance, A0 is the initial sound wave amplitude, and A is the received sound wave amplitude.

[0277] Vibration modal shift: Extract the vibration frequency and modal shape through frequency domain analysis (e.g., Fast Fourier Transform, FFT), and calculate the shift from the baseline state. For example, use the following formula:

[0278] Where Δf is the vibration frequency shift, is the current vibration frequency, is the baseline vibration frequency.

[0279] 3. Call data-damage association algorithm:

[0280] Algorithm description: The pre-set data-damage association algorithm is an algorithm based on machine learning or physical model, which can associate the extracted feature quantities with damage types. For example, use machine learning algorithms such as Support Vector Machine (SVM) or Decision Tree.

[0281] Feature fusion: Fuse the extracted strain abnormal fluctuation value, sound wave propagation attenuation coefficient and vibration modal shift to form a comprehensive feature vector.

[0282] Mapping relationship library: The mapping relationship library is a pre-set database that stores the mapping relationship between different damage types (such as microcracks, interface separation) and feature quantities. For example, microcracks may cause the sound wave propagation attenuation coefficient to increase and the vibration modal shift to increase.

[0283] 4. Automatic identification of hidden damage:

[0284] Comparison analysis: Compare the comprehensive feature vector with the data in the mapping relationship library to identify the most matching damage type. Damage localization: Combine the spatial coordinates of the sensors to determine the specific location of the damage. For example, if multiple sensors detect similar feature quantity changes, the damage area can be determined through triangulation. Damage degree evaluation: Evaluate the severity of the damage according to the size of the feature quantity. For example, the larger the sound wave propagation attenuation coefficient, the more serious the microcrack.

[0285] Step SD00, combined with the sensor coordinates and feature quantity distribution gradient of the identified hidden damage, through the pre-set binding of the sensor spatial coordinates and the model three-dimensional geometric coordinates, the three-dimensional range of the damage area in the model is inverted, and the fine positioning is realized through the pre-set positioning accuracy algorithm correction, and the severity of the damage is evaluated based on the size and distribution range of the feature quantity.

[0286] The specific process is as follows: first, the precise mapping of sensor spatial coordinates and digital twin model three-dimensional geometric coordinates is completed. This process is realized through a preset coordinate conversion matrix and calibrated using field measurement data (such as using a total station), ensuring that the spatial error between the physical sensor and the virtual model is less than a preset value (for example, ±2mm). At the same time, multi-modal damage features are extracted from real-time monitoring data, including: strain anomalies (such as local strain sudden increase exceeding baseline by 50%), acoustic emission signal energy and ringing count, and vibration modal frequency shift (such as a decrease in fundamental frequency of more than 3%). According to the preset trigger threshold and logic (such as at least two of the three features simultaneously exceeding the limit), the system automatically identifies and marks the sensor with signal anomalies and its three-dimensional coordinates. Then, taking the trigger sensor as the center, a spatial interpolation algorithm (such as Kriging interpolation) is used to calculate the continuous distribution of feature quantities in the sensor network, and then generate the feature quantity spatial gradient field. By analyzing the maximum modulus value direction and the dense area of the contour line of the gradient field, the damage propagation direction and boundary are preliminarily determined. Combined with the gradient field information and the trigger sensor coordinates, a preset inversion algorithm (such as an iterative reconstruction algorithm based on elastic wave travel time tomography) is called to deduce the initial three-dimensional morphology of the damage area (such as an ellipsoid or a crack surface) in the digital twin model, and calculate its geometric center, principal axis size and volume.

[0287] Finally, the initial results are optimized and corrected using a preset positioning accuracy algorithm. For example, an extended Kalman filter is used to fuse multiple rounds of monitoring data, constantly updating the estimated value of the damage state to reduce the influence of random noise; at the same time, based on the uncertainty model of the sensor network layout, the 95% confidence interval of the damage location and size is calculated. The output results include the precise geometric parameters of the damage area (such as center coordinates (1.200±0.003m, 0.500±0.003m, 0.300±0.005m), equivalent diameter 0.05±0.002m) and positioning uncertainty. Based on the normalized feature quantity comprehensive amplitude and damage volume, according to the preset damage level matrix (such as mapping "high amplitude and large volume" to "severe" level), the quantitative evaluation and classification of damage severity are finally completed.

[0288] Step SE00, based on the type of damage identified, the positioning results and the severity of the evaluation, the preset multi-level warning is started according to the preset damage level division standard.

[0289] The specific operation is as follows: First, preset damage level classification standards and early warning response mechanisms are established. Based on the percentage of characteristic quantities (such as strain, acoustic emission energy, and frequency shift) exceeding preset thresholds and the damage volume, damage is classified into three levels: minor (10%-30% exceeding limits), moderate (30%-60% exceeding limits), and severe (over 60% exceeding limits). Each level corresponds to an early warning level (such as yellow, orange, and red), and the response measures for each level are clearly defined. For example, a yellow warning prompts routine inspection, an orange warning requires detailed assessment and planned maintenance, and a red warning triggers emergency shutdown and emergency response procedures. Next, the system monitors sensor data in real time and continuously identifies, locates, and assesses damage through a digital twin model. When the assessment result meets the triggering conditions for any damage level, the system automatically activates the corresponding level of early warning. Finally, according to the preset notification strategy, early warning information is pushed to designated personnel in real time through an integrated alarm platform (such as SMS, email, audible and visual alarms, and system interface pop-ups). For example, a yellow alert is sent to routine maintenance personnel; an orange alert simultaneously notifies the maintenance team and project manager; and a red alert is broadcast to all personnel and triggers on-site audible and visual alarms to ensure timely response and accountability.

[0290] Based on the same inventive concept, embodiments of the present invention provide a high-solid-waste, low-carbon, high-performance grouting material and its intelligent control preparation system, including a memory and a processor. The memory stores information that can be run on the processor to implement the following... Figure 1 The procedure for the method shown.

[0291] 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 digital and intelligent control preparation method of high solid waste low carbon high performance grouting material, characterized in that, The application relates to a method for automatically generating a precise mixing scheme of industrial solid waste, and belongs to the technical field of industrial solid waste utilization. The method comprises the following steps: acquiring XRF chemical composition and XRD mineral composition data of industrial solid waste raw materials, and constructing a material gene database containing key composition parameters, process parameters and performance indexes; based on the material gene database, performing a pseudo-mixing ratio experiment by using a pre-trained performance prediction model, screening a candidate mixing ratio scheme through a pre-set high-throughput virtual calculation process, predicting working property, strength development and shrinkage performance, and outputting corresponding performance prediction results; inputting the performance prediction results as a fitness function into a multi-objective optimization algorithm, and outputting an optimal material gene combination through reverse search under pre-set multi-objective constraints; based on the optimal material gene combination, automatically generating a precise mixing scheme in combination with a pre-set mixing rule library, and pre-setting basic process parameters such as stirring speed, stirring time and feeding sequence according to material characteristics and target performance requirements, and then starting a stirring process; continuously collecting key process parameters in the stirring process through a pre-set real-time monitoring system, and dynamically comparing and analyzing the collected real-time process data with a pre-set target range in a digital twin model; based on the comparison and analysis results, automatically adjusting the addition amount of an external agent or mixing water through a pre-set intelligent control algorithm to control the real-time process parameters within the target range; based on the material gene database, performing a pseudo-mixing ratio experiment by using a pre-trained performance prediction model, screening a candidate mixing ratio scheme through a pre-set high-throughput virtual calculation process, predicting working property, strength development and shrinkage performance including: the XRF chemical composition, XRD mineral composition and particle size data of the industrial solid waste stored in the material gene database are called, and the activity prediction results and gradation optimization conclusions are synchronously generated through a pre-set comprehensive characteristic analysis algorithm of the solid waste; meanwhile, external input engineering actual requirements are quantified into corresponding performance constraint indexes through a pre-set engineering requirement conversion method; the activity prediction results, gradation optimization conclusions and quantified performance constraint indexes are mapped into the accurate range of core formula parameters of the grouting material through a pre-set multi-dimensional parameter correlation algorithm; the accurate range of the core formula parameters of the grouting material is taken as an input variable and a horizontal boundary of a pre-set orthogonal combination algorithm, the parameter combination is systematically covered and the key interval is strengthened through the pre-set orthogonal combination algorithm, and a pseudo-mixing ratio scheme set covering a potential optimal performance space is generated; the pseudo-mixing ratio scheme set is batch-predicted by using a pre-set performance prediction model, and prediction data of each scheme is output, wherein the prediction data includes working property, strength development and shrinkage performance data, and the pre-set performance prediction model is trained based on the material gene database and takes a hydration kinetics equation and a rheological constitutive relation as physical constraints; 2. The method according to claim 1, wherein the method is characterized by, a pre-set high-throughput virtual calculation process is started to perform parallel simulation verification on the prediction data, and finally, performance prediction results of working property, strength development and shrinkage performance corresponding to each pseudo-mixing ratio scheme are formed. The method for automatically generating a precise mixing scheme of industrial solid waste comprises the following steps: XRF chemical composition and XRD mineral composition data of industrial solid waste raw materials are acquired, process parameters and corresponding performance index data are synchronously collected, multi-source heterogeneous data is normalized and outlier is removed, and a structured basic data set is formed; Key component parameter features and process parameter features are extracted from the structured basic data set, principal components with cumulative contribution rate greater than or equal to a preset contribution rate are reserved through principal component analysis algorithm, and an associated mapping between the features and the performance index is established, so as to complete the construction of the material gene database; Based on the established associated mapping, when the key components of a new batch of solid waste exceed a preset component fluctuation warning threshold, a small batch verification experiment is automatically triggered, and the material gene database is updated in real time after the verification result meets the standard; Based on the updated material gene database and the associated mapping, when new solid waste data from a new production site is input, a deep neural network transfer learning algorithm based on an attention mechanism is used to automatically fine-tune the model parameters to adapt to the new solid waste data from the new production site through preset similarity matching.

3. The method according to claim 2, wherein the method is characterized by, When new solid waste data from a new production site is input, a deep neural network transfer learning algorithm based on an attention mechanism is used to automatically fine-tune the model parameters to adapt to the new solid waste data from the new production site through preset similarity matching, including: Geographical information, production process parameters, environmental conditions and historical fluctuation data of the new solid waste from the new production site are collected, and a preset multi-dimensional quantification processing method is used to obtain a quantized feature vector; The quantized feature vector and the internal features of the solid waste are fused through a gated recurrent unit network to generate a composite feature vector containing context information; The composite feature vector is input into a deep neural network based on an attention mechanism, the weights of the bottom layer general feature extraction layer are frozen, and a preset similarity matching algorithm is used to automatically select a sample subset with the closest component features in the historical database to guide the fine-tuning of the top layer production site adaptation parameter layer, and the fine-tuning process uses preset fine-tuning parameters to ensure that the parameter adjustment range is controlled within a preset range; The performance of the fine-tuned model is evaluated through a preset precision standard, and if the performance does not meet the standard, an incremental learning process is triggered, and after supplementing external disturbance experimental data, the model is fine-tuned again, including updating the model parameters using an incremental learning algorithm, and re-evaluating until the standard is met.

4. The method according to claim 1, wherein the method is characterized by, The accurate range of the core formulation parameters of the grouting material is taken as the input variables and the horizontal boundaries of the preset orthogonal combination algorithm, and the parameter combinations are systematically covered and the key intervals are strengthened through the preset orthogonal combination algorithm to generate a set of quasi-proportioning schemes covering the potential optimal performance space, including: According to a preset parameter level gradient rule, the horizontal gradients are divided within the accurate range of the core formulation parameters of the grouting material; A preset orthogonal table matching algorithm is called to match the corresponding preset orthogonal table based on the number of parameters and the number of horizontal gradients to generate a basic combination set covering the horizontal balance of each parameter, achieving systematic coverage of parameter combinations; Through a preset key interval determination algorithm, the parameter interaction key intervals that significantly affect workability, strength development and shrinkage performance are identified; For the identified key intervals, the number of parameter horizontal gradients in the intervals is increased according to a preset strengthening gradient rule, and the strengthened combination set of the key intervals is generated through the orthogonal combination algorithm; The basic combination set is merged with the reinforced combination set, invalid schemes are removed by a preset redundancy removal algorithm, and finally a set of quasi-proportioning schemes covering the potential optimal performance space is generated.

5. The method according to claim 1, wherein the method is characterized by, The performance prediction results are input into a multi-objective optimization algorithm as a fitness function, and the optimal material gene combination is output through reverse search under preset multi-objective constraints. According to a preset fitness function construction rule, the workability, strength development, and shrinkage performance data in the performance prediction results are taken as core inputs to construct a multi-objective fitness function. The multi-objective fitness function and the preset multi-objective constraints are input into a preset multi-objective optimization algorithm, and the reverse search is started based on preset running parameters. During the running of the multi-objective optimization algorithm, a Pareto front is dynamically generated through a preset Pareto optimal solution screening rule, and the iteration is stopped when a preset stop condition is reached. From the final Pareto front, a unique optimal scheme is selected through a preset optimal solution strategy rule.

6. The intelligent preparation method of the high-solid-waste low-carbon high-performance grouting material according to claim 5, characterized in that, Based on the optimal material gene combination, an accurate proportioning scheme is automatically generated by combining a preset proportioning rule library, and the basic process parameters are preset according to the material properties and target performance requirements, and then the mixing process is started. The preset intelligent proportioning rule library is called to convert the core formula parameters in the optimal material gene combination into single raw material proportioning amounts with a precision of a preset design amount, forming a standardized proportioning scheme. The material property data and target performance indicators corresponding to the optimal material gene combination are extracted and input into a preset material property-process parameter correlation model to automatically output adaptive basic process parameters. A small-batch pre-mixing verification process is started, real-time rheological data and homogeneity indicators during the pre-mixing process are collected, and the deviation from the preset process adaptation threshold is compared. If the deviation exceeds the threshold, the mixing time or the feeding interval is dynamically corrected through a preset fine-tuning algorithm, and the process parameters are updated. The accurate proportioning scheme verified by the pre-verification and the optimized process parameters are transmitted to the full-automatic mixing control system, and the system performs raw material metering, staged feeding, and variable-speed mixing operations according to a preset time sequence.

7. The method according to claim 6, wherein the method is characterized by, After the addition amount of the admixture or mixing water is automatically adjusted by the preset intelligent control algorithm, the maintenance monitoring and life prediction steps are included, which specifically include: Through a preset distributed sensing network, external environmental parameters and internal state data of the component are synchronously collected at a preset sampling frequency, and the data are fused based on a preset environmental sensitivity weight algorithm. The external environmental parameters include real-time temperature, humidity, diurnal temperature difference, and wind speed, and the internal state data of the component include hydration heat release rate and surface strain. The fused data are real-time connected to the digital twin model of the engineering component, an improved hydration-transport-mechanical coupling algorithm is called, and based on the fused data of the connected model, the short-term strength development process of the component is dynamically simulated, the shrinkage and cracking risk is real-time predicted, and the long-term durability decay trend is predicted, and a quantitative performance evolution curve and risk warning value are generated. When the predicted shrinkage and cracking risk warning value exceeds the preset threshold, the digital twin model triggers the built-in maintenance decision module, matches the adaptive scheme from the preset environment-maintenance measure rule library based on the real-time external environmental parameters, and sends control instructions to the execution system. The actual maintenance process data and performance detection results of the component are returned to the digital twin model at a preset period, deviation calculation is performed on the predicted performance evolution curve and risk early warning value, key parameters in the coupling algorithm are corrected through a preset error back propagation algorithm, and the material gene database is updated synchronously.

8. A digital and intelligent preparation system for high solid waste, low carbon and high performance grouting material, characterized in that, The computer readable storage medium stores a program, and the program can be loaded and executed by the processor to implement the method for digitally preparing the high-solid-waste low-carbon high-performance grouting material according to any one of claims 1 to 7.

Images