A silica-fume-steel fiber modified geopolymer repair material, a preparation method and a performance optimization method thereof

Abstract

The application relates to the technical field of building materials, in particular to a silica fume-steel fiber modified geopolymer repair material and a preparation method and performance optimization method thereof. The repair material comprises cementitious materials, an alkali activator and other components; wherein the alkali activator is added with an additive, and the additive comprises nanometer silicon dioxide, nanometer calcium carbonate, carbon nanotubes and graphene oxide. The application realizes the dynamic balance of "retardation and rapid setting" through the cementitious materials, so that the setting time of the slurry can be flexibly regulated according to the construction requirements; the cementitious materials form a dense gel network of C-A-S-H and N-A-S-H interlaced with each other under the action of the alkali activator; meanwhile, the nanometer silicon dioxide additionally improves the mechanical properties; the synergistic effect of the steel fiber and the nanometer additive improves the splitting tensile strength; and the synergistic hydration of the cementitious material system and the double densification effect of the nanometer material significantly improve the corrosion resistance.

Description

Technical Field

[0001] This application relates to the field of building materials technology, specifically to a silica fume-steel fiber modified geopolymer repair material and its preparation method and performance optimization method. Background Technology

[0002] Concrete structures inevitably experience cracking, spalling, and damage during service, requiring timely repair to restore their functionality and durability. Traditional cement-based repair materials suffer from drawbacks such as slow setting and hardening, low early strength, poor bonding between new and old surfaces, shrinkage cracking, high risk of chloride ion penetration, and high carbon emissions. Geopolymers, as inorganic cementitious materials formed from aluminosilicate raw materials (such as fly ash and slag) under strong alkali activation, possess advantages such as rapid hardening and early strength, high temperature resistance, chemical resistance, low permeability, and the ability to utilize industrial solid waste, making them a promising green repair material. However, existing geopolymer repair materials still face several challenges. One is the difficulty in controlling workability and setting time. The rapid alkali activation reaction often leads to a rapid decrease in slurry fluidity (poor workability) and an excessively short setting time, making it difficult to meet the on-site workability requirements. Typically, water-reducing agents are relied upon to improve fluidity, but these agents are costly, their effectiveness may be unstable, and they may introduce other adverse effects (such as air entrainment and excessive retarding). Secondly, insufficient toughness leads to cracking: as a brittle material, geopolymers are prone to cracking when constrained by shrinkage or under load, affecting the integrity and durability of the repaired structure. Thirdly, interfacial bonding and chloride ion penetration resistance need improvement: the interface between new and old concrete is a weak point in the repaired structure, requiring improved bond strength and inhibition of the diffusion of harmful media such as chloride ions. Fourthly, performance stability and mix design rely heavily on experience: raw material fluctuations are significant, and mix design mainly relies on extensive trial mixing and engineer experience, lacking a deep understanding of the microscopic mechanisms and high-precision prediction tools, making it difficult to guarantee material performance stability. Therefore, there is an urgent need to develop a geopolymer repair material with adjustable workability, controllable setting time, excellent toughness, strong bonding, high impermeability, stable performance, and a green and low-carbon profile, along with its efficient optimization methods. Summary of the Invention

[0003] To address the challenges of controlling the setting time of existing geopolymer repair materials, which requires water-reducing agents but can easily lead to other adverse effects, this application provides a silica fume-steel fiber modified geopolymer repair material, its preparation method, and its performance optimization method.

[0004] In a first aspect, this application provides a silica fume-steel fiber modified geopolymer repair material, employing the following technical solution:

[0005] A silica fume-steel fiber modified geopolymer repair material comprises the following components in parts by weight:

[0006] Cementitious materials: 30 parts slag, 50-70 parts fly ash, and 0-20 parts silica fume, wherein the total weight of the slag, fly ash, and silica fume is 100 parts.

[0007] Alkali activator: 20 parts sodium silicate solution, 6.5 parts sodium hydroxide solution, 0.05-4.55 parts additive, and 10-16 parts water; the sodium silicate solution has a modulus of 2, and the sodium hydroxide solution has a sodium hydroxide mass fraction of 40%; the additive includes at least one of nano-silica, nano-calcium carbonate, carbon nanotubes, and graphene oxide.

[0008] Other: 45 parts standard sand, 0-12 parts steel fiber.

[0009] By adopting the above technical solution, this invention optimizes the composition ratio of 30 parts slag and 50-70 parts fly ash, effectively extending the setting time of the system by utilizing the retarding effect of fly ash, thus avoiding the problem of excessively rapid early hardening leading to a short construction window. Simultaneously, this invention abandons traditional chemical water-reducing agents. The fluidity and setting time are regulated through the following synergistic mechanisms: First, the adjustable dosage of silica fume within the range of 0-20 parts can fill the interparticle gaps and release more free water, thereby improving the fluidity of the fly ash geopolymer slurry while accelerating the reaction and setting through its high activity; second, the optimized design of the alkali activator modulus achieves precise control of the dissolution rate and condensation speed, combined with the rapid hydration characteristics provided by the 30 parts slag content, achieving a dynamic balance between "retarded setting and rapid setting" within an adjustable range, allowing the setting time of the slurry to be flexibly adjusted according to construction needs.

[0010] The multi-component composite effect of the cementitious material system in this invention is the core of achieving enhanced compressive strength. Slag provides an abundant CaO source, generating high-strength CASH gel under alkali activation; fly ash provides a large amount of active silica and alumina, forming a three-dimensional NASH network framework; silica fume further contributes additional active silica sources. Under the action of an alkali activator, the three cementitious materials form a dense gel network where CASH and NASH are interwoven. Simultaneously, the micro-filling effect of additives such as nano-silica further densifies the gel matrix, improving the microstructure characteristics and enhancing mechanical properties by increasing the formation of additional geopolymer products.

[0011] In this invention, the synergistic effect of the addition of steel fibers and nano-additives is key to improving the splitting tensile strength. Nano-silica not only increases the micro-density of the geopolymer gel but also enhances the interfacial bonding between the steel fibers and the matrix, enabling the steel fibers to more effectively reinforce the bridging cracks. Through the random distribution of steel fibers in the three-dimensional network and the interfacial strengthening of the nano-additives, the repair material of this invention can absorb a large amount of energy under tension through the debonding, pull-out, and fracture processes of the fibers, significantly improving the ductility and crack resistance of the material.

[0012] This invention significantly enhances corrosion resistance through the synergistic hydration of the cementitious material system and the dual densification effect of nanomaterials. The composite structure of CASH and NASH gels inherently possesses a lower permeability coefficient and higher chemical stability than traditional CSH gels; the addition of silica fume further fills the capillary pores of the gel network; the nano-filling effect of the nano-additives reduces the total porosity of the matrix, greatly inhibiting the penetration of corrosive media (water, chloride ions, sulfates, etc.) into the material interior. Simultaneously, the uniform distribution of steel fibers in the matrix acts as a "physical barrier" to block crack penetration channels. Therefore, the repair material prepared by this invention exhibits excellent durability under typical corrosive environments such as chloride ion penetration, sulfate attack, and carbonization.

[0013] This application achieves a dynamic balance between "slow setting and fast setting" through the use of slag, fly ash, and silica fume in the cementitious materials, allowing the setting time of the slurry to be flexibly adjusted according to construction needs. Under the action of an alkali activator, the three cementitious materials form a dense gel network of interwoven CASH and NASH. At the same time, the micro-filling effect of additives such as nano-silica further densifies the gel matrix, improves the microstructure characteristics by increasing the formation of additional geological polymerization products, and enhances mechanical properties. The addition of steel fibers and the synergistic effect of nano-additives improve the splitting tensile strength. The corrosion resistance is significantly improved through the synergistic hydration of the cementitious material system and the dual densification effect of nanomaterials.

[0014] Preferably, the additive is nano-silica.

[0015] By employing the above-mentioned technical solutions, compared with other nanomaterials, nano-silica exhibits the most comprehensive and stable improvement in compressive strength. While carbon nanotubes can significantly improve flexural strength, their difficulty in dispersion in alkali-activated systems limits the maximization of the reinforcing effect; graphene oxide inhibits early strength development; and nano-calcium carbonate primarily contributes to improved durability. In contrast, nano-silica demonstrates excellent improvement in compressive strength, splitting tensile strength, flexural strength, and durability, making it the most comprehensive additive for enhancing overall performance.

[0016] The reinforcing effect of nano-silica on geopolymer materials stems from the following synergistic mechanisms: First, nano-silica particles can fill the nanoscale pores in the gel matrix, making the matrix more compact. Well-dispersed nanoparticles exhibit smaller particle sizes, increasing the physical filling effect, thereby producing a dense microstructure, better bonding behavior, and simultaneous improvement in micro and macro mechanical properties. Second, nano-silica dissolves under alkaline-activated conditions, releasing active silica to participate in the geopolymerization reaction, generating additional NASH gel and increasing the gel content of the system. Third, nano-silica not only improves the micro-density of the geopolymer gel but also enhances the bonding force between reinforcing materials such as steel fibers and the matrix, thereby further increasing the improvement in splitting tensile and flexural strength. Fourth, nano-silica particles can accelerate the early geopolymerization reaction, especially better-dispersed nano-silica, which can promote the early dissolution and hydration process of precursors, improving micromechanical properties.

[0017] Preferably, the additive is nano-silica and nano-calcium carbonate.

[0018] By adopting the above technical solution, when nano-silica and nano-calcium carbonate are combined, this invention preferably uses this combination as a binary additive system. Its synergistic improvement effect on mechanical properties and durability is superior to other binary combinations. The specific mechanism is as follows:

[0019] Nano-silica typically has a particle size of 10-50 nm, primarily filling nanoscale pores; while nano-calcium carbonate typically has a particle size of 50-200 nm, primarily filling submicron-scale pores. When used in combination, they form a multi-scale filling network from the nanoscale to the submicron scale, resulting in a matrix density significantly higher than that of using a single nanomaterial. Nano-silica mainly contributes to improved compressive strength, while nano-calcium carbonate excels in durability (especially resistance to acid rain corrosion and carbonation). When combined, nano-silica enhances the overall mechanical properties of the matrix through filling and interfacial reinforcement, while nano-calcium carbonate inhibits the penetration of corrosive media by accelerating polymerization and forming a denser microstructure, thus achieving a comprehensive improvement in both mechanical and durability properties.

[0020] Nano-silica provides an additional source of active silicon, while nano-calcium carbonate decomposes under alkaline activation conditions to generate calcium ions. These calcium ions interact with the aluminosilicate gel in the geopolymer system, promoting the formation of CASH gel and further enhancing the density of the gel network. The combination of the two simultaneously improves the thoroughness of the geopolymerization reaction and the stability of the gel product, thereby enabling the material to maintain higher and more stable mechanical properties throughout its service life.

[0021] In the slag-fly ash-silica fume system of this invention, the CaO provided by the slag and the calcium ions released by nano-calcium carbonate synergistically promote the formation of CASH gel; the active silica-alumina source provided by the fly ash generates a NASH network under the catalysis of nano-silica. The CASH / NASH composite gel structure formed by both exhibits high strength and excellent chemical stability.

[0022] Preferably, the additive is nano-silica, nano-calcium carbonate, and graphene oxide.

[0023] By adopting the above technical solution, when nano-silica, nano-calcium carbonate, and graphene oxide are used in combination, this invention preferably uses this ternary additive system. Its comprehensive improvement in mechanical properties, durability, and interfacial properties is significantly better than any binary combination. The specific mechanism is as follows:

[0024] The combination of these three elements forms a full-scale reinforcement system from nanoparticles to nanosheets: nano-silica fills nanoscale pores, nano-calcium carbonate fills submicron-scale pores and provides a nucleation effect, while graphene oxide, as a two-dimensional sheet, provides bridging and crack-resistant functions on a larger scale. Together, they construct a full-scale reinforcement network covering three-dimensional space.

[0025] Graphene oxide forms strong interfacial bonds with geopolymer gels through its abundant oxygen-containing functional groups on its surface. In the ternary system, graphene oxide also functions as a dispersant, promoting the uniform dispersion of nano-silica and nano-calcium carbonate in the matrix through its surface adsorption.

[0026] The three additives complement each other in their reinforcing mechanism: nano-silica mainly fills the voids and provides an active silicon source, nano-calcium carbonate mainly densifies the pores and induces crystal nucleation, and graphene oxide mainly strengthens the interface and blocks cracks. The composite gel system formed by the three has higher structural order and density. When nano-silica, nano-calcium carbonate, and graphene oxide are incorporated, the geopolymer repair material not only shows a significant increase in compressive strength, but also a greater increase in splitting tensile strength due to the bridging effect of graphene oxide and the reinforcement of the fiber-matrix interface. Furthermore, its corrosion resistance is excellent under corrosive environments such as chloride ion penetration and sulfate attack due to the densification effect contributed by the three additives.

[0027] The formulation of this invention contains 50-70 parts fly ash, and the low activity and high porosity of fly ash are its performance shortcomings. The active silicon source effect of nano-silica can effectively stimulate the activity of fly ash; the nucleation effect of nano-calcium carbonate provides nucleation sites for the dissolution and repolymerization of fly ash; the layered structure of graphene oxide can strengthen the connection between fly ash particles through its two-dimensional geometric properties. The combined effect of these three factors significantly improves the material performance under high fly ash content.

[0028] Secondly, this application provides a method for preparing a silica fume-steel fiber modified geopolymer repair material, employing the following technical solution:

[0029] A method for preparing a silica fume-steel fiber modified geopolymer repair material, comprising the following steps:

[0030] Step 1: Mix the prescribed amounts of sodium silicate solution, sodium hydroxide solution, additives, and water, and adjust the temperature to 20-30℃ to obtain the alkali activator;

[0031] Step 2: Mix the slag, fly ash, and silica fume according to the formula, then add the standard sand according to the formula and mix, and then add the alkali activator and mix to obtain a uniform slurry;

[0032] Step 3: Add the prescribed amount of steel fiber to the homogeneous slurry and stir to obtain silica fume-steel fiber modified geopolymer repair material.

[0033] Preferably, in step one, the additive comprises hydrophobic nano-silica;

[0034] Step 1: Mix the prescribed amount of hydrophobic nano-silica and water, and adjust the temperature to 20-30℃ to obtain alkali activator A; mix the prescribed amounts of sodium silicate solution and sodium hydroxide solution, and adjust the temperature to 20-30℃ to obtain alkali activator B;

[0035] Step 2: Mix the slag, fly ash, and silica fume according to the formula, then add the standard sand according to the formula and mix. Next, add the alkali activator A and mix, then add the alkali activator B and mix to obtain a uniform slurry.

[0036] By employing the above technical solution, hydrophobic nano-silica reacts with water under high-speed stirring to form water droplets that are encapsulated by hydrophobic nano-silica particles, creating a "liquid powder" structure, thus obtaining "dry water." When added to a geopolymer system, the reduction in viscosity and yield shear stress is manifested in a higher flow diameter, mainly due to the combined effect of particle size coarsening in the precursor and the flowability of the hydrophobic nano-silica. The addition of conventional nanomaterials typically increases the viscosity of the slurry, but the "dry water" technology introduces hydrophobic nano-silica into the system in a pre-encapsulated form, avoiding direct adsorption of water by particles and an increase in system viscosity. Instead, it reduces the yield stress and plastic viscosity of the slurry.

[0037] This invention involves mixing hydrophobic nano-silica alone with water to form alkali activator A, and mixing sodium silicate solution and sodium hydroxide solution to form alkali activator B. In step two, alkali activator A is added first and then stirred, followed by alkali activator B and stirred again. The core principle of this design is that the hydrophobic modified layer on the surface of the hydrophobic nano-silica may be chemically eroded or damaged when in prolonged contact with high-concentration strong alkali solutions (especially 40% sodium hydroxide solution), leading to a decrease in its hydrophobic properties and dispersion performance. By adding the alkali activator step by step, with solution A added first, the hydrophobic nano-silica preferentially contacts the cementing material to form a coating layer before the high-concentration alkali activator B is introduced, effectively protecting the functional integrity of the hydrophobic nano-silica and maximizing the "dry water" enhancement effect.

[0038] When alkali activator A (a mixture of hydrophobic nano-silica and water) is first added to the cementitious material, water droplets encapsulated in the hydrophobic nano-silica are released on the surface of the cementitious material, forming a localized wetting and reaction interface, creating conditions for the initiation of the geopolymerization reaction. Subsequently, when alkali activator B (sodium silicate + sodium hydroxide) is added, the high-concentration alkali solution rapidly diffuses across the already formed wetting interface, triggering a more uniform and complete geopolymerization reaction. This avoids the problem of uneven reaction caused by excessively high localized alkali concentrations that may occur when the alkali activator is added all at once.

[0039] Thirdly, this application provides a method for optimizing the performance of silica fume-steel fiber modified geopolymer repair materials, employing the following technical solution:

[0040] A method for performance optimization of a silica fume-steel fiber modified geopolymer repair material includes the following steps:

[0041] The raw material characteristic parameters of the cementitious material are determined, including the active silica-alumina molar ratio, glass phase content, and average particle size of the powder. The distribution range of the raw material characteristic parameters is statistically analyzed to obtain the fluctuation characteristics.

[0042] A pre-built proportioning-performance prediction model is invoked, using the raw material characteristic parameters, silica fume content, steel fiber content, and additive content as inputs to obtain performance indicators. These performance indicators include workability, mechanical properties, and durability. Workability includes flowability, mechanical properties include compressive strength, splitting tensile strength, and bond strength, and durability includes the chloride ion diffusion coefficient of the material bulk. Combining the fluctuation characteristics, the distribution range of these performance indicators under raw material fluctuations is evaluated. With the performance indicators meeting preset requirements as a constraint, the range of proportions with the lowest failure probability is determined.

[0043] Composite specimens are prepared based on the specified component range. The composite specimen is composed of a preset matrix and a repair layer formed by the repair material covering the surface of the matrix. Preset accelerated environmental tests are carried out, and preset detection technology is used to track interface damage. The damage degradation rate of the interface is extracted based on a preset interface damage evolution model. Climate data is obtained, and the service life curve of the interface is generated based on the damage degradation rate and the climate data. The component range that meets the preset service life requirements is retained as the feasible region for proportioning.

[0044] Based on the service life curve of the interface, with the goal of maximizing the service life of the interface and with the workability and mechanical properties meeting preset requirements as hard constraints, the optimal part ratio range is deduced within the feasible range of the ratio. The optimal part ratio range is then output to the ratio design stage to adjust the parts of silica fume, steel fiber and additives, thereby achieving pre-prediction and source control of durability.

[0045] By adopting the above technical solution, this method reduces the failure probability by considering the fluctuation characteristics of raw materials for robust optimization; it generates a service life curve by combining accelerated environmental testing and interface damage tracking, and uses the lowest damage rate as the target to deduce the optimal ratio. This realizes the transformation from experience-based trial mixing to data-driven pre-prediction and source control of durability, significantly improving material reliability and service life.

[0046] The steps for constructing the ratio-performance prediction model are as follows:

[0047] Historical data of multiple batches of industrial raw materials were collected, and the characteristic parameters, silica fume content, steel fiber content, and additive content of each batch of raw materials were measured. The actual values ​​of performance indicators were also measured to form a training sample set.

[0048] The correlation between the raw material characteristic parameters and the actual values ​​of the performance indicators in the training sample set is analyzed. Based on the correlation, the network topology is determined, and a mechanism-guided initial network is constructed. The raw material characteristic parameters, the amount of silica fume, the amount of steel fiber, and the amount of each additive are used as inputs, and the performance indicators are used as outputs. A preset physical constraint term is embedded in the loss function.

[0049] The training sample set is divided into meta-training tasks and meta-testing tasks according to raw material batches. Based on the meta-training tasks and the initial network, a model-independent meta-learning algorithm is used to train the meta-model and obtain the meta-model. The meta-testing task is used to verify the generalization ability of the meta-model.

[0050] Based on the meta-test task, the meta-model is iterated and verified a preset number of times to ensure that the prediction determination coefficient of the compressive strength at a preset age in the mechanical properties meets a preset first threshold and the prediction determination coefficient of the bond strength meets a preset second threshold, and the verification meta-model is output.

[0051] The validation meta-model is used as the ratio-performance prediction model.

[0052] By adopting the above technical solution, this method uses a mechanism-guided neural network to embed physical constraints and combines meta-learning (MAML) to divide tasks for training according to raw material batches, enabling the model to quickly adapt to raw material fluctuations and achieve high-precision prediction of mechanical properties (compressive strength, bonding strength) with small samples (with the coefficient of determination meeting the standard), significantly improving generalization and prediction reliability.

[0053] Based on the aforementioned fluctuation characteristics, the distribution range of the performance indicators under raw material fluctuations is evaluated. With the performance indicators meeting preset requirements as a constraint, the interval of the lowest possible failure probability is determined, including:

[0054] Based on the fluctuation characteristics, the distribution of the raw material characteristic parameters is fitted to obtain the probability density function of each raw material characteristic parameter, thus forming the joint probability distribution of the raw material characteristic parameters;

[0055] Based on the joint probability distribution, a preset random sampling is performed to generate multiple sets of raw material characteristic parameter samples; multiple combinations of component numbers are selected, which consist of the number of parts of silica fume, steel fiber, and additives; the ratio-performance prediction model is called, and multiple sets of performance index prediction values ​​are obtained by taking the multiple sets of raw material characteristic parameter samples and the multiple combinations of component numbers as input.

[0056] The proportion of predicted values ​​of the multiple sets of performance indicators that do not meet the preset requirements is determined as the failure probability.

[0057] Using the lowest failure probability as the screening objective, the combination of parts with the lowest failure probability is selected from the multiple combinations of parts, and the number of parts is determined based on the combination of parts.

[0058] By employing the above technical solution, this method quantifies the probability of performance failure under raw material fluctuations through fitting the raw material parameter distribution, random sampling, and model prediction, and selects the range of proportions with the lowest failure probability as the objective. This achieves a shift from deterministic design to probabilistic robust optimization, significantly improving the reliability and pass rate of the formulation under raw material fluctuations.

[0059] Composite specimens are prepared based on the specified component range. Each composite specimen consists of a predetermined substrate and a repair layer formed by the repair material covering the surface of the substrate. Predetermined accelerated environmental tests are conducted, and predetermined detection techniques are used to track interface damage. The damage degradation rate of the interface is extracted based on a predetermined interface damage evolution model. Climate data is obtained, and a service life curve of the interface is generated based on the damage degradation rate and the climate data. The component range that meets the predetermined service life requirements is retained as the feasible mixing region, including:

[0060] Based on the specified fraction range, a composite specimen was prepared, which was composed of the preset substrate and the repair material covering the surface of the preset substrate. A salt solution wet-dry cycle accelerated test was carried out.

[0061] At different cycles of the accelerated test, in-situ X-ray computed tomography was performed on the same interface location of the same composite specimen to obtain three-dimensional structural evolution sequence images of the interface region.

[0062] The digital volume correlation technique is used to perform correlation calculation on the sequence images, calculate the three-dimensional displacement field and strain tensor of the interface, identify the microcrack initiation location by the preset strain criterion, and invert and calibrate the critical stress and corresponding number of cycles based on the preset interface mechanical model.

[0063] Using the critical stress and the corresponding number of cycles, combined with the interface opening displacement when the specimen completely fails, the fracture parameters of the preset interface mechanical model are calibrated, a preset interface damage evolution model is established, and the damage degradation rate is extracted from the preset interface damage evolution model.

[0064] Acquire climate data of the target service environment, convert the damage degradation rate into an equivalent service time scale based on a preset environmental equivalence conversion relationship, generate an interface service life curve from the preset environmental equivalence conversion relationship, and mark the fraction interval and the corresponding interface service life curve as the ratio feasible region.

[0065] By employing the above technical solution, this method uses in-situ CT and digital volume correlation technology to non-destructively track the initiation and evolution of microcracks at the interface, calibrate the parameters of the cohesive force model, and establish a damage degradation model. Combined with accelerated testing and climate data, it generates service life curves, providing quantitative durability data for the feasible domain of the mix design, and achieving accurate prediction of the repair interface life.

[0066] In summary, this application has the following beneficial effects:

[0067] 1. This application achieves a dynamic balance between "retarded setting and rapid setting" through slag, fly ash, and silica fume in the cementitious materials, allowing the setting time of the slurry to be flexibly adjusted according to construction needs; the three cementitious materials form a dense gel network of interwoven CASH and NASH under the action of an alkali activator; at the same time, the micro-filling effect of additives such as nano-silica further densifies the gel matrix, improves the microstructure characteristics by increasing the formation of additional geological polymerization products, and enhances mechanical properties; the addition of steel fibers and the synergistic effect of nano-additives improve splitting tensile strength; and the corrosion resistance is significantly improved through the synergistic hydration of the cementitious material system and the dual densification effect of nanomaterials.

[0068] 2. This application achieves data-driven pre-prediction and source control of durability by using raw material fluctuation statistics, ratio-performance prediction model, minimum failure probability optimization, in-situ CT / DVC tracking of interface damage and generation of service life curve, thereby improving the reliability and life of repair materials. Attached Figure Description

[0069] Figure 1 This is a schematic flowchart illustrating a method for optimizing the performance of a silica fume-steel fiber modified geopolymer repair material according to an embodiment of this application. Detailed Implementation

[0070] The raw materials in this application include the following:

[0071] Slag: Commercially available S95 grade slag powder;

[0072] Fly ash: Grade II or above fly ash;

[0073] Steel fiber: Straight steel fiber;

[0074] The present application will be further described in detail below with reference to embodiments and comparative examples.

[0075] Example 1

[0076] A method for preparing a silica fume-steel fiber modified geopolymer repair material includes the following steps:

[0077] Step 1: Mix 200g sodium silicate solution, 65g sodium hydroxide solution, 15g nano silica, and 130g water at 300rpm for 5 minutes, and adjust the temperature to 25℃ (20-30℃ is acceptable) to obtain the alkali activator; the sodium silicate solution has a modulus of 2, and the sodium hydroxide solution has a sodium hydroxide mass fraction of 40%.

[0078] Step 2: Mix 300g slag, 600g fly ash, and 100g silica fume at 60rpm for 2 minutes, then add 450g standard sand and mix at 60rpm for 1 minute. Finally, add alkali activator and mix to obtain a uniform slurry.

[0079] Step 3: Add 80g of steel fiber to the uniform slurry and stir at 60rpm for 2min to obtain silica fume-steel fiber modified geopolymer repair material.

[0080] Application Example 1

[0081] The steps for repairing cracked old concrete using the silica fume-steel fiber modified geopolymer repair material prepared above are as follows:

[0082] (1) Treatment of old concrete interface: Remove floating dust, oil stains and loose particles from both sides and inside the crack, and wash it clean with a high-pressure water gun. After the surface is dry, use an angle grinder to cut a V-shaped groove along the crack direction (groove depth 20mm, groove width 30mm), and roughen the groove wall with a chisel. Finally, blow away the residue in the groove with compressed air, and pre-wet it with clean water until it is saturated and dry.

[0083] (2) Preparation of repair materials: The silica fume-steel fiber modified geopolymer repair materials were prepared on-site according to the proportions and methods of Example 1.

[0084] (3) Material filling: Fill the crack groove with silica fume-steel fiber modified geopolymer repair material, compact it with a trowel and smooth it repeatedly to ensure that the material is in close contact with the groove wall and is slightly higher than the original concrete surface.

[0085] (4) Curing: After filling, cover the repaired area with plastic film and cure for 7 days at 20±2℃ and relative humidity ≥95%. After curing, grind the raised part to make it flat and obtain the repaired concrete.

[0086] The old concrete has a compressive strength of 68.1 MPa, a splitting tensile strength of 4.80 MPa, and a chloride ion diffusion coefficient of 4.68 × 10⁻⁶. -12 m 2 / s.

[0087] Examples 2-5

[0088] Examples 2-5 are based on the preparation method of Example 1, with adjustments made to the content of each component of the silica fume-steel fiber modified geopolymer repair material. The specific adjustments are shown in Table 1.

[0089] Performance testing

[0090] The silica fume-steel fiber modified geopolymer repair materials and the repaired concrete of Examples 1-5 were subjected to the following performance tests, and the test results are shown in Table 1.

[0091] 1. Final setting time

[0092] The final setting time of silica fume-steel fiber modified geopolymer repair material was determined according to T / CECS1556-2024.

[0093] 2. Plastic viscosity

[0094] The plastic viscosity of silica fume-steel fiber modified geopolymer repair materials was determined according to GB / T10247-2008.

[0095] 3. Compressive strength and splitting tensile strength

[0096] The compressive strength and splitting tensile strength of the repaired concrete were determined according to GB / T50081-2019.

[0097] 4. Chloride ion diffusion coefficient

[0098] The chloride ion diffusion coefficient of the repaired concrete was determined according to GB / T50082-2024.

[0099] Table 1. Content (g) of each component and performance test results of silica fume-steel fiber modified geopolymer repair materials in Examples 1-5

[0100]

[0101] Referring to Table 1, a comparison of Examples 1-5 shows that the silica fume-steel fiber modified geopolymer repair material obtained using the above formula has good performance. This is because the present invention optimizes the composition ratio of 30 parts slag and 50-70 parts fly ash, utilizing the retarding effect of fly ash to effectively extend the setting time of the system, avoiding the problem of excessively rapid early hardening leading to a short construction window. Simultaneously, the present invention abandons traditional chemical water-reducing agents. The fluidity and setting time are regulated through the following synergistic mechanisms: First, the adjustable dosage of silica fume within the range of 0-20 parts can fill the interparticle gaps and release more free water, thereby improving the fluidity of the fly ash geopolymer slurry while accelerating the reaction and setting through its high activity; second, the optimized design of the alkali activator modulus achieves precise control of the dissolution rate and condensation speed, combined with the rapid hydration characteristics provided by the 30 parts slag content, achieving a dynamic balance between "retarded setting and rapid setting" within an adjustable range, allowing the setting time of the slurry to be flexibly adjusted according to construction needs.

[0102] The multi-component composite effect of the cementitious material system in this invention is the core of achieving enhanced compressive strength. Slag provides an abundant CaO source, generating high-strength CASH gel under alkali activation; fly ash provides a large amount of active silica and alumina, forming a three-dimensional NASH network framework; silica fume further contributes additional active silica sources. Under the action of an alkali activator, the three cementitious materials form a dense gel network where CASH and NASH are interwoven. Simultaneously, the micro-filling effect of additives such as nano-silica further densifies the gel matrix, improving the microstructure characteristics and enhancing mechanical properties by increasing the formation of additional geopolymer products.

[0103] In this invention, the synergistic effect of the addition of steel fibers and nano-additives is key to improving the splitting tensile strength. Nano-silica not only increases the micro-density of the geopolymer gel but also enhances the interfacial bonding between the steel fibers and the matrix, enabling the steel fibers to more effectively reinforce the bridging cracks. Through the random distribution of steel fibers in the three-dimensional network and the interfacial strengthening of the nano-additives, the repair material of this invention can absorb a large amount of energy under tension through the debonding, pull-out, and fracture processes of the fibers, significantly improving the ductility and crack resistance of the material.

[0104] This invention significantly enhances corrosion resistance through the synergistic hydration of the cementitious material system and the dual densification effect of nanomaterials. The composite structure of CASH and NASH gels inherently possesses a lower permeability coefficient and higher chemical stability than traditional CSH gels; the addition of silica fume further fills the capillary pores of the gel network; the nano-filling effect of the nano-additives reduces the total porosity of the matrix, greatly inhibiting the penetration of corrosive media (water, chloride ions, sulfates, etc.) into the material interior. Simultaneously, the uniform distribution of steel fibers in the matrix acts as a "physical barrier" to block crack penetration channels. Therefore, the repair material prepared by this invention exhibits excellent durability under typical corrosive environments such as chloride ion penetration, sulfate attack, and carbonization.

[0105] Examples 6-8

[0106] Example 6 is based on the preparation method of Example 1, but the method is adjusted as follows:

[0107] Step 1: Mix 200g sodium silicate solution, 65g sodium hydroxide solution, 30g nano calcium carbonate and 130g water at 300rpm for 5min, and adjust the temperature to 25℃ (20-30℃ is acceptable) to obtain the alkali activator; the sodium silicate solution has a modulus of 2, and the sodium hydroxide solution has a sodium hydroxide mass fraction of 40%.

[0108] Example 7 is based on the preparation method of Example 1, but the method is adjusted as follows:

[0109] Step 1: Mix 200g sodium silicate solution, 65g sodium hydroxide solution, 3g carbon nanotubes and 130g water at 300rpm for 5min, and adjust the temperature to 25℃ (20-30℃ is acceptable) to obtain the alkali activator; the sodium silicate solution has a modulus of 2, and the sodium hydroxide solution has a sodium hydroxide mass fraction of 40%.

[0110] Example 8 is based on the preparation method of Example 1, but the method is adjusted as follows:

[0111] Step 1: Mix 200g sodium silicate solution, 65g sodium hydroxide solution, 0.5g graphene oxide and 130g water at 300rpm for 5min, and adjust the temperature to 25℃ (20-30℃ is acceptable) to obtain the alkali activator; the sodium silicate solution has a modulus of 2, and the sodium hydroxide solution has a sodium hydroxide mass fraction of 40%.

[0112] The repaired concrete from Examples 6-8 was subjected to the performance tests described above, and the test results are shown in Table 2.

[0113] Table 2 Performance test results for Examples 1 and 6-8

[0114]

[0115] Referring to Table 2, a comparison of Examples 1 and 6-8 shows that the silica fume-steel fiber modified geopolymer repair material obtained by selecting nano-silica as the additive has a better effect. This is because, compared with other nanomaterials, nano-silica has the most comprehensive and stable effect in improving compressive strength. Although carbon nanotubes can significantly improve flexural strength, their dispersion difficulties in alkali-activated systems limit the maximization of the reinforcing effect; graphene oxide has an inhibitory effect on early strength development; and nano-calcium carbonate mainly contributes to improved durability. In contrast, nano-silica exhibits excellent improvement effects in compressive strength, splitting tensile strength, flexural strength, and durability, making it the additive with the most comprehensive performance improvement.

[0116] The reinforcing effect of nano-silica on geopolymer materials stems from the following synergistic mechanisms: First, nano-silica particles can fill the nanoscale pores in the gel matrix, making the matrix more compact. Well-dispersed nanoparticles exhibit smaller particle sizes, increasing the physical filling effect, thereby producing a dense microstructure, better bonding behavior, and simultaneous improvement in micro and macro mechanical properties. Second, nano-silica dissolves under alkaline-activated conditions, releasing active silica to participate in the geopolymerization reaction, generating additional NASH gel and increasing the gel content of the system. Third, nano-silica not only improves the micro-density of the geopolymer gel but also enhances the bonding force between reinforcing materials such as steel fibers and the matrix, thereby further increasing the improvement in splitting tensile and flexural strength. Fourth, nano-silica particles can accelerate the early geopolymerization reaction, especially better-dispersed nano-silica, which can promote the early dissolution and hydration process of precursors, improving micromechanical properties.

[0117] Examples 9-12

[0118] Example 9 is based on the preparation method of Example 1, but the method is adjusted as follows:

[0119] Step 1: Mix 200g sodium silicate solution, 65g sodium hydroxide solution, 15g nano silica, 30g nano calcium carbonate and 130g water at 300rpm for 5min, and adjust the temperature to 25℃ (20-30℃ is acceptable) to obtain the alkali activator; the sodium silicate solution has a modulus of 2, and the sodium hydroxide solution has a sodium hydroxide mass fraction of 40%.

[0120] Example 10 is based on the preparation method of Example 1, but the method is adjusted as follows:

[0121] Step 1: Mix 200g sodium silicate solution, 65g sodium hydroxide solution, 15g nano silica, 3g carbon nanotubes and 130g water at 300rpm for 5min, and adjust the temperature to 25℃ (20-30℃ is acceptable) to obtain the alkali activator; the sodium silicate solution has a modulus of 2, and the sodium hydroxide solution has a sodium hydroxide mass fraction of 40%.

[0122] Example 11 is based on the preparation method of Example 1, but the method is adjusted as follows:

[0123] Step 1: Mix 200g sodium silicate solution, 65g sodium hydroxide solution, 15g nano silica, 0.5g graphene oxide and 130g water at 300rpm for 5min, and adjust the temperature to 25℃ (20-30℃ is acceptable) to obtain the alkali activator; the sodium silicate solution has a modulus of 2, and the sodium hydroxide solution has a sodium hydroxide mass fraction of 40%.

[0124] Example 12 is based on the preparation method of Example 1, but the method is adjusted as follows:

[0125] Step 1: Mix 200g sodium silicate solution, 65g sodium hydroxide solution, 15g nano silica, 30g nano calcium carbonate, 0.5g graphene oxide, and 130g water at 300rpm for 5min, and adjust the temperature to 25℃ (20-30℃ is acceptable) to obtain the alkali activator; the sodium silicate solution has a modulus of 2, and the sodium hydroxide solution has a sodium hydroxide mass fraction of 40%.

[0126] The repaired concrete from Examples 9-12 was subjected to the performance tests described above, and the test results are shown in Table 3.

[0127] Table 3 Performance test results for Examples 1 and 9-12

[0128]

[0129] Referring to Table 3, a comparison of Examples 1 and 9-12 shows that changing the additive formulation from individual addition to a combination of two additives results in a better effect for the silica fume-steel fiber modified geopolymer repair material. Among the two combinations, the combination of nano-silica and nano-calcium carbonate yields the best performance. This is because nano-silica typically has a particle size of 10-50 nm, primarily filling nanoscale pores; while nano-calcium carbonate typically has a particle size of 50-200 nm, primarily filling submicron-scale pores. When used in combination, they form a multi-scale filling network from nanoscale to submicron scale, resulting in a significantly higher matrix density than when using a single nanomaterial. Nano-silica mainly contributes to improved compressive strength, while nano-calcium carbonate excels in durability (especially resistance to acid rain corrosion and carbonation). When the two are combined, nano-silica enhances the overall mechanical properties of the matrix through filling and interfacial reinforcement, while nano-calcium carbonate inhibits the penetration of corrosive media by accelerating the polymerization reaction and forming a denser microstructure, thereby achieving a comprehensive improvement in mechanical properties and durability.

[0130] Nano-silica provides an additional source of active silicon, while nano-calcium carbonate decomposes under alkaline activation conditions to generate calcium ions. These calcium ions interact with the aluminosilicate gel in the geopolymer system, promoting the formation of CASH gel and further enhancing the density of the gel network. The combination of the two simultaneously improves the thoroughness of the geopolymerization reaction and the stability of the gel product, thereby enabling the material to maintain higher and more stable mechanical properties throughout its service life.

[0131] In the slag-fly ash-silica fume system of this invention, the CaO provided by the slag and the calcium ions released by nano-calcium carbonate synergistically promote the formation of CASH gel; the active silica-alumina source provided by the fly ash generates a NASH network under the catalysis of nano-silica. The CASH / NASH composite gel structure formed by both exhibits high strength and excellent chemical stability.

[0132] The additive, a three-phase mixture of nano-silica, nano-calcium carbonate, and graphene oxide, yielded a silica fume-steel fiber modified geopolymer repair material with good performance. This is because the three components combine to form a full-scale reinforcement system from nanoparticles to nanosheets: nano-silica fills nanoscale pores, nano-calcium carbonate fills submicron-scale pores and provides a nucleation effect, while graphene oxide, as a two-dimensional sheet, provides bridging and crack-resistant functions on a larger scale. The three components synergistically construct a full-scale reinforcement network covering three-dimensional space.

[0133] Graphene oxide forms strong interfacial bonds with geopolymer gels through its abundant oxygen-containing functional groups on its surface. In the ternary system, graphene oxide also functions as a dispersant, promoting the uniform dispersion of nano-silica and nano-calcium carbonate in the matrix through its surface adsorption.

[0134] The three additives complement each other in their reinforcing mechanism: nano-silica mainly fills the voids and provides an active silicon source, nano-calcium carbonate mainly densifies the pores and induces crystal nucleation, and graphene oxide mainly strengthens the interface and blocks cracks. The composite gel system formed by the three has higher structural order and density. When nano-silica, nano-calcium carbonate, and graphene oxide are incorporated, the geopolymer repair material not only shows a significant increase in compressive strength, but also a greater increase in splitting tensile strength due to the bridging effect of graphene oxide and the reinforcement of the fiber-matrix interface. Furthermore, its corrosion resistance is excellent under corrosive environments such as chloride ion penetration and sulfate attack due to the densification effect contributed by the three additives.

[0135] The formulation of this invention contains 50-70 parts fly ash, and the low activity and high porosity of fly ash are its performance shortcomings. The active silicon source effect of nano-silica can effectively stimulate the activity of fly ash; the nucleation effect of nano-calcium carbonate provides nucleation sites for the dissolution and repolymerization of fly ash; the layered structure of graphene oxide can strengthen the connection between fly ash particles through its two-dimensional geometric properties. The combined effect of these three factors significantly improves the material performance under high fly ash content.

[0136] Examples 13-16

[0137] Example 13 is based on the preparation method of Example 1, but the method is adjusted as follows:

[0138] Step 1: Mix 15g of hydrophobic nano-silica and 130g of water at 1000rpm for 2min, and adjust the temperature to 25℃ (20-30℃ is acceptable) to obtain alkali activator A; Mix 200g of sodium silicate solution and 65g of sodium hydroxide solution at 300rpm for 3min, and adjust the temperature to 25℃ (20-30℃ is acceptable) to obtain alkali activator B;

[0139] Step 2: Mix 300g of slag, 600g of fly ash, and 100g of silica fume at 60rpm for 2 minutes. Then add 450g of standard sand and mix at 60rpm for 1 minute. Next, add alkali activator A and mix at 60rpm for 2 minutes. Then add alkali activator B and mix at 60rpm for 2 minutes to obtain a uniform slurry.

[0140] Example 14 is based on the preparation method of Example 1, but the method is adjusted as follows:

[0141] Step 1: Mix 15g of hydrophobic nano-silica and 130g of water at 1000rpm for 2min, and adjust the temperature to 25℃ (20-30℃ is acceptable) to obtain alkali activator A; Mix 200g of sodium silicate solution and 65g of sodium hydroxide solution at 300rpm for 3min, and adjust the temperature to 25℃ (20-30℃ is acceptable) to obtain alkali activator B;

[0142] Step 2: Mix 300g of slag, 600g of fly ash, and 100g of silica fume at 60rpm for 2 minutes. Then add 450g of standard sand and mix at 60rpm for 1 minute. Next, add alkali activator A and alkali activator B directly and mix at 60rpm for 2 minutes to obtain a uniform slurry.

[0143] Example 15 is based on the preparation method of Example 1, but the method is adjusted as follows:

[0144] Step 1: Mix 15g of hydrophobic nano-silica and 130g of water at 1000rpm for 2min, and adjust the temperature to 25℃ (20-30℃ is acceptable) to obtain alkali activator A; Mix 200g of sodium silicate solution, 65g of sodium hydroxide solution, and 30g of nano-calcium carbonate at 300rpm for 3min, and adjust the temperature to 25℃ (20-30℃ is acceptable) to obtain alkali activator B;

[0145] Step 2: Mix 300g of slag, 600g of fly ash, and 100g of silica fume at 60rpm for 2 minutes. Then add 450g of standard sand and mix at 60rpm for 1 minute. Next, add alkali activator A and mix at 60rpm for 2 minutes. Then add alkali activator B and mix at 60rpm for 2 minutes to obtain a uniform slurry.

[0146] Example 16 is based on the preparation method of Example 1, but the method is adjusted as follows:

[0147] Step 1: Mix 15g of hydrophobic nano-silica and 130g of water at 1000rpm for 2min, and adjust the temperature to 25℃ (20-30℃ is acceptable) to obtain alkali activator A; Mix 200g of sodium silicate solution, 65g of sodium hydroxide solution, 30g of nano-calcium carbonate and 0.5g of graphene oxide at 300rpm for 3min, and adjust the temperature to 25℃ (20-30℃ is acceptable) to obtain alkali activator B;

[0148] Step 2: Mix 300g of slag, 600g of fly ash, and 100g of silica fume at 60rpm for 2 minutes. Then add 450g of standard sand and mix at 60rpm for 1 minute. Next, add alkali activator A and mix at 60rpm for 2 minutes. Then add alkali activator B and mix at 60rpm for 2 minutes to obtain a uniform slurry.

[0149] The repaired concrete from Examples 13-16 was subjected to the performance tests described above, and the test results are shown in Table 4.

[0150] Table 4 Performance test results for Examples 1 and 13-16

[0151]

[0152] Referring to Table 4, a comparison of Examples 1 and 13-16 shows that hydrophobic nano-silica, when reacted with water to form alkali activator A, and other components to form alkali activator B, and when added separately to both alkali activator A and alkali activator B, results in a better-performing silica fume-steel fiber modified geopolymer repair material. This is because, under high-speed stirring, hydrophobic nano-silica forms water droplets that are encapsulated by hydrophobic nano-silica particles, creating a "liquid powder" structure, thus achieving "dry water." When added to the geopolymer system, the reduction in viscosity and yield shear stress is manifested in a higher flow diameter, mainly due to the combined effect of particle size coarsening in the precursor and the flowability of the hydrophobic nano-silica. The addition of conventional nanomaterials typically increases the viscosity of the slurry, while the "dry water" technology introduces hydrophobic nano-silica into the system in a pre-encapsulated form, avoiding direct adsorption of water by particles and an increase in system viscosity, thereby reducing the yield stress and plastic viscosity of the slurry.

[0153] This invention involves mixing hydrophobic nano-silica alone with water to form alkali activator A, and mixing sodium silicate solution and sodium hydroxide solution to form alkali activator B. In step two, alkali activator A is added first and then stirred, followed by alkali activator B and stirred again. The core principle of this design is that the hydrophobic modified layer on the surface of the hydrophobic nano-silica may be chemically eroded or damaged when in prolonged contact with high-concentration strong alkali solutions (especially 40% sodium hydroxide solution), leading to a decrease in its hydrophobic properties and dispersion performance. By adding the alkali activator step by step, with solution A added first, the hydrophobic nano-silica preferentially contacts the cementing material to form a coating layer before the high-concentration alkali activator B is introduced, effectively protecting the functional integrity of the hydrophobic nano-silica and maximizing the "dry water" enhancement effect.

[0154] When alkali activator A (a mixture of hydrophobic nano-silica and water) is first added to the cementitious material, water droplets encapsulated in the hydrophobic nano-silica are released on the surface of the cementitious material, forming a localized wetting and reaction interface, creating conditions for the initiation of the geopolymerization reaction. Subsequently, when alkali activator B (sodium silicate + sodium hydroxide) is added, the high-concentration alkali solution rapidly diffuses across the already formed wetting interface, triggering a more uniform and complete geopolymerization reaction. This avoids the problem of uneven reaction caused by excessively high localized alkali concentrations that may occur when the alkali activator is added all at once.

[0155] This application also provides a method for optimizing the performance of silica fume-steel fiber modified geopolymer repair materials, including the following steps:

[0156] Step S1: Determine the raw material characteristic parameters of the cementitious material, including the active silicon-aluminum molar ratio, glass phase content, and average particle size of the powder. Statistically analyze the distribution range of the raw material characteristic parameters to obtain fluctuation characteristics.

[0157] Among them, the raw material characteristic parameters refer to the intrinsic physicochemical parameters characterizing the chemical activity and physical morphology of the cementitious material. In this application, they specifically refer to the active silica-alumina molar ratio, glass phase content, and average particle size, used to quantify batch-to-batch variations in raw material quality. The active silica-alumina molar ratio refers to the molar ratio of active silica to active alumina in the cementitious material that can be dissolved by an alkali activator and participate in the polymerization reaction, determined using selective dissolution combined with inductively coupled plasma spectroscopy. The glass phase content refers to the mass fraction of amorphous glass phase in the cementitious material, determined using X-ray diffraction combined with the Rietveld refinement method, reflecting the potential reactivity of the raw material under alkali-activated conditions. The average particle size refers to the equivalent volume average particle size (D50) of the cementitious material powder, in μm, determined using a laser particle size analyzer. The fluctuation characteristics refer to the statistical distribution characteristics of the characteristic parameters of each raw material in multiple batches of industrial raw materials. In this application, they specifically refer to the mean, standard deviation, and coefficient of variation obtained after statistical analysis of the characteristic parameters of at least 10 consecutive batches of raw materials.

[0158] The necessary procedures are as follows:

[0159] (1) The chemical composition of each batch of slag, fly ash, and silica fume was determined. Active silica and active alumina were extracted by selective dissolution method. The molar ratio of active silica to alumina was calculated using the following formula:

[0160] .

[0161] In the formula, Mass of active silica, g; Mass of activated alumina, g; Molar mass of silicon dioxide; : Molar mass of aluminum oxide.

[0162] (2) The glass phase content of each batch of raw materials was determined by the XRD Rietveld refinement method.

[0163] (3) The average particle size of each batch of raw materials was determined by laser particle size analysis (ISO13320).

[0164] (4) Collect the above three parameters from no less than 10 industrial batches consecutively, and calculate the arithmetic mean μ, standard deviation σ, and coefficient of variation for each parameter:

[0165] .

[0166] coefficient of variation It is a dimensionless quantity used to measure the degree of dispersion between batches.

[0167] (5) Based on the actual measured distribution range (e.g.) The fluctuation range of each raw material characteristic parameter is represented by the mean, standard deviation, coefficient of variation, and distribution range, which together constitute the fluctuation characteristics.

[0168] Step S2: Call the pre-built proportion-performance prediction model, using raw material characteristic parameters, silica fume content, steel fiber content, and additive content as performance inputs to obtain performance indicators. Performance indicators include workability, mechanical properties, and durability. Workability includes flowability, mechanical properties include compressive strength, splitting tensile strength, and bond strength, and durability includes the chloride ion diffusion coefficient of the material bulk. Combined with fluctuation characteristics, evaluate the distribution range of performance indicators under raw material fluctuations, and use the performance indicators meeting preset requirements as constraints to solve for the range of parts with the lowest failure probability.

[0169] Among them, the ratio-performance prediction model refers to a data-driven prediction model that takes raw material characteristic parameters and ratio parameters as inputs and material performance indicators as outputs. In this application, it specifically refers to the mechanism-guided meta-learning network model constructed and verified through steps S21 to S24.

[0170] Performance indicators refer to a set of quantitative parameters for evaluating the overall performance of repair materials, including workability (flowability, unit mm), mechanical properties (28-day compressive strength, splitting tensile strength, bond strength, unit MPa), and durability (material bulk chloride ion diffusion coefficient, unit 10). -12 m 2 / s). Failure probability refers to the probability that, given a combination of proportions, the predicted performance value will not meet the preset requirements due to fluctuations in the characteristic parameters of the raw materials. The proportion range refers to the range of values ​​for the proportions of silica fume, steel fiber, and additives while meeting the target of minimizing the failure probability.

[0171] The necessary procedures are as follows:

[0172] (1) Obtain the fluctuation characteristics from step S1 and determine the statistical distribution type of the raw material characteristic parameters.

[0173] (2) Call the ratio-performance prediction model trained in steps S21 to S24.

[0174] (3) Using raw material characteristic parameters (active silicon-aluminum molar ratio, glass phase content, average particle size of powder) and part combination (parts of silica fume, parts of steel fiber, parts of additives) as model input, batch prediction of various performance indicators is performed.

[0175] (4) Regarding the specific execution process of combining the fluctuation characteristics to evaluate the distribution range of performance indicators under raw material fluctuations and using the performance indicators meeting the preset requirements as constraints to solve the minimum failure probability interval, please refer to the subsequent steps S2A to S2D. It will not be elaborated here.

[0176] Step S3: Prepare composite specimens based on the component range. The composite specimens are composed of a preset matrix and a repair layer formed by repair material covering the surface of the matrix. Conduct preset accelerated environment tests, use preset detection technology to track the damage at the interface between the repair layer and the matrix, extract the damage degradation rate of the interface based on the preset interface damage evolution model, obtain climate data, generate the service life curve of the interface based on the damage degradation rate and climate data, and retain the component range that meets the preset service life requirements as the feasible region for the proportioning.

[0177] The composite specimen refers to a two-layer structure specimen composed of a pre-designed substrate and a repair layer formed by repair material covering the substrate surface. In this application, it specifically refers to a cubic specimen with dimensions of 100mm × 100mm × 100mm. Accelerated environmental testing refers to a test method that accelerates the material degradation process by intensifying environmental factors under laboratory conditions. In this application, it specifically refers to the salt solution wet-dry cycle accelerated test, with the cycle regime being: immersion in a 3%~5% NaCl solution at room temperature for 12 hours, followed by drying at 60℃±2℃ for 12 hours, constituting one complete cycle. Pre-designed detection technology: In this application, it specifically refers to in-situ X-ray computed tomography (X-rayCT) technology, combined with a digital volume correlation (DVC) algorithm, used to track the three-dimensional damage evolution of the interface without damaging the specimen. Pre-designed interface damage evolution model: Refers to a pre-designed mathematical model describing the evolution of interface damage variables between the repair material and the substrate with the number of cycles. In this application, it specifically refers to a degradation model established based on a cohesive force model and a Weibull distribution. Damage degradation rate: refers to the rate at which the damage variable at the interface increases with the number of wet-dry cycles. It is calculated by a preset interface damage evolution model after extracting the interface strain field and displacement field from CT sequence images using DVC technology. Interface service life curve: refers to a quantitative curve describing the change in the degree of damage accumulation at the interface with equivalent service time. The horizontal axis represents equivalent service time (unit: year), and the vertical axis represents the damage variable (value range: 0~1). Fulfillable domain: refers to the set of proportions of interfaces that, after accelerated environmental testing, meet the preset service life requirements and achieve satisfactory workability and mechanical properties. It is the search space for optimization in step S4.

[0178] Step S3A involves preparing a composite specimen based on a predetermined component range. The composite specimen consists of a predetermined substrate and a repair layer formed by a repair material covering the substrate surface. A predetermined accelerated environmental test is conducted, and predetermined detection technology is used to track interface damage. The damage degradation rate of the interface is extracted based on a predetermined interface damage evolution model, and climate data is obtained. Based on the damage degradation rate and climate data, a service life curve of the interface is generated. The component range that meets the predetermined service life requirements is retained as the feasible region for the proportioning. This includes the following detailed process:

[0179] (1) Based on the part range, a composite specimen composed of a matrix and a repair layer formed by a repair material covering the surface of the matrix was prepared, and a salt solution wet-dry cycle accelerated test was carried out.

[0180] Specifically, the following steps are performed: prepare a geopolymer repair slurry according to the typical proportions (such as the median and boundary values) within the range of fractions obtained in step S2, pour it onto the surface of the pretreated old concrete substrate to form a composite specimen, and cure it for 28 days at a temperature of 20℃±2℃ and a relative humidity of ≥95%; place the cured composite specimen in a salt solution wet-dry cycle test chamber to perform accelerated environmental testing. The cycle regime is to soak in a 3%~5% NaCl solution at room temperature for 12 hours, followed by drying at 60℃±2℃ for 12 hours, which constitutes a complete cycle, and record the number of cycles.

[0181] (2) Under different cycles of accelerated testing, in-situ X-ray computed tomography was performed on the same interface position of the same composite specimen to obtain three-dimensional structural evolution sequence images of the interface region.

[0182] Specifically, at preset cycle number nodes of the accelerated test (such as cycles 0, 5, 10, 20, 30, and 50), the composite specimen is placed in an in-situ X-ray CT device to scan and acquire three-dimensional volumetric data of the interface area, with a spatial resolution of not less than [missing information]. This yields a sequence of CT images.

[0183] (3) The digital volume correlation technology is used to perform correlation calculation on the sequence images, calculate the three-dimensional displacement field and strain tensor of the interface, identify the microcrack initiation location by the preset strain criterion, and invert and calibrate the critical stress and corresponding number of cycles based on the preset interface mechanical model.

[0184] Specifically, the following steps are taken: A digital volume correlation algorithm is used to perform correlation calculations on CT volume data from adjacent cycles to calculate the three-dimensional displacement field and strain tensor of the interface region; the microcrack initiation location is identified using a preset strain criterion, and the interface opening displacement at the initiation point is extracted. And the corresponding critical cycle number. Let the critical cycle number for microcrack initiation be denoted as . Measured by DVC The corresponding interface opening displacement is .

[0185] The preset strain criterion is that the equivalent strain of the interface exceeds a preset threshold. , For interface equivalent strain; For strain tensor; The Frobenius double dot product of the strain tensor, i.e., the sum of squares of its components, is used to represent the strain tensor. (Using Einstein's summation convention); The preset threshold is determined by the statistical upper limit of the undamaged baseline specimen. Interface opening displacement. It is extracted directly from the displacement field of the initiation point by DVC.

[0186] A three-dimensional finite element model of the composite specimen was established, and the bilinear cohesion model was embedded into the interface elements, using the values ​​extracted in step (3). Using the interface opening displacement-cycle number evolution curve measured by DVC as the calibration target, the interface stiffness is obtained by inversion optimization by minimizing the L2 norm of the difference between the finite element prediction and the measured evolution curve. With peak intensity .

[0187] (4) Using the critical stress and the corresponding number of cycles, combined with the interface opening displacement when the specimen completely fails, the fracture parameters of the preset interface mechanical model are calibrated, the preset interface damage evolution model is established, and the damage degradation rate is extracted from the preset interface damage evolution model.

[0188] Specifically:

[0189] Based on the bilinear cohesive force model, the relationship between the interfacial traction force and the interfacial opening displacement is as follows:

[0190] .

[0191] In the formula, The force is the interfacial traction force, in MPa. For the interface opening displacement, units ; Interface stiffness, in MPa / ; For the critical opening displacement, unit ; For complete failure opening displacement, unit ; The value represents the peak intensity at the interface, in MPa.

[0192] Measured by DVC The corresponding interface opening displacement Combined with the interface opening displacement when the specimen completely fails Calibration of interfacial fracture energy Among them, the completely failed opening displacement Critical opening displacement measured directly by DVC The interface stiffness is extracted from step (3). With peak intensity Fracture energy determined by finite element inversion calibration Calculate using the triangle area formula:

[0193] .

[0194] The area under the traction-displacement curve of the bilinear cohesive model is a triangle with a base length of . Gao Wei Therefore, the fracture energy (energy dissipated per unit area) is the base multiplied by the height divided by 2, i.e. .

[0195] Because the bilinear cohesive model has different expressions for the traction force in the elastic rising segment and the softening falling segment, the interfacial cumulative dissipated energy density... The interface needs to be displaced according to the current loop count. The calculation is performed in segments according to the current stage.

[0196] When the interface is in a flexible phase ( At this time, traction force For the opening displacement from 0 to Integrate to obtain the cumulative energy dissipated during this stage:

[0197] .

[0198] When the interface enters the softening stage ( When this occurs, the energy accumulated in the elastic segment needs to be superimposed, and the traction force on the softened segment needs to be applied. from arrive integral:

[0199] .

[0200] Superimposed elastic segment cumulative energy (Depend on ), thus obtaining the total cumulative dissipated energy of the softened section.

[0201] When the interface completely fails ( When the traction force drops to zero, the accumulated dissipated energy reaches its saturation value, i.e., the fracture energy. .

[0202] Combining the above three sections, the cumulative energy dissipation at the interface during the Nth cycle is obtained as follows:

[0203] .

[0204] Within the framework of continuous damage mechanics, damage variables must satisfy boundedness (values ​​[0,1]), monotonically non-decreasing property, and physical interpretability. The ratio of cumulative dissipated energy density G(N) to the total fracture energy Gc at the interface is defined as the damage variable, which simultaneously satisfies the above conditions: when N=0, G(0)=0, therefore D=0 (no damage); when the interface completely fails, G→Gc, therefore D→1; and G(N) monotonically increases, ensuring that D monotonically does not decrease. Thus, energy-type damage variables are obtained:

[0205] .

[0206] The energy-type damage variable was calculated for each cycle number. Measured sequence and number of loops Perform nonlinear regression, fit the data using a Weibull distribution, and determine the number of feature iterations. With shape parameters This ensures that the coefficient of determination meets a preset threshold, thereby establishing the evolutionary pattern of the cumulative damage variable:

[0207] .

[0208] In the formula, For the number of feature loops ( (Number of loops at time) For shape parameters ( This indicates accelerated damage. For exponential degradation, (Indicates damage deceleration).

[0209] The Weibull distribution originates from the Weakest Link Theory and is applicable to the successive activation and connection of interfaces composed of numerous micro-defects under the drive of thermal and wet stress and salt crystallization pressure. By directly mapping the failure probability to the degree of interface damage, the evolution law of the cumulative damage variable mentioned above is obtained.

[0210] The rate of damage and degradation of the interface is determined by... Differentiation yields:

[0211]

[0212] Derivation Explanation: Let ,but By the chain rule Substitute That's it.

[0213] (5) Obtain climate data of the target service environment, convert the damage degradation rate into an equivalent service time scale based on the preset environmental equivalent conversion relationship, and generate the interface service life curve from the preset environmental equivalent conversion relationship.

[0214] Specifically, this involves acquiring climate data (annual average temperature, humidity, rainfall frequency, and salt spray deposition rate) of the target project's service environment. Environmental equivalence conversion requires comprehensive consideration of the field cycling frequency and the temperature acceleration effect. Based on the Time-Temperature-Humidity Superposition (TTSP) principle, the equivalent conversion from laboratory accelerated cycling to the field service environment can be decomposed into the combined effect of the annual average dry-wet cycle frequency and the temperature acceleration factor, thus yielding the environmental equivalent cycling frequency:

[0215] .

[0216] In the formula, Equivalent laboratory cycle frequency (times / year) represents the conversion factor for converting the average annual natural cycle frequency in the field to the equivalent laboratory cycle frequency based on the temperature acceleration effect. The average annual frequency of natural wet and dry cycles in the local area (times / year) is obtained from climate data (average number of rainy days per year, humidity fluctuation cycle). The Arrhenius temperature acceleration factor reflects the rate of reaction accelerated by laboratory temperature relative to field temperature.

[0217] .

[0218] Interface degradation involves ion diffusion and chemical reactions, both of which follow an Arrhenius temperature dependence. Due to laboratory temperature... Higher than the ambient temperature Acceleration factor That is, the interfacial damage caused by a single laboratory cycle is equivalent to A single on-site cycle; conversely, a single on-site cycle is only equivalent to The equivalent laboratory cycle frequency is the field frequency divided by the acceleration factor. This factor is then used to... Laboratory conditions are mapped to average annual temperature conditions in the field to ensure that the equivalent conversion is interpretable from a chemical kinetic perspective.

[0219] In the formula, The activation energy of the interface degradation reaction (in J / mol) is determined by the Arrhenius graphical method from at least three sets of accelerated tests at different temperatures. This is the universal gas constant; ( ( ) represents the laboratory drying temperature; The local annual average thermodynamic temperature (in K).

[0220] Based on environmental equivalence conversion relationships, the number of laboratory cycles was calculated. Converted to equivalent service time :

[0221] .

[0222] Will Substitution The service life curve of the generated interface:

[0223] .

[0224] Extrapolate until the damage variable reaches a preset failure threshold (e.g.) Corresponding service life:

[0225] .

[0226] Based on the service life curve, let hour ,Will Substitution get Then, the design service life can be obtained through inverse algebraic solution. The explicit expression for this formula. This formula has a good definition for all positive parameters and directly corresponds to the ultimate durability index in engineering structural design.

[0227] The component quantity range and the corresponding interface service life curve are marked as the feasible region of the ratio.

[0228] Specifically: if the service life of the composite specimen interface corresponding to the number of samples meets the preset requirements (e.g., If the condition is met, then the interval containing that number of parts is retained as the feasible region for matching; otherwise, the interval is discarded.

[0229] Step S4: Based on the service life curve of the interface, with the goal of maximizing the service life of the interface and with the workability and mechanical properties meeting the preset requirements as hard constraints, the optimal part ratio range is deduced within the feasible range of the proportion. The optimal part ratio range is then output to the proportion design stage to adjust the proportions of silica fume, steel fiber and additives, thereby achieving pre-prediction and source control of durability.

[0230] In this context, "hard constraints" refer to boundary conditions that must be strictly met and cannot be violated during the optimization process; in this application, this specifically refers to the requirement that both workability and mechanical performance indicators must be met simultaneously. The optimal proportion range refers to the range of proportions of silica fume, steel fiber, and additives within the feasible proportion region that maximizes the service life of the interface while satisfying all hard constraints. The proportioning design stage refers to the formulation determination phase during the production or on-site preparation of repair materials; in this application, it specifically refers to the process step of adjusting the actual weighing amounts of silica fume, steel fiber, and additives according to the optimal proportion range.

[0231] The necessary procedures are as follows:

[0232] (1) Obtain the set of service life curves of the feasible region of the ratio and the corresponding interface from step S3;

[0233] (2) Establish a constrained single-objective robust optimization model. The objective function is to maximize the service life of the interface:

[0234] .

[0235] In the formula, decision variables Let be a vector consisting of the number of parts silica ash, the number of parts steel fiber, and the number of parts additives, and Belongs to the feasible region of proportioning The service life of the interface. It is calculated based on the preset interface damage evolution model and the environmental equivalence conversion relationship.

[0236] Due to Weibull parameters and With proportion vector Since there is no explicit analytical relationship derived from first principles, a second-order polynomial response surface methodology is used to establish a surrogate model for quickly predicting interface durability parameters under different mix proportions during the optimization process. This method, through a small number of typical mix proportion experiments, fits complex nonlinear mappings with low-order polynomials, significantly reducing computational costs while maintaining accuracy. The response surface form is as follows:

[0237] .

[0238] .

[0239] The input dimension of the response surface methodology is consistent with the dimension of the proportioning parameters: when the additive is a single component, the input dimension... (Parts of silica fume, parts of steel fiber, and parts of additives); when the additives are... When seeding, the input dimension is expanded to (The number of parts silica fume, steel fiber, and each additive) will be used to expand the response surface coefficient accordingly. Where, The number of parts of silica ash. This refers to the number of steel fiber parts. The number of parts for each additive; The response surface coefficients are obtained by least squares fitting. To fit the residuals, the response surface model is used as a fast predictor for the pre-defined interface damage evolution model in S4 optimization.

[0240] The hard constraint is: .

[0241] In the formula, the values ​​with caps represent the performance prediction values ​​obtained by calling the ratio-performance prediction model in step S2.

[0242] (3) The particle swarm optimization algorithm is used to find the global optimal solution within the feasible region. The population size is set. Maximum number of iterations The particle velocity update formula is:

[0243] .

[0244] .

[0245] In the formula, For the first The first particle The generation Dimensional speed; For the corresponding position; Inertia weight; For cognitive and social learning factors; Uniform random numbers; :particle Historically best position; The global optimal location is determined by truncating the location boundary according to the upper and lower limits of the feasible region.

[0246] The particle velocity update follows the standard particle swarm iteration rules, and the search direction is guided by individual optimality and global optimality to obtain the global optimal solution;

[0247] (4) At the optimal ratio point The boundaries are successively expanded within the neighborhood of each dimension, and the failure probability of each ratio point in the expanded superrectangular region is evaluated by Monte Carlo simulation. The maximum boundary value that still meets the preset threshold in each dimension is taken to form the optimal number interval of axis alignment.

[0248] (5) Output the optimal part range to the proportion design stage, adjust the actual dosage of silica fume, steel fiber and additives, so that the durability of the repair material can be predicted in advance and controlled at the source during the preparation stage.

[0249] The steps for constructing the ratio-performance prediction model are as follows:

[0250] Step S21: Collect historical data of multiple batches of industrial raw materials, determine the characteristic parameters of each batch of raw materials, the amount of silica fume, the amount of steel fiber, and the amount of additives, and determine the actual values ​​of performance indicators to form a training sample set.

[0251] Historical data refers to raw material testing records and corresponding proportions and performance test data accumulated in previous production or testing, and in this application specifically refers to data from no fewer than 50 different batches of industrial raw materials. The training sample set refers to the input-output paired data set used for training the machine learning model, and each sample includes raw material characteristic parameters, proportion parameters, and the actual values ​​of the corresponding performance indicators.

[0252] The necessary procedures are as follows:

[0253] (1) Randomly select no less than 50 different batches of slag, fly ash, and silica fume samples from the production line or raw material warehouse;

[0254] (2) Determine the characteristic parameters of each batch of raw materials according to the method in step S1;

[0255] (3) Record the number of parts of silica fume, steel fiber, and additives for each batch of raw materials;

[0256] (4) Determine the fluidity, compressive strength, splitting tensile strength, bond strength and chloride ion diffusion coefficient according to relevant standards to obtain the actual values ​​of the performance indicators;

[0257] (5) Pair the raw material characteristic parameters, proportion parameters and actual performance index values ​​of each batch to form a training sample set.

[0258] Step S22: Analyze the correlation between the raw material characteristic parameters and the actual values ​​of performance indicators in the training sample set, determine the network topology based on the correlation, construct the mechanism-guided initial network, take the raw material characteristic parameters and the number of silica fume, steel fiber and each additive as input, take the performance indicators as output, and embed the preset physical constraint term in the loss function.

[0259] The correlation law refers to the statistical correlation or causal relationship between raw material characteristic parameters and performance indicators. In this application, it specifically refers to the input-output sensitivity ranking obtained by quantification through Pearson correlation coefficient, mutual information, or grey relational analysis. The network topology refers to the connection method and dimensional configuration of nodes in each layer of the neural network. In this application, it specifically refers to the dynamically configured nodes in the input layer (based on 6 nodes for 3 raw material characteristic parameters + 3 proportioning parameters; when there are multiple additives, the input nodes are configured according to the number of each additive, adjusted to 5+m (3 raw material characteristic parameters + silica fume + steel fiber + each additive)), 3 hidden layers, and 5 nodes in the output layer (corresponding to 5 performance indicators). The mechanism-guided initial network refers to a neural network that incorporates prior knowledge of materials science on the basis of traditional data-driven networks. In this application, it specifically refers to the initial weights of the network being generated from synthetic samples constrained by the physical relationship between gel production amount and strength, and set after pre-training, rather than random initialization. The physical constraint term refers to the penalty term embedded in the loss function that reflects the basic laws of materials science. In this application, it specifically refers to the monotonicity constraint and the gel mass conservation constraint.

[0260] The necessary procedures are as follows:

[0261] (1) Normalize the training sample set, use Pearson correlation coefficient and mutual information analysis to analyze the correlation between the characteristic parameters of each raw material and the actual values ​​of the performance index, and sort the input features according to their sensitivity to determine their importance.

[0262] (2) Determine network topology based on association patterns: The number of nodes in the input layer is dynamically configured (the basic configuration is 6 nodes; when there are multiple additives, the input nodes are configured according to the number of portions of each additive, and the number of nodes is adjusted to...). The system has 3 hidden layers (with 128, 256, and 128 nodes respectively), 5 output layers, ReLU activation function, and Softplus activation function for output layer stress resistance to ensure non-negative prediction values.

[0263] (3) Constructing a mechanism-guided initial network. The amount of geopolymer gel generated mainly depends on the total amount of active silica and aluminum in the system. In the slag-fly ash-silica ternary cementitious system, slag and fly ash provide the basic active silica and aluminum sources, with contents of respectively... and Silica fume, as a highly active microsilica powder, has a specific dosage. It has a significant enhancing effect on gel formation, and the enhancing effect can be corrected by a modifier. Description: The active aluminum phase in fly ash is also affected by the dosage, with a correction factor of [missing value]. Based on the principle of stoichiometric linear superposition, the total gel formation is the sum of the contributions from each source:

[0264] .

[0265] In the formula, The active silicon-aluminum content (mass fraction) of the raw material. This refers to the number of parts of silica ash. The fly ash content is determined by subtracting 30 parts of slag and silica fume from 100 parts of cementitious materials. This is the gel yield coefficient; The coefficient for enhancing the activity of silicon-aluminum alloys is then used to generate synthetic samples based on the gel-strength power law. :

[0266] .

[0267] In the formula, The network is obtained by fitting training data. The initial mechanism-guided network is pre-trained using synthetic samples: all layers except the output layer are frozen, and the output layer weights converge only on the distribution constrained by the gel-strength physical relationship, driven solely by synthetic samples; after pre-training, the layers are unfrozen, and end-to-end fine-tuning is performed using real training samples. Pre-training causes the network's initial parameters to converge near the physical relationship, replacing random initialization.

[0268] Improvement layer: Generates synthetic samples based on this physical relationship, and pre-trains the network to make the initial weights approximate the physical trend. Finalization layer: Obtains the initial network parameters guided by the mechanism.

[0269] (4) Embed physical constraint terms in the loss function. The base layer uses mean squared error loss:

[0270] .

[0271] In the formula, The weighting coefficient for the k-th performance indicator is... This is the actual value. This is the network prediction value. The total number of samples.

[0272] The improved layer introduces a monotonicity penalty term and a gel conservation penalty term. The monotonicity penalty term is defined as follows:

[0273] .

[0274] In the formula, As a range weight, it is activated only within a reasonable doping range covered by the training data (e.g., silica ash content of 0-20 parts per cubic centimeter). Outside this range, it is not activated. To avoid punishing a physically reasonable downward trend.

[0275] The gel conservation penalty term is defined as follows:

[0276] .

[0277] In the formula, This is the theoretical gel formation amount calculated based on the stoichiometry of the geopolymer reaction. This represents the predicted gel formation amount for the intermediate layers of the network. The theoretical gel formation amount is calculated using the following formula:

[0278] .

[0279] That is, the total amount of gel formed. Consistent. In the formula, The active silicon-aluminum content (mass fraction, dimensionless). This is the gel yield coefficient (dimensionless).

[0280] Shaping layer:

[0281] The total loss function is: .

[0282] In the formula, , The penalty coefficient is 0.01 to 0.1, calibrated by sensitivity analysis on the validation set. The initial network is trained via backpropagation.

[0283] Step S23: Divide the training sample set into meta-training tasks and meta-testing tasks according to raw material batches. Based on the meta-training tasks and the initial network, use a model-independent meta-learning algorithm to train and obtain a meta-model. Verify the generalization ability of the meta-model based on the meta-testing tasks.

[0284] The training process includes: (1) Task sampling: Randomly sampling raw materials from the training sample set according to batch. Individual training task Each task contains the batch of raw materials. (2) Inner loop adaptation: for each task From the current global parameters Departure, Execution Step gradient descent:

[0285] .

[0286] In the formula, For the inner loop learning rate, The total loss function for step S22;

[0287] (3) Outer loop meta-optimization: Calculate the meta-loss for all meta-training tasks. Perform gradient descent on global parameters:

[0288] .

[0289] In the formula, The meta-learning rate;

[0290] (4) Iterate through steps (2)-(3) until the meta-loss converges.

[0291] In this application, the meta-training task refers to a set of multiple sub-tasks used for the first stage of meta-learning training. Specifically, it refers to K tasks (K≥10) randomly selected from the training sample set according to raw material batches, with each task containing several samples from that batch. The meta-testing task refers to a set of sub-tasks used for the second stage of meta-learning to verify generalization ability. Specifically, it refers to M tasks (M≥5) randomly selected from the training sample set according to raw material batches, with no overlap with the raw material batches of the meta-training task. The model-independent meta-learning algorithm refers to a meta-learning framework independent of model architecture. In this application, it specifically refers to Model-Independent Meta-Learning (MAML), which learns an initial parameter sensitive to task distribution by performing a small number of gradient descent iterations on each task, enabling new tasks to adapt quickly with only a small number of samples. The meta-model refers to the set of network parameters with rapid adaptability obtained after MAML training.

[0292] Step S24: Based on the meta-test task, perform a preset number of iterations to verify the meta-model, ensuring that the coefficient of determination for predicting the compressive strength at a preset age meets a preset first threshold, and the coefficient of determination for predicting the bond strength meets a preset second threshold. Output the verified meta-model. Wherein: the coefficient of determination is a statistical indicator used to evaluate the predictive accuracy of a regression model, denoted as... Dimensionless. Preset first threshold: refers to the minimum accuracy that the validation meta-model must achieve in predicting 28-day compressive strength. The value is set to 0.92 in this application. The preset second threshold refers to the minimum threshold that the meta-model must reach to verify the accuracy of its bond strength prediction. The value is set to 0.88 in this application.

[0293] The necessary procedures are as follows:

[0294] (1) Load the meta-model parameters obtained in step S23.

[0295] (2) For each new batch of raw materials in the meta-test task, perform a small amount of gradient updates (such as 5-step gradient descent) using only 1 to 5 samples of the batch to obtain the adapted task-specific model.

[0296] (3) Using the adapted model, predict the 28-day compressive strength and bond strength of this batch of raw materials under each ratio, and calculate the coefficient of determination between the predicted and actual values:

[0297] .

[0298] In the formula, This is the actual value. For predicted values, The average of the actual values. This represents the number of test samples.

[0299] (4) Verify the 28-day compressive strength of no less than 10 meta-test tasks separately. And the bond strength If the condition is met, the output validation meta-model is used as the matching-performance prediction model; if not, return to step S22 to adjust the network topology or physical constraints and retrain.

[0300] Step S25: Use the validation meta-model as the matching-performance prediction model.

[0301] Based on the aforementioned fluctuation characteristics, the distribution range of the performance indicators under raw material fluctuations is evaluated. With the performance indicators meeting preset requirements as a constraint, the interval of the lowest possible failure probability is determined, including:

[0302] Step S2A: Based on the fluctuation characteristics, the distribution of the raw material characteristic parameters is fitted to obtain the probability density function of each raw material characteristic parameter, thus forming the joint probability distribution of the raw material characteristic parameters.

[0303] The probability density function refers to the function that describes the probability distribution of a continuous random variable across different intervals. In this application, it specifically refers to the statistical distribution density expression that each of the raw material characteristic parameters follows. The joint probability distribution refers to the function that describes the probability distribution of multiple random variables taking values ​​simultaneously. In this application, it specifically refers to the overall distribution of the three raw material characteristic parameters after coupling through a Copula function.

[0304] The necessary procedures are as follows:

[0305] (1) Obtain the fluctuation characteristics from step S1, including the mean, standard deviation and sample data of each raw material characteristic parameter;

[0306] (2) Perform a normality test on the characteristic parameters of each raw material. If they follow a normal distribution, then the probability density function is:

[0307] .

[0308] In the formula, These correspond to the active silicon-aluminum molar ratio, glass phase content, and average particle size of the powder, respectively. For the first Raw material characteristic parameters; The mean; The standard deviation is denoted as .

[0309] If the distribution does not follow a normal distribution, then kernel density estimation (KDE) is used for fitting.

[0310] (3) Marginal distribution fitting and independence test were performed on the characteristic parameters of each raw material; for parameters with correlation, the marginal probability density functions of each characteristic parameter of the raw material were coupled into a joint probability distribution using the Gaussian Copula function:

[0311] .

[0312] In the formula, Correlation coefficient matrix Three-dimensional standard normal joint CDF; It is the inverse CDF (quantile function) of the standard normal distribution. For the first Marginal CDF of the term parameters. Correlation coefficient matrix. elements Kendall's stau rank correlation coefficient Through relational formulas It was estimated.

[0313] Step S2B: Based on the joint probability distribution, perform preset random sampling to generate multiple sets of raw material characteristic parameter samples; select multiple combinations of component numbers, which consist of the number of parts of silica fume, steel fiber, and additives; call the proportion-performance prediction model, using the multiple sets of raw material characteristic parameter samples and multiple combinations of component numbers as inputs, to obtain multiple sets of predicted performance index values.

[0314] Among them, preset random sampling refers to a numerical method that randomly selects representative samples according to a joint probability distribution, and in this application, it specifically refers to Latin hypercube sampling (LHS) or Monte Carlo random sampling (MCS). The number combination refers to the discrete combination of the proportions of the three components, silica fume, steel fiber, and additives, and in this application, it specifically refers to grid points or random sampling points selected within their respective allowable ranges according to a preset step size.

[0315] The necessary procedures are as follows:

[0316] (1) Based on the joint probability distribution obtained in step S2A, N sets of raw material characteristic parameter samples (N≥1000) are generated by Latin hypercube sampling so that the samples cover the main probability quality range of each raw material characteristic parameter (±3σ range for normal distribution, and confidence interval with cumulative probability not less than 99.7% for non-normal distribution).

[0317] (2) Within the range of 0~20 parts of silica fume, 0~12 parts of steel fiber, and 0.05~4.55 parts of additives, select M groups of parts (M≥100) according to the preset step size, or use Sobol sequence to generate space filling design samples;

[0318] (3) For each group of parts, pair it with N groups of raw material characteristic parameter samples to form N×M input vectors;

[0319] (4) Call the ratio-performance prediction model output in step S24, input the above N×M groups of vectors in batches, and obtain the corresponding set of predicted performance index values.

[0320] Step S2C: Calculate the proportion of predicted performance indicators that do not meet the preset requirements and determine it as the failure probability.

[0321] Among them, the preset requirements refer to the minimum technical indicators that the repair material must meet. In this application, they specifically refer to the hard constraint conditions: flowability ≥ 140 mm, 28-day compressive strength ≥ 60 MPa, splitting tensile strength ≥ 5.0 MPa, bond strength ≥ 5.0 MPa, and chloride ion diffusion coefficient ≤ 5.0 × 10-12 m2 / s.

[0322] The necessary procedures are as follows:

[0323] (1) For each combination of samples, determine whether the predicted performance index values ​​of the N groups meet the preset requirements. If any index does not meet the requirements, mark the sample as invalid.

[0324] (2) Count the number of failed samples in this group of components. ;

[0325] (3) Calculate the failure probability of this combination of components: .

[0326] The failure probability is a dimensionless quantity with a value range of [0,1].

[0327] (4) Perform the above statistics on each of the M group combinations to obtain the failure probability corresponding to each group combination.

[0328] Step S2D involves selecting the combination of parts with the lowest failure probability from multiple combinations of parts, using the lowest failure probability as the screening objective; and determining the number of parts range based on the combination of parts.

[0329] The component combination refers to the set of discrete proportioning points whose failure probability is lower than a preset threshold after statistical analysis in step S2C. Each combination is determined by the three-dimensional coordinates of the number of silica fume, steel fiber, and additives. The component range refers to the result after converting the discrete component combinations into a continuous range of values, allowing for continuous adjustment within the range during the proportioning design process.

[0330] Necessity Explanation: The component combinations generated in step S2B are discrete grid points (e.g., 10 parts silica fume / 8 parts steel fiber / 2.0 parts nano silica), while step S3 and subsequent proportioning design steps require a continuous component adjustment range (e.g., 8-15 parts silica fume) to flexibly fine-tune within the range according to batch fluctuations of raw materials. Therefore, although step S2D is summarized as "screening component range" in the claims, its implementation necessarily includes a data processing step of converting discrete qualified points into a continuous range. The detailed disclosure of this step is given below:

[0331] The necessary procedures are as follows:

[0332] (1) Failure probability screening of discrete component combinations:

[0333] Will The component combinations are sorted in ascending order of failure probability, and the failure probability is then filtered. The combination of the number of parts with a preset threshold (value of 0.05), or the combination of the parts with the lowest failure probability. Group( This forms a set of discrete combinations with low failure probability. ,in The number of parts of silica ash. This refers to the number of steel fiber parts. This refers to the number of additives.

[0334] (2) Determine the number of parts interval based on the combination of parts:

[0335] For the selected set of discrete number combinations Cluster analysis was performed to identify densely distributed regions in the three-dimensional ratio space of silica fume, steel fiber, and additives. The minimum and maximum values ​​of each main cluster in each dimension were extracted as continuous boundaries to form the ratio interval.

[0336] Specifically: using the K-means algorithm or the DBSCAN algorithm to... Perform clustering. If using the K-means algorithm, determine the optimal number of clusters based on the elbow rule or silhouette coefficient. (generally Using Euclidean distance as a similarity metric Divided into Cluster If the DBSCAN algorithm is used, set the neighborhood radius. (e.g., 1.5 times the step size of each dimension) and minimum number of samples (e.g. 5) The core clusters are divided based on density reachability and noise points are removed.

[0337] (3) Determining the boundary of continuous values:

[0338] Extract the minimum and maximum values ​​of each valid principal cluster in each dimension, and use them as the axis-aligned boundaries of the continuous interval:

[0339] .

[0340] .

[0341] .

[0342] The above boundaries form a continuous interval of parts:

[0343] .

[0344] (4) Interval optimization and output:

[0345] If multiple clusters correspond to multiple number intervals, the interval containing the most original number combinations (i.e., the interval corresponding to the largest cluster) is selected first, or the interval with the longest service life is selected based on the interface service life prediction value; the selected number interval is used as the output of step S2 and passed to step S3.

[0346] This specific embodiment is merely an explanation of this application and is not intended to limit it. After reading this specification, those skilled in the art can make modifications to this embodiment without contributing any inventive step, but such modifications are protected by patent law as long as they fall within the scope of the claims of this application.

Claims

1. A silica-fume-steel-fiber-modified geopolymer repair material, characterized in that, The components include the following parts by weight: Cementitious materials: 30 parts slag, 50-70 parts fly ash, and 0-20 parts silica fume, wherein the total weight of the slag, fly ash, and silica fume is 100 parts. Alkali activator: 20 parts sodium silicate solution, 6.5 parts sodium hydroxide solution, 0.05-4.55 parts additive, and 10-16 parts water; the sodium silicate solution has a modulus of 2, and the sodium hydroxide solution has a sodium hydroxide mass fraction of 40%; the additive includes at least one of nano-silica, nano-calcium carbonate, carbon nanotubes, and graphene oxide. Other: 45 parts standard sand, 0-12 parts steel fiber.

2. The silica fume-steel fiber modified geopolymer repair material according to claim 1, characterized in that: The additive is nano-silica.

3. The silica fume-steel fiber modified geopolymer repair material according to claim 1, characterized in that: The additives are nano-silica and nano-calcium carbonate.

4. The silica fume-steel fiber modified geopolymer repair material according to claim 1, characterized in that: The additives are nano-silica, nano-calcium carbonate, and graphene oxide.

5. The method for preparing the silica fume-steel fiber modified geopolymer repair material according to any one of claims 1-4, characterized in that, Includes the following steps: Step 1: Mix the prescribed amounts of sodium silicate solution, sodium hydroxide solution, additives, and water, and adjust the temperature to 20-30℃ to obtain the alkali activator; Step 2: Mix the slag, fly ash, and silica fume according to the formula, then add the standard sand according to the formula and mix, and then add the alkali activator and mix to obtain a uniform slurry; Step 3: Add the prescribed amount of steel fiber to the homogeneous slurry and stir to obtain silica fume-steel fiber modified geopolymer repair material.

6. The preparation method of the silica fume-steel fiber modified geopolymer repair material according to claim 5, characterized in that: In step one, the additive includes hydrophobic nano-silica; Step 1: Mix the prescribed amount of hydrophobic nano-silica and water, and adjust the temperature to 20-30℃ to obtain alkali activator A; mix the prescribed amounts of sodium silicate solution and sodium hydroxide solution, and adjust the temperature to 20-30℃ to obtain alkali activator B; Step 2: Mix the slag, fly ash, and silica fume according to the formula, then add the standard sand according to the formula and mix. Next, add the alkali activator A and mix, then add the alkali activator B and mix to obtain a uniform slurry.

7. The method for optimizing the performance of the silica fume-steel fiber modified geopolymer repair material according to any one of claims 1-4, characterized in that, Includes the following steps: The raw material characteristic parameters of the cementitious material are determined, including the active silica-alumina molar ratio, glass phase content, and average particle size of the powder. The distribution range of the raw material characteristic parameters is statistically analyzed to obtain the fluctuation characteristics. A pre-built proportioning-performance prediction model is invoked, using the raw material characteristic parameters, silica fume content, steel fiber content, and additive content as inputs to obtain performance indicators. These performance indicators include workability, mechanical properties, and durability. Workability includes flowability, mechanical properties include compressive strength, splitting tensile strength, and bond strength, and durability includes the chloride ion diffusion coefficient of the material bulk. Combining the fluctuation characteristics, the distribution range of these performance indicators under raw material fluctuations is evaluated. With the performance indicators meeting preset requirements as a constraint, the range of proportions with the lowest failure probability is determined. Composite specimens are prepared based on the specified component range. The composite specimen is composed of a preset matrix and a repair layer formed by the repair material covering the surface of the matrix. Preset accelerated environmental tests are carried out, and preset detection technology is used to track interface damage. The damage degradation rate of the interface is extracted based on a preset interface damage evolution model. Climate data is obtained, and the service life curve of the interface is generated based on the damage degradation rate and the climate data. The component range that meets the preset service life requirements is retained as the feasible region for proportioning. Based on the service life curve of the interface, with the goal of maximizing the service life of the interface and with the workability and mechanical properties meeting preset requirements as hard constraints, the optimal part ratio range is deduced within the feasible range of the ratio. The optimal part ratio range is then output to the ratio design stage to adjust the parts of silica fume, steel fiber and additives, thereby achieving pre-prediction and source control of durability.

8. The method for performance optimization of the silica fume-steel fiber modified geopolymer repair material according to claim 7, characterized in that, The steps for constructing the ratio-performance prediction model are as follows: Historical data of multiple batches of industrial raw materials were collected, and the characteristic parameters, silica fume content, steel fiber content, and additive content of each batch of raw materials were measured. The actual values ​​of performance indicators were also measured to form a training sample set. The correlation between the raw material characteristic parameters and the actual values ​​of the performance indicators in the training sample set is analyzed. Based on the correlation, the network topology is determined, and a mechanism-guided initial network is constructed. The raw material characteristic parameters, the amount of silica fume, the amount of steel fiber, and the amount of each additive are used as inputs, and the performance indicators are used as outputs. A preset physical constraint term is embedded in the loss function. The training sample set is divided into meta-training tasks and meta-testing tasks according to raw material batches. Based on the meta-training tasks and the initial network, a model-independent meta-learning algorithm is used for training to obtain a meta-model. The generalization ability of the meta-model is verified based on the meta-test task. Based on the meta-test task, the meta-model is iterated and verified a preset number of times to ensure that the prediction determination coefficient of the compressive strength at a preset age in the mechanical properties meets a preset first threshold and the prediction determination coefficient of the bond strength meets a preset second threshold, and the verification meta-model is output. The validation meta-model is used as the ratio-performance prediction model.

9. The method for optimizing the performance of the silica fume-steel fiber modified geopolymer repair material according to claim 7, characterized in that, Based on the aforementioned fluctuation characteristics, the distribution range of the performance indicators under raw material fluctuations is evaluated. With the performance indicators meeting preset requirements as a constraint, the interval of the lowest possible failure probability is determined, including: Based on the fluctuation characteristics, the distribution of the raw material characteristic parameters is fitted to obtain the probability density function of each raw material characteristic parameter, thus forming the joint probability distribution of the raw material characteristic parameters; Based on the joint probability distribution, a preset random sampling is performed to generate multiple sets of raw material characteristic parameter samples; multiple combinations of component numbers are selected, which consist of the number of parts of silica fume, steel fiber, and additives; the ratio-performance prediction model is called, and multiple sets of performance index prediction values ​​are obtained by taking the multiple sets of raw material characteristic parameter samples and the multiple combinations of component numbers as input. The proportion of predicted values ​​of the multiple sets of performance indicators that do not meet the preset requirements is determined as the failure probability. Using the lowest failure probability as the screening objective, the combination of parts with the lowest failure probability is selected from the multiple combinations of parts, and the number of parts is determined based on the combination of parts.

10. The method for performance optimization of the silica fume-steel fiber modified geopolymer repair material according to claim 9, characterized in that, Composite specimens are prepared based on the specified component range. Each composite specimen consists of a predetermined substrate and a repair layer formed by the repair material covering the surface of the substrate. Predetermined accelerated environmental tests are conducted, and predetermined detection techniques are used to track interface damage. The damage degradation rate of the interface is extracted based on a predetermined interface damage evolution model. Climate data is obtained, and a service life curve of the interface is generated based on the damage degradation rate and the climate data. The component range that meets the predetermined service life requirements is retained as the feasible mixing region, including: Based on the specified fraction range, a composite specimen was prepared, which was composed of the preset substrate and the repair material covering the surface of the preset substrate. A salt solution wet-dry cycle accelerated test was carried out. At different cycles of the accelerated test, in-situ X-ray computed tomography was performed on the same interface location of the same composite specimen to obtain three-dimensional structural evolution sequence images of the interface region. The digital volume correlation technique is used to perform correlation calculation on the sequence images, calculate the three-dimensional displacement field and strain tensor of the interface, identify the microcrack initiation location by the preset strain criterion, and invert and calibrate the critical stress and corresponding number of cycles based on the preset interface mechanical model. Using the critical stress and the corresponding number of cycles, combined with the interface opening displacement when the specimen completely fails, the fracture parameters of the preset interface mechanical model are calibrated, a preset interface damage evolution model is established, and the damage degradation rate is extracted from the preset interface damage evolution model. Acquire climate data of the target service environment, convert the damage degradation rate into an equivalent service time scale based on a preset environmental equivalence conversion relationship, generate an interface service life curve from the preset environmental equivalence conversion relationship, and mark the fraction interval and the corresponding interface service life curve as the ratio feasible region.

Images