Straw-dynamic reversible cross-linking agent hybrid system-oriented machine learning optimization collaborative network structure construction method

Abstract

The invention discloses a machine learning optimization collaborative network structure construction method for a straw three-component-dynamic reversible cross-linking agent hybrid system. According to the method, structural characteristic parameters of cellulose, hemicellulose and lignin in straw, characteristics such as reversible bond types, bond energy and response conditions of a dynamic reversible cross-linking agent and technological parameters jointly construct a characteristic database, and the characteristic database is jointly used as input parameters of a hybrid system; a structure-performance prediction model of a hybrid system is established by using a Gaussian process regression model, and multi-objective optimization of material performance, network collaboration and processing energy consumption is realized based on Bayesian optimization. According to the optimized proportion of the three components, the type of the dynamic cross-linking agent, the addition amount and the process conditions, the three components of the straw and the dynamic reversible cross-linking agent are subjected to actual compounding and network construction, and therefore the collaborative network structure material with the optimal performance is obtained. According to the method, the structural design efficiency of the dynamic cross-linking hybrid system can be remarkably improved, experimental trial and error are reduced, the biomass material with high mechanical property, high network synergy and low processing energy consumption is obtained, and the method is suitable for the field of overall high-value utilization of agricultural and forestry wastes such as straw.

Description

Technical Field

[0001] This invention relates to the field of biomass material design and intelligent manufacturing technology, specifically to a method for constructing high-performance materials for a hybrid system of straw three components (lignin, cellulose, and hemicellulose). In particular, this invention relates to a method that integrates dynamic reversible chemistry and machine learning optimization to achieve precise design and performance control of the synergistic network structure of a hybrid system composed of straw three components and a dynamic reversible crosslinking agent. Background Technology

[0002] Straw is a high-yield, highly renewable agricultural biomass resource, whose main structural components are cellulose, hemicellulose, and lignin. Traditional high-value utilization technologies for straw (such as pulping, biorefining, and composite material preparation) mostly rely on separating these three components and then utilizing them separately. However, the separation process suffers from problems such as high energy consumption, large consumption of chemical reagents, and low selectivity, and often leads to the degradation of lignin structure, making it difficult to achieve efficient utilization of all components.

[0003] In recent years, the "three-component integrated utilization" strategy has attracted attention, aiming to avoid separation and directly utilize the original structure of straw. However, this strategy faces three major challenges: (1) poor interfacial compatibility between the three components, which easily leads to phase separation; (2) the complex and heterogeneous structure of lignin makes it difficult to optimize through artificial experience; (3) the system involves multiple variables such as raw material structure, crosslinking agent type, and process parameters, and the traditional trial-and-error method is inefficient and difficult to achieve the comprehensive optimization of performance and cost.

[0004] Machine learning technology offers a novel approach for performance prediction and formulation optimization of complex material systems. However, there are currently no publicly available reports on the application of machine learning to the specific system of "straw three-component hybrid system—dynamic reversible crosslinking—synergistic network structure". Therefore, there is an urgent need for a method that can systematically integrate experiments, dynamic chemistry, and data-driven modeling to achieve the intelligent design and construction of high-performance synergistic networks for all components of straw. Summary of the Invention

[0005] This invention provides a machine learning optimization collaborative network structure construction method for a straw three-component-dynamic reversible crosslinking agent hybrid system. By constructing a feature database containing three-component structural parameters, dynamic reversible crosslinking agent bond characteristics, and process conditions, performance prediction is performed based on Gaussian process regression (GPR), and the optimal network structure parameters of the three-component-dynamic crosslinking agent hybrid system are obtained through Bayesian optimization, thereby achieving comprehensive optimization of material strength, toughness, and energy consumption.

[0006] The technical solution of the present invention includes the following four steps.

[0007] Step 1: Construct a hybrid architecture feature database The structural characteristic parameters of the three components of straw (cellulose crystallinity CrI, crystal form ratio, lignin S / G ratio, β-O-4 bond content, phenolic hydroxyl content, hemicellulose side chain structure, etc.) were entered into the database. Simultaneously, processing parameters (hot pressing temperature, pressure, holding time, swelling time, etc.) were entered into the database. The bond energy, opening and closing conditions, response pH, and temperature of the dynamic reversible crosslinking agent are considered "virtual characteristic parameters" at the algorithm level, used only for machine learning model building. These three types of parameters together constitute a complete characteristic database for the hybrid system.

[0008] Step 2: Construct a performance prediction model for the hybrid system based on GPR Gaussian process regression (GPR) is used to establish a feature-to-performance mapping model.

[0009] Let the input feature vector be:

[0010] Model output performance (Including strength, modulus, synergy index, energy consumption, etc.), satisfying:

[0011] in: Mean function:

[0012] Kernel function (using RBF kernel):

[0013] After training, GPR can output performance predictions and uncertainties (confidence intervals) for optimization.

[0014] Step 3: Solving for parameters of a multi-objective cooperative network based on Bayesian optimization Using GPR as the surrogate model, Bayesian optimization is employed to solve the following objective:

[0015] Bayesian optimization selects the next set of candidate parameters using the following Expected Improvement (EI) criterion:

[0016] The final parameters obtained include the ratio of the three components, the type and dosage of the dynamic crosslinking agent, and the optimal network construction temperature and pressure.

[0017] Step 4: Construct the actual hybrid system material based on the optimization results Based on the optimal scheme in step 3, the three-component straw raw materials and dynamic reversible crosslinking agents were actually compounded and networked. Construction strategies included "crosslinking before hot pressing" and "in-situ crosslinking-shear mixing-hot pressing." The resulting materials underwent systematic performance evaluation, including: the improvement in mechanical properties, the reversibility and stability of the dynamic network (verifiable through oscillatory shear rheology), cyclic reconstruction capability (performance retention rate after ≥3 hot pressing-swelling cycles), processing energy consumption comparison, and the cost-performance ratio (CPI). Finally, an optimized synergistic network structure material was obtained.

[0018] Beneficial effects of the present invention 1. Achieve green and efficient utilization of all components: By directly and specifically binding the three components of straw with a dynamic reversible cross-linking agent, a synergistic network is constructed, completely avoiding the high-energy-consuming and high-polluting separation steps. This achieves the utilization of straw resources in a full-component, low-loss, and high-value manner, which is in line with the principles of circular economy.

[0019] 2. Dynamic interface enhancement and performance improvement: The dynamic crosslinking agent acts as a "smart bridge" and drives the migration and rearrangement of the three components during processing through reversible fracture and recombination. This effectively inhibits phase separation and significantly improves the interfacial bonding strength and network uniformity, thereby synergistically enhancing the material's mechanical strength, toughness and thermal stability.

[0020] 3. Data-driven design greatly improves R&D efficiency: By using machine learning to mine the complex mapping relationship between "structure-process-performance" from multi-dimensional experimental data, the material R&D is transformed from an "experience-based trial and error" mode to a "prediction-driven" mode, which can reduce the amount of experimental work by about 80%, shorten the R&D cycle, and reduce R&D costs.

[0021] 4. Achieving multi-objective comprehensive optimization: Through the multi-objective optimization framework of machine learning, multiple key indicators such as the strength, toughness, energy consumption, and cost of materials can be weighed and optimized simultaneously to obtain the best comprehensive performance formula and process solution, overcoming the limitations of traditional single-objective optimization.

[0022] 5. Imparting repairable and reconfigurable properties to materials: Based on a network constructed by dynamic and reversible chemistry, the network structure can be reorganized and its performance restored by mild stimulation (heat, humidity, etc.) when the material is damaged or needs to be reprocessed. It has excellent recyclability and recycling potential, extends the service life of the material, and reduces waste generation. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating the overall technical approach of the present invention.

[0024] Figure 2 This is a schematic diagram of the synergistic network formed by the three components of straw and the dynamic reversible crosslinking agent.

[0025] Figure 3 A schematic diagram of a machine learning-driven collaborative network optimization design.

[0026] Figure 4 This is a schematic diagram illustrating the performance changes during the material cyclic reconstruction process. Detailed Implementation

[0027] Example 1: Construction of a Hybrid System Feature Database Three types of straw raw materials from different sources (rice straw, corn straw, and wheat straw) were selected and preliminarily processed, including impurity removal, drying at 40–60℃ to a moisture content of ≤10 wt%, and crushing to 20–60 mesh.

[0028] The contents of the three components were determined by NREL method, and the following structural parameters were determined by FTIR, 2D-HSQC, solid-state NMR and XRD: cellulose crystallinity: 42–58%; cellulose Iα / Iβ ratio: 0.32–0.45; lignin S / G ratio: 0.7–1.4; lignin β-O-4 bond retention: 35–55%; phenolic hydroxyl content: 1.8–2.6 mmol / g; hemicellulose side chain ratio (Ara / Xyl): 0.12–0.28; The "characteristic parameters" of the dynamic reversible crosslinking agent are entered into the database according to the algorithm requirements, including: borate ester bond: bond energy 27-32 kJ / mol, pH response range 7-10; imine (Schiff base) bond: bond energy 50-60 kJ / mol, reversible at pH 4-6; disulfide bond: reversibly broken at reducing agent concentration of 1-5 mM; Diels-Alder bond: broken at 110-140℃; metal-polyphenol coordination: coordination constant 10²–10. 4 ; In addition, processing variables are included as a third type of input: hot pressing temperature: 120-190℃, pressure: 4-12 MPa, holding time: 5-20 min, initial ratio: lignin 18-32%, cellulose 40-55%, hemicellulose 15-32%, and dynamic crosslinking agent addition (actually used in the network formation stage): 2-12 wt%. The three types of characteristic variables mentioned above together constructed the initial database of the hybrid system, which included a total of 52 sets of sample data.

[0029] Example 2: Performance Prediction and Bayesian Optimization Based on GPR Using the database from Example 1 as input, a Gaussian process regression model (GPR) is established.

[0030] Let the input feature vector be:

[0031] The kernel function used in the model is the RBF kernel.

[0032] After training, it is used to predict the following properties: tensile strength (MPa), tensile modulus, toughness (kJ·m⁻²), cooperative network index (interfacial binding degree + dynamic bond density), and processing energy consumption (kWh / kg).

[0033] The model prediction accuracy (with 10% cross-validation allowance) reached: Strength: R² = 0.91; Toughness: R² = 0.88; Energy consumption: R² = 0.86.

[0034] The Bayesian optimization objective function is defined as follows:

[0035]

[0036] Among them, weight It can be allocated according to needs (example: 0.35 / 0.35 / 0.2 / 0.1).

[0037] EI (Expected Improvement) was used as the sampling criterion:

[0038] After 40 iterations, the optimized solution was output: the optimal three-group ratio: lignin 24%, cellulose 51%, hemicellulose 25%; the optimal dynamic crosslinking agent type: borate ester + metal-polyphenol composite system; the addition amount: 6 wt%; the hot pressing temperature: 162℃; the pressure: 9 MPa; the holding time: 12 min.

[0039] Example 3: Preparation of materials based on optimization results The target material was prepared according to the optimization results of Example 2. The three components of crushed straw were mixed evenly with 24% lignin, 51% cellulose, and 25% hemicellulose. An ethanol / water system (1:1) containing borate ester crosslinking agent was prepared with a concentration of 6wt%. A solution containing Fe³⁺–polyphenol coordination agent was prepared so that the final Fe³⁺:polyphenol molar ratio was 1:3. The straw powder swelled in the above crosslinking agent mixed solution for 1.5 h. The mixture was filtered and lightly dried to a moisture content of 15–20%. The mixture was then hot-pressed at 162°C, 9 MPa, for 12 min.

[0040] The properties of the material are as follows:

Claims

1. A method for constructing a machine learning-optimized collaborative network structure for a straw three-component-dynamic reversible crosslinking agent hybrid system, characterized in that, The method uses a hybrid system composed of three components of straw and a dynamically reversible crosslinking agent as the object of machine learning optimization, and includes the following steps: S1: Construct a feature database for the hybrid system. The structural feature parameters of cellulose, hemicellulose and lignin in straw, along with the bond type, bond energy, opening and closing conditions, number of reaction sites and related process parameters of the dynamic reversible crosslinking agent, are entered into the database and used as model input. S2: A structure-performance prediction model for the three-component dynamic crosslinking agent hybrid system is established using Gaussian process regression (GPR). The model output includes mechanical properties, network synergy index, and processing energy consumption. S3: Based on the Bayesian optimization method, the structure-performance prediction model is optimized in multiple objectives to obtain the optimal combination of structural parameters of the hybrid system. The optimization objectives include at least: maximizing material properties, maximizing network synergy, and minimizing processing energy consumption. S4: Based on the optimal combination of structural parameters obtained in S3, select the corresponding three-part distribution ratio, dynamic reversible crosslinking agent type and addition amount, and process conditions to actually compound and construct the straw three components and dynamic reversible crosslinking agent, thereby obtaining the optimized synergistic network structure material.

2. The method according to claim 1, characterized in that, The dynamic reversible crosslinking agent includes at least one of the following: borate ester, imine bond, disulfide bond, Diels-Alder bond, or metal-polyphenol coordination system. When its characteristic parameters are entered into the database, they include bond energy, opening and closing conditions, response temperature, response pH, or coordination constant.

3. The method according to claim 1, characterized in that, The three structural characteristic parameters described in S1 include at least one of the following: cellulose crystallinity, cellulose crystal form ratio, lignin S / G ratio, β-O-4 bond retention, phenolic hydroxyl content, hemicellulose side chain structure, or inorganic ash content.

4. The method according to claim 1, characterized in that, The S4 web forming method is selected from hot pressing, melt mixing, extrusion, swelling-drying reconstruction or a combination thereof.

Images