Machine learning based corrosion resistant conductive composite formulation optimization method and system

Abstract

The application provides a corrosion-resistant conductive composite material formula optimization method and system based on machine learning, and relates to the technical field of composite material design and intelligent manufacturing. The method comprises the following steps: collecting and preprocessing the material formula, preparation process and performance index of the corrosion-resistant conductive composite material, encoding the material formula and preparation process into a structured feature vector according to a material component and process hierarchical coding system, calculating a performance label according to the performance index, and then establishing a performance prediction model; generating a plurality of groups of candidate combinations by using a feasible region-based random sampling method, and predicting each group of candidate combinations by using the performance prediction model; obtaining an optimal solution set by using a non-dominated sorting-based multi-objective evolutionary algorithm; and filtering out a formula and process combination from the optimal solution set according to the performance requirements of the target application scene, and decoding to obtain an optimized material formula and preparation process scheme. The application effectively realizes multi-performance collaborative prediction and efficient screening of candidate formulas.

Description

Technical Field

[0001] This invention relates to the field of composite material design and intelligent manufacturing technology, and in particular to a method and system for optimizing the formulation of corrosion-resistant and conductive composite materials based on machine learning. Background Technology

[0002] Corrosion-resistant conductive composite materials are a class of functional materials that combine electrical conductivity and corrosion resistance. They are widely used in grounding systems, power equipment, petrochemicals, marine engineering, and aerospace—fields where high weather resistance and electrical stability are required. Currently, the development of these materials mainly relies on experience-driven and trial-and-error methods. Researchers typically select polymer matrices, conductive fillers, and functional additives based on existing experience, repeatedly prepare samples by adjusting component content and process parameters, and test their electrical conductivity, mechanical properties, and corrosion resistance to ultimately select formulations that meet application requirements.

[0003] However, traditional R&D models have significant limitations. On the one hand, the performance of corrosion-resistant conductive composite materials is affected by the coupling effects of multiple factors, such as matrix type, type and content of conductive filler, functional additive system, mixing temperature, molding pressure, and heat treatment conditions. The parameter space is huge, and it is difficult to quickly converge to the optimal formulation region by relying on trial and error. On the other hand, in practical applications, materials usually need to meet multiple indicators such as conductivity, mechanical properties, and corrosion resistance at the same time. There are often mutual constraints between different properties. For example, increasing the content of conductive filler is beneficial to enhance conductivity, but it may lead to a decrease in mechanical properties. Traditional methods are difficult to achieve an effective balance and synergistic optimization among multiple objectives.

[0004] In recent years, materials genome engineering and machine learning technologies have received widespread attention in the field of materials research and development. By collecting and structuring existing literature, patents, and experimental data, and using machine learning models to establish the mapping relationship between material composition, process parameters, and performance, rapid prediction of material properties can be achieved to a certain extent. However, existing methods still generally suffer from problems such as inconsistent data representation and classification coding, insufficient comprehensive performance synergistic optimization capabilities, and low efficiency in candidate formulation generation and global optimization under high-dimensional constraints. Therefore, there is an urgent need for an intelligent formulation design method for corrosion-resistant conductive composite materials. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention constructs a standardized data representation and classification coding system based on the data correlation between material composition, preparation process, and material properties. It then utilizes machine learning to establish a performance surrogate prediction model, and combines this with a genetic algorithm to perform multi-objective iterative optimization under constraints of component content and process parameters. This results in a machine learning-based method and system for optimizing the formulation of corrosion-resistant conductive composite materials, particularly for the intelligent screening and optimization of polymer-based corrosion-resistant conductive composite material formulations. The aim is to solve problems in existing technologies, such as reliance on experience-based material selection and trial-and-error methods in the development of corrosion-resistant conductive composite materials, inconsistent data representation, difficulty in effectively modeling the relationship between formulation, process, and performance, difficulty in coordinating the optimization of multiple performance objectives, and difficulty in directly converting optimization results into practically feasible formulation schemes. This invention achieves multi-performance co-prediction and efficient screening of candidate formulations.

[0006] On the one hand, this invention proposes a method for optimizing the formulation of corrosion-resistant conductive composite materials based on machine learning, which includes the following process:

[0007] Several sets of raw sample data of corrosion-resistant conductive composite materials were collected. Each set of raw sample data includes material formulation, preparation process and performance indicators.

[0008] Each set of original sample data is preprocessed, and the preprocessed material formulation and preparation process are encoded into structured feature vectors according to the material composition and process hierarchical coding system. Performance labels are calculated based on the preprocessed performance indicators.

[0009] Using structured feature vectors as input and performance labels as output, a machine learning-based performance prediction model for corrosion-resistant conductive composite materials is established.

[0010] Based on the preset formulation constraints, a random sampling method based on the feasible region is used to generate several sets of candidate formulation and process combinations.

[0011] A machine learning-based performance prediction model for corrosion-resistant and conductive composite materials was used to predict each candidate combination of formulation and process. A multi-objective evolutionary algorithm based on non-dominated sorting was then used to screen the Pareto optimal solution set.

[0012] Based on the performance requirements of the target application scenario, one or more formulation and process combinations are selected from the Pareto optimal solution set. Then, according to the material composition and process hierarchical coding system, the selected formulation and process combinations are decoded to obtain the optimized material formulation and preparation process scheme.

[0013] Furthermore, the material formulation includes, but is not limited to: polymer matrix components, conductive filler components, and functional additive components, as well as the content information of each component;

[0014] The preparation process includes, but is not limited to: process type and process parameters;

[0015] The performance indicators include, but are not limited to: electrical conductivity data, mechanical properties data, and corrosion resistance data.

[0016] Furthermore, the specific methods of preprocessing include: standardization and data cleaning;

[0017] For any original sample data, the polymer matrix components, conductive filler components and functional additive components in the material formulation are respectively merged into standardized coding categories, and the content information of each component in the material formulation is uniformly converted into mass fraction to form a standardized material formulation.

[0018] Based on the preset process terminology dictionary and keyword matching rules, the preparation process information is categorized into standardized process flows. For preparation process information that cannot be directly matched, the standardized process flow to which the preparation process information belongs is determined through manual matching.

[0019] Data cleaning is performed on standardized material formulations, standardized process flows, and performance indicators to obtain pre-processed material formulations, preparation processes, and performance indicators.

[0020] Furthermore, the specific method for performing data cleaning on standardized material formulations, standardized process flows, and performance indicators to obtain pre-processed material formulations, preparation processes, and performance indicators is as follows:

[0021] Standardized material formulations, standardized process flows, and performance indicators are broken down into several independent fields, and missing value checks are performed on each field:

[0022] If no missing fields are found, the standardized material formulation, standardized process flow, and performance indicators for that group will be retained.

[0023] If any fields are missing, they will be filled in by backtracking and verification, and the completed standardized material formulas, standardized process flows and performance indicators will be retained; for missing fields that cannot be filled in, the set of standardized material formulas, standardized process flows and performance indicators will be deleted.

[0024] Based on preset numerical ranges and preset logical consistency rules, anomalies are identified in the retained standardized material formulas, standardized process flows, and performance indicators. The identified abnormal fields are then backtracked and corrected to obtain the corrected standardized material formulas, standardized process flows, and performance indicators. If correction is not possible, the set of standardized material formulas, standardized process flows, and performance indicators is deleted.

[0025] The revised standardized material formulation and standardized process flow are used as the pretreated material formulation and preparation process, and the revised performance indicators are used as the pretreated performance indicators.

[0026] Furthermore, the specific method for encoding the pretreated material formulation and preparation process into a structured feature vector based on the material composition and process hierarchical coding system, and calculating the performance label based on the pretreated performance indicators, is as follows:

[0027] The material composition and process layer coding system includes: a matrix material coding module, a conductive filler coding module, a functional additive coding module, and a process flow coding module;

[0028] Using the matrix material coding module, the polymer matrix components in the pretreated material formulation are binary indicated and coded to obtain the matrix material coding vector;

[0029] Using a conductive filler coding module, the conductive filler components in the pretreated material formulation are binary indicated and coded to obtain a conductive filler coding vector;

[0030] The functional additive coding module is used to perform binary indicator coding on the functional additive components in the pretreated material formulation to obtain the functional additive coding vector;

[0031] Using the process flow coding module, the pre-treated preparation process is binary indicated and coded to obtain the process flow coding vector;

[0032] The matrix material encoding vector, conductive filler encoding vector, functional additive encoding vector, and process flow encoding vector together form a structured feature vector.

[0033] The conductivity, mechanical properties, and corrosion resistance data from the pre-processed performance indicators are weighted and summed to obtain the conductivity label. Mechanical property labels and corrosion resistance label .

[0034] Furthermore, the machine learning-based performance prediction model for corrosion-resistant conductive composite materials employs a machine learning regression model to model the mapping relationship between structured feature vectors and performance labels.

[0035] Furthermore, the formulation constraints include: the content range of each component in the polymer matrix component, conductive filler component, and functional additive component; the range of process parameters for the preparation process; and the total amount of components; wherein each of the polymer matrix component, conductive filler component, and functional additive component includes at least one component.

[0036] Furthermore, the specific method for using a machine learning-based performance prediction model for corrosion-resistant conductive composite materials to predict each candidate combination of formulation and process, and then using a multi-objective evolutionary algorithm based on non-dominated sorting to screen and obtain the Pareto optimal solution set, is as follows:

[0037] Each candidate combination of formulation and process is treated as a candidate solution;

[0038] Using a machine learning-based performance prediction model for corrosion-resistant and conductive composite materials, the performance of each candidate solution is predicted, and the predicted conductivity label, predicted mechanical property label, and predicted corrosion resistance label of each candidate solution are obtained.

[0039] The predicted conductivity, mechanical properties, and corrosion resistance labels of each candidate solution are normalized to obtain the normalized conductivity, mechanical properties, and corrosion resistance of each candidate solution, and a multi-objective optimization function is constructed.

[0040] The multi-objective optimization function is: taking the normalized electrical conductivity, mechanical properties and corrosion resistance as objectives, and maximizing the three objectives simultaneously;

[0041] Set population size Maximum number of generations The range of crossover probability values, the range of mutation probability values, and the stagnation threshold; wherein the population size is... Same as the number of candidate solutions;

[0042] Treat each candidate solution as an individual, obtain the initial population, and begin evolution. In each round of evolution, perform the following process:

[0043] Based on the dominance conditions, all individuals in the current population are sorted into non-dominated categories. Based on the dominance relationships between individuals, all individuals are divided into multiple Pareto levels. Individuals in the first level are taken as non-dominated solutions and constitute the Pareto front of the current generation.

[0044] Calculate the crowding distance for each individual within the same non-dominant class;

[0045] A tournament selection strategy based on non-dominance rank and crowding distance is adopted to select from all individuals in the current population. There are 1 parent generation individual; among them, individuals with lower non-dominance levels are preferred as parent generation individuals, and when the non-dominance levels are the same, individuals with larger crowding distances are preferred as parent generation individuals.

[0046] Based on the aforementioned crossover probability and mutation probability ranges, the crossover and mutation probabilities in the current evolutionary process are updated. Then, the updated crossover and mutation probabilities are used to perform crossover and mutation on the parent individuals to generate... Individual offspring;

[0047] For the generated Each offspring individual undergoes constraint repair to obtain an offspring population, which is then merged with the current population to form a joint population.

[0048] The joint population is ordered by non-dominance, and the crowding distance of each individual within the same non-dominance level is calculated. The top individuals are selected according to the criterion of priority based on non-dominance level and secondarily on crowding distance. Each individual serves as the next generation of the population;

[0049] Repeat the above steps until the maximum number of generations is reached. The iteration terminates when the improvement in the Pareto front is less than the stagnation threshold for several consecutive generations, and the Pareto optimal solution set is output. .

[0050] On the other hand, this invention proposes a machine learning-based formulation optimization system for corrosion-resistant conductive composite materials, the system comprising:

[0051] The sample data acquisition module is used to collect raw sample data of several sets of corrosion-resistant conductive composite materials. Each set of raw sample data includes material formulation, preparation process and performance indicators.

[0052] The preprocessing and coding module is used to preprocess each set of original sample data, and encode the preprocessed material formulation and preparation process into structured feature vectors according to the material composition and process hierarchical coding system, and calculate performance labels based on the preprocessed performance indicators;

[0053] The prediction model building module is used to build a machine learning-based performance prediction model for corrosion-resistant conductive composite materials, taking structured feature vectors as input and performance labels as output.

[0054] The candidate combination generation module is used to generate several sets of candidate formulation and process combinations based on a random sampling method based on the feasible region, according to preset formulation constraints.

[0055] The multi-objective optimization module is used to predict the performance of corrosion-resistant conductive composite materials using a machine learning-based model, and to select the Pareto optimal solution set by using a multi-objective evolutionary algorithm based on non-dominated sorting.

[0056] The decoding output module is used to select one or more formulation and process combinations from the Pareto optimal solution set according to the performance requirements of the target application scenario, and decode the selected formulation and process combinations according to the material composition and process hierarchical coding system to obtain the optimized material formulation and preparation process scheme.

[0057] The beneficial effects of adopting the above technical solution are as follows:

[0058] Compared with existing technologies, the method of this invention collects formulation, process, and performance data from literature, patents, and experimental records. After cleaning, standardization, and manual verification, a standardized classification and coding system is established, covering matrix materials, conductive fillers, functional additives, and process flows. This system can transform multi-source heterogeneous data into structured feature vectors. Furthermore, under constraints of component content, material type, and process parameters, a "formulation-process-performance" training dataset is constructed, with conductivity label C, mechanical property label L, and corrosion resistance label R as outputs. Based on this, a machine learning-based surrogate prediction model is established to achieve rapid and accurate prediction of conductivity, mechanical properties, and corrosion resistance.

[0059] This invention further incorporates a genetic algorithm to perform global iterative optimization driven by a multi-performance comprehensive fitness function. This mechanism enables efficient synergistic optimization of multiple performance objectives (conductivity, mechanical properties, and corrosion resistance) and rapid screening of candidate formulations, while satisfying component content constraints and process parameter constraints. Finally, the optimization results are decoded in reverse to obtain the actual material system, the content of each component, and the preparation process parameters.

[0060] In summary, the method of this invention effectively solves the problems of long R&D cycle, difficulty in finding the optimal parameters in high-dimensional space, and difficulty in taking multiple performance indicators into account in the traditional trial-and-error method. It can achieve synergistic optimization of electrical conductivity, mechanical properties and corrosion resistance, and improve the efficiency of formulation development and the engineering applicability of optimization results. Attached Figure Description

[0061] Figure 1 This is a flowchart of the machine learning-based formulation optimization method for corrosion-resistant conductive composite materials in an embodiment of the present invention;

[0062] Figure 2 The following are comparison charts of the prediction results and actual results of the proxy prediction model for different performance labels in the embodiments of the present invention; wherein, (a) is a comparison chart of the prediction results and actual results of the proxy prediction model for the conductivity performance label C; (b) is a comparison chart of the prediction results and actual results of the proxy prediction model for the mechanical performance label L; and (c) is a comparison chart of the prediction results and actual results of the proxy prediction model for the corrosion resistance performance label R.

[0063] Figure 3The diagrams are Pareto front evolution diagrams during the multi-objective evolutionary optimization process in this embodiment of the invention; wherein, (a) is the Pareto front evolution diagram in the CL objective space; (b) is the Pareto front evolution diagram in the CR objective space; (c) is the Pareto front evolution diagram in the LR objective space; and (d) is the Pareto front evolution diagram for the three objectives C, L, and R.

[0064] Figure 4 This is a structural diagram of the machine learning-based formulation optimization system for corrosion-resistant conductive composite materials in an embodiment of the present invention. Detailed Implementation

[0065] To facilitate understanding of this application, specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and embodiments. The following embodiments are illustrative of the invention but are not intended to limit its scope. Rather, these embodiments are provided to provide a more thorough and complete understanding of the disclosure of this application.

[0066] Example 1:

[0067] This embodiment presents a machine learning-based method for optimizing the formulation of corrosion-resistant conductive composite materials, such as... Figure 1 As shown, the method includes the following steps:

[0068] Several sets of raw sample data of corrosion-resistant conductive composite materials were collected. Each set of raw sample data includes material formulation, preparation process and performance indicators.

[0069] The material formulation includes, but is not limited to: polymer matrix components, conductive filler components, and functional additive components, as well as the content information of each component.

[0070] The preparation process includes, but is not limited to, process type and process parameters.

[0071] The performance indicators include, but are not limited to: electrical conductivity data, mechanical properties data, and corrosion resistance data.

[0072] In this embodiment, material formulations, preparation processes, and performance indicators of known corrosion-resistant conductive composite materials were collected from various sources, including literature, patents, product information, and experimental records, as raw sample data. The conductivity data collected in this embodiment is expressed as conductivity (S / m), and the mechanical property data includes tensile strength and elongation at break; the corrosion resistance data includes corrosion rate (mm / a), polarization resistance (Ω·cm²), charge transfer resistance (Ω·cm²), and pitting density (pits / cm²).

[0073] It should be noted that when collecting raw sample data, data sources, test conditions, and auxiliary descriptive information of known corrosion-resistant conductive composite materials can also be collected simultaneously according to actual needs. This information is used for data traceability and auxiliary explanation, but does not participate in subsequent standardized modeling and analysis. Therefore, in this embodiment, each piece of raw sample data includes at least: data source information, polymer matrix material type, conductive filler type, functional additive type, content of each component, preparation process type, preparation process parameters, conductivity performance index, mechanical performance index, corrosion resistance performance index, test conditions, and auxiliary descriptive information.

[0074] Each set of original sample data is preprocessed, and the preprocessed material formulation and preparation process are encoded into structured feature vectors according to the material composition and process hierarchical coding system. Performance labels are then calculated based on the preprocessed performance indicators.

[0075] In this embodiment, by standardizing, cleaning, and manually verifying the original sample set, the material names, component expressions, content units, process descriptions, and performance index units are unified, resulting in standardized structured feature vectors and performance labels.

[0076] The preprocessing includes standardization and data cleaning.

[0077] For any original sample data, the polymer matrix components, conductive filler components, and functional additive components in the material formulation are respectively grouped into standardized coding categories, and the content information of each component in the material formulation is uniformly converted into mass fraction to form a standardized material formulation.

[0078] In standardized material formulations, the first The polymer matrix component is denoted as Polymer matrix components The content information is recorded as ; will the first The conductive filler component is denoted as conductive filler components The content information is recorded as ; will the first The functional additive components are denoted as Functional additive components The content information is recorded as .

[0079] In this embodiment, the material and process information in the original samples is merged and mapped according to a preset classification system. Specifically, the names of similar polymer matrix materials, similar conductive fillers, and similar functional additives from samples from different sources are uniformly merged. In particular, this embodiment categorizes polymer matrices into thermoplastic polymers. thermoplastic elastomers Rubber matrix and thermosetting resins The conductive filler components were grouped into carbon-based carbon black. Carbon-based graphite Carbon-based – Nano-carbon and conductive polymers Functional additive components are grouped into flame retardants. , reinforced fiber Dispersants / coupling agents and antioxidants / stabilizers .

[0080] In this embodiment, the component content in the standardized sample set is expressed in three ways: mass fraction, parts by mass, and volume fraction. These three methods need to be uniformly converted to mass fraction (wt%). For components expressed as mass fraction, their numerical value is directly used as the standardized content; for components expressed as parts by mass (phr), let the first... The mass fractions of each component are Then its converted mass fraction for:

[0081] (1);

[0082] in, This represents the total number of components in the material formulation. Indicates the first The mass fractions of each component.

[0083] For components expressed as volume fractions, let the first... The volume fraction of each component is The density is Then its converted mass fraction for:

[0084] (2);

[0085] in, Indicates the first Volume fraction of each component; Indicates the first The density of each component.

[0086] When the material formula is the first Each component by mass and volume When using mixed descriptions, you can first determine based on The volume is converted into mass, and then converted into mass fraction according to formula (1).

[0087] The above conversion achieves a unified expression of component content in samples from different sources, ensuring consistency between subsequent feature construction and model input.

[0088] Based on the preset process terminology dictionary and keyword matching rules, the preparation process information is grouped into standardized process flows. For preparation process information that cannot be directly matched, the standardized process flow to which the preparation process information belongs is determined through manual matching.

[0089] In this embodiment, for the terms describing the process type in the preparation process information, a unified classification is performed using a combination of a preset process terminology dictionary and rule matching. The terminology used in this embodiment is... The standardized process flow is denoted as This embodiment categorizes the standardized process flow into melt blending. Solution blending Hot pressing – carbonization Specifically, "melt blending," "internal mixing," "open milling," or "twin-screw extrusion" are categorized under melt blending. Solvent dispersion-evaporation, emulsion blending, or in-situ polymerization are grouped under solution blending. "Hot pressing", "high temperature carbonization" or "graphitization treatment" are grouped together as hot pressing-carbonization. For process descriptions whose categories cannot be uniquely determined through a preset process terminology dictionary or keyword matching rules, their corresponding standardized process flow shall be determined by manual review.

[0090] Data cleaning is performed on standardized material formulations, standardized process flows, and performance indicators to obtain pre-processed material formulations, preparation processes, and performance indicators.

[0091] The specific method for data cleaning of standardized material formulations, standardized process flows, and performance indicators to obtain pre-processed material formulations, preparation processes, and performance indicators is as follows:

[0092] Standardized material formulations, standardized process flows, and performance indicators are broken down into several independent fields, and missing value checks are performed on each field:

[0093] If no missing fields are found, the standardized material formulation, standardized process flow, and performance indicators for that group will be retained.

[0094] If any fields are missing, they will be filled in by backtracking and verification, and the completed standardized material formulas, standardized process flows, and performance indicators will be retained. For missing fields that cannot be filled in, the set of standardized material formulas, standardized process flows, and performance indicators will be deleted.

[0095] In this embodiment, standardized material formulations, standardized process flows, and performance indicators are designated as core fields, while data sources, testing conditions, and auxiliary descriptive information are designated as non-core fields. A rule-based data cleaning method is used to process missing fields. When a core field is missing, the original literature, product manuals, or experimental records are manually reviewed and supplemented. If the core field can be supplemented after review, the supplemented sample data is retained and proceeds to the subsequent standardization process. Sample data for which the core field cannot be supplemented after review is discarded. When a non-core field is missing, the sample data is retained, and the corresponding field is marked as null.

[0096] Based on the preset numerical range and preset logical consistency rules, anomalies are identified in the retained standardized material formulas, standardized process flows and performance indicators. The identified abnormal fields are then backtracked and corrected to obtain the corrected standardized material formulas, standardized process flows and performance indicators. If correction is not possible, the set of standardized material formulas, standardized process flows and performance indicators are deleted.

[0097] In this embodiment, data anomaly identification and screening are performed by combining preset numerical ranges, logical consistency constraints, and statistical distribution of similar samples. The identification criteria for the abnormal fields include the conventional physical property range of materials, the reporting interval of publicly available literature, product technical specifications, and statistical results of similar samples. Specifically, the mass fraction of each component in a standardized material formulation should be greater than or equal to 0 and less than or equal to 100 wt%, and the total content of all components should meet the sum constraint within a preset error range. Considering that the recording accuracy and expression methods of data from different sources may differ, this embodiment sets the correctable numerical range to be no more than 2 wt% of the absolute value of the total component content deviating from 100 wt% based on the conventional data accuracy: when the deviation value is within the correctable numerical range, it is determined to be a correctable deviation, and the mass fraction of each component is automatically normalized and corrected; when the deviation value exceeds the correctable numerical range, it is determined to be abnormal data, and the field containing the data is marked as an abnormal field.

[0098] For standardized process flows, the process parameters included should meet the reasonable value range of the corresponding process. For example, for melt blending... The mixing temperature is 120℃-180℃, and the mixing time is 10min-30min; for solution blending, the mixing temperature is 20℃-80℃, the mixing time is 30min-240min, and the drying temperature is 40℃-120℃; for hot pressing-carbonization... The hot-pressing temperature is 120℃-220℃, the hot-pressing pressure is 5MPa-30MPa, the holding time is 5min-60min, the carbonization temperature is 400℃-1200℃, and the carbonization time is 30min-240min. In this embodiment, the correctable numerical range for the process parameters is set as follows: the deviation from the upper and lower limits of the corresponding reasonable value range shall not exceed 10%. When the deviation is within the correctable numerical range, the field is marked as data to be verified and corrected through original literature, product manuals, or experimental records. When the deviation exceeds the correctable numerical range, it is determined to be significantly outside the reasonable range, and the field containing this data is marked as an abnormal field.

[0099] The logical consistency constraints mentioned are: inconsistencies between the total amount of formulation components and the total amount constraint; mismatch between process parameters and process categories; mismatch between performance index units and test conditions; and inconsistent values ​​for the same field in the text, tables, and figures for the same sample. Inconsistent fields are corrected through unit standardization, naming standardization, and cross-checking of information from the same source. Specifically, temperature is standardized to °C, time to min, pressure to MPa, and content to mass fraction; different descriptions of the same material or process are standardized in naming; and when there are conflicts in the text description, table data, and figure descriptions, the information that appears repeatedly and is consistent in the original data source shall prevail.

[0100] If the identified abnormal fields can be verified through original documents, product manuals, or experimental records as unit conversion errors, entry errors, or format errors, they are determined to be data that can be retrospectively corrected. For data that can be retrospectively corrected, the consistency information in the original document text, tables, figure descriptions, product manuals, and experimental records should be checked first, and the correction should be completed through unit unification, format correction, or numerical backfilling. Data that cannot be verified or has substantial logical conflicts should be deleted.

[0101] Data cleaning is performed on standardized material formulations, standardized process flows, and performance indicators to obtain pre-processed material formulations, preparation processes, and performance indicators.

[0102] In this embodiment, the standardized sample dataset obtained after the above preprocessing can improve the consistency, comparability and reusability of multi-source heterogeneous data, and provide a standardized data foundation for subsequent modeling and optimization.

[0103] The specific method for encoding the pretreated material formulation and preparation process into a structured feature vector based on the material composition and process hierarchical coding system, and calculating the performance label based on the pretreated performance indicators, is as follows:

[0104] The material composition and process layer coding system includes: a matrix material coding module, a conductive filler coding module, a functional additive coding module, and a process flow coding module.

[0105] In this embodiment, by establishing a unified classification and coding system, the original text-based formulation and process information can be converted into structured input suitable for machine learning processing. In this system, the matrix material coding module is used to characterize different polymer matrix material categories; the conductive filler coding module is used to characterize different conductive filler categories; the functional additive coding module is used to characterize different functional additive categories; and the process flow coding module is used to characterize different preparation process flow categories.

[0106] Using the matrix material coding module, the polymer matrix components in the pretreated material formulation are binary indicated and coded to obtain the matrix material coding vector.

[0107] Using a conductive filler coding module, the conductive filler components in the pretreated material formulation are binary indicated and coded to obtain a conductive filler coding vector.

[0108] Using a functional additive coding module, the functional additive components in the pretreated material formulation are binary indicated and coded to obtain the functional additive coding vector.

[0109] Using the process flow coding module, the pre-treated preparation process is binary indicated and coded to obtain the process flow coding vector.

[0110] The matrix material encoding vector, conductive filler encoding vector, functional additive encoding vector, and process flow encoding vector constitute a structured feature vector.

[0111] In this embodiment, a matrix material coding module is used. Conductive filler coding module Functional Additive Coding Module and process flow coding module Each input categorical variable is structured using binary indicator coding, and the resulting coding is vectorized. Specifically, each coding module is arranged sequentially according to a preset category order. The matrix material coding module, for example, handles thermoplastic polymers... thermoplastic elastomers Rubber matrix and thermosetting resins Binary indication encoding is performed; the conductive filler encoding module encodes carbon-based carbon black. Carbon-based graphite Carbon-based – Nano-carbon and conductive polymers Binary indication coding is performed; the functional additive coding module encodes flame retardants. , reinforced fiber Dispersants / coupling agents and antioxidants / stabilizers Binary indicator coding is performed; the process flow coding module performs melt blending. Solution blending Hot pressing – carbonization Binary indicator encoding is performed. In each encoding module, the position corresponding to the data category is assigned a value of 1, and the remaining positions are assigned a value of 0. Then, the encoding results are sequentially concatenated according to the matrix material encoding module, conductive filler encoding module, functional additive encoding module, and process flow encoding module to form a unified structured feature vector, thereby realizing the structured expression of the original formulation information and process information. After encoding processing, each group of sample data forms a corresponding structured feature vector, which in this embodiment is a 15-dimensional structured feature vector, used to characterize the formulation composition and preparation process information of the corrosion-resistant conductive composite material.

[0112] The conductivity, mechanical properties, and corrosion resistance data from the pre-processed performance indicators are weighted and summed to obtain the conductivity label. Mechanical property labels and corrosion resistance label .

[0113] In this embodiment, the collected conductivity data is only the conductivity value; therefore, the pre-processed conductivity value is directly used to represent the conductivity label. For mechanical property labels used to characterize the overall mechanical property level of materials The tensile strength and elongation at break after pretreatment are obtained by weighted fusion after normalization, and are expressed as:

[0114] (3);

[0115] in, This is the dimensionless value of the tensile strength (MPa) after normalization; The normalized dimensionless value of elongation at break (%); and All are weighting coefficients, and satisfy the following conditions: By mapping multidimensional mechanical performance data to a single comprehensive evaluation index, the stability and consistency of the model training and optimization process are improved. In this embodiment, the weighting coefficient is set to [value missing] after calculation using the entropy weighting method and correction based on expert experience in grounding material application scenarios. .

[0116] Corrosion resistance label A multi-index comprehensive evaluation strategy is adopted for corrosion rate. (mm / a), polarization resistance (Ω·cm²) charge transfer resistance (Ω·cm²) and pitting density (number of samples / cm²), firstly, the corrosion performance data of different dimensions are uniformly mapped to an interval through normalization. Then, the entropy weight method combined with expert experience is used to determine the weight of each corrosion performance data. Finally, the corrosion resistance label is obtained through linear weighting. , is represented as:

[0117] (4);

[0118] in, This is the dimensionless value of the corrosion rate after normalization. This is the dimensionless value of the polarization resistance after normalization. This is the dimensionless value of the charge transfer resistance after normalization. This is the dimensionless value of the pitting density after normalization; , , and All are weighting coefficients, and satisfy the following conditions: In this embodiment, the weighting coefficients are set as follows: (Calculated using the entropy weighting method and corrected based on expert experience regarding grounding material application scenarios). In formula (4), the corrosion rate and pitting density All of these are negative indicators and need to be positiveized, i.e., represented as and polarization resistor and charge transfer resistance As a positive indicator, it directly participates in the weighting.

[0119] It should be noted that for some samples where only a few corrosion performance data are available, normalization and weighted calculations are performed based on the actual available corrosion performance data, and the actual obtained corrosion performance data is normalized and redistributed to ensure the consistency and comparability of corrosion resistance performance characterization of samples from different sources.

[0120] Using structured feature vectors as input and performance labels as output, a machine learning-based performance prediction model for corrosion-resistant conductive composite materials is established.

[0121] In this embodiment, the structured feature vectors and performance labels obtained above are used to establish a machine learning-based performance prediction model for corrosion-resistant conductive composite materials, so as to realize the mapping from formulation variables and process variables to conductivity, mechanical properties and corrosion resistance.

[0122] The machine learning-based performance prediction model for corrosion-resistant conductive composite materials employs a machine learning regression model to model the mapping relationship between structured feature vectors and performance labels.

[0123] In this embodiment, the machine learning-based performance prediction model for corrosion-resistant conductive composite materials employs a machine learning regression model to establish formulation and process variables, and correlates them with conductivity performance labels. Mechanical property labels and corrosion resistance label The mapping relationship between them. The machine learning regression model can be any one of the following: gradient boosting decision tree model, random forest model, support vector machine model, extreme gradient boosting model, or neural network model. By establishing a performance prediction model for corrosion-resistant conductive composite materials, it is possible to avoid physical sample preparation and experimental testing for a large number of candidate solutions one by one, thereby improving the efficiency of formula optimization.

[0124] In this embodiment, a gradient boosting decision tree (GBDT) model is used to construct a machine learning-based performance prediction model for corrosion-resistant conductive composite materials. The input of this model is a 15-dimensional structured feature vector, and the output is a conductivity label. Mechanical property labels and corrosion resistance label The model parameters were set as follows: learning rate of 0.1, number of trees of 100, maximum depth of 5, and five-fold cross-validation was used to evaluate the model performance. The results are as follows. Figure 2 As shown, the coefficient of determination for predicting electrical conductivity is displayed. The coefficient of determination for predicting mechanical properties is 0.98. The coefficient of determination for predicting corrosion resistance is 0.95. The value of 0.81 indicates that the model can meet the requirements of high-frequency calls and fast evaluation.

[0125] Based on the preset formulation constraints, a random sampling method based on the feasible region is used to generate several sets of candidate formulation and process combinations.

[0126] The formulation constraints include: the content range of each component in the polymer matrix component, conductive filler component, and functional additive component; the range of process parameters for the preparation process; and the total amount of components; wherein each of the polymer matrix component, conductive filler component, and functional additive component includes at least one component.

[0127] In this embodiment, the formulation constraints are as follows: the polymer matrix component includes a thermoplastic polymer. thermoplastic elastomers Rubber matrix and thermosetting resins At least one component; the conductive filler component includes carbon-based carbon black. Carbon-based graphite Carbon-based – Nano-carbon and conductive polymers At least one component; the functional additive component includes flame retardants. , reinforced fiber Dispersants / coupling agents and antioxidants / stabilizers At least one component in the polymer matrix; the content range of each component in the polymer matrix is ​​as follows: The content range of each component in the conductive filler composition is as follows: The content range of each component in the functional additives is as follows: . by melt blending For example, its process parameter range is set as follows: mixing temperature The range of values ​​is Mixing time The range of values ​​is The sum of the contents of each component is 100 wt%. Under the above constraints, the population size is set to 100, and a random sampling method based on the feasible region is used to generate candidate formulations and process combinations, which can ensure that the candidate schemes have practical feasibility and engineering rationality.

[0128] A machine learning-based performance prediction model for corrosion-resistant and conductive composite materials was used to predict each candidate combination of formulation and process. A multi-objective evolutionary algorithm based on non-dominated sorting was then used to screen the Pareto optimal solution set.

[0129] In this embodiment, a performance prediction model for corrosion-resistant and conductive composite materials based on machine learning is used to predict the performance of the generated candidate formulation and process combinations. A multi-objective optimization model is established with conductive performance labels, mechanical performance labels, and corrosion resistance performance labels as objectives. A multi-objective evolutionary algorithm based on non-dominated sorting is used to iteratively screen the candidate formulation-process combinations to obtain the Pareto optimal solution set.

[0130] The specific method for using a machine learning-based performance prediction model for corrosion-resistant conductive composite materials to predict each candidate combination of formulation and process, and then using a multi-objective evolutionary algorithm based on non-dominated sorting to obtain the Pareto optimal solution set is as follows:

[0131] Each candidate combination of formulation and process is treated as a candidate solution. .

[0132] In this embodiment, each candidate formulation and process combination is treated as a candidate solution and represented in vector form. Its components are the decision variables, including: matrix material category, conductive filler category, functional additive category, process flow category, component content, and process parameters. The category variables are discrete variables, while the content variables and process parameters are continuous variables. The candidate solution is denoted as... ,in, Indicates the dimension of decision variables; They represent the 1st, 2nd, and 3rd respectively. One decision variable.

[0133] Using a machine learning-based performance prediction model for corrosion-resistant and conductive composite materials, for each candidate solution Perform performance prediction to obtain candidate solutions Predicted conductivity label Predicted mechanical properties label and labels predicting corrosion resistance .

[0134] For each candidate solution Predicted conductivity label Predicted mechanical properties label and labels predicting corrosion resistance Normalization is performed to obtain the normalized electrical conductivity, mechanical properties, and corrosion resistance of each candidate solution, and a multi-objective optimization function is constructed.

[0135] In this embodiment, to eliminate the differences in the dimensions and numerical ranges of different performance indicators, the predicted conductivity label is... Predicted mechanical properties label and labels predicting corrosion resistance After normalization, we get:

[0136] (5);

[0137] (6);

[0138] (7);

[0139] in, The normalized conductivity; To predict the minimum value of the conductivity label; To predict the maximum value of the conductivity label; The normalized mechanical properties; To predict the minimum value of the mechanical property label; To predict the maximum value of the mechanical property label; The normalized corrosion resistance; To predict the minimum value for corrosion resistance labels; To predict the maximum value of the corrosion resistance label.

[0140] The multi-objective optimization function is defined as follows: taking the normalized electrical conductivity, mechanical properties, and corrosion resistance as objectives, and maximizing all three objectives simultaneously, expressed as:

[0141] (8);

[0142] in, Represents a multi-objective optimization function; This means that the normalized electrical conductivity, mechanical properties, and corrosion resistance are maximized simultaneously.

[0143] Set population size Maximum number of generations The range of crossover probability values, the range of mutation probability values, and the stagnation threshold; wherein the population size is... The same as the number of candidate solutions.

[0144] Treat each candidate solution as an individual, obtain the initial population, and begin evolution. In each round of evolution, perform the following process:

[0145] Based on the dominance conditions, all individuals in the current population are sorted into non-dominated categories. Based on the dominance relationships between individuals, all individuals are divided into multiple Pareto levels. Individuals in the first level are taken as non-dominated solutions and constitute the Pareto front of the current generation.

[0146] The dominance condition is: for any two individuals and ,definition Dominate The conditions are:

[0147] (9);

[0148] Based on the defined dominance relationships, individuals are non-dominated and ranked, dividing all individuals into multiple Pareto levels. The higher the level, the better the candidate solution corresponding to that individual is in a multi-objective context.

[0149] To maintain the uniform distribution of the solution set in the target space, the crowding distance of each body within the same non-dominated level is calculated, and the th... Crowding distance of individuals Defined as:

[0150] (10);

[0151] in, and Indicates that all individuals in the first... The maximum and minimum values ​​on each target; and They represent according to the number The target values ​​of adjacent individuals after sorting the targets; the larger the crowding distance, the more sparse the candidate solution corresponding to that individual is, and it should be retained first.

[0152] A tournament selection strategy based on non-dominance rank and crowding distance is adopted to select from all individuals in the current population. There are 1 parent individuals; among them, individuals with lower non-dominance levels are preferred as parent individuals, and when the non-dominance levels are the same, individuals with larger crowding distances are preferred as parent individuals.

[0153] To improve the algorithm's early-stage global search capability and later-stage local exploitation capability, adaptive crossover probability and adaptive mutation probability are adopted.

[0154] Based on the aforementioned crossover probability and mutation probability ranges, the crossover and mutation probabilities in the current evolutionary process are updated. Then, the updated crossover and mutation probabilities are used to perform crossover and mutation on the parent individuals to generate... Individual offspring.

[0155] The method for updating the crossover probability and mutation probability in the current evolutionary iteration process is as follows:

[0156] (11);

[0157] (12);

[0158] in, Indicates the current evolutionary process; Indicates the first The crossover probability during generational evolution; This indicates the upper limit of the range of values ​​for the crossover probability; This indicates the lower bound of the range of values ​​for the crossover probability; Indicates the first The probability of mutation during the evolutionary process; This represents the upper limit of the range of possible mutation probabilities. This represents the lower bound of the mutation probability range. As the number of generations increases, the crossover probability gradually decreases while the mutation probability gradually increases, in order to enhance the ability to search for local optimal neighborhoods in later stages.

[0159] For the generated Each offspring individual undergoes constraint repair to obtain an offspring population, which is then merged with the current population to form a joint population.

[0160] In this embodiment, for the offspring individuals generated after crossover and mutation, constraint repair is performed to ensure that they meet the actual constraints of the formulation and process. Specifically, continuous variables in each offspring individual are repaired, as shown below:

[0161] (13);

[0162] in, Represents continuous variables before repair; Represents the continuous variable after repair; and They represent the first The lower and upper bounds of a continuous variable.

[0163] When the sum of the contents of each component in the offspring does not meet the total amount constraint, the component contents are normalized and corrected, as follows:

[0164] (14);

[0165] in, This represents the set of all component content variables; Represents the normalized corrected th One content variable; Indicates the repaired number One content variable; Indicates the repaired number There are several content variables; in this embodiment, after normalization correction, the total content of each component is 100%.

[0166] The joint population is ordered by non-dominance, and the crowding distance of each individual within the same non-dominance level is calculated. The top individuals are selected according to the criterion of priority based on non-dominance level and secondarily on crowding distance. Each individual becomes part of the next generation of the population.

[0167] Repeat the above steps until the maximum number of generations is reached. The iteration terminates when the improvement in the Pareto front is less than the stagnation threshold for several consecutive generations, and the Pareto optimal solution set is output. .

[0168] In this embodiment, the multi-objective evolutionary optimization adopts a multi-objective evolutionary algorithm based on non-dominated sorting, with a population size of... Set to 100, maximum number of generations. The number of generations is set to 100. Parent individuals are generated using a binary tournament selection strategy. The crossover probability uses an adaptive decreasing method, with an initial value of 0.90 and a termination value of 0.60; the mutation probability uses an adaptive increasing method, with an initial value of 0.02 and a termination value of 0.10, to balance the algorithm's global search capability in the early stages with its local exploitation capability in the later stages. The termination condition is reaching the maximum number of generations. In addition, it also includes cases where the improvement in the Pareto front for 10 consecutive generations is less than the stagnation threshold. In this embodiment, the stagnation threshold is set to 1×10. -4 For offspring individuals generated after crossover and mutation, boundary repair is performed according to the preset upper and lower limits of variables, and the total content of each component is normalized to ensure that the repaired candidate solutions meet the constraints of categorical variable values, continuous variable range constraints, and the condition that the total content of each component is 100wt%.

[0169] In this embodiment, as Figure 3 As shown, the Pareto optimal solution set obtained through the multi-objective evolutionary optimization method. In the process, each candidate solution is labeled with conductivity properties. Mechanical property labels and corrosion resistance label The three objectives reach a non-dominated optimal state, and different candidate solutions reflect performance trade-offs.

[0170] Based on the performance requirements of the target application scenario, one or more formulation and process combinations are selected from the Pareto optimal solution set. Then, according to the material composition and process hierarchical coding system, the selected formulation and process combinations are decoded to obtain the optimized material formulation and preparation process scheme.

[0171] In this embodiment, based on the performance requirements of the target application scenario or preset screening conditions, and combined with the distribution characteristics of candidate formulations and process combinations on the three objectives of conductivity, mechanical properties, and corrosion resistance, one or more of the following are selected from the Pareto optimal solution set: a performance-balanced scheme, a high conductivity preferred scheme, and a high mechanical property-corrosion resistance preferred scheme. These are then used for subsequent experimental verification and engineering application screening. Specifically, when the application scenario prioritizes only one type of performance, a representative candidate scheme is output; when the application scenario requires comparative screening of candidate schemes with different performance focuses, two or more representative candidate schemes are output as the final optimized material formulation and preparation process scheme.

[0172] To screen the representative candidate solutions, a machine learning-based performance prediction model for corrosion-resistant conductive composite materials was first used to analyze the Pareto optimal solution set. For each solution, performance prediction is performed, resulting in predicted conductivity, mechanical properties, and corrosion resistance labels. These labels are then normalized to obtain the normalized conductivity, mechanical properties, and corrosion resistance of each solution, denoted as […]. .

[0173] (1) The performance balance scheme refers to the scheme that, under the condition that the conductivity, mechanical properties and corrosion resistance all meet the preset threshold constraints, selects the Pareto optimal solution set. The solution with the most balanced distribution of the three objectives is selected. This selection can be based on the following balance index. :

[0174] (15);

[0175] in, As an intermediate variable, .

[0176] Select the option that satisfies the threshold constraint and makes The minimum solution is used as the performance balancing scheme, thereby ensuring good coordination among the three performance aspects.

[0177] (2) The preferred high conductivity option refers to: in the mechanical property label and corrosion resistance label Under the premise that none of them are lower than the preset lower limit, from the Pareto optimal solution set Select conductivity label The largest solution. When multiple solutions have similar conductivity, the solution with better performance in other objectives (mechanical or corrosion resistance) can be selected as the final output solution.

[0178] (3) High mechanical properties – corrosion resistance preferred option refers to: on the conductivity label Under the premise of meeting basic usage requirements, mechanical performance labels and corrosion resistance label Perform a joint evaluation and extract from the Pareto optimal solution set The solution with the best overall performance is selected. The joint evaluation function can be expressed as:

[0179] (16);

[0180] in, and Let be the weight coefficient, and satisfy... ; This is a joint evaluation function.

[0181] Based on this, from the Pareto optimal solution set Select the joint evaluation function that satisfies the conductivity constraints. The optimal solution is the one with the highest mechanical properties and corrosion resistance.

[0182] From the Pareto optimal solution set After selecting representative candidate schemes, the structured feature vectors corresponding to each representative candidate scheme are reverse mapped according to the material composition and process hierarchical coding system to obtain the corresponding polymer matrix components, conductive filler components and functional auxiliary components, the content information of each component and the preparation process.

[0183] This embodiment will ultimately yield three representative candidate schemes, which are presented in Table 1. The performance prediction results of the three representative candidate schemes are also given in Table 1.

[0184] Table 1 Representative Candidate Solutions

[0185]

[0186] To verify the effectiveness of the machine learning-based formulation optimization method for corrosion-resistant conductive composite materials proposed in this embodiment (hereinafter referred to as the method of this embodiment), the performance balancing schemes in Table 1 were selected for experimental preparation and performance testing.

[0187] Samples were prepared according to the material formulation and process parameters of the performance-balanced scheme: polyethylene was used as the matrix material, with a matrix content of 58.0 wt%; carbon black was used as the main conductive filler, with a content of 38.0 wt%; graphite was used as the secondary conductive filler, with a content of 2.5 wt%; the functional additive system included flame retardant, dispersant, and antioxidant, with a total content of 1.5 wt%, of which flame retardant was 0.7 wt%, dispersant was 0.5 wt%, and antioxidant was 0.3 wt%. After the components were mixed evenly, they were melt-blended at 160℃ for 20 min, and then molded to obtain a composite material sample with dimensions of 100 mm × 100 mm × 2 mm.

[0188] The performance of the prepared composite material sample was tested: the conductivity was measured by the four-probe method, and the result was 6200 S / m; the tensile strength was measured by the universal testing machine, and the result was 85.5 MPa; the elongation at break was 138.6%; the corrosion resistance test was carried out by neutral salt spray test according to GB / T 10125-2021 standard. After 500h of continuous spraying, the composite material sample was subjected to the following three tests: (1) the average corrosion rate was obtained by converting the corrosion weight loss data; (2) the charge transfer resistance was tested in 3.5% NaCl solution by electrochemical impedance spectroscopy (EIS); (3) the pitting density R4 was statistically analyzed by observing the sample surface with a metallographic microscope. According to formula (4), the above three obtainable sub-indicators were normalized and weighted. Since the corrosion rate was obtained by actual measurement in this verification, Charge transfer resistance and pitting density Three indicators, polarization resistance not obtained. Therefore, the original weights were redistributed, and after normalization and weighted calculation, the corrosion resistance label was obtained. .

[0189] Comparing the measured results with the predicted values ​​in the corresponding range in Table 1, it can be seen that the conductivity, tensile strength and corrosion resistance comprehensive score all fall within the predicted range, indicating that the machine learning-based corrosion-resistant conductive composite material performance prediction model established in this embodiment has high prediction accuracy, and the candidate formulations optimized by the genetic algorithm have good engineering feasibility.

[0190] The three representative candidate solutions mentioned above can be applied to engineering scenarios with different requirements for conductivity, mechanical properties, or corrosion resistance. The experimental results further demonstrate the effectiveness and reliability of the method in this embodiment in the intelligent optimization of corrosion-resistant conductive composite material formulations.

[0191] Example 2:

[0192] This embodiment presents a machine learning-based system for optimizing the formulation of corrosion-resistant conductive composite materials, such as... Figure 4 As shown, the system includes:

[0193] The sample data acquisition module is used to collect raw sample data of several sets of corrosion-resistant conductive composite materials. Each set of raw sample data includes material formulation, preparation process and performance indicators.

[0194] The preprocessing and coding module is used to preprocess each set of original sample data, and encode the preprocessed material formulation and preparation process into structured feature vectors according to the material composition and process hierarchical coding system, and calculate performance labels based on the preprocessed performance indicators.

[0195] The prediction model building module is used to build a machine learning-based performance prediction model for corrosion-resistant conductive composite materials, taking structured feature vectors as input and performance labels as output.

[0196] The candidate combination generation module is used to generate several sets of candidate formulation and process combinations based on a random sampling method based on the feasible region, according to preset formulation constraints.

[0197] The multi-objective optimization module is used to predict the performance of corrosion-resistant conductive composite materials using a machine learning-based model, predict each candidate combination of formulation and process, and use a multi-objective evolutionary algorithm based on non-dominated sorting to obtain the Pareto optimal solution set.

[0198] The decoding output module is used to select one or more formulation and process combinations from the Pareto optimal solution set according to the performance requirements of the target application scenario, and decode the selected formulation and process combinations according to the material composition and process hierarchical coding system to obtain the optimized material formulation and preparation process scheme.

[0199] Example 3:

[0200] This embodiment proposes an electronic device, including: one or more processors, and a memory, wherein the memory is used to store instructions, and when the instructions are executed by the one or more processors, the one or more processors execute the machine learning-based corrosion-resistant conductive composite material formulation optimization method.

[0201] The electronic device may be a mobile phone, computer, or tablet computer, etc., and includes a memory and a processor. The memory stores a computer program, which, when executed by the processor, implements the machine learning-based method for optimizing the formulation of corrosion-resistant conductive composite materials as described in the embodiments. It is understood that the electronic device may also include input / output (I / O) interfaces and communication components.

[0202] The processor is used to execute all or part of the steps in the machine learning-based method for optimizing the formulation of corrosion-resistant conductive composite materials as described in the above embodiments. The memory is used to store various types of data, which may include, for example, instructions for any application or method in the electronic device, as well as application-related data.

[0203] The processor can be implemented as an Application Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Programmable Logic Device (PLD), Field Programmable Gate Array (FPGA), controller, microcontroller, microprocessor, or other electronic components, and is used to execute the machine learning-based formulation optimization method for corrosion-resistant conductive composite materials described in the above embodiments.

[0204] Example 4:

[0205] This embodiment proposes a computer-readable storage medium that stores executable instructions. When these instructions are executed, if they are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium.

[0206] The computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the machine learning-based corrosion-resistant conductive composite material formulation optimization method described in the various embodiments of this application.

[0207] The aforementioned storage media include: flash memory, hard disks, multimedia cards, card-type memory (e.g., SD (Secure Digital Memory Card) or DX (Memory Data Register, MDR) memory), random access memory (RAM), static random-access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic storage, disks, optical discs, servers, APP (Application) application stores, and other media capable of storing program verification codes. These media store computer programs, which, when executed by a processor, can implement the various steps of the aforementioned machine learning-based corrosion-resistant conductive composite material formulation optimization method.

[0208] Example 5:

[0209] This embodiment proposes a computer program product, including a computer program or instructions, which, when executed by a processor, implements the aforementioned machine learning-based method for optimizing the formulation of corrosion-resistant conductive composite materials.

[0210] Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or part of the technical solution, can be embodied in the form of a computer program product.

[0211] The various embodiments in this application are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0212] The scope of protection of this application is not limited to the embodiments described above. Obviously, those skilled in the art can make various modifications and variations to this disclosure without departing from the scope and spirit of this disclosure. If such modifications and variations fall within the scope of this disclosure and its equivalents, then the intent of this disclosure also includes these modifications and variations.

Claims

1. A method for optimizing the formulation of corrosion-resistant conductive composite materials based on machine learning, characterized in that, This method includes the following steps: Several sets of raw sample data of corrosion-resistant conductive composite materials were collected. Each set of raw sample data includes material formulation, preparation process and performance indicators. Each set of original sample data is preprocessed, and the preprocessed material formulation and preparation process are encoded into structured feature vectors according to the material composition and process hierarchical coding system. Performance labels are calculated based on the preprocessed performance indicators. Using structured feature vectors as input and performance labels as output, a machine learning-based performance prediction model for corrosion-resistant conductive composite materials is established. Based on the preset formulation constraints, a random sampling method based on the feasible region is used to generate several sets of candidate formulation and process combinations. A machine learning-based performance prediction model for corrosion-resistant and conductive composite materials was used to predict each candidate combination of formulation and process. A multi-objective evolutionary algorithm based on non-dominated sorting was then used to screen the Pareto optimal solution set. Based on the performance requirements of the target application scenario, one or more formulation and process combinations are selected from the Pareto optimal solution set. Then, according to the material composition and process hierarchical coding system, the selected formulation and process combinations are decoded to obtain the optimized material formulation and preparation process scheme.

2. The method for optimizing the formulation of corrosion-resistant conductive composite materials based on machine learning according to claim 1, characterized in that, The material formulation includes, but is not limited to: polymer matrix components, conductive filler components, and functional additive components, as well as the content information of each component; The preparation process includes, but is not limited to: process type and process parameters; The performance indicators include, but are not limited to: electrical conductivity data, mechanical properties data, and corrosion resistance data.

3. The method for optimizing the formulation of corrosion-resistant conductive composite materials based on machine learning according to claim 2, characterized in that, The specific methods of preprocessing include: standardization and data cleaning; For any original sample data, the polymer matrix components, conductive filler components and functional additive components in the material formulation are respectively merged into standardized coding categories, and the content information of each component in the material formulation is uniformly converted into mass fraction to form a standardized material formulation. Based on the preset process terminology dictionary and keyword matching rules, the preparation process information is categorized into standardized process flows. For preparation process information that cannot be directly matched, the standardized process flow to which the preparation process information belongs is determined through manual matching. Data cleaning is performed on standardized material formulations, standardized process flows, and performance indicators to obtain pre-processed material formulations, preparation processes, and performance indicators.

4. The method for optimizing the formulation of corrosion-resistant conductive composite materials based on machine learning according to claim 3, characterized in that, The specific method for data cleaning of standardized material formulations, standardized process flows, and performance indicators to obtain pre-processed material formulations, preparation processes, and performance indicators is as follows: Standardized material formulations, standardized process flows, and performance indicators are broken down into several independent fields, and missing value checks are performed on each field: If no missing fields are found, the standardized material formulation, standardized process flow, and performance indicators for that group will be retained. If any fields are missing, they will be filled in by backtracking and verification, and the completed standardized material formulas, standardized process flows and performance indicators will be retained; for missing fields that cannot be filled in, the set of standardized material formulas, standardized process flows and performance indicators will be deleted. Based on the preset numerical range and preset logical consistency rules, anomalies are identified in the retained standardized material formulas, standardized process flows and performance indicators, and the identified abnormal fields are backtracked and corrected to obtain the corrected standardized material formulas, standardized process flows and performance indicators. If it cannot be corrected, then delete the standardized material formula, standardized process flow and performance indicators of that group; The revised standardized material formulation and standardized process flow are used as the pretreated material formulation and preparation process, and the revised performance indicators are used as the pretreated performance indicators.

5. The method for optimizing the formulation of corrosion-resistant conductive composite materials based on machine learning according to claim 4, characterized in that, The specific method for encoding the pretreated material formulation and preparation process into a structured feature vector based on the material composition and process hierarchical coding system, and calculating the performance label based on the pretreated performance indicators, is as follows: The material composition and process layer coding system includes: a matrix material coding module, a conductive filler coding module, a functional additive coding module, and a process flow coding module; Using the matrix material coding module, the polymer matrix components in the pretreated material formulation are binary indicated and coded to obtain the matrix material coding vector; Using a conductive filler coding module, the conductive filler components in the pretreated material formulation are binary indicated and coded to obtain a conductive filler coding vector; The functional additive coding module is used to perform binary indicator coding on the functional additive components in the pretreated material formulation to obtain the functional additive coding vector; Using the process flow coding module, the pre-treated preparation process is binary indicated and coded to obtain the process flow coding vector; The matrix material encoding vector, conductive filler encoding vector, functional additive encoding vector, and process flow encoding vector constitute a structured feature vector. The conductivity, mechanical properties, and corrosion resistance data from the pre-processed performance indicators are weighted and summed to obtain the conductivity label. Mechanical property labels and corrosion resistance label .

6. The method for optimizing the formulation of corrosion-resistant conductive composite materials based on machine learning according to claim 5, characterized in that, The machine learning-based performance prediction model for corrosion-resistant conductive composite materials employs a machine learning regression model to model the mapping relationship between structured feature vectors and performance labels.

7. The method for optimizing the formulation of corrosion-resistant conductive composite materials based on machine learning according to claim 6, characterized in that, The formulation constraints include: the content range of each component in the polymer matrix component, conductive filler component, and functional additive component; the range of process parameters for the preparation process; and the total amount of components; wherein each of the polymer matrix component, conductive filler component, and functional additive component includes at least one component.

8. The method for optimizing the formulation of corrosion-resistant conductive composite materials based on machine learning according to claim 7, characterized in that, The specific method for using a machine learning-based performance prediction model for corrosion-resistant conductive composite materials to predict each candidate formulation and process combination, and then using a multi-objective evolutionary algorithm based on non-dominated sorting to obtain the Pareto optimal solution set, is as follows: Each candidate combination of formulation and process is treated as a candidate solution; Using a machine learning-based performance prediction model for corrosion-resistant and conductive composite materials, the performance of each candidate solution is predicted, and the predicted conductivity label, predicted mechanical property label, and predicted corrosion resistance label of each candidate solution are obtained. The predicted conductivity, mechanical properties, and corrosion resistance labels of each candidate solution are normalized to obtain the normalized conductivity, mechanical properties, and corrosion resistance of each candidate solution, and a multi-objective optimization function is constructed. The multi-objective optimization function is: taking the normalized electrical conductivity, mechanical properties and corrosion resistance as objectives, and maximizing the three objectives simultaneously; Set population size Maximum number of generations The range of crossover probability values, the range of mutation probability values, and the stagnation threshold; wherein the population size is... Same as the number of candidate solutions; Treat each candidate solution as an individual, obtain the initial population, and begin evolution. In each round of evolution, perform the following process: Based on the dominance conditions, all individuals in the current population are sorted into non-dominated categories. Based on the dominance relationships between individuals, all individuals are divided into multiple Pareto levels. Individuals in the first level are taken as non-dominated solutions and constitute the Pareto front of the current generation. Calculate the crowding distance for each individual within the same non-dominant class; A tournament selection strategy based on non-dominance rank and crowding distance is adopted to select from all individuals in the current population. There are 1 parent generation individual; among them, individuals with lower non-dominance levels are preferred as parent generation individuals, and when the non-dominance levels are the same, individuals with larger crowding distances are preferred as parent generation individuals. Based on the aforementioned crossover probability and mutation probability ranges, the crossover and mutation probabilities in the current evolutionary process are updated. Then, the updated crossover and mutation probabilities are used to perform crossover and mutation on the parent individuals to generate... Individual offspring; For the generated Each offspring individual undergoes constraint repair to obtain an offspring population, which is then merged with the current population to form a joint population. The joint population is ordered by non-dominance, and the crowding distance of each individual within the same non-dominance level is calculated. The top individuals are selected according to the criterion of priority based on non-dominance level and secondarily on crowding distance. Each individual serves as the next generation of the population; Repeat the above steps until the maximum number of generations is reached. The iteration terminates when the improvement in the Pareto front is less than the stagnation threshold for several consecutive generations, and the Pareto optimal solution set is output. .

9. A machine learning-based formulation optimization system for corrosion-resistant conductive composite materials, used to implement the machine learning-based formulation optimization method for corrosion-resistant conductive composite materials according to any one of claims 1-8, characterized in that, The system includes: The sample data acquisition module is used to collect raw sample data of several sets of corrosion-resistant conductive composite materials. Each set of raw sample data includes material formulation, preparation process and performance indicators. The preprocessing and coding module is used to preprocess each set of original sample data, and encode the preprocessed material formulation and preparation process into structured feature vectors according to the material composition and process hierarchical coding system, and calculate performance labels based on the preprocessed performance indicators; The prediction model building module is used to build a machine learning-based performance prediction model for corrosion-resistant conductive composite materials, taking structured feature vectors as input and performance labels as output. The candidate combination generation module is used to generate several sets of candidate formulation and process combinations based on a random sampling method based on the feasible region, according to preset formulation constraints. The multi-objective optimization module is used to predict the performance of corrosion-resistant conductive composite materials using a machine learning-based model, and to select the Pareto optimal solution set by using a multi-objective evolutionary algorithm based on non-dominated sorting. The decoding output module is used to select one or more formulation and process combinations from the Pareto optimal solution set according to the performance requirements of the target application scenario, and decode the selected formulation and process combinations according to the material composition and process hierarchical coding system to obtain the optimized material formulation and preparation process scheme.

Images