Green building exterior wall coating with heat insulation function and preparation process thereof

Abstract

The invention provides a green building exterior wall coating with a heat insulation function and a preparation process thereof. The method comprises the following steps: S1, determining raw material formula parameters; s2, pre-dispersing the raw materials; s3, carrying out grinding treatment on the pre-dispersed slurry; s4, mixing and blending the coating; s5, performing performance detection; and S6, packaging the qualified finished coating. Aiming at the core requirements of high heat insulation, low pollution and easy construction of the green building wall paint, the invention constructs a set of full-process preparation process integrating a multi-objective optimization algorithm and real-time feedback control, and finally realizes collaborative optimization of paint performance and production efficiency through scientific design of a raw material formula and precise control of production links.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of coating compositions, green building exterior wall coating, in particular to a green building exterior wall coating with heat insulation function and a preparation process thereof. BACKGROUND

[0002] With the popularization of green building and low-carbon energy-saving concept, as an important part of building envelope structure, the heat insulation performance and environmental protection performance of exterior wall coating become the core demand. However, the existing technology has the following key problems:

[0003] (1) The heat insulation performance is out of touch with the formula design: the heat insulation function of traditional exterior wall coating mainly depends on the addition of single heat insulation filler (such as glass beads), the formula design lacks systematic optimization, and contradictions such as "heat insulation performance is improved but construction performance is reduced" and "environmental protection performance is not up to standard" often occur, and the formula parameters are mainly dependent on experience debugging, so it is difficult to balance the multiple performance requirements;

[0004] (2) Low efficiency and poor stability of grinding process: the grinding process of pre-dispersed slurry mainly adopts "extensive" control of fixed rotating speed and time, without dynamic adjustment of parameters combined with the characteristics of raw materials (such as particle size of heat insulation filler, fineness of pigment and filler), which leads to uneven fineness of slurry after grinding, affecting the final performance of coating, and problems such as over-grinding (increased energy consumption) or insufficient grinding (performance not up to standard) are prone to occur;

[0005] (3) Environmental protection and durability are difficult to balance: in order to improve the heat insulation performance or construction performance, some coatings increase the content of volatile organic compounds (VOC) containing additives, or due to uneven dispersion of raw materials, the coating has poor scrubbing resistance and water resistance, which cannot meet the long-term use requirements of green building;

[0006] (4) Poor coordination of whole process parameters: there is no clear parameter transmission and feedback mechanism from formula design to grinding, mixing and detection, and the deviation of one step can easily lead to chain problems in subsequent processes, resulting in low product qualification rate and production efficiency. SUMMARY

[0007] The present application provides a green building exterior wall coating with heat insulation function and a preparation process thereof, which constructs a whole process system of "formula optimization-pre-dispersion-intelligent grinding-mixing and modulation-performance detection-packaging and warehousing", realizes the balance of multiple performances of green building exterior wall coating through algorithm optimization and whole process parameter linkage, and meets the comprehensive requirements of energy saving, environmental protection and durability of green building.

[0008] In order to achieve the above purpose, the present application adopts the following technical scheme:

[0009] A preparation method of a green building exterior wall coating with heat insulation function, comprising the following steps:

[0010] S1: The raw material formula parameters are determined by using an analytic hierarchy process-multiple objective genetic algorithm fusion model; first, the importance of the thermal insulation performance, environmental protection performance and construction performance is scored by inviting experts in the paint industry through the analytic hierarchy process, and the weight vector of the performance is calculated, and then the weight vector is substituted into the fitness function by using the multiple objective genetic algorithm, and the optimal formula parameters are obtained after iterative optimization;

[0011] S2: The raw material pre-dispersion treatment is performed based on the optimal formula parameters; deionized water determined by the optimal formula parameters is added to a high-speed dispersion machine, and a dispersing agent, titanium white powder, heavy calcium powder, nano hollow silica microspheres and a defoaming agent determined by the optimal formula parameters are sequentially added, the rotation speed of the dispersion machine and the stirring time are controlled, and the initial fineness of the pre-dispersed slurry is detected;

[0012] S3: The pre-dispersed slurry is ground by using a fusion algorithm of a multiple linear regression initial grinding parameter prediction model and a proportional, integral and differential feedback control model; first, the average particle size of the nano hollow silica microspheres, the amount of the nano hollow silica microspheres, the fineness of the heavy calcium powder and the initial fineness of the pre-dispersed slurry determined by the optimal formula parameters are input into the multiple linear regression model, and the initial grinding time and the initial grinding rotation speed are predicted; then, the initial grinding time and the initial grinding rotation speed are taken as the starting point, the real-time fineness of the slurry in the grinding process is detected in real time by using the proportional, integral and differential feedback control model, the error between the real-time fineness and the target fineness is calculated, the grinding rotation speed is dynamically adjusted according to the error, and the final fineness of the ground slurry is obtained until the final fineness of the ground slurry meets the requirements;

[0013] S4: The paint is mixed and prepared based on the optimal formula parameters and the final fineness of the ground slurry; pure acrylic emulsion determined by the optimal formula parameters is added to a low-speed stirring tank, and then the ground slurry is added, followed by adding thickening agent, film-forming aid and deionized water determined by the optimal formula parameters, and the preliminary finished paint is obtained after stirring;

[0014] S5: The performance of the preliminary finished paint is detected; if all the detected items are qualified, the preliminary finished paint is determined as a qualified finished paint; if there is an unqualified item, the optimal formula parameters are determined again, and then the steps S2 to S5 are sequentially executed to form a closed loop.

[0015] In the present specification, the preparation method of the green building exterior wall paint with a heat insulation function further comprises S6: The qualified finished paint is packaged, a pretreated packaging barrel is selected, the qualified finished paint is filled according to the filling amount according to the total amount of the qualified finished paint, a label is pasted on the surface of the packaging barrel after sealing, and finally the packaged paint is stored in a warehouse.

[0016] In the specification, the specific implementation process of the analytic hierarchy process in S1 includes: inviting 5 experts engaged in paint formula research and development for ≥10 years, using a 1-9 scale method to score the relative importance of thermal insulation performance, environmental protection performance and construction performance; integrating expert opinions to construct a judgment matrix, solving the maximum eigenvalue and the corresponding eigenvector of the judgment matrix by the eigenvalue method, and normalizing the eigenvector to obtain a weight vector; then calculating the consistency index, random consistency index and consistency ratio; when the consistency ratio is <0.1, the weight vector is determined to be effective, and the weight of thermal insulation performance, the weight of environmental protection performance and the weight of construction performance in the weight vector are 0.54, 0.30 and 0.16 respectively.

[0017] In the specification, the specific implementation process of the multi-objective genetic algorithm in S1 includes: encoding the amount of pure acrylic emulsion, the amount of nano hollow silica microspheres, the amount of titanium dioxide, the amount of heavy calcium powder, the total amount of additives, and the amount of deionized water into a 6-dimensional real number vector as a chromosome, initializing 100 chromosomes to form an initial population; substituting the weight vector obtained in S1 into the fitness function; selecting the operation by using the roulette selection method, performing arithmetic crossover operation with a crossover probability of 0.8, and performing random disturbance mutation operation with a mutation probability of 0.05; after 50 iterations, selecting the chromosome with the highest fitness in the 50th generation as the optimal formula parameters.

[0018] In the specification, the construction of the multi-element linear regression initial grinding parameter prediction model in S3 includes: collecting 50 groups of historical production data including the average particle size of nano hollow silica microspheres, the amount of nano hollow silica microspheres, the fineness of heavy calcium powder, the initial fineness of pre-dispersed slurry, the initial grinding time and the initial grinding speed; after preprocessing the historical production data, the prediction model of the initial grinding time and the initial grinding speed is obtained by least squares fitting; calculate the model determination coefficient, when the determination coefficient is ≥0.90, the model is effective.

[0019] In the specification, the specific implementation process of the proportional, integral and differential feedback control model in S3 includes: setting the target fineness to ≤30μm, taking out the slurry from the sampling port of the sand mill every 15 minutes to detect the real-time fineness; calculating the error between the real-time fineness and the target fineness; setting the proportional coefficient to 0.8, the integral coefficient to 0.05 and the differential coefficient to 0.1, obtaining the speed adjustment amount through the control amount calculation relationship; the actual grinding speed is the sum of the initial grinding speed and the speed adjustment amount; repeat the above process until the real-time fineness ≤30μm, stop grinding, and get the final fineness.

[0020] In the specification, the parameter transfer process of S1 to S3 is specifically: the average particle size of the nano hollow silica microspheres and the amount of the nano hollow silica microspheres determined by the optimal formula parameters directly affect the time regression coefficient and the rotation speed regression coefficient of the multiple linear regression model; the initial fineness of the pre-dispersed slurry obtained in S2 directly affects the time regression coefficient and the rotation speed regression coefficient.

[0021] In the specification, the specific feedback adjustment process of the closed loop in S5 includes: if the thermal conductivity in the heat insulation performance detection is > 0.04 W / (m·K), return to S1 step, increase the upper limit of the amount range of the nano hollow silica microspheres in the multi-objective genetic algorithm, and reiterate optimization; if the volatile organic compound content in the environmental protection performance detection is > 50 g / L, return to S1 step, replace the film additive type to be an environmentally friendly dodecanol ester, and at the same time, adjust the total amount range of the additive from 3-5 parts to 2.5-4.5 parts, and re-determine the optimal formula parameters; if the adhesion grade in the physical and mechanical performance detection is > 1 grade, return to S1 step, adjust the amount range of the pure acrylic emulsion from 30-40 parts to 32-42 parts, and re-execute S1 to S5 steps; until all detection items are qualified, the closed loop ends.

[0022] In the specification, the pretreatment operation of the packaging barrel in S6 includes: selecting a 20L or 10L galvanized iron barrel, coating an epoxy resin anti-rust layer on the inner wall of the barrel to prevent corrosion, first washing the barrel with deionized water, drying in a 60℃ oven for 2 hours, and detecting the moisture content in the barrel by weighing method to ensure that the moisture content is ≤0.1g.

[0023] A kind of green building outer wall coating with heat insulation function, which is prepared by the preparation method of the green building outer wall coating with heat insulation function described in any one of the above.

[0024] In summary, the present application has at least the following beneficial effects:

[0025] (1) synergistic improvement of thermal insulation performance and environmental protection performance: through systematic formula optimization, the content of volatile organic compounds (VOC) and harmful substances is strictly controlled to meet the environmental protection standards of green buildings while the thermal insulation function of the coating is strengthened (which can effectively reduce the heat transfer coefficient of the building outer wall and reduce the heat exchange between indoor and outdoor), avoiding the contradiction of "meeting the thermal insulation standard but exceeding the environmental protection standard";

[0026] (2) intelligentization and high efficiency of preparation process: the "prediction-feedback" fusion algorithm is used in the grinding process, and the parameters are dynamically adjusted according to the characteristics of the raw materials, which avoids the waste of energy and the fluctuation of performance caused by extensive control, improves the grinding efficiency and the stability of slurry fineness, and at the same time, the parameter transfer is clear throughout the whole process, which reduces the rework caused by parameter deviation and improves the production qualification rate;

[0027] (3) Coating workability and durability optimization: The ratio of additives to film-forming substances in the formula is scientifically calculated to ensure that the coating viscosity, solid content, and other construction parameters are compatible with conventional exterior wall construction processes (such as spraying and rolling), and because the raw materials are evenly dispersed, the durability indicators such as coating adhesion, scrub resistance, and water resistance are significantly improved, extending the service life of the coating;

[0028] (4) Full-process traceability and controllability: From algorithm optimization of formula design to parameter detection and feedback at each step, a complete "design-production-detection" closed loop is formed, and if a certain link does not meet the performance standards, it can be precisely traced back and adjusted (such as optimizing the thermal insulation filler parameters when the thermal insulation is insufficient), improving the flexibility and reliability of the technical solution;

[0029] (5) Raw material utilization rate improvement: Through precise formula parameters and process control, excessive addition of raw materials (such as not increasing the amount of fillers to compensate for insufficient grinding) or waste (such as avoiding excessive grinding that causes raw material loss) is reduced, reducing production costs while reducing waste emissions, in line with the concept of low-carbon production. BRIEF DESCRIPTION OF DRAWINGS

[0030] Figure 1 is a flowchart of the preparation method of the green building exterior wall coating with thermal insulation function involved in the present application.

[0031] Figure 2 is a flowchart of the multi-objective genetic algorithm (MOGA) optimization involved in the present application.

[0032] Figure 3 is a flowchart of the raw material pre-dispersion treatment involved in the present application.

[0033] Figure 4 is a flowchart of the coating mixing and conditioning involved in the present application. DETAILED DESCRIPTION

[0034] The embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0035] As shown in Figure 1 , the present embodiment provides a preparation method of a green building exterior wall coating with thermal insulation function, comprising the following:

[0036] This scheme is aimed at the preparation of a green building exterior wall coating with thermal insulation function, and a full-process technology system of "formula optimization-pre-dispersion-intelligent grinding-mixing conditioning-performance detection-packaging into warehouse" is constructed, the core innovation lies in the optimization of two key algorithms driven links, and the close connection of each step parameter is realized:

[0037] (1) Formulation design stage: The "analytic hierarchy process (AHP)-multi-objective genetic algorithm (MOGA)" fusion model is adopted, with thermal insulation, environmental protection and workability as objective functions. The mass fraction and key characteristic parameters (such as thermal insulation filler particle size and pigment fineness) of film-forming substances, thermal insulation fillers, pigments and fillers, additives and solvents are systematically optimized to output a scientifically suitable formulation scheme.

[0038] (2) Pre-dispersion stage: The amount of raw materials is controlled according to the formula parameters. The solid raw materials and liquid media are initially mixed by a high-speed disperser. The fineness of the pre-dispersion slurry is detected to provide qualified initial materials for subsequent grinding.

[0039] (3) Intelligent grinding process: Taking the formula parameters and the fineness of the pre-dispersed slurry as input, the initial grinding time and speed are first predicted by the multiple linear regression (MLR) algorithm, and then the speed is adjusted in real time by the PID feedback control algorithm to ensure that the fineness of the slurry is accurately met and to avoid the defects of coarse control.

[0040] (4) Mixing and performance testing: The qualified slurry is mixed with film-forming substances and remaining additives to adjust the viscosity and solid content of the coating. Then, the product quality is verified through comprehensive performance testing (heat insulation, environmental protection, durability, etc.). The test results can be fed back to the formulation stage for dynamic adjustment.

[0041] (5) Packaging and warehousing: Qualified coatings are filled, sealed and labeled in a standardized manner. The labeling content covers the key parameters of the formula and performance indicators to ensure the traceability of the product storage and use process.

[0042] S1: Determination of raw material formulation parameters

[0043] In the preparation of green building exterior wall coatings, the raw material formulation is the core foundation determining the thermal insulation performance, environmental friendliness, and workability of the final product. This step utilizes a fusion model of the Analytic Hierarchy Process (AHP) and Multi-Objective Genetic Algorithm (MOGA) to combine subjective experience with objective data, achieving scientific optimization of formulation parameters. The output results (the dosage of each raw material and key characteristic parameters) will directly serve as the core input for subsequent steps (especially the S3 grinding treatment), ensuring the continuity of the process chain.

[0044] 1.1 Model Building: Design of the Correlation Between Objective Function and Decision Variables

[0045] To balance the multiple performance requirements of coatings, the model is based on the principle of "prioritizing thermal insulation, supplementing with environmental protection, and taking into account construction", and constructs three mutually constrained objective functions, and clarifies the physical meaning and scope of the decision variables.

[0046] Objective function definition:

[0047] 1. Thermal insulation performance target: ,in The thermal conductivity of the coating (unit: W / (m·K)) is required according to GB / T10294-2008. This indicator is primarily determined by the amount and particle size of the heat-insulating filler (nano-hollow silica microspheres); the higher the hollowness of the microspheres and the greater the amount used, the better. The smaller.

[0048] 2. Environmental performance targets: ,in The content of volatile organic compounds (unit: g / L) is required according to GB18582-2020. This indicator is determined by the type and amount of film-forming aids and solvents. Among film-forming aids, dodecyl alcohol esters are more environmentally friendly than glycol ethers, so they are preferred.

[0049] 3. Construction performance targets: ,in The viscosity of the coating is expressed in mPa·s. For optimal application viscosity (determined through industry application experience), the following requirements are required. This indicator is controlled by the amount of thickener and the solid content; the higher the amount of hydroxyethyl cellulose, the lower the solid content. The larger.

[0050] Decision variable definition (all are mass fractions, and the sum must satisfy constraints): The amount (parts) of pure acrylic emulsion, as a film-forming substance, affects the adhesion and water resistance of the coating, ranging from [30, 40] (too little will result in incomplete film formation, too much will increase VOCs). : Dosage of nano-hollow silica microspheres (parts), core heat insulation filler, range [10,20] (too little will result in insufficient heat insulation, too much will cause the coating to crack easily); : Amount (parts) of titanium dioxide (rutile type) to provide hiding power, range [5,10] (too little will result in insufficient hiding power, too much will increase costs); : Dosage (parts) of heavy calcium carbonate powder (800 mesh) to adjust coating cost and hardness, range [10, 15]; Total amount of additives (parts), including dispersant ( ), defoamer ( Thickener ), film-forming aids ( ),Right now The range is [3,5] (too little will lead to uneven dispersion or insufficient defoaming, too much will increase VOCs). The amount (parts) of deionized water used is used as a solvent to adjust viscosity, with the following constraints: (Total mass parts: 100).

[0051] 1.2 Model Training: Co-optimization Process of AHP and MOGA

[0052] Model training is divided into two stages: “Target Weight Determination (AHP)” and “Recipe Parameter Optimization (MOGA). By combining subjective weights with objective algorithms, the practicality of the optimization results is ensured.

[0053] 1. Using the Analytic Hierarchy Process (AHP) to determine target weights:

[0054] Step 1: Construct a judgment matrix. Invite 5 coating industry experts (with ≥10 years of experience in formulation development) to rate the importance of the three objectives using a 1-9 scale (1 = equally important, 9 = extremely important). For example, thermal insulation performance ( ) and environmental performance ( The importance ratio of ) is 2 ( Slightly important), and construction performance ( The ratio is 3 ( (This is of higher importance), and the final judgment matrix is ​​obtained by integrating expert opinions:

[0055] ;

[0056] Step 2: Calculate the weight vector. Solve using the eigenvalue method. Maximum eigenvalue , corresponding feature vector The weights are obtained after normalization. (Insulation) (Environmental friendly), (construction).

[0057] Step 3: Consistency Check. Calculate the consistency index. The random consistency index RI = 0.58 (3rd order matrix), and the consistency ratio CR = CI / RI = 0.0078 < 0.1, indicating that the matrix is ​​valid.

[0058] 2. Multi-objective genetic algorithm (MOGA) optimization (process as follows) Figure 2 (as shown)

[0059] Step 1: Population Initialization. Randomly generate 100 initial solutions (chromosomes), each chromosome being a 6-dimensional vector. This satisfies the constraints on the range of each variable and the total number of parts.

[0060] Step 2: Fitness Function Construction. The three objective functions are weighted and integrated into a single-objective fitness function (higher values ​​indicate better fitness): ;in, , Calculated using empirical formulas (such as...) (Based on historical data fitting).

[0061] Step 3: Genetic manipulation.

[0062] Selection: A roulette wheel selection method is used, where chromosomes with higher fitness have a higher probability of being selected; Crossover: Arithmetic crossover is performed on the selected chromosomes, for example, in the parent generation. offspring Crossover probability 0.8; Mutation: Randomly select one gene from the chromosome and randomly perturb it within its range (e.g., ... (From 15 to 15±1), the mutation probability is 0.05;

[0063] Step 4: Iterative optimization. Repeat the above operation for 50 generations, retaining the top 20% of chromosomes in fitness in each generation. The chromosome with the highest fitness in the 50th generation is the optimal solution.

[0064] 1.3 Model Application and Output: Determination of Optimal Formulation Parameters

[0065] The trained model was applied to actual formulation design. By inputting basic raw material parameters (such as 90% hollowness of nano-hollow silica microspheres and 95% whiteness of titanium dioxide), the optimal formulation parameters were obtained as follows (all parameters will be directly used in subsequent steps):

[0066] One part (pure acrylic emulsion, solid content 50%); Parts (nano hollow silica microspheres, average particle size) ); Parts (titanium dioxide, whiteness) ); (Calcium carbonate powder, fineness) item); (Additives:) Dispersant One part of defoamer, Thickener, (film-forming aids); Parts (deionized water, conductivity) Simultaneously output key performance prediction values: , , All of them meet the target requirements.

[0067] S2: Raw material pre-dispersion treatment

[0068] Pre-dispersion is the process of initially mixing solid raw materials (pigments, fillers, heat-insulating fillers) with a liquid medium (water, dispersant). Its core objective is to break up raw material agglomerates and form a uniform pre-dispersion slurry (fineness). This lays the foundation for efficient grinding in S3. This step strictly controls the amount of raw materials according to the formula parameters output by S1, and ensures that the dispersion effect matches the grinding requirements of S3 through testing.

[0069] 2.1 Distributed Equipment and Preliminary Preparations

[0070] Equipment selection: Use a 100L high-speed disperser (power 5.5kW, speed range 0~3000r / min), with a dispersion disc diameter of 150mm (to ensure uniform mixing);

[0071] Raw material pretreatment: 15 parts of nano-hollow silica microspheres determined in S1 ( ), 8 parts titanium dioxide ( ), 12 portions of heavy calcium powder ( Each particle is passed through a 120-mesh sieve to remove mechanical impurities (to avoid impurities affecting the subsequent grinding accuracy).

[0072] The dispersant (0.8 parts, 40% active ingredient) and defoamer (0.5 parts, 30% active ingredient) should be diluted with deionized water in advance (mass ratio 1:1) to facilitate uniform dispersion.

[0073] Deionized water (S1) The test involves two phases: 18 portions in this step and 6.8 portions in step S4, to measure conductivity. (Ensure dispersion effect without ion interference).

[0074] 2.2 Pre-dispersion operation process (controlled step-by-step in sequence, the process is as follows) Figure 3 (As shown)

[0075] 1. Preparation of dispersant aqueous solution:

[0076] Add 18 parts of deionized water to the dispersion tank, turn on the disperser, and set the initial speed to 800 rpm (low-speed stirring to avoid splashing). While stirring, slowly add 0.8 parts of the dispersant (diluted), and continue stirring for 5 minutes until the dispersant is completely dissolved. At this point, measure the solution concentration using a handheld refractometer and calculate the concentration of the effective component of the dispersant.

[0077] The concentration must be ≥1.5% to ensure that the dispersant can effectively encapsulate the solid particles; if it does not meet the standard, add 0.1 parts of dispersant and stir for 2 minutes before retesting.

[0078] 2. Addition of pigments, fillers, and insulating fillers:

[0079] Maintain a rotation speed of 800 rpm, first add 8 parts of titanium dioxide (rutile type), and stir for 3 minutes (titanium dioxide has a high density, so it disperses preferentially to avoid sedimentation); then add 12 parts of heavy calcium carbonate powder (800 mesh), and stir for 3 minutes (heavy calcium carbonate powder is easy to disperse and forms a synergistic filling with titanium dioxide); finally add 15 parts of nano-hollow silica microspheres ( Stir for 5 minutes (microspheres are light and easily aggregate, so extending the stirring time ensures initial dispersion). After adding each ingredient, check the inner wall of the dispersion tank for residue. If residue is found, rinse with a small amount of deionized water (≤0.5 parts) to avoid material loss.

[0080] 3. Defoaming and high-speed dispersion: Add 0.5 parts of defoamer (after dilution), increase the speed of the disperser to 1200 r / min (high-speed shearing breaks up agglomerates), and continue stirring for 10 minutes. At this time, the temperature of the slurry will rise to 35-40℃ due to shearing action. The temperature needs to be controlled ≤45℃ (to avoid defoamer inactivation).

[0081] 4. Fineness testing of pre-dispersed slurry: The fineness of the slurry is tested using a scraper fineness gauge (range 0-100μm): A small amount of slurry is dropped into the groove of the fineness gauge, and scraped vertically across it with a scraper. The maximum particle size of the undispersed particles is observed within 30 seconds. If the test result... (If the initial conditions for S3 grinding are met), then pre-dispersion is complete; if If the high-speed stirring is extended by 5 minutes, the test should be repeated until the standard is met.

[0082] 2.3 Output Results

[0083] A pre-dispersed slurry was generated (approximately 54.3 parts in total, including raw materials and 18 parts water). Key parameter: fineness. (Directly used as input variables for S3 Algorithm 1); Solid content (This provides a reference for concentration control during the S3 grinding process).

[0084] S3: Slurry grinding process based on a "prediction-feedback" fusion algorithm

[0085] Slurry pre-dispersed by S2 (fineness) Further grinding is required. This ensures uniform dispersion of the thermal insulation filler and pigments, improving coating stability. This step employs two correlation algorithms: "Multiple Linear Regression (MLR) Initial Parameter Prediction" and "Proportional, Integral, and Derivative Feedback Control (PID Feedback Control)." Using the formula parameters output by S1 as input, the grinding parameters are adjusted in real time to achieve a highly efficient and energy-saving grinding process.

[0086] 3.1 Algorithm 1: Multiple Linear Regression (MLR) Initial Grinding Parameter Prediction Model

[0087] The core function of this model is to predict the initial grinding time and speed in advance based on the characteristics of the raw materials and the pre-dispersion results, avoiding blind trial and error and providing basic parameters for subsequent PID control. Its input comes directly from the formula parameters of S1 and the pre-dispersion results of S2, and its output serves as the initial conditions for Algorithm 2.

[0088] 3.1.1 Model Construction: Physical Relationship Between Input and Output

[0089] The model input variables are all key factors affecting grinding efficiency: Average particle size (nm) of hollow silica nanospheres, derived from S1 (The smaller the particle size, the easier it is to grind, but excessive grinding may damage the hollow structure.) The amount (parts) of nano-hollow silica microspheres from S1 (The more you use, the greater the grinding resistance and the longer it takes.) The fineness (mesh) of the heavy calcium carbonate powder, from S1 Mesh (the larger the mesh number, the smaller the initial particle size, and the shorter the grinding time);

[0090] Initial fineness (μm) of S2 predispersed slurry, as shown in the test results from S2 (example). The greater the initial fineness, the longer the grinding time required.

[0091] The model output variables are the initial control parameters of the grinding process: Initial grinding time (min) refers to the estimated time to reach the target fineness; Initial grinding speed (r / min) refers to the initial operating speed of the sand mill (the higher the speed, the higher the grinding efficiency, but the greater the energy consumption).

[0092] Based on the physical correlation of the above variables, a linear regression model is constructed:

[0093] (1)

[0094] (2)

[0095] in: , , , For time regression coefficients (units: min / nm, min / part, min·mesh, min / μm); , , , The rotational speed regression coefficients (units: r / (min·nm), r / (min·part), r·mesh / min, r / (min·μm)); , For constant terms (unit: min, r / min).

[0096] 3.1.2 Model Training: Fitting and Validation of Historical Data

[0097] The model was trained using 50 sets of historical production data (each set containing the above-mentioned input and output variables), and the coefficients were fitted using the least squares method to verify the model's accuracy.

[0098] Data preprocessing: Remove outliers (such as sudden increases in grinding time due to equipment failure), and standardize input variables (e.g., ...). Standardized to ), to avoid the influence of dimensions.

[0099] Coefficient fitting: Linear regression was performed using Python's scikit-learn library to obtain the fitting coefficients.

[0100] (For every 1 nm increase in particle size, the initial time increases by 0.02 min). =0.5 (For every additional serving, the time increases by 0.5 minutes). =1000 (For every 1 / mesh increase in the reciprocal of the fineness, the time increases by 1000 min·mesh, that is, 800 mesh takes about 2.08 min less time than 600 mesh). =0.8 (For every 1 μm increase in initial fineness, the time increases by 0.8 min). =5;

[0101] (For every 1nm increase in particle size, the initial rotation speed decreases by 1.2r / min, because larger particles require lower rotation speeds to avoid breakage.) (For every additional unit of dosage, increase the rotation speed by 8 r / min to improve efficiency). (For every 1 mesh increase in the reciprocal of the fineness, the rotation speed decreases by 500 r·mesh / min, meaning that fine fillers require a lower rotation speed.) (For every 1 μm increase in initial fineness, the rotation speed increases by 2 r / min). =1000.

[0102] Model validation: Calculating the coefficient of determination ( ) and 0.90 ( This indicates that the model can explain more than 90% of the variable changes, and the fitting effect is good.

[0103] 3.1.3 Model Application and Output: Calculation of Initial Parameters

[0104] Substitute the actual parameters of S1 and S2 into formulas (1) and (2) to calculate the initial grinding parameters:

[0105] initial time : Initial speed : Output result: , , which serves as the initial input for Algorithm 2.

[0106] 3.2 Algorithm 2: Real-time Grinding Parameter Adjustment Model Based on PID Feedback Control

[0107] The model outputs Algorithm 1. , Starting from this point, the grinding speed is dynamically adjusted by detecting the error between the fineness of the slurry and the target value in real time, thus solving the problem of "initial prediction deviation" (such as the fluctuation of grinding efficiency caused by the difference in raw material batches) and ensuring that the final fineness meets the standard.

[0108] 3.2.1 Model Construction: Dynamic Correlation between Error and Control Variable

[0109] Define core variables: The fineness of the slurry at time t (μm) was measured by a scraper fineness gauge (sampled once every 15 minutes, with a sample volume of 5 mL, and the average value was taken after 3 tests). Target fineness (industry standard requirements); error (μm): If This indicates that the fineness is not up to standard and grinding needs to be intensified; if This indicates that the standard has been met; control quantity Speed ​​adjustment amount (r / min), i.e., actual operating speed. This indicates an acceleration. (This indicates a slowdown).

[0110] The PID control model adjusts the proportional (P), integral (I), and derivative (D) components in a coordinated manner. The formula is:

[0111] (3)

[0112] in: =0.8 (proportional coefficient): directly responds to the current error. The larger, The larger the value, the faster the error is reduced; =0.05 (integral coefficient): Accumulates historical errors and eliminates steady-state deviations (if small errors exist for a long time, they are gradually adjusted through the integral term). =0.1 (differential coefficient): reflects the rate of change of error and suppresses overshoot (e.g., when the error decreases rapidly, reduce the adjustment range to avoid excessive grinding).

[0113] 3.2.2 Model Training: PID Parameter Tuning Process

[0114] The PID parameters are determined by trial and error to ensure system stability and rapid response.

[0115] Proportional circuit debugging: First set =0、 =0, gradually increase Until the system experiences slight fluctuations ( =1.0 (speed fluctuation ±50r / min), then reduce by 20% to get =0.8; Integral stage debugging: maintain =0.8、 =0, gradually increase steady-state error ≤2μm When the value is 0.05, the error stabilizes within 1 μm); Differential element adjustment: maintain =0.8、 =0.05, gradually increase Up to overshoot ≤5% When the value is 0.1, the overshoot is 3%.

[0116] 3.2.3 Model Application and Interaction Process

[0117] The grinding process uses Algorithm 1 , Starting with PID control and combining it with real-time adjustments, the specific steps are as follows:

[0118] 1. Minutes 0-15: At initial speed Run the test and take samples after 15 minutes.

[0119] (Detected 3 times: 44μm, 45μm, 46μm, average 45μm); Error Integral term (The error is constant at 15 μm for the first 15 minutes); Differential term (The error increases from 0 to 15 μm with a positive rate of change); control quantity Actual speed (Increase the rotation speed to speed up grinding).

[0120] 2. 15-30 minutes: Press Run the test and take samples after 30 minutes.

[0121] (Detected 3 times: 37μm, 38μm, 39μm, average 38μm); Error Integral term (The error is 8 μm in the last 15 minutes); Differential term (Error decreases, rate of change becomes negative); control quantity Actual speed (Maintain a high speed, as errors still exist).

[0122] 3. 30-45 minutes: Press Run the test and take samples after 45 minutes.

[0123] (Detected 3 times: 31μm, 32μm, 33μm, average 32μm); Error Integral term (The error is 2 μm in the last 15 minutes); Differential term (The error continues to decrease); control quantity Actual speed (Reducing the speed appropriately to avoid over-grinding).

[0124] 4. 45-55 minutes: Press Running, 55 minutes ( Sampling and testing at specific times:

[0125] (Meets standards), error Stop grinding and output the final fineness. , as the input parameter for S4 hybrid modulation.

[0126] Step connection and algorithm fusion logic

[0127] 1. Parameter transfer from S1 to S3: Particle size of hollow silica microspheres output by S1 Dosage Fineness of heavy calcium carbonate powder The initial fineness of the output of S2 Together, they serve as inputs to S3 Algorithm 1, and the initial grinding parameters are directly determined by formulas (1) and (2). , This achieves a seamless connection between "formula design" and "process parameters".

[0128] 2. S3 internal algorithm linkage: Output of Algorithm 1 , These are the initial conditions for Algorithm 2. Algorithm 2 corrects the prediction deviation of Algorithm 1 through real-time error adjustment (Formula 3). The two work together to improve grinding efficiency by 10% (from the original 60 minutes to 55 minutes) and increase the fineness qualification rate from 90% to 98%.

[0129] 3. Impact of S3 on subsequent steps: the final fineness of the S3 output. This will directly affect the dispersion uniformity of the S4 mixture (the smaller the particle size, the easier it is for the additives to disperse during mixing), and will ultimately be verified through the performance testing of S5 (such as heat insulation and scrub resistance), forming a closed-loop feedback.

[0130] S4: Coating Mixing and Preparation (Integrating grinding slurry with other components to control final performance)

[0131] Mixing and preparation is the process of combining the properly ground slurry with film-forming substances (pure acrylic emulsion), remaining additives, and deionized water in a specific ratio to form a homogeneous and stable coating system. This step requires strict control of the raw material addition based on the formulation parameters of S1, and using the grinding fineness of S3 as a prerequisite. Viscosity and solid content are adjusted to ensure the coating meets the application and performance requirements. The specific process is as follows: Figure 4 As shown.

[0132] 4.1 Mixing equipment and raw material preparation

[0133] Equipment selection: 500L low-speed mixing tank (power 3kW, speed range 0-600r / min), equipped with paddle stirring blades (to avoid excessive shear force that could damage the nanosphere structure);

[0134] Raw material verification: Film-forming substance: S1 One batch of pure acrylic emulsion (50% solids content, pH value tested at 7.5-8.5 to ensure stability); Grinding slurry: S3 output. Slurry (approximately 54.3 parts in total); Remaining additives: S1 Thickener (hydroxyethyl cellulose, pre-soaked in 5 times the volume of deionized water for 2 hours), Film-forming aid (dodecyl alcohol ester, purity ≥99%); deionized water: the remaining 6.8 parts from S1 (from the same batch as used in S2). ).

[0135] 4.2 Hybrid Modulation Operation Procedure (Stage-by-Stage Control)

[0136] 1. Pretreatment of pure acrylic emulsion: Add 35 parts of pure acrylic emulsion to a mixing tank, turn on the mixer, set the speed to 300 r / min (low speed mixing to avoid emulsion breakage), and mix for 5 minutes to make the emulsion evenly dispersed. At this time, the emulsion temperature is measured to be 25℃ (room temperature, which meets the process requirements).

[0137] 2. Addition and mixing of grinding slurry: Under stirring at 300 rpm, slowly pump the S3 grinding slurry (29 μm) into the mixing tank through a pipeline, controlling the addition rate at 5 L / min (to be added in about 10 minutes) to avoid excessive local concentration leading to agglomeration. After addition is complete, increase the stirring speed to 500 rpm (to enhance mixing intensity) and stir for 20 minutes to form an "emulsion-slurry" mixture system (at this point, the system is a grayish-white homogeneous fluid).

[0138] 3. Addition and dispersion of additives: Keep the rotation speed at 500 r / min, first add the swollen thickener (0.6 parts), stir for 8 minutes (the thickener needs to be fully dispersed to avoid the formation of "fish eyes"); then add 2.3 parts of film-forming aid, stir for 8 minutes (the film-forming aid needs to be fully compatible with the emulsion to improve the continuity of film formation).

[0139] 4. Viscosity and solid content adjustment: Add 6.8 parts of deionized water, reduce the rotation speed to 400 r / min, and stir for 15 minutes. At this time, measure the key indicator: viscosity. The results were obtained using a rotational viscometer (NDJ-5S type, rotor No. 2, 60 r / min). (in S1 prediction) Within the specified range, meeting construction requirements); solid content According to GB / T1725-2007, the determination of nonvolatile matter content in paints, varnishes and plastics, take 2g of sample and dry it in an oven at 105℃ for 3 hours, calculate: The theoretical calculated value of S1 is 48%, while the actual value is within the acceptable range of 46% to 50%, so no adjustment is needed.

[0140] 5. Homogenization treatment: Reduce the rotation speed to 200 r / min and stir for 5 minutes (to reduce shear force and avoid introducing air bubbles), and finally form a uniform coating system without particles or air bubbles, which is the preliminary finished coating.

[0141] 4.3 Output Results

[0142] Generate a preliminary finished coating (total quantity 100 parts), key parameters: Fineness: follow S3. (Because no new solid particles are added during the mixing stage, the fineness remains unchanged); viscosity Solid content (As a basic parameter for S5 performance testing).

[0143] S5: Performance testing of finished coatings (verifying the optimization goals of S1 and forming a closed-loop feedback).

[0144] Performance testing is a crucial step in evaluating whether a coating meets design requirements. It is necessary to comprehensively verify indicators such as thermal insulation, environmental friendliness, and physical and mechanical properties. The test results are directly related to the optimization goals of S1. If the standards are not met, feedback will be sent to S1 to adjust the formula to ensure that the final product is qualified.

[0145] 5.1 Detection Items and Methods (corresponding to the objective function of S1)

[0146] 1. Thermal insulation performance testing (verification) According to GB / T10294-2008 Determination of Steady-State Thermal Resistance and Related Properties of Thermal Insulation Materials - Protective Hot Plate Method, the DRPL-III type protective hot plate method thermal resistance tester was used.

[0147] Sample preparation: The coating was applied to a 5mm thick cement mortar substrate (300mm × 300mm), and the dry film thickness was controlled at 1.5mm (the wet film thickness was measured to be 3mm using a wet film comb). (Conversion), and maintain for 7 days at 23℃ and 50% humidity;

[0148] Test results: (Better than the predicted value of S1 by 0.038, meeting the insulation target).

[0149] 2. Environmental performance testing (verification) ):

[0150] VOC content: Detected according to GB18582-2020 using a GC-2014 gas chromatograph. A 10g sample was taken and the result was... (Better than the predicted value of S1 by 45); Free formaldehyde: Detected according to GB / T23993-2009 using a 722 spectrophotometer. Take a 5g sample and measure the free formaldehyde. (Meets environmental protection requirements).

[0151] 3. Physical and mechanical property testing (to assist in verifying workability and durability):

[0152] Adhesion: According to GB / T9286-1998, use a cross-cut tester (1mm spacing) to test. After the cross-cut is completed, use 3M tape to adhere and peel off. The adhesion grade is 0 (better than ≤1 grade). Scrub resistance: According to GB / T9266-2009, use a JISK5600 scrub resistance tester (500g load) to test. The scrub resistance is ≥5000 cycles (6200 cycles). Water resistance: According to GB / T1733-1993, immerse the test plate in 23℃ deionized water for 96 hours. Observe for no bubbling or peeling (qualified).

[0153] 4. Appearance and storage stability:

[0154] Appearance: When observed at a distance of 1m from the sample under natural light, the coating showed no lumps or sedimentation, and the color was uniform (ΔE≤1 compared to the standard color card); Storage stability: After sealing the coating and placing it in a 50℃ oven for 30 days, there was no layering or hardening, and the viscosity change rate was ≤5% (qualified).

[0155] 5.2 Feedback of test results (closed-loop control)

[0156] All test items passed, confirming this batch of paint as "qualified finished paint"; if any item fails to meet the requirements (e.g., assuming...), the batch is considered qualified finished paint. If the result is not satisfactory, return to step S1 to adjust the formula (e.g., increase the amount of nano-hollow silica microspheres to 16 parts), and repeat steps S2-S5 until the result is satisfactory.

[0157] S6: Qualified finished paint packaging (ensuring product quality and complete information)

[0158] Packaging is the final step in the manufacturing process. It requires standardized filling, sealing, and labeling to ensure the coating's quality remains stable during storage and transportation, while also clearly conveying product information to users (including the core parameters of S1 and the test results of S5).

[0159] 6.1 Preparations before packaging

[0160] Packaging drum selection: 20L / drum (main specification) and 10L / drum (secondary specification) galvanized iron drums (with an epoxy resin anti-rust coating on the inner wall to prevent reaction with the paint); Drum treatment: Rinse the inside of the drum with deionized water and dry in a 60℃ oven for 2 hours to ensure that there are no impurities inside the drum (visual inspection) and the moisture content is ≤0.1g (weighing method: empty drum weight). Quality after drying , ).

[0161] 6.2 Filling and Sealing

[0162] Homogenization: Pump the qualified S5 finished coating into the pre-filling mixing tank and stir at 200 rpm for 5 minutes (to ensure uniformity before filling); Automatic filling: Use a GF-200 automatic filling machine and set the filling volume to 20L drums. 10L bucket For every 10 barrels filled, one barrel is randomly selected for weighing (the weight of a 20L barrel of paint is approximately 24kg, as the density of the paint is approximately 1.2kg / L). If the error exceeds the range (e.g., 20.3L), the flow rate parameters of the filling machine are adjusted (from 10L / min to 9L / min) and the filling is repeated. Sealing test: Immediately after filling, the cap (with rubber sealing ring) is added and tightened with a wrench (torque 30N·m). The container is then inverted for 30 minutes to observe for any leakage (qualified).

[0163] 6.3 Label Identification (Associating S1 and S5 Parameters)

[0164] Affix a label to the bucket, including: Product Name: Green Building Exterior Wall Coating with Heat Insulation Function; Key Formula Parameters: 35 parts pure acrylic emulsion, 15 parts nano hollow silica microspheres ( ), titanium dioxide 8 parts, etc.; performance indicators (from S5): Adhesion rating: 0; Storage information: Shelf life: 12 months; Storage temperature: 5-35℃; Relative humidity: ≤70%; Standard: GB / T9755-2014 Synthetic resin emulsion exterior wall coatings.

[0165] 6.4 Warehousing: The packaged paints are stored in the finished product warehouse in batches, with a stacking height of ≤3 layers (to avoid deformation of the bottom buckets under pressure). The warehouse is equipped with a thermometer and hygrometer (for real-time monitoring) and a ventilation system, and is kept away from fire sources and corrosive substances to complete the entire preparation process.

[0166] Summary of the overall process flow logic:

[0167] 1. Parameter transfer chain: S1 formulation parameters → S2 pre-dispersed raw material dosage → S3 initial grinding conditions → S4 mixing and modulation ratio → S5 performance testing standards → S6 labeling information, forming a complete data chain of "design-production-testing-labeling".

[0168] 2. Feedback loop: The test results of S5 are directly fed back to S1. If the performance does not meet the standards, the formula is adjusted to ensure that the final product meets the optimization goals of S1.

[0169] 3. Algorithm Synergy: S1's multi-objective optimization algorithm provides scientific formulas for the entire process, while S3's "prediction-feedback" algorithm ensures grinding efficiency. Together, they improve process stability (from 85% to 98% pass rate) and raw material utilization (reducing losses by 5%).

[0170] Through meticulous design and close integration of each step, the final product not only meets the insulation and environmental protection requirements of green buildings, but also achieves a balance between performance, efficiency and cost through a traceable parameter chain and optimization algorithm.

[0171] In some embodiments, the introduction of Bayesian networks addresses the uncertainty caused by raw material fluctuations and process disturbances. By establishing a correlation model of "raw materials-process-performance" through probabilistic reasoning, the parameters of each step are transmitted more accurately, thereby improving the performance compliance rate.

[0172] Determination of raw material formulation parameters based on "Bayesian network-multi-objective optimization" fusion

[0173] This step is the core design phase of the process. It uses Bayesian networks (BNs) to handle the uncertainties in the relationship between raw material parameters and performance, providing probabilistic constraints for the multi-objective genetic algorithm (MOGA), ultimately outputting an optimal formulation that balances thermal insulation, environmental friendliness, and workability. Its output parameters will directly serve as the input benchmark for S2 pre-dispersion and S3 grinding, forming the basis for the overall process continuity.

[0174] 1.1 Bayesian Network Model Construction: Data-Driven Probabilistic Association

[0175] The core of Bayesian networks is to learn the conditional probabilities between variables through historical data, thereby achieving probabilistic predictions of "raw material parameters → performance indicators". Model construction requires three stages: "variable selection - structure learning - parameter learning" to ensure that the network can truly reflect the process rules.

[0176] (1) Variable selection and discretization:

[0177] Input variables (raw material parameters): Based on process experience and correlation analysis, four variables with significant impact on performance were selected (all from the S1 initial parameter pool):

[0178] A: The amount of nano-hollow silica microspheres used (parts) is physically represented by the proportion of heat insulation filler, and is discrete into three intervals: [10,13), [13,17), and [17,20] (the interval division is based on the distribution peak of historical data).

[0179] B: The particle size (nm) of the nano-hollow silica microspheres is a key parameter affecting the thermal insulation effect, and is discrete as [50,70), [70,90), [90,100] (refer to the particle size distribution range provided by the supplier).

[0180] C: Amount of pure acrylic emulsion (parts), core of film-forming material, discrete as [30,33), [33,37), [37,40] (taking into account both film integrity and cost);

[0181] D: Film-forming aid dosage (parts), affecting VOC and film-forming properties, discrete as [1.8, 2.1), [2.1, 2.5), [2.5, 3.0] (based on environmental standards and construction temperature requirements).

[0182] Output variables (performance metrics):

[0183] Thermal insulation performance ( Binarized values ​​are {pass, fail};

[0184] Environmental performance ( The binary value is {pass, fail}.

[0185] (2) Network structure learning:

[0186] The PC algorithm (based on conditional independence test) was used to learn the dependencies between variables from 500 sets of historical production data (laboratory formulation and performance records from 2019 to 2023). In the data preprocessing stage, outliers (such as sudden increases in VOC due to equipment failure) were removed, and missing values ​​were filled (using the KNN algorithm, k=5). The final network structure is as follows:

[0187] A and B directly affect (The thermal insulation performance is mainly determined by the thermal insulation filler).

[0188] C and D directly affect (VOCs mainly come from emulsions and film-forming aids).

[0189] There is a weak correlation between A and C (excessive filler may require increasing the amount of emulsion to ensure film formation).

[0190] (3) Calculation of Conditional Probability Table (CPT):

[0191] The conditional probability of each node is calculated using maximum likelihood estimation, for example:

[0192] When A=[13,17) (medium dosage) and B=[70,90) (medium particle size), (In 120 tests conducted on this combination in historical data, the insulation performance met the standards in 110 tests).

[0193] When C=[33,37) and D=[2.1,2.5), (VOCs met the standard in 88 out of 100 tests).

[0194] These probability values ​​will serve as "reliability metrics" for subsequent optimization.

[0195] 1.2 Optimization Process of Bayesian Network and MOGA

[0196] Bayesian networks do not replace MOGA, but rather modify MOGA's fitness function by providing a "performance achievement probability", reducing invalid searches and making the optimization direction more focused on the high-reliability recipe region.

[0197] (1) Probabilistic screening of the initial population: After MOGA randomly generates 100 initial recipes (chromosomes), the "joint achievement probability" of each recipe is calculated through a Bayesian network. (Based on the conditional independence of network structure). For example:

[0198] formula of , ,but ;

[0199] Eliminate The formula (28 in total) retains 72 high-potential individuals for iteration, reducing unnecessary computation by 30%.

[0200] (2) Probability-weighted modification of the fitness function: The original fitness function F only considers the performance optimization objective. After modification, probability weights are introduced, making it easier to retain the "high performance and high reliability" formula.

[0201] (4) Among them, (weight) =0.54, =0.30, =0.16). For example:

[0202] formula If F = 0.85 and P = 0.90, then F' = 0.765; Formula If F = 0.88 and P = 0.70, then F' = 0.616;

[0203] Obviously It is more likely to be selected for the next generation, avoiding the optimization of formulas that have "excellent theoretical performance but are difficult to meet in practice".

[0204] (3) Optimize the output and verification of results:

[0205] After 50 iterations, the optimal formula not only meets the performance indicators but also has high reliability: Formula parameters: A = 15 parts ( C=35 parts, D=2.3 parts (other parameters are the same as before); Performance prediction: , , Probability Guarantee: , (Verified via Bayesian network).

[0206] Slurry grinding treatment based on "prediction-feedback + Bayesian diagnosis"

[0207] The grinding process is a key step in determining the uniformity of coating dispersion. By introducing a Bayesian network, the causes of abnormalities can be diagnosed in real time and the control strategy can be adjusted to ensure the accuracy of the connection with S4 mixing modulation.

[0208] Construction of Bayesian Network Anomaly Diagnosis Model

[0209] The model uses "decreased grinding efficiency" as the diagnostic target, infers the cause of the abnormality through real-time parameters, and provides a basis for adjusting PID parameters, avoiding energy waste or damage to the microsphere structure caused by blindly increasing the speed.

[0210] 1. Variable definition and network structure:

[0211] Input variables (real-time monitoring data): : Current grinding speed (r / min); t: Grinding time (min); Fineness decrease over the past 15 minutes ( ),Right now ; Microsphere size (nm, from S1);

[0212] Output variables (anomaly types): Z: {Normal, Microsphere Agglomeration, Zirconium Bead Wear} (Three typical anomalies summarized based on historical fault records). Network structure display: and It directly affects Z (the decrease in efficiency is directly related to changes in rotational speed and fineness). Z is indirectly associated by influencing the probability of reunion.

[0213] 2. Application process for abnormal diagnosis:

[0214] When grinding for 15 minutes, it was detected (Normal size should be ≥10μm), triggering diagnosis:

[0215] Input real-time parameters: , , , ; Calculate the posterior probability: , , Diagnostic conclusion: The most likely cause of the decreased efficiency is microsphere aggregation (aggregates require higher shear force to break them up).

[0216] Synergistic Adjustment of Bayesian Diagnostics and PID Control

[0217] Based on the diagnostic results, the PID parameters are adjusted accordingly, rather than simply increasing the speed: If the diagnosis is "microsphere agglomeration" (requiring enhanced shear): the proportional coefficient is adjusted. Increase from 0.8 to 0.9 (to accelerate speed response) while keeping the integral and derivative coefficients unchanged; if diagnosed as "zirconium bead wear" (decreased grinding ability): in addition to increasing In addition, the grinding time needs to be extended. (Increased from 55 min to 60 min).

[0218] This batch was adjusted after being diagnosed with "microsphere aggregation". actual speed (2 r / min higher than the original adjustment value), test after 20 minutes. (Returned to normal), finally reaching its peak at 55 minutes. It saves 8% of energy compared to when it is not diagnosed.

[0219] Performance failure tracing and feedback based on Bayesian networks

[0220] Performance testing is not merely about verifying results; it's also about using Bayesian networks to reverse-engineer the causes of substandard performance, forming a closed loop of "detection-diagnosis-adjustment." For example, when thermal insulation performance... When a problem is found to be non-compliant, the root cause must be traced quickly to avoid rework of the entire process.

[0221] Bayesian diagnostic network construction and inference

[0222] The network uses "performance not meeting standards" as the result and works backward to deduce the most likely cause of raw material or process deviation. The node definitions are as follows:

[0223] Root node (potential cause): The hollowness of the nanospheres is insufficient (<90%). Grinding fineness exceeds standard ( ); The emulsion solids content is low (<48%).

[0224] Intermediate node (observable deviation): Measured particle size of microspheres (Insufficient hollowness is often accompanied by excessively large particle size); : (Exceeding the standard by 2μm); Emulsion solids content = 46% (2% low).

[0225] Leaf nodes (non-compliant results): ( ).

[0226] Reasoning process:

[0227] 1. Input leaf node evidence 2. Calculate the posterior probability of each root node: , , 3. Verify intermediate nodes: Trace back the records in S3. ( If true, confirm. This is the main cause; 4. Adjust the strategy: Return to S3, extend the grinding time to 60 minutes, keep the other step parameters unchanged, and restart production. (qualified).

[0228] Algorithm synergy value: Bayesian networks act as a "glue," making the optimization results of AHP-MOGA more reliable and the control of MLR-PID more precise, ultimately improving the coating performance compliance rate and reducing unit energy consumption.

[0229] By leveraging the probabilistic reasoning capabilities of Bayesian networks, the adaptability to complex industrial environments is enhanced, providing a replicable intelligent solution for the large-scale production of green building coatings.

[0230] A green building exterior wall coating with heat insulation function is prepared by applying the preparation method of the green building exterior wall coating with heat insulation function described in any one of the above-mentioned methods.

[0231] I. Determination of Raw Material Formulation Parameters (S1)

[0232] All raw materials are used in parts by weight, totaling 100 parts.

[0233] (a) Setting the objective function and constraints

[0234] 1. Core objective function: Thermal insulation performance objective: ,in The thermal conductivity of the coating (unit: W / (m·K)) must meet the following requirements. (Refer to GB / T10294-2008 standard); Environmental performance targets: ,in The volatile organic compound (VOC) content of the coating (unit: g / L) must meet the following requirements. (Refer to GB18582-2020 standard); Construction performance targets: ,in The viscosity of the coating is expressed in mPa·s. (Industry-leading application viscosity) must meet the following requirements. .

[0235] 2. Decision Variables and Constraints: Let the mass fraction of each raw material be... (Pure acrylic emulsion) (Nano-hollow silica microspheres) (Titanium dioxide) (Calcium carbonate powder) (Total dosage of adjuvants, including) Dispersant, Defoamer Thickener Film-forming aid), $x_6$ (deionized water), with the following constraints: ; Range of values ​​for each variable: , , , , , .

[0236] (II) Bayesian Network Probability Prediction

[0237] Construct a Bayesian network of "raw material parameters - performance compliance probability", and select 4 key raw material parameters as parent nodes (A: Dosage, B: Nanosphere particle size, C: Dosage, D: Dosage), 2 performance indicators for child nodes ( : qualified, : (Qualified), a conditional probability table is calculated using 500 sets of historical data. For example, when A=[13,17), B=[70,90), C=[33,37), D=[2.1,2.5), , The probability of the combined formulation meeting the standard is: .

[0238] (III) Optimization of Multi-Objective Genetic Algorithm

[0239] 1. Fitness Function Correction: By introducing Bayesian probability weights, the corrected fitness function is as follows:

[0240] ;in = 0.54 (heat insulation weight), = 0.30 (environmental protection weight), = 0.16 (construction weight).

[0241] 2. Iterative optimization: The initial population is 100. After 50 generations of genetic operations (crossover probability 0.8, mutation probability 0.05), the optimal formula is finally obtained: = 35 parts (pure acrylic emulsion, solid content 50%), = 15 parts (nano microspheres, particle size 75 nm), = 8 parts (titanium dioxide, whiteness 96%), = 12 parts (heavy calcium powder, 800 mesh), = 4.2 parts ( = 0.8, , , ), parts (deionized water, conductivity 5 μS / cm), and , .

[0242] II. Pretreatment of raw materials for predispersion (S2)

[0243] (I) Pretreatment and equipment preparation

[0244] 1. Raw material pretreatment: Pass parts of nano microspheres, parts of titanium dioxide, parts of heavy calcium powder through a 120-mesh sieve respectively to remove impurities; parts of dispersant, parts of defoamer are diluted with deionized water at a ratio of 1:1; parts of deionized water are divided into 18 parts (used in this step) and 6.8 parts (used in S4). <00​​​​​​​​​​​​​

[0249] 2. Addition of pigments, fillers and heat insulation fillers: Maintain a speed of 800 r / min, first add 8 parts of titanium dioxide and stir for 3 minutes; then add 12 parts of heavy calcium carbonate powder and stir for 3 minutes; finally add 15 parts of nanospheres and stir for 5 minutes. After each addition of a raw material, observe the tank wall for any residue (rinse with ≤0.5 parts of deionized water if there is any residue).

[0250] 3. Defoaming and high-speed dispersion: Add the diluted defoamer, increase the rotation speed to 1200 r / min, stir for 10 minutes, and monitor the slurry temperature to ≤45℃ (to avoid defoamer failure).

[0251] 4. Fineness test: The fineness of the slurry was tested using a 0-100μm scraper fineness meter, and the average value of 3 tests was taken. ,Require This step ultimately yields... The pre-dispersed slurry was prepared, and the solid content of the slurry was calculated simultaneously.

[0252] .

[0253] III. Slurry Grinding Process (S3)

[0254] (I) Prediction of initial grinding parameters for MLR

[0255] 1. Model Construction: Based on S1 (Nanosphere particle size), A = 15 parts (Nanosphere dosage) Mesh (fineness of heavy calcium carbonate powder), and S2 Using (initial fineness) as input, construct the initial grinding time. and rotational speed MLR model: ; ;in =0.02、 =0.5、 =1000、 =0.8、 =5; =-1.2、 =8、 =-500、 =2、 =1000.

[0256] 2. Parameter calculation: Substituting the data, we get: ; ;

[0257] (ii) Real-time adjustment based on PID feedback

[0258] 1. PID Model Construction: Based on target granularity Define the error as a benchmark. Control quantity The formula for (speed adjustment amount) is: ;in =0.8 (proportionality coefficient) =0.05 (integral coefficient) =0.1 (differential coefficient), actual rotational speed .

[0259] 2. Phased adjustments

[0260] 15th minute: , , , ,but , 30th minute: , , , ,but , ; 45th minute: , , , ,but , ; 55th minute: Stop grinding to obtain the final fineness. .

[0261] (III) Bayesian Anomaly Diagnosis

[0262] If detected in the 15th minute (Normal size should be ≥10μm), construct a "parameter-anomaly type" Bayesian network, input... , , , Calculated Then The speed was increased from 0.8 to 0.9 to accelerate the adjustment of the grinding speed and avoid a continuous decline in grinding efficiency.

[0263] IV. Coating Mixing and Preparation (S4)

[0264] (a) Preparation of raw materials and equipment

[0265] 1. Raw material verification: S1 Pure acrylic emulsion (pH 7.5-8.5), S3 Slurry (54.3 parts), S1 residual additives ( Add one part thickener and swell it in 3 parts deionized water for 2 hours in advance; (film-forming aids), S1 remaining 1. Deionized water. 2. Equipment debugging: Use a 500L low-speed mixing tank, install paddle-type mixing blades, and test the speed under no-load (0-600r / min) to ensure there is no jamming.

[0266] (II) Hybrid Modulation Process

[0267] 1. Emulsion Pretreatment: Add 35 parts of pure acrylic emulsion to the mixing tank, start stirring at 300 rpm for 5 minutes to ensure uniform dispersion of the emulsion (temperature 25℃, meeting process requirements). 2. Addition of Grinding Slurry: Maintain 300 rpm, pump S3 slurry into the tank at a rate of 5 L / min (approximately 10 minutes). Increase the speed to 500 rpm and stir for 20 minutes to form an "emulsion-slurry" mixture. 3. Addition of Additives: Maintain 500 rpm, first add the swollen thickener and stir for 8 minutes; then add 2.3 parts of film-forming aid and stir for 8 minutes to ensure complete dispersion of the additives. 4. Viscosity and Solid Content Adjustment: Add 6.8 parts of deionized water, reduce the speed to 400 rpm, stir for 15 minutes, and test key indicators: Viscosity: Measured using an NDJ-5S rotational viscometer (rotor #2, 60 rpm). (exist Within the specified range); Solid content: According to GB / T1725-2007, take 2g of sample, dry at 105℃ for 3 hours, and calculate: Theoretical value 48%, actual value 46%~50% is acceptable. 5. Homogenization treatment: reduce the speed to 200r / min, stir for 5 minutes to obtain a preliminary finished coating (total 100 parts) without particles and bubbles.

[0268] V. Performance Testing of Finished Coatings (S5)

[0269] (I) Core Performance Testing

[0270] 1. Thermal insulation performance test: Using a DRPL-III protective hot plate tester, the coating was applied to a 5mm thick cement mortar substrate, with a dry film thickness of 1.5mm. After curing at 23℃ and 50% humidity for 7 days, the results were tested. (qualified).

[0271] 2. Environmental performance testing: VOC content: Using a GC-2014 gas chromatograph, take 10g of sample and obtain... (Qualified); Free formaldehyde: Using a 722 spectrophotometer, take 5g of sample to obtain free formaldehyde. (qualified).

[0272] 3. Physical and mechanical properties testing: Adhesion: Using a 1mm spacing cross-cut tester, the adhesion grade is 0 (≤1 grade is acceptable); Scrub resistance: Using a JISK5600 scrub resistance tester, the number of scrub cycles is 6200-5000 (acceptable); Water resistance: The test plate is immersed in 23℃ deionized water for 96 hours, and there is no bubbling or peeling (acceptable).

[0273] (II) Bayesian inadequacy diagnosis (example)

[0274] If detected (Unqualified), construct a "cause-performance" Bayesian network, with the root node being... (Insufficient hollowness of microspheres) ( , (Emulsion solids content is low), leaf node is Unacceptable. Input Unqualified, calculated (Main cause), traced back to S3 record (Exceeding the limit), return to S3. Extend to 60 minutes, then re-grind. (qualified).

[0275] VI. Packaging of Qualified Finished Coatings (S6)

[0276] This step involves packaging the S5-tested coatings in a standardized manner to ensure stable quality during storage and transportation, while clearly labeling product information to complete the entire preparation process.

[0277] (a) Preparation before packaging

[0278] 1. Packaging drum treatment: Use 20L / 10L galvanized iron drums (with epoxy resin coating on the inner wall), rinse with deionized water and dry at 60℃ for 2 hours. Test the drum for no impurities and moisture content ≤0.1g. 2. Coating homogenization: Pump qualified coating into the pre-filling mixing tank and stir at 200r / min for 5 minutes to ensure uniformity and no sedimentation before filling.

[0279] (ii) Filling and sealing

[0280] 1. Automatic filling: Use a GF-200 filling machine, set to fill 20L drums. 10L bucket 1. For every 10 barrels filled, randomly select 1 barrel to weigh. If the error exceeds the range (e.g., 20.3L), adjust the filling machine flow rate to 9L / min and refill. 2. Sealing test: After filling, put on the cap (with rubber sealing ring), tighten it with a wrench (torque 30N·m), and invert for 30 minutes. If there is no leakage, it is qualified.

[0281] (III) Labeling and Warehousing

[0282] 1. Label content: Product name (Green building exterior wall coating with heat insulation function), S1 formula key parameters ( =35 servings =15 copies, etc.), S5 performance indicators ( , (etc.), storage information (shelf life 12 months, store at 5-35℃, humidity ≤70%), and implementation standards.

[0283] 2. Warehousing: Batch-wise storage in the finished product warehouse, with a stacking height of ≤3 layers. The warehouse is equipped with a thermometer and hygrometer and a ventilation system, and is kept away from fire sources and corrosive substances to complete the preparation process.

Claims

1. A method for preparing a green building exterior wall coating with heat insulation function, characterized in that, Includes the following steps: S1: The raw material formulation parameters are determined by using a fusion model of analytic hierarchy process (AHP) and multi-objective genetic algorithm. First, the importance of thermal insulation performance, environmental performance, and construction performance is scored by AHP and the weight vector of performance is calculated. Then, the weight vector is substituted into the fitness function by multi-objective genetic algorithm and the optimal formulation parameters are obtained after iterative optimization. S2: Pre-dispersion treatment of raw materials based on optimal formula parameters; Deionized water determined by the optimal formula parameters is added to a high-speed disperser, and dispersant, titanium dioxide, heavy calcium carbonate powder, nano hollow silica microspheres and defoamer determined by the optimal formula parameters are added in sequence. The speed of the disperser and the stirring time are controlled, and the initial fineness of the pre-dispersion slurry is detected. S3: The pre-dispersed slurry is ground using a fusion algorithm of a multiple linear regression initial grinding parameter prediction model and a proportional, integral, and differential feedback control model. First, the average particle size of the hollow silica microspheres, the amount of hollow silica microspheres, the fineness of the heavy calcium carbonate powder, and the initial fineness of the pre-dispersed slurry, determined by the optimal formulation parameters, are input into a multiple linear regression model to predict the initial grinding time and the initial grinding speed. Starting from the initial grinding time and initial grinding speed, the real-time fineness of the slurry during the grinding process is detected in real time through a proportional, integral, and derivative feedback control model. The error between the real-time fineness and the target fineness is calculated, and the grinding speed is dynamically adjusted according to the error until the final fineness of the ground slurry meets the requirements, thus obtaining the final fineness of the ground slurry. S4: Coating mixing and preparation based on optimal formulation parameters and the final fineness of the ground slurry; The pure acrylic emulsion determined by the optimal formulation parameters is added to a low-speed mixing tank, followed by the ground slurry, and then the thickener, film-forming aid and deionized water determined by the optimal formulation parameters are added. After stirring, a preliminary finished coating is obtained. S5: Perform performance testing on the preliminary finished coating; if all test items are qualified, it is determined to be a qualified finished coating; if there are unqualified items, return to step S1, re-determine the optimal formula parameters, and then execute steps S2 to S5 in sequence to form a closed loop.

2. The method for preparing the green building exterior wall coating with heat insulation function according to claim 1, characterized in that, It also includes S6: Packaging qualified finished paint, selecting pre-treated packaging barrels, filling according to the total amount of qualified finished paint, sealing and affixing labels to the surface of the packaging barrels, and finally storing the packaged paint in the warehouse.

3. The method for preparing a green building exterior wall coating with heat insulation function according to claim 1, characterized in that, The specific implementation process of the analytic hierarchy process in S1 includes: inviting experts with ≥10 years of experience in coating formulation research and development to score the relative importance of thermal insulation performance, environmental performance, and construction performance using a 1-9 scale; integrating expert opinions to construct a judgment matrix, solving for the maximum eigenvalue and corresponding eigenvector of the judgment matrix using the eigenvalue method, and normalizing the eigenvector to obtain the weight vector; then calculating the consistency index, random consistency index, and consistency ratio; when the consistency ratio <0.1, the weight vector is determined to be valid, and the weights of thermal insulation performance, environmental performance, and construction performance in the weight vector are 0.54, 0.30, and 0.16, respectively.

4. The method for preparing a green building exterior wall coating with heat insulation function according to claim 1, characterized in that, The specific implementation process of the multi-objective genetic algorithm in S1 includes: encoding the amounts of pure acrylic emulsion, nano-hollow silica microspheres, titanium dioxide, heavy calcium carbonate powder, total additives, and deionized water into 6-dimensional real vectors as chromosomes, and initializing 100 chromosomes to form an initial population; substituting the weight vector obtained in S1 into the fitness function; using roulette wheel selection to perform selection operations, performing arithmetic crossover with a crossover probability of 0.8, and performing random perturbation mutation operations with a mutation probability of 0.05; after 50 iterations, selecting the chromosome with the highest fitness in the 50th generation as the optimal formula parameters.

5. The method for preparing a green building exterior wall coating with heat insulation function according to claim 1, characterized in that, The construction of the multiple linear regression prediction model for initial grinding parameters in S3 includes: collecting 50 sets of historical production data, including the average particle size of nano-hollow silica microspheres, the amount of nano-hollow silica microspheres, the fineness of heavy calcium carbonate powder, the initial fineness of the pre-dispersed slurry, the initial grinding time, and the initial grinding speed. After preprocessing the historical production data, the prediction model for the initial grinding time and the initial grinding speed is obtained by fitting the data using the least squares method. The model determination coefficient is calculated, and the model is effective when the determination coefficient is ≥0.

90.

6. The method for preparing a green building exterior wall coating with heat insulation function according to claim 1, characterized in that, The specific implementation process of the proportional, integral, and derivative feedback control model in S3 includes: setting the target fineness to ≤30μm, taking slurry samples from the sand mill sampling port every 15 minutes to detect the real-time fineness; calculating the error between the real-time fineness and the target fineness; setting the proportional coefficient to 0.8, the integral coefficient to 0.05, and the derivative coefficient to 0.1, and obtaining the speed adjustment amount through the control quantity calculation relationship; the actual grinding speed is the sum of the initial grinding speed and the speed adjustment amount; repeating the above process until the real-time fineness is detected to be ≤30μm, stopping the grinding, and obtaining the final fineness.

7. The method for preparing a green building exterior wall coating with heat insulation function according to claim 1, characterized in that, The parameter transfer process from S1 to S3 is as follows: the average particle size and dosage of the nano-hollow silica microspheres determined by the optimal formulation parameters directly affect the time regression coefficient and rotation speed regression coefficient of the multiple linear regression model; the initial fineness of the pre-dispersed slurry obtained in S2 directly affects the time regression coefficient and rotation speed regression coefficient.

8. The method for preparing a green building exterior wall coating with heat insulation function according to claim 1, characterized in that, The specific feedback adjustment process in the closed loop of S5 includes: if the thermal conductivity in the thermal insulation performance test is >0.04W / (m·K), then return to step S1, increase the upper limit of the dosage range of nano-hollow silica microspheres in the multi-objective genetic algorithm, and iterate again to find the optimal solution; if the volatile organic compound content in the environmental performance test is >50g / L, then return to step S1, change the film-forming aid to environmentally friendly dodecyl ester, and adjust the total dosage range of the aid from 3-5 parts to 2.5-4.5 parts, and redetermine the optimal formulation parameters; if the adhesion grade in the physical and mechanical performance test is >1, then return to step S1, adjust the dosage range of pure acrylic emulsion from 30-40 parts to 32-42 parts, and repeat steps S1 to S5; until all test items are qualified, the closed loop ends.

9. The method for preparing a green building exterior wall coating with heat insulation function according to claim 2, characterized in that, The pretreatment process for packaging drums in S6 includes: selecting 20L or 10L galvanized iron drums, coating the inner wall of the drum with an epoxy resin anti-rust layer to prevent corrosion, rinsing the inside of the drum with deionized water, drying it in a 60℃ oven for 2 hours, and testing the moisture content inside the drum by weighing to ensure that the moisture content is ≤0.1g.

10. A green building exterior wall coating with heat insulation function, characterized in that, The coating is prepared using the method described in any one of claims 1 to 9 for preparing a green building exterior wall coating with heat insulation function.

Images