Multi-objective optimization method, device and equipment for heat treatment process of high-strength steel

Abstract

The application relates to a multi-objective optimization method, device and equipment for a high-strength steel heat treatment process, which comprises obtaining phase change kinetics parameters according to a continuous cooling transformation curve and combining experimental design to obtain microstructure and mechanical property data, constructing a related data set covering key heat treatment parameters, microstructure characteristics and final mechanical properties, and establishing a mapping relationship between a parameter space and macroscopic properties to determine an optimized basic process window; a prediction model is constructed and trained by using a machine learning regression method to realize rapid prediction from input process parameters to output organization state and mechanical properties; the prediction model is further embedded in a genetic algorithm to perform reverse optimization of process parameters with target performance as a constraint, and an optimal or approximately optimal heat treatment process parameter combination is output. The application can reduce experimental cost and period, and realize efficient optimization and accurate design of the heat treatment process under a complex nonlinear relationship.

Description

Technical Field

[0001] This application relates to the field of high-strength steel heat treatment process optimization technology, and more specifically to a multi-objective optimization method, apparatus and equipment for high-strength steel heat treatment process. Background Technology

[0002] With the rapid development of aerospace, high-end equipment and other fields, key load-bearing components such as aircraft landing gear and main load-bearing structural components are facing increasingly stringent service performance requirements. These components need to achieve excellent fracture toughness and fatigue resistance at the same time on the basis of ultra-high strength, that is, to achieve the core goal of synergistic optimization of strength and toughness. Under this requirement, ultra-high strength steels such as 300M steel have become the preferred materials for such key components due to their excellent comprehensive mechanical properties.

[0003] To fully unleash the strength and toughness potential of 300M steel and meet the aforementioned performance indicators, heat treatment is the core process for controlling its final microstructure and mechanical properties. Currently, the traditional quenching and low-temperature tempering processes widely used in the industry mainly rely on the dispersed precipitation of carbides in the martensitic matrix for their strengthening mechanism. However, this single martensitic-carbide microstructure, while imparting ultra-high strength to the material, is limited by the inherent brittleness of the matrix, making it difficult to further improve toughness and achieve a higher level of strength-toughness synergy.

[0004] To overcome the bottleneck in improving toughness, quenching and partitioning processes, as well as quenching, partitioning, and tempering processes, have emerged. The core logic of these new processes is to improve the toughness of materials by controlling the distribution of carbon elements to introduce and stabilize residual austenite with phase transformation-induced plasticity. However, these processes involve multiple key parameters such as quenching temperature, partitioning temperature, and partitioning time. These parameters have a highly nonlinear and strongly coupled complex mapping relationship with the final microstructure (such as martensite lath morphology, residual austenite content, and carbide precipitation characteristics) and macroscopic mechanical properties. This complex relationship between process, microstructure, and properties is difficult to describe accurately using simple physical models or empirical formulas.

[0005] Currently, the optimization of such complex process systems still relies heavily on trial and error based on expert experience. This method requires repeated sample preparation, heat treatment, and performance testing, which is time-consuming and costly. At the same time, it is limited by the number of experiments, making it difficult to achieve efficient searching in a large parameter space and easily missing the optimal process. Furthermore, it only follows the forward exploration path from process to performance and lacks the reverse design capability to deduce process parameters from the target performance, making it difficult to meet the needs of precise performance control. Summary of the Invention

[0006] This application provides a multi-objective optimization method, apparatus, and equipment for the heat treatment process of high-strength steel. It enables rapid simulation prediction of heat treatment process parameters and performance-oriented reverse design without the need for extensive full-cycle physical experiments. This improves the efficiency and accuracy of high-strength steel material design and process development, thereby solving the problems of traditional material heat treatment research and development, which rely on experience-based trial and error, have long experimental cycles, are costly, and are difficult to accurately analyze and control complex nonlinear relationships under multi-parameter coupling.

[0007] In a first aspect, this application provides a multi-objective optimization method for the heat treatment process of high-strength steel. The method includes: generating preset phase transformation kinetic parameters based on the continuous cooling transformation curve of the high-strength steel material to be treated obtained by the thermal expansion method, and generating the microstructure evolution law of the high-strength steel material to be treated under different cooling conditions based on the preset phase transformation kinetic parameters; constructing a preset correlation dataset based on the microstructure evolution law and the microstructure and mechanical property data of multiple sets of samples with different process states prepared by a preset experimental design, and performing preset processing on the preset correlation dataset; constructing a preset prediction model based on the preset correlation dataset using a machine learning regression method to characterize the nonlinear mapping relationship between heat treatment process parameters and microstructure evolution and final performance under multi-parameter coupling conditions, and performing preset training on the preset prediction model; and embedding the trained preset prediction model as an optimization problem evaluator into an optimization algorithm to construct a process parameter reverse design model with target performance as a constraint and perform iterative optimization using a genetic algorithm to output the optimal or near-optimal combination of heat treatment process parameters that satisfies the target performance.

[0008] In one alternative of the first aspect, generating the microstructure evolution law of the high-strength steel material to be treated under different cooling conditions based on the preset phase transformation kinetic parameters includes: analyzing the characteristic points of the continuous cooling transformation curve of the high-strength steel material to be treated under different cooling rates, based on the continuous cooling transformation curve of the high-strength steel material to be treated measured by the thermal expansion method, to generate phase transformation kinetic parameters including the martensitic transformation start temperature, the martensitic transformation end temperature, and the bainitic transformation range; and determining the temperature range of the heat treatment process based on the phase transformation kinetic parameters, and within the temperature range, using the partitioning temperature as an optimization variable, and using the quenching temperature, partitioning time, and cooling rate as auxiliary variables, constructing multiple heat treatment process combinations using full factorial experimental design or partial factorial experimental design.

[0009] In one optional embodiment of the first aspect, based on the microstructure evolution law and the microstructure and mechanical property data of multiple sets of samples prepared using a preset experimental design under different process conditions, a preset associated dataset is constructed and the preset associated dataset is pre-processed, including: collecting microstructure characteristics including austenite lath size, retained austenite volume fraction, and carbon concentration of multiple sets of samples prepared using a preset experimental design under different process conditions, as well as mechanical property indicators including yield strength, tensile strength, elongation, and impact toughness; uniformly numbering and structuring the microstructure characteristics and mechanical property indicators to construct a multidimensional associated dataset including process parameters, microstructure characteristics, and mechanical properties; and performing missing value imputation, outlier removal, duplicate data cleaning, and data consistency verification on the multidimensional associated dataset, and numericalizing and standardizing all features of the multidimensional associated dataset to eliminate differences in the dimensions of different physical quantities.

[0010] In one alternative embodiment of the first aspect, based on the preset associated dataset, a preset prediction model is constructed using a machine learning regression method to characterize the nonlinear mapping relationship between process parameters, microstructure evolution, and final performance under multi-parameter coupling conditions. This includes: constructing physically meaningful derived features based on the original features of the multidimensional associated dataset, wherein the derived features include one or more of the following: the difference between partitioning temperature and quenching temperature, the product of partitioning time and temperature, normalized temperature parameters, and features reflecting the thermodynamic driving force for carbon partitioning from martensite to austenite; using the multidimensional associated dataset and derived features as input features and one or more mechanical performance indicators as prediction targets, and employing an extreme gradient boosting method to construct a positive prediction model for process and performance; and using the multidimensional associated dataset and derived features as input features and one or more microstructure features as prediction targets, and employing an extreme gradient boosting method to construct a prediction model for process and microstructure.

[0011] In one optional embodiment of the first aspect, the preset prediction model is preset-trained, comprising: dividing the multidimensional association dataset into a training set, a validation set, and a test set according to a preset ratio; wherein the preset prediction model is iteratively trained on the training set using a forward step-by-step addition ensemble learning training method, and in each iteration, a new base learner is constructed to fit the residual between the current model prediction result and the true value, and the output of the base learner in each iteration is accumulated; based on the validation set, the preset hyperparameters of the preset prediction model are optimized using a grid search or Bayesian optimization method to generate an optimal combination of hyperparameters, wherein the preset hyperparameters include one or more of the following: maximum depth of a single decision tree, learning rate or step size reduction factor, number of base estimators, and regularization parameter; and based on the test set, the preset prediction model is performance-evaluated using a preset evaluation index to determine the degree of deviation between the model prediction result and the true value, and after the preset evaluation index meets the preset accuracy requirement, a trained preset prediction model is generated, wherein the preset evaluation index includes one or more of the following: coefficient of determination, mean square error, or root mean square error.

[0012] In one optional embodiment of the first aspect, the step of embedding the trained preset prediction model as an optimization problem evaluator into the optimization algorithm, constructing a process parameter reverse design model constrained by the target performance, and iteratively optimizing it using a genetic algorithm includes: setting at least one optimization objective, constraint condition, and design variable to be optimized according to the target performance; within the parameter space defined by the constraint condition, setting the maximum number of generations, population size, crossover probability, and mutation probability of a single-objective or multi-objective genetic algorithm, and randomly generating an initial population composed of multiple individuals, wherein each individual represents a set of heat treatment process parameter combinations through an encoding method; evaluating the fitness of each individual in the initial population, generating a fitness evaluation result, and determining whether a preset termination condition is met based on the fitness evaluation result; if so, selecting the individual with the highest fitness from all iterations, decoding it to obtain and output the optimal or near-optimal heat treatment process parameter combination that meets the target performance requirements; if not, performing a genetic operation on the current population; and iteratively iterating the fitness evaluation and genetic operation process until the preset termination condition is met.

[0013] In one optional embodiment of the first aspect, fitness evaluation is performed on each individual in the initial population to generate fitness evaluation results, including: decoding the encoding of each individual in the initial population into specific process parameter values; inputting the decoded process parameter combination into a trained preset prediction model and receiving the corresponding prediction results output by the preset prediction model; constructing a fitness function according to the optimization objective and constraints, wherein when the prediction result satisfies the constraints, the target performance prediction result is used as the fitness value; when the prediction result does not satisfy the constraints, a penalty term is applied to the fitness value to reduce the fitness value of individuals that do not satisfy the constraints; and, based on the fitness value, determining whether a preset termination condition is met; if so, selecting the individual with the highest fitness value from all iterations, decoding it to obtain and output the optimal or near-optimal combination of heat treatment process parameters that meets the target performance requirements; otherwise, performing genetic operations on the current population.

[0014] In one optional embodiment of the first aspect, the genetic operation performed on the current population includes: selecting superior individuals from the current population as parent individuals for reproduction according to a preset selection strategy and a preset selection probability based on the individual's fitness value; performing a crossover operation on the selected parent individuals according to a preset crossover probability to generate offspring individuals, wherein the crossover operation includes recombination of different process parameter combinations by randomly selecting a crossover position in the parent individual's encoding string and swapping a portion of the encoding after the crossover position; performing a mutation operation on the offspring individuals according to a preset mutation probability by randomly changing the value or encoding bit of at least one process parameter in the individual's encoding string; merging the new individuals generated through selection, crossover, and mutation with the individuals with the highest fitness selected from the current population according to a preset elite retention strategy to generate a new generation population; and using the new generation population as the current population, updating the generation number, and returning to perform fitness evaluation.

[0015] In one optional embodiment of the first aspect, when performing preset processing on the preset associated dataset, the method further includes: constructing a quantitative response relationship model between preset performance indicators and preset process parameters based on the preset associated dataset, and determining the changing trend of performance indicators with process parameters through the quantitative response relationship model; performing correlation analysis on the correspondence between performance changes and organizational evolution based on the changing trend and the organizational evolution law of the preset associated dataset, and generating correlation analysis results; and generating a range of values ​​for heat treatment process parameters that meet the target performance requirements based on the changing trend and the correlation analysis results, and using the range of values ​​as a preset initial optimization window.

[0016] In a second aspect, this application provides a multi-objective optimization device for high-strength steel heat treatment process to implement the method described in any one of the first aspects, comprising: a feature evolution module, used to generate preset phase transformation kinetic parameters based on the continuous cooling transformation curve of the high-strength steel material to be treated obtained by the thermal expansion method, and to generate the microstructure evolution law of the high-strength steel material to be treated under different cooling conditions based on the preset phase transformation kinetic parameters; a data construction module, used to construct a preset correlation dataset based on the microstructure evolution law and the microstructure and mechanical property data of multiple sets of samples with different process states prepared by a preset experimental design, and to perform preset processing on the preset correlation dataset; a model construction module, used to construct a preset prediction model based on the preset correlation dataset using a machine learning regression method to characterize the nonlinear mapping relationship between heat treatment process parameters and microstructure evolution and final performance under multi-parameter coupling conditions, and to perform preset training on the preset prediction model; and an optimization output module, used to embed the trained preset prediction model as an optimization problem evaluator into an optimization algorithm, construct a process parameter reverse design model with target performance as a constraint, and perform iterative optimization using a genetic algorithm to output the optimal or near-optimal combination of heat treatment process parameters that satisfies the target performance.

[0017] Thirdly, this application provides an electronic device, comprising: a processor adapted to execute a computer program; and a computer-readable storage medium storing a computer program, wherein when the computer program is executed by the processor, it implements the method described in any one of the first aspects.

[0018] It should be understood that the above general description and the following detailed description are merely exemplary and do not limit this application. Attached Figure Description

[0019] The accompanying drawings, which are incorporated herein and form part of this specification, illustrate one or more embodiments of the present application and, together with the description, serve to explain the principles of the present application and to enable those skilled in the art to make and use the present application.

[0020] Figure 1 This is a flowchart illustrating an exemplary multi-objective optimization method according to some embodiments of this application.

[0021] Figure 2 This is a flowchart illustrating an exemplary method for generating organizational evolution patterns according to some embodiments of this application.

[0022] Figure 3 This is a schematic diagram of the continuous cooling transformation curve of an exemplary high-strength steel material according to some embodiments of this application. Figure 4These are electron microscope images of lath martensite morphology at two exemplary partitioning temperatures according to some embodiments of this application; wherein (a) is a partitioning temperature of 300°C and (b) is a partitioning temperature of 400°C.

[0023] Figure 5 These are electron micrographs of carbide morphologies at two exemplary partitioning temperatures according to some embodiments of this application; wherein (a) is a partitioning temperature of 300°C and (b) is a partitioning temperature of 400°C.

[0024] Figure 6 These are electron microscope images of the residual austenite morphology at two exemplary partitioning temperatures according to some embodiments of this application; wherein (a) is a partitioning temperature of 300°C and (b) is a partitioning temperature of 400°C.

[0025] Figure 7 This is a schematic diagram of an exemplary XRD diffraction pattern of residual austenite at different partition temperatures according to some embodiments of this application.

[0026] Figure 8 This is a schematic diagram of the residual austenite content at different partitioning temperatures according to some embodiments of this application.

[0027] Figure 9 This is an exemplary stress-strain curve of 300M steel at different fractionation temperatures according to some embodiments of this application.

[0028] Figure 10 This is a flowchart illustrating an exemplary method for constructing and processing associated datasets according to some embodiments of this application.

[0029] Figure 11 This is a flowchart illustrating an exemplary method for constructing a preset prediction model according to some embodiments of this application.

[0030] Figure 12 This is a schematic diagram of an exemplary XGBoost model according to some embodiments of this application.

[0031] Figure 13 This is a flowchart illustrating an exemplary preset prediction model training method according to some embodiments of this application.

[0032] Figure 14 This is a flowchart illustrating an exemplary iterative optimization method according to some embodiments of this application.

[0033] Figure 15 This is a schematic diagram of an exemplary genetic algorithm flow according to some embodiments of this application.

[0034] Figure 16This is a flowchart illustrating an exemplary initial optimization window generation method according to some embodiments of this application.

[0035] Figure 17 This is a schematic diagram of the module connections of an exemplary multi-objective optimization device according to some embodiments of this application.

[0036] Figure 18 This is a schematic diagram of the structure of an exemplary electronic device according to some embodiments of this application. Detailed Implementation

[0037] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments may be implemented in various forms and should not be construed as limited to the examples set forth herein; rather, the description of these embodiments is intended to make this application more comprehensive and complete, and to fully convey the concept of the exemplary embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to provide a deeper understanding of embodiments of this application.

[0038] To facilitate understanding of the technical solutions provided in this application, the relevant terms are explained below.

[0039] It should be noted that the terminology used in the implementation section of this application is only for explaining the embodiments of this application and is not intended to limit this application.

[0040] For example, the term "and / or" in this article simply describes the relationship between related objects, indicating that three relationships can exist. For instance, A and / or B can represent: A alone, A and B simultaneously, and B alone. The term "at least one" simply describes the combination relationship of listed objects, indicating that one or more can exist. For instance, at least one of the following: A, B, C can represent the following combinations: A alone, B alone, C alone, A and B simultaneously, A and C simultaneously, B and C simultaneously, and A, B, and C simultaneously. The term "multiple" refers to two or more. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0041] For example, the term "correspondence" can indicate a direct or indirect correspondence between two things, or an association between them, or a relationship of instruction and being instructed, configuration and being configured, etc. The term "instruction" can be direct, indirect, or indicate an association. For example, A instructing B can mean A directly instructs B, for example, B can be obtained through A; it can also mean A indirectly instructs B, for example, A instructs C, B can be obtained through C; or it can mean an association between A and B. The terms "predefined" or "preconfigured" can refer to pre-stored codes, tables, or other relevant information that can be used for instruction in the device, or it can refer to something agreed upon by a protocol. "Protocol" can refer to standard protocols in the field. The term "when..." can be interpreted as "if," "when," or "in response," etc. Similarly, depending on the context, the phrases "if determined" or "if detected (the condition or event of the statement)" can be interpreted as "when determined" or "in response to determined" or "when detected (the condition or event of the statement)" or "in response to detected (the condition or event of the statement)" and similar descriptions. The terms "first," "second," "third," "fourth," "A," "B," etc., are used to distinguish different objects, not to describe a specific order. The terms "includes" and "has," and any variations thereof, are intended to cover non-exclusive inclusion.

[0042] Currently, the optimization of complex process systems involving technology, microstructure, and properties still heavily relies on trial-and-error methods based on expert experience. However, this method has fundamental flaws: First, each complete process experiment requires multiple steps, including sample preparation, heat treatment, microstructure characterization, and performance testing. This is not only time-consuming and labor-intensive but also leads to high material and testing costs, significantly extending the R&D cycle. Second, the potential process space formed by combinations of process parameters is extremely vast, while the number of experiments that can actually be conducted is limited. This makes this exploration method inherently sparse and inefficient, easily overlooking globally optimal combinations of process parameters. Third, this method follows a forward exploration path from process to performance and lacks reverse design capabilities. That is, when faced with clear design goals such as optimal toughness while meeting specific strength requirements, it is impossible to quickly determine the optimal process parameters from the target performance. This has become an obstacle to the precise customization of material properties.

[0043] Although auxiliary tools such as statistical regression and finite element simulation have been attempted to be applied to process optimization, they all have obvious limitations. Statistical methods are difficult to accurately characterize the highly nonlinear relationship between process, microstructure and performance, and the prediction accuracy is insufficient. Finite element simulation relies heavily on idealized material parameters and physical assumptions, which not only has high computational costs, but also has limited reliability in predicting the final performance.

[0044] Therefore, in order to solve the above-mentioned technical problems, this application provides a multi-objective optimization method 100 for the heat treatment process of high-strength steel. This multi-objective optimization method 100 aims to deeply integrate data and material process knowledge and combine machine learning methods to build a complete technical system from forward accurate prediction to performance-oriented reverse design, fundamentally solving the core deficiencies of traditional methods in terms of efficiency, accuracy and reverse design capabilities.

[0045] Machine learning methods, especially data-driven modeling techniques, offer a novel approach to solving these challenges. Gradient boosting decision trees, neural networks, and other machine learning models excel at mining complex mapping relationships from high-dimensional, nonlinear data, enabling the construction of high-precision process and performance prediction models based on historical experimental data. More importantly, combining these prediction models with intelligent optimization algorithms such as genetic algorithms can build a complete reverse design system from target performance to optimal process, driving the transformation of process development from traditional trial-and-error to modern intelligent design. While ensuring high prediction accuracy, it can significantly reduce the number of experimental iterations, significantly improve R&D efficiency, and enable rapid development of customized processes for specific performance requirements.

[0046] Figure 1 A flowchart illustrating an exemplary multi-objective optimization method according to some embodiments of this application is shown.

[0047] refer to Figure 1 As shown. The multi-objective optimization method 100 includes at least the following steps 101 to 104.

[0048] 101: Based on the continuous cooling transformation curve of the high-strength steel material to be treated obtained by the thermal expansion method, preset phase transformation kinetic parameters are generated, and the microstructure evolution law of the high-strength steel material to be treated under different cooling conditions is generated based on the preset phase transformation kinetic parameters.

[0049] Figure 2 A flowchart illustrating an exemplary method for generating organizational evolution patterns according to some embodiments of this application is shown. Figure 3 A schematic diagram of the continuous cooling transformation curve of an exemplary high-strength steel material according to some embodiments of this application is shown.

[0050] refer to Figure 2 and Figure 3 As shown. Specifically, in step 101, the microstructure evolution of the high-strength steel material to be treated under different cooling conditions is generated according to preset phase transformation kinetic parameters, including at least the following steps 101a to 101b.

[0051] 101a: Based on the continuous cooling transformation (CCT) curves of the high-strength steel material to be treated, measured by the thermal expansion method under different cooling rates, the characteristic points of the continuous cooling transformation curves as a function of temperature are analyzed to generate phase transformation kinetic parameters including the martensitic transformation start temperature, the martensitic transformation end temperature, and the bainitic transformation range, thereby elucidating the microstructure evolution law of the high-strength steel material to be treated under different cooling conditions.

[0052] The thermal expansion test process includes heating the high-strength steel sample to be treated to the austenitizing temperature range (e.g., 850℃~950℃), holding it at the temperature for a preset time (e.g., 10min~30min), and then cooling it at different cooling rates (e.g., 1℃ / s, 5℃ / s, 10℃ / s, 20℃ / s), and recording the expansion curve of the sample length as a function of temperature. The microstructure evolution law includes at least the relationship between the martensite volume fraction and the cooling rate, the relationship between the retained austenite content and the partitioning temperature, and the relationship between the bainite transformation amount and time.

[0053] The extraction of these feature points includes identifying the locations of abrupt changes in the slope of the expansion curve by performing first-order or second-order derivative analysis, determining the temperature points where the slope changes significantly as the initiation or termination temperatures of the phase transformation, and determining the bainite transformation initiation and termination temperature ranges by combining the curve change trends under different cooling rates.

[0054] 101b: Determine the temperature range of the heat treatment process based on the phase transformation kinetic parameters, and use the partitioning temperature as the main optimization variable within the temperature range, while using the quenching temperature, partitioning time and cooling rate as auxiliary variables. Construct multiple heat treatment process combinations using full factorial experimental design or partial factorial experimental design.

[0055] Specifically, based on the phase transformation kinetic parameters, quenching and partitioning process experiments are designed, with partitioning temperature as the core optimization variable. Multiple levels are selected for experiments within the temperature range where significant carbon partitioning behavior may occur. At the same time, quenching temperature, partitioning time, etc. can be used as auxiliary variables. A series of samples with different process states are prepared by adopting full factorial or partial factorial experimental design to cover the parameter space with research value.

[0056] The method for determining the temperature range includes using the martensitic transformation start temperature as the reference lower limit of the quenching termination temperature, using the bainitic transformation start temperature as the reference upper limit of the bainitic transformation start temperature, and determining the temperature range in which carbon partitioning may occur between the martensitic transformation start temperature and the bainitic transformation start temperature. Within this temperature range, the partitioning temperature is used as the main optimization variable, and the quenching temperature, partitioning time, and cooling rate are used as auxiliary variables.

[0057] Furthermore, multiple heat treatment process combinations were constructed and corresponding samples were prepared using full factorial experimental design or partial factorial experimental design. When the number of variables was small, full factorial experimental design was used to obtain the response relationship under complete parameter combinations. When the number of variables was large, partial factorial experimental design or orthogonal experimental design was used to reduce experimental costs.

[0058] In actual implementation, each heat treatment process combination corresponds to one sample, which is prepared through the following steps: austenitizing treatment is performed according to the set quenching temperature and cooling to the target partitioning temperature according to the set cooling rate. The sample is held at the partitioning temperature for a preset time and then finally cooled to room temperature.

[0059] 102: Based on the evolution of the structure and the microstructure and mechanical property data of multiple sets of samples with different process states prepared by the pre-designed experimental design, a pre-designed correlation dataset is constructed and the pre-designed correlation dataset is pre-processed.

[0060] Figure 4 Electron micrographs of lath martensite morphology at two partition temperatures are shown for some embodiments of this application. Figure 5 Electron micrographs of carbide morphologies at two partitioning temperatures are shown for some embodiments of this application. Figure 6 Electron micrographs of the residual austenite morphology at two exemplary partitioning temperatures are shown for some embodiments of this application. Figure 7 This paper illustrates an exemplary XRD diffraction pattern of residual austenite at different partition temperatures, representing some embodiments of this application. Figure 8 This illustration shows a schematic diagram of the residual austenite content at different partitioning temperatures, representing some embodiments of this application. Figure 9 The stress-strain curves of an exemplary 300M steel at different fractionation temperatures are shown in some embodiments of this application.

[0061] refer to Figures 4 to 9 As shown. In step 102, all obtained samples undergo comprehensive microstructure characterization and mechanical property testing. The system collects microstructure and macroscopic mechanical property data, and constructs a structured dataset linking process parameters, microstructure characteristics, and mechanical properties. Specifically, this includes using equipment such as scanning electron microscopes and X-ray diffractometers to quantitatively obtain key microstructure characteristics such as martensite lath size, retained austenite volume fraction, and carbon concentration; and obtaining core mechanical property indicators such as yield strength, tensile strength, elongation, and impact toughness of the samples according to standard testing methods.

[0062] Figure 10 The diagram illustrates a flowchart of an exemplary method for constructing and processing associated datasets according to some embodiments of this application.

[0063] refer to Figure 10 As shown. Specifically, in step 102, based on the evolution law of the tissue and the microstructure and mechanical property data of multiple sets of samples with different process states prepared by a preset experimental design, a preset correlation dataset is constructed and the preset correlation dataset is subjected to preset processing, including at least the following steps 102a to 102c.

[0064] 102a: Collect microstructure characteristics and mechanical property indicators of multiple groups of samples with different process states prepared according to the laws of tissue evolution and the pre-designed experimental design.

[0065] The microstructure characteristics include one or more of the following: martensite lath size, retained austenite volume fraction, carbon concentration in retained austenite, bainite volume fraction, and carbide size and distribution characteristics; the mechanical properties include one or more of the following: yield strength, tensile strength, elongation after fracture, and impact absorption energy.

[0066] 102b: Based on the microstructure characteristics and mechanical performance indicators, a unified numbering and structured organization is carried out to construct a multidimensional associated dataset containing process parameters, microstructure characteristics and mechanical properties.

[0067] 102c: Based on the multidimensional associated dataset, perform missing value imputation, outlier removal, duplicate data cleaning, and data consistency verification. Also, perform numerical and standardization processing on all features of the multidimensional associated dataset to eliminate differences in the dimensions of different physical quantities.

[0068] In the imputation of missing values, if the missing percentage is below a preset threshold (e.g., 5%), mean imputation or interpolation methods are used; if the missing percentage is above the preset threshold, the corresponding sample is deleted. For outlier removal, box plotting (IQR method) or the 3σ criterion is used to identify and remove or replace outliers. For duplicate data cleaning, completely duplicate samples are deleted, and partially duplicate samples are averaged and merged. For data consistency verification, the process parameters and tissue data are checked for matching, and the data units are verified for consistency. Normalization is used in this standardization process to eliminate the influence of units and accelerate model convergence.

[0069] 103: Based on the pre-set associated dataset, a pre-set prediction model is constructed using machine learning regression method to characterize the nonlinear mapping relationship between heat treatment process parameters and microstructure evolution and final performance under multi-parameter coupling conditions, and the pre-set prediction model is pre-trained.

[0070] In step 103, a data-driven method is introduced to model the nonlinear coupling relationship between complex heat treatment process parameters and material microstructure and properties, so as to achieve high-precision prediction and reverse optimization design of target properties.

[0071] Figure 11 A flowchart illustrating an exemplary preset prediction model construction method according to some embodiments of this application is shown. Figure 12 A schematic diagram of an exemplary XGBoost model is shown, representing some embodiments of this application.

[0072] refer to Figure 11 and Figure 12 As shown. Specifically, in step 103, based on a preset associated dataset, a preset prediction model is constructed using machine learning regression methods to characterize the nonlinear mapping relationship between process parameters and microstructure evolution and final performance under multi-parameter coupling conditions, including at least the following steps 103a to 103c.

[0073] 103a: Construct physically meaningful derived features based on the original features of a multidimensional associated dataset.

[0074] Among them, the derived features include one or more of the following: the difference between the partitioning temperature and the quenching temperature, the product of the partitioning time and the temperature, the normalized temperature parameter, and the dimensionless parameter that reflects the thermodynamic driving force driving the carbon partitioning from martensite to austenite. Through the above derived features, the input variables have both experimental data attributes and physical mechanism constraints, so as to enhance the model's ability to learn physical mechanisms.

[0075] 103b: Using multidimensional associated datasets and derived features as input features, and one or more mechanical performance indicators among yield strength, tensile strength, elongation and impact toughness as prediction targets, the Extreme Gradient Boosting (XGBoost) method is used to construct a positive prediction model from process parameters to mechanical properties.

[0076] The positive prediction model belongs to the ensemble learning algorithm. It constructs multiple decision trees (CART) in sequence and accumulates their prediction results to approximate the nonlinear mapping relationship between complex process parameters and mechanical properties with high accuracy. In actual implementation, in addition to the extreme gradient boosting method, this application can also use random forest, support vector regression or neural network methods to build the model.

[0077] Specifically, the input to the forward prediction model is a preprocessed feature vector x (such as quenching temperature, partitioning temperature, partitioning time, and cooling rate) and the corresponding true performance values ​​y (such as yield strength, tensile strength, elongation, and impact toughness). The forward prediction model learns through a forward step-by-step additive approach: the first tree (Tree1) is directly trained based on the original data (x, y), and its predicted output is... The second tree (Tree2) no longer fits the original y, but instead fits the residual predicted by the first tree. This process corrects the prediction error from the previous round; each subsequent tree follows this rule, striving to fit the residual between the current model's cumulative prediction and the true value; finally, for a given input x, the predicted output of the forward prediction model is... Output values ​​for all K decision trees The sum of the weights of the corresponding leaf nodes, i.e. .

[0078] This mechanism enables the positive prediction model built with XGBoost to accurately learn the complex interaction and non-monotonic dependence between features and targets in an ensemble manner through a tree-by-tree error correction process. It is very suitable for handling high-dimensional, nonlinear, small-sample datasets such as process parameters to mechanical properties in this application. At the same time, the positive prediction model can provide feature importance ranking, intuitively revealing the relative contribution of each process parameter to the final performance, and providing direct guidance for physical mechanism analysis and process optimization.

[0079] 103c: Using multidimensional associated datasets and derived features as input features, and taking one or more of the microstructure features such as martensite lath size, retained austenite volume fraction and carbon concentration as prediction targets, an extreme gradient boosting method is used to construct a prediction model of process parameters to microstructure evolution, so as to achieve quantitative prediction of material microstructure state.

[0080] Specifically, in order to further reveal the regulatory laws of process parameters on microstructure and provide interpretable intermediate variables, a machine learning prediction model can be trained in parallel, which takes process parameters as input and key microstructure features (such as martensite lath size, retained austenite volume fraction and carbon concentration) as output. The construction process of this model is the same as step 103b, and its prediction results can serve as a bridge to understand the microstructure mechanism of final performance changes.

[0081] Therefore, the machine learning model constructed through steps 103b and 103c of this application enables the rapid and accurate prediction of the corresponding microstructure evolution trend and final mechanical properties by inputting any set of process parameters. This transforms the traditional physical verification mode from experiment to measurement into an efficient simulation prediction virtual computing mode, providing a reference digital model for process optimization.

[0082] Figure 13 The diagram illustrates a flowchart of an exemplary preset prediction model training method according to some embodiments of this application.

[0083] refer to Figure 13 As shown. Specifically, in step 103, the preset prediction model is preset trained, which includes at least the following steps 103d to 103f.

[0084] 103d: Divide the multidimensional associated dataset into training, validation, and test sets according to a preset ratio.

[0085] The preset ratio can be 6:2:2 or 7:2:1. In this application, the exemplary reference is set to 7:2:1, that is, the complete dataset is randomly divided into training set, validation set and test set in a 7:2:1 ratio. Then, the XGBoost model is trained through the training set, and the training process is monitored and hyperparameters are tuned through the validation set. The training objective is to minimize the loss function (such as mean squared error) between the predicted result and the true value. The preset prediction model is iteratively trained on the training set using a forward step-by-step addition ensemble learning training method. In each iteration, a new base learner (regression tree) is built, the residual between the current model prediction result and the true value is used as the fitting target, weights are assigned to the newly generated base learner and accumulated into the overall model, thereby gradually reducing the overall prediction error and improving the model fitting accuracy.

[0086] 103e: Based on the validation set, the preset hyperparameters of the preset prediction model are optimized using grid search or Bayesian optimization methods to generate the optimal combination of hyperparameters.

[0087] The key hyperparameters that need to be configured when training the XGBoost model include: max_depth (the maximum depth of a single decision tree, used to control the model complexity and the ability to capture interaction relationships), learning_rate (the learning rate or step size reduction factor, used to control the contribution weight of each tree to the final model and prevent overfitting), n_estimators (the number of base estimators, i.e., the total number of decision trees), and regularization parameters. These hyperparameters are tuned on the validation set through grid search or Bayesian optimization to obtain the best performance.

[0088] 103f: Based on the test set, the performance of the pre-trained prediction model is evaluated using preset evaluation metrics. The degree of deviation between the model's prediction results and the true values ​​is determined. After the preset evaluation metrics meet the preset accuracy requirements, the pre-trained prediction model is generated.

[0089] The preset evaluation indicators include one or more of the following: coefficient of determination, mean square error, or root mean square error. These indicators are used to quantify the prediction accuracy of the model and ensure that the model can make accurate predictions for unseen combinations of heat treatment process parameters. When the preset evaluation indicators reach the thresholds set by the designers (e.g., R² ≥ 0.90 or RMSE is below the preset range), the model training is considered complete, and a preset prediction model for subsequent multi-objective optimization is generated. Otherwise, the process returns to step 103d to retrain or adjust the model structure.

[0090] In some examples of this application, physical constraints of phase transformation dynamics are introduced during model training. The range of martensitic transformation temperature, the bainitic transformation interval, and the thermodynamic driving force of carbon partitioning are embedded as constraint terms. Then, by applying a penalty term that violates physical laws to the prediction results, the model output satisfies the physical constraints of phase transformation dynamics, thereby improving the physical interpretability of the model under small sample conditions.

[0091] In some examples of this application, during model training, phase transformation constraints based on the continuous cooling transformation curve can also be introduced. By constructing a physical constraint function that includes the phase transformation start temperature, termination temperature, and transformation range, the prediction results are constrained so that the model output satisfies the material phase transformation law.

[0092] 104: The pre-trained prediction model is embedded into the optimization algorithm as an evaluation tool for the optimization problem. A reverse design model of process parameters with the target performance as a constraint is constructed and iterative optimization is performed using a genetic algorithm to output the optimal or near-optimal combination of heat treatment process parameters that meet the target performance.

[0093] In step 104, the two models trained in step 103 are introduced into the optimization framework as fast evaluation functions to replace traditional experiments or numerical simulations and realize the reverse design of heat treatment process parameters.

[0094] Figure 14 A flowchart illustrating an exemplary iterative optimization method according to some embodiments of this application is shown; Figure 15 A schematic diagram of an exemplary genetic algorithm flow is shown for some embodiments of this application.

[0095] refer to Figure 14 and Figure 15 As shown. Specifically, in step 103, the trained preset prediction model is embedded in the optimization algorithm as an optimization problem evaluator, a reverse design model of process parameters constrained by the target performance is constructed, and iterative optimization is performed using a genetic algorithm, including at least the following steps 104a to 104e.

[0096] 104a: Based on the target performance, set at least one optimization objective, constraints, and design variables to be optimized.

[0097] The optimization objective can be single-objective or multi-objective, including but not limited to maximizing tensile strength, maximizing elongation, maximizing the strength-ductility product (strength × elongation), or simultaneously optimizing strength and ductility properties in the case of multiple objectives; the constraints include but are not limited to yield strength, tensile strength, elongation, or impact toughness meeting preset threshold ranges, constraints on the range of process parameters (such as upper and lower limits of quenching temperature, partitioning temperature, partitioning time, and cooling rate), and microstructure constraints (such as the retained austenite volume fraction being within a preset range); the design variables to be optimized include one or more of quenching temperature, partitioning temperature, partitioning time, and cooling rate, and are uniformly expressed as design variables.

[0098] For example, a typical optimization proposition can be stated as: "When the tensile strength is not less than..." Under these conditions, find ways to improve impact toughness The proposition "maximizing the combination of process parameters" can be formally defined as: the optimization objective is to maximize the predicted impact toughness result. The constraint condition is the tensile strength prediction result. (Strength constraints) and all process parameters (such as quenching temperature) , distribution temperature Allocation time The design variables must be within a pre-defined overall performance optimization window (process feasibility constraint), and the design variables are the set of process parameters to be optimized. .

[0099] 104b: Within the parameter space defined by constraints, set the maximum number of generations, population size, crossover probability, and mutation probability for a single-objective or multi-objective genetic algorithm, and randomly generate an initial population consisting of multiple individuals.

[0100] The process parameters are represented by real number encoding or binary encoding. Each individual corresponds to a complete set of heat treatment process parameters. The individual chromosome is composed of multiple gene loci, and each gene locus corresponds to a process variable.

[0101] 104c: Perform fitness assessment on each individual in the initial population, generate fitness assessment results, and apply the fitness assessment results.

[0102] Specifically, fitness assessment is performed on each individual in the initial population to generate fitness assessment results, including at least the following steps 1041 to 1043.

[0103] 1041: Decode the encoding of each individual in the initial population into specific process parameter values.

[0104] 1042: Combine the decoded process parameters, input them into the trained preset prediction model, and receive the corresponding prediction results output by the preset prediction model. Construct a fitness function based on the optimization objective and constraints. In single-objective optimization, the fitness function is a function of the objective performance. In multi-objective optimization, a weighted summation method or Pareto ranking method is used to construct the fitness evaluation system. Regarding constraint handling, when the prediction result satisfies all constraints, the function value of the objective performance is directly used as the fitness value. When the prediction result does not satisfy the constraints, a penalty function mechanism is introduced to apply a penalty term to the fitness value to reduce the fitness value of individuals that do not meet the constraints. This penalty term is positively correlated with the degree of constraint violation, thereby reducing the probability of that individual being selected.

[0105] For example, if Then fitness Otherwise, fitness is , where α is a large penalty coefficient, and the fitness value quantifies the degree to which the set of process parameters meets the target performance requirements.

[0106] 1043: Based on the fitness value, determine if the preset termination condition is met. If so, select the individual with the highest fitness value from all iterations, decode it to obtain and output the optimal or near-optimal combination of heat treatment process parameters that meet the target performance requirements. If not, perform genetic operations on the current population.

[0107] The preset termination condition may include, but is not limited to, reaching the maximum number of generations. The optimal fitness value changes less than the change threshold set by the designer for several consecutive generations, or the fitness reaches the preset target value, which is set by the designer according to actual needs.

[0108] 104d: Determine whether the preset termination condition is met. If yes, select the individual with the highest fitness from all iterations, decode it to obtain and output the optimal or near-optimal combination of heat treatment process parameters that meet the target performance requirements; otherwise, perform genetic operations on the current population.

[0109] Specifically, performing genetic operations on the current population includes at least the following steps 1044 to 1048.

[0110] 1044: Based on the fitness value of an individual, select superior individuals from the current population with a preset selection probability according to a preset selection strategy for reproduction; wherein, the preset selection strategy includes, but is not limited to, roulette wheel selection, tournament selection, or sorting selection strategies.

[0111] 1045: Generate offspring individuals by performing crossover operations on the selected parent individuals according to the preset crossover probability. The crossover operation includes randomly selecting a crossover position in the encoding string of the parent individual and swapping the encoding after the crossover position to achieve recombination of different process parameter combinations.

[0112] For example, in a single-point crossover, a pair of parent individuals is randomly selected and the portion of their encoded strings after a certain point is swapped to generate two new offspring individuals in order to explore new combinations of parameters.

[0113] 1046: Perform mutation operations on offspring individuals according to a preset mutation probability. This is done by randomly changing the value or coding bit of at least one process parameter in the individual's coding string to enhance population diversity and avoid getting trapped in local optima. The preset random mutation probability is set by the designer according to actual needs.

[0114] 1047: The new individuals generated through selection, crossover, and mutation are merged with the individuals with the highest fitness selected from the current population according to the preset elite retention strategy to generate a new generation of population;

[0115] 1048: Return to the current population as the new generation and update the generation number for fitness evaluation.

[0116] 104e: Iterative fitness evaluation and genetic operation process until the preset termination condition is met.

[0117] When the termination condition is met, the individual with the highest fitness value or located at the Pareto optimal front is selected from all iterations, decoded, and the corresponding combination of heat treatment process parameters is obtained and output as the optimal or near-optimal solution that meets the target performance requirements.

[0118] In addition to using a genetic algorithm, this application can also use NSGA-II, multi-objective particle swarm optimization, or Bayesian optimization methods to achieve the above-mentioned multi-objective optimization process.

[0119] In some examples of this application, when the preset prediction model outputs, multiple predictions are made for the same input heat treatment process parameter combination to obtain multiple sets of prediction results. The prediction mean, prediction variance, or standard deviation are calculated based on these multiple sets of prediction results as a measure of the uncertainty of the prediction results. During the optimization process, the uncertainty measure is used as one of the auxiliary optimization indicators or constraints. Under the premise of meeting the target performance requirements, the heat treatment process parameter combination corresponding to the prediction result with the lower uncertainty measure is selected first. When the uncertainty measure corresponding to the prediction result is higher than the measurement threshold preset by the designer, a penalty term is applied to the prediction result or its fitness value is reduced to avoid the optimization algorithm converging to the high uncertainty region.

[0120] In some examples of this application, when the requirements include multiple competing objectives, this application can be extended to multi-objective optimization, that is, using a multi-objective genetic algorithm, whose output is no longer a single solution, but a set of Pareto optimal solutions. Each solution in the Pareto optimal solution set represents an optimal trade-off between multiple performance indicators that cannot be further improved simultaneously. Designers can choose the most suitable combination of heat treatment process parameters from this solution set according to the actual emphasis.

[0121] Figure 16 A flowchart illustrating an exemplary initial optimization window generation method according to some embodiments of this application is shown.

[0122] refer to Figure 16 As shown. In some embodiments of this application, when performing preset processing on a preset associated dataset, at least the following steps 110 to 112 are included.

[0123] 110: Based on the preset associated dataset, construct a quantitative response relationship model between preset performance indicators and preset process parameters, and determine the changing trend of performance indicators with process parameters through the quantitative response relationship model.

[0124] Specifically, process parameters including quenching temperature, partitioning temperature, partitioning time, and cooling rate are used as independent variables, and mechanical performance indicators including yield strength, tensile strength, elongation, and impact toughness are used as dependent variables. A quantitative response relationship model is constructed using regression analysis methods, including but not limited to multiple linear regression models, multinomial regression models, response surface model (RSM), or machine learning-based regression models (such as random forest regression or gradient boosting regression). Based on this quantitative response relationship model, each process parameter is scanned or interpolated within its value range to obtain response curves or response surfaces of mechanical performance indicators as a function of a single or multiple variables. Key features are extracted by analyzing the changing trends of the response curves or response surfaces. These key features include, but are not limited to, one or more of the following: monotonically increasing or decreasing intervals, extreme points (maximum or minimum values), inflection points or sensitive intervals (regions where performance is sensitive to parameter changes), and synergistic or competitive effects under the coupling of multiple parameters.

[0125] 111: Based on the changing trends and combined with the organizational evolution patterns of the pre-set associated dataset, conduct correlation analysis on the correspondence between performance changes and organizational evolution, and generate correlation analysis results.

[0126] Specifically, the microstructure characteristics, including martensite lath size, retained austenite volume fraction, and carbon concentration, will be correlated with mechanical properties. This correlation analysis includes, but is not limited to, Pearson correlation coefficient analysis, Spearman rank correlation analysis, principal component analysis (PCA), grey relational analysis, or sensitivity analysis. Under multi-parameter coupling conditions, key microstructure characteristics that significantly affect target performance and their dominant mechanisms will be identified, such as the effect of retained austenite volume fraction on elongation, the contribution of martensite lath refinement to strength improvement, and the influence of carbon fractionation on the synergistic effect of strength and plasticity. Furthermore, combined with the microstructure evolution laws generated in step 101, the microstructure transformation paths within different process parameter ranges will be explained, thereby establishing a mapping relationship between process parameters, microstructure state, and performance response, and generating corresponding correlation analysis results.

[0127] 112: Based on the changing trends and correlation analysis results, generate the range of values ​​for heat treatment process parameters that meet the target performance requirements and use the range of values ​​as the preset initial optimization window.

[0128] Specifically, a performance threshold range is set based on the target performance index, and a region of heat treatment process parameter combinations that meet the performance threshold is selected from the response curves or response surfaces mentioned above. Then, the region of heat treatment process parameter combinations is further selected based on the correlation analysis results to eliminate parameter ranges that may lead to unfavorable structures (such as excessive martensite or unstable retained austenite). The effective value range of each parameter in the heat treatment process parameters is determined to form a sub-region of a multi-dimensional parameter space, where each dimension corresponds to a process parameter and the range of the sub-region is significantly smaller than the original global search space. This multi-dimensional parameter sub-region is defined as a preset initial optimization window and used as a constraint condition for the initial population generation and search range of the genetic algorithm in step 104, thereby reducing the optimization search dimension, improving optimization efficiency, and reducing invalid searches.

[0129] In some examples of this application, during the iterative optimization process in step 104, the preset initial optimization window is dynamically adjusted according to the fitness distribution of the current population in order to gradually shrink the search space and improve the optimization convergence speed.

[0130] Therefore, through the technical solution constructed above, this application realizes the transformation from passive process trial and error to performance verification to active performance input to process output. Designers only need to input a quantitative performance target, and the internal prediction model and optimization algorithm can be automatically called to output process parameter recommendations after global optimization in a short time, thereby improving the efficiency and accuracy of new material research and development and process customization.

[0131] The preferred embodiments of this application have been described in detail above with reference to the accompanying drawings. However, this application is not limited to the specific details of the embodiments described above. Within the scope of the technical concept of this application, various simple modifications can be made to the technical solutions of this application, and these simple modifications all fall within the protection scope of this application. For example, the various specific technical features described in the specific embodiments described above can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, this application will not describe the various possible combinations separately. Furthermore, various different embodiments of this application can also be arbitrarily combined, as long as they do not violate the spirit of this application, they should also be considered as the content disclosed in this application.

[0132] It should also be understood that, in the various method embodiments of this application, the order of the processes mentioned above does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0133] The methods provided in the embodiments of this application have been described above. The apparatus provided in the embodiments of this application will be described below.

[0134] Figure 17 A schematic diagram of the module connections of an exemplary multi-objective optimization apparatus according to some embodiments of this application is shown.

[0135] refer to Figure 17 As shown, the multi-objective optimization device 200 for the high-strength steel heat treatment process of this application includes:

[0136] The feature evolution module 201 is used to generate preset phase transformation kinetic parameters based on the continuous cooling transformation curve of the high-strength steel material to be treated obtained by the thermal expansion method, and to generate the microstructure evolution law of the high-strength steel material to be treated under different cooling conditions based on the preset phase transformation kinetic parameters.

[0137] The data construction module 202 is used to construct a preset correlation dataset based on the microstructure and mechanical property data of multiple sets of samples with different process states prepared by a preset experimental design, and to perform preset processing on the preset correlation dataset.

[0138] The model building module 203 is used to construct a preset prediction model based on a preset associated dataset, using machine learning regression methods to characterize the nonlinear mapping relationship between heat treatment process parameters and microstructure evolution and final performance under multi-parameter coupling conditions, and to perform preset training on the preset prediction model.

[0139] The optimized output module 204 is used to embed the trained preset prediction model as an optimization problem evaluator into the optimization algorithm, construct a process parameter reverse design model with target performance as a constraint, and use a genetic algorithm for iterative optimization to output the optimal or near-optimal combination of heat treatment process parameters that meet the target performance.

[0140] It should be understood that the device embodiments and method embodiments can correspond to each other, and similar descriptions can be referred to the method embodiments. To avoid repetition, they will not be repeated here. Specifically, the multi-objective optimization device 200 can correspond to the corresponding subject in the multi-objective optimization method 100 of the present application embodiments, and each unit in the multi-objective optimization device 200 is for implementing the corresponding process in the multi-objective optimization method 100. For the sake of brevity, they will not be repeated here.

[0141] It should also be understood that the various units in the multi-objective optimization device 200 involved in the embodiments of this application are based on logical functional division. In practical applications, the function of one unit can also be implemented by multiple units, or the function of multiple units can be implemented by one unit. Furthermore, these functions can also be implemented with the assistance of one or more other units. For example, some or all of the multi-objective optimization device 200 can be merged into one or more additional units. Furthermore, some units(s) in the multi-objective optimization device 200 can be further divided into multiple functionally smaller units, which can achieve the same operation without affecting the technical effects of the embodiments of this application. Moreover, the multi-objective optimization device 200 can also include other units. In practical applications, these functions can also be implemented with the assistance of other units, and can be implemented collaboratively by multiple units.

[0142] It should also be understood that the terms "module" or "unit" used in the embodiments of this application refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0143] For example, the multi-objective optimization apparatus 200 involved in the embodiments of this application, and the method of the embodiments of this application, can be constructed and implemented by running a computer program (including program code) capable of executing the steps involved in the corresponding method on a general-purpose computing device including processing elements and storage elements such as a central processing unit (CPU), random access memory (RAM), and read-only memory (ROM). The computer program can be recorded on, for example, a computer-readable storage medium and loaded into an electronic device through the computer-readable storage medium. The computer program is used to implement the corresponding method of the embodiments of this application. In other words, the units mentioned above can be implemented in hardware, in software instructions, or in a combination of hardware and software. Specifically, the steps of the method embodiments in the embodiments of this application can be completed by the integrated logic circuits of the hardware in the processor and / or in software instructions. The steps of the method disclosed in the embodiments of this application can be directly embodied as being executed by a hardware decoding processor, or by a combination of hardware and software in the decoding processor. Optionally, the software can be located in a mature storage medium in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. The software in the memory can be run by the processor to perform the steps described in the method embodiments above.

[0144] Figure 18 A schematic diagram of the structure of an exemplary electronic device according to some embodiments of this application is shown.

[0145] refer to Figure 18 As shown, the electronic device 300 includes at least a processor 310 and a computer-readable storage medium 320. The processor 310 and the computer-readable storage medium 320 can be connected via a bus or other means. The computer-readable storage medium 320 stores a computer program 321, which includes computer instructions. The processor 310 executes the computer instructions stored in the computer-readable storage medium 320. The processor 310 is the computing and control core of the electronic device 300, and is suitable for implementing one or more computer instructions, specifically for loading and executing one or more computer instructions to achieve a corresponding method flow or function.

[0146] As an example, processor 310 may also be referred to as a central processing unit (CPU). Processor 310 may include, but is not limited to: general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete component gate or transistor logic devices, discrete hardware components, etc.

[0147] As an example, the computer-readable storage medium 320 may be a high-speed RAM memory or a non-volatile memory, such as at least one disk storage device; optionally, it may also be at least one computer-readable storage medium located remotely from the aforementioned processor 310. Specifically, the computer-readable storage medium 320 includes, but is not limited to, volatile memory and / or non-volatile memory. The non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory may be random access memory (RAM), which serves as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDR SDRAM), Enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), and Direct Rambus RAM (DR RAM).

[0148] refer to Figure 18 As shown, the electronic device 300 may also include a transceiver 330.

[0149] The processor 310 can control the transceiver 330 to communicate with other devices; specifically, it can send information or data to other devices or receive information or data sent by other devices. The transceiver 330 may include a transmitter and a receiver. The transceiver 330 may further include antennas, and the number of antennas may be one or more.

[0150] It should be understood that the various components in the electronic device 300 are connected through a bus system, which includes a data bus, a power bus, a control bus, and a status signal bus. It is worth noting that the electronic device 300 can be any type of electronic device with data processing capabilities; the computer-readable storage medium 320 stores first computer instructions; the processor 310 loads and executes the first computer instructions stored in the computer-readable storage medium 320 to implement the corresponding steps in the method embodiments of this application; in specific implementations, the first computer instructions in the computer-readable storage medium 320 are loaded and executed by the processor 310, and to avoid repetition, this will not be described further here.

[0151] According to another aspect of this application, embodiments of this application provide a chip. This chip can be an integrated circuit chip with signal processing capabilities, capable of implementing or executing the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The chip can also be referred to as a system-on-a-chip (SoC), system-on-a-chip (SoC), chip system, or system-on-chip, etc. This chip can be applied to various electronic devices capable of mounting chips, enabling the device with the chip mounted to execute the corresponding steps in the methods or logic block diagrams disclosed in the embodiments of this application. For example, the chip may be suitable for implementing one or more computer instructions, specifically suitable for loading and executing one or more computer instructions to achieve a corresponding method flow or corresponding function.

[0152] According to another aspect of this application, embodiments of this application provide a computer-readable storage medium (Memory). This computer-readable storage medium is a computer's memory device used to store programs and data. It is understood that the computer-readable storage medium here may include both built-in storage media within the computer and, of course, extended storage media supported by the computer. The computer-readable storage medium provides storage space that stores the operating system of an electronic device. This storage space contains computer instructions suitable for loading and execution by a processor. When these computer instructions are read and executed by the processor of the computer device, they cause the computer device to perform the corresponding steps in the methods or logic diagrams disclosed in the embodiments of this application.

[0153] According to another aspect of this application, embodiments of this application provide a computer program product or computer program. The computer program product or computer program includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform corresponding steps in the methods or logic block diagrams disclosed in the embodiments of this application. In other words, when the solutions provided in this application are implemented using software, they can be implemented in whole or in part as a computer program product or computer program. The computer program product or computer program includes one or more computer instructions. When these computer program instructions are loaded and executed on a computer, all or part of the processes of the embodiments of this application are run or the functions of the embodiments of this application are implemented.

[0154] It is worth noting that the computer involved in this application can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions involved in this application can be stored in a computer-readable storage medium, or can be transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means.

[0155] Those skilled in the art will recognize that the units and process steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. In other words, those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of protection of this application.

[0156] Finally, it should be noted that the above content is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims. For example, the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. Furthermore, various different embodiments of this application can also be arbitrarily combined, as long as they do not violate the basic idea of ​​this application, and they should also be considered as the content disclosed in this application.

Claims

1. A multi-objective optimization method for the heat treatment process of high-strength steel, characterized in that, The method includes: Based on the continuous cooling transformation curve of the high-strength steel material to be treated obtained by the thermal expansion method, preset phase transformation kinetic parameters are generated, and the microstructure evolution law of the high-strength steel material to be treated under different cooling conditions is generated based on the preset phase transformation kinetic parameters. Based on the aforementioned tissue evolution law and the microstructure and mechanical property data of multiple sets of samples with different process states prepared using a preset experimental design, a preset correlation dataset is constructed and the preset correlation dataset is subjected to preset processing. Based on the pre-defined associated dataset, a pre-defined prediction model is constructed using machine learning regression methods to characterize the nonlinear mapping relationship between heat treatment process parameters and microstructure evolution and final performance under multi-parameter coupling conditions, and the pre-defined prediction model is then pre-trained; and, The trained pre-predictive model is embedded into the optimization algorithm as an evaluation tool for the optimization problem. A reverse design model of process parameters with the target performance as a constraint is constructed and iterative optimization is performed using a genetic algorithm to output the optimal or near-optimal combination of heat treatment process parameters that meet the target performance.

2. The method according to claim 1, characterized in that, The microstructure evolution of the high-strength steel material to be treated under different cooling conditions is generated based on the preset phase transformation kinetic parameters, including: Based on the continuous cooling transformation curves of the high-strength steel material to be treated, measured using the thermal expansion method under different cooling rates, the characteristic points of the continuous cooling transformation curves as a function of temperature are analyzed to generate phase transformation kinetic parameters including the martensitic transformation start temperature, the martensitic transformation end temperature, and the bainitic transformation range; and, The temperature range of the heat treatment process is determined based on the phase transformation kinetic parameters. Within the temperature range, the partitioning temperature is used as the optimization variable, while the quenching temperature, partitioning time, and cooling rate are used as auxiliary variables. Multiple heat treatment process combinations are constructed using full factorial experimental design or partial factorial experimental design.

3. The method according to claim 1 or 2, characterized in that, Based on the aforementioned tissue evolution patterns and the microstructure and mechanical property data of multiple sets of samples prepared using a pre-designed experimental setup under different processing conditions, a pre-defined correlation dataset is constructed and pre-processed, including: Microstructural characteristics, including austenite lath size, retained austenite volume fraction, and carbon concentration, were collected from multiple groups of samples prepared according to the aforementioned microstructure evolution law and using a pre-designed experimental setup. Mechanical property indicators, including yield strength, tensile strength, elongation, and impact toughness, were also collected. Based on the aforementioned microstructure characteristics and mechanical performance indicators, a unified numbering and structured organization is performed to construct a multidimensional associated dataset containing process parameters, microstructure characteristics, and mechanical properties; and, Based on the multidimensional associated dataset, missing value imputation, outlier removal, duplicate data cleaning, and data consistency verification are performed. All features of the multidimensional associated dataset are then numericalized and standardized to eliminate differences in the dimensions of different physical quantities.

4. The method according to claim 3, characterized in that, Based on the pre-defined associated dataset, a pre-defined predictive model is constructed using machine learning regression methods to characterize the nonlinear mapping relationship between process parameters and microstructure evolution and final performance under multi-parameter coupling conditions, including: Based on the original features of the multidimensional associated dataset, physically meaningful derived features are constructed, wherein the derived features include one or more of the following: the difference between partitioning temperature and quenching temperature, the product of partitioning time and temperature, normalized temperature parameter, and features reflecting the thermodynamic driving force driving carbon partitioning from martensite to austenite. Using the multidimensional associated dataset and derived features as input features, and one or more mechanical performance indicators as prediction targets, an extreme gradient boosting method is employed to construct a positive prediction model for process and performance; and... Using the multidimensional associated dataset and derived features as input features, and one or more micro-organism features as prediction targets, an extreme gradient boosting method is used to construct a process and organization prediction model.

5. The method according to claim 1 or 4, characterized in that, Pre-training the preset prediction model includes: The multidimensional association dataset is divided into training set, validation set and test set according to a preset ratio. The training set is trained iteratively using a forward step-by-step addition ensemble learning training method. In each iteration, a new base learner is built to fit the residual between the current model prediction result and the true value, and the output of the base learner in each iteration is accumulated. Based on the validation set, the preset hyperparameters of the preset prediction model are optimized using grid search or Bayesian optimization methods to generate an optimal combination of hyperparameters. The preset hyperparameters include one or more of the following: maximum depth of a single decision tree, learning rate or step size reduction factor, number of base estimators, and regularization parameter; and... Based on the test set, the performance of the preset prediction model is evaluated using preset evaluation indicators. The degree of deviation between the model prediction result and the true value is determined. After the preset evaluation indicators meet the preset accuracy requirements, the trained preset prediction model is generated. The preset evaluation indicators include one or more of the following: coefficient of determination, mean square error, or root mean square error.

6. The method according to claim 5, characterized in that, The step of embedding the trained pre-defined prediction model as an optimization problem evaluator into the optimization algorithm, constructing a reverse design model of process parameters with target performance as a constraint, and using a genetic algorithm for iterative optimization includes: Based on the target performance, set at least one optimization objective, constraints, and design variables to be optimized; Within the parameter space defined by the constraints, the maximum number of generations, population size, crossover probability, and mutation probability of a single-objective or multi-objective genetic algorithm are set, and an initial population consisting of multiple individuals is randomly generated, wherein each individual represents a set of heat treatment process parameters through an encoding method. The fitness of each individual in the initial population is evaluated, a fitness evaluation result is generated, and based on the fitness evaluation result, it is determined whether a preset termination condition is met. If so, the individual with the highest fitness is selected from all iterations, decoded, and the optimal or near-optimal combination of heat treatment process parameters that meets the target performance requirements is output. If not, genetic operations are performed on the current population; and... The fitness evaluation and genetic operations are iterated repeatedly until the preset termination conditions are met.

7. The method according to claim 6, characterized in that, Fitness assessment is performed on each individual in the initial population to generate fitness assessment results, including: The encoding of each individual in the initial population is decoded into specific process parameter values; The decoded process parameters are combined and input into a pre-trained prediction model. The corresponding prediction results output by the pre-trained prediction model are received. A fitness function is constructed based on the optimization objective and constraints. When the prediction result satisfies the constraints, the target performance prediction result is used as the fitness value. When the prediction result does not satisfy the constraints, a penalty term is applied to the fitness value to reduce the fitness value of individuals that do not satisfy the constraints. Based on the fitness value, it is determined whether the preset termination condition is met. If so, the individual with the highest fitness value is selected from all iterations, and it is decoded to obtain and output the optimal or near-optimal combination of heat treatment process parameters that meet the target performance requirements. If not, genetic operations are performed on the current population.

8. The method according to claim 6 or 7, characterized in that, The genetic operations performed on the current population include: Based on the individual's fitness value, superior individuals are selected from the current population with a preset selection probability according to a preset selection strategy for reproduction. According to a preset crossover probability, a crossover operation is performed on the selected parent individual to generate a child individual. The crossover operation includes randomly selecting a crossover position in the encoding string of the parent individual and swapping the part of the encoding after the crossover position to achieve recombination of different combinations of process parameters. The offspring individuals are mutated according to a preset mutation probability by randomly changing the value or encoding bit of at least one process parameter in the individual's encoding string. New individuals generated through selection, crossover, and mutation are merged with the highest-fitting individuals in the current population, selected according to a pre-defined elite retention strategy, to generate a new generation of the population; and... The new generation of the population is used as the current population, and after updating the generation number, it is returned to perform fitness evaluation.

9. The method according to claim 1 or 6, characterized in that, When performing preset processing on the preset associated dataset, the following is also included: Based on the preset associated dataset, a quantitative response relationship model between preset performance indicators and preset process parameters is constructed, and the changing trend of performance indicators with process parameters is determined through the quantitative response relationship model. Based on the aforementioned trend and the organizational evolution patterns of the preset associated dataset, a correlation analysis is performed on the relationship between performance changes and organizational evolution, generating correlation analysis results; and, Based on the changing trends and correlation analysis results, a range of values ​​for heat treatment process parameters that meet the target performance requirements is generated, and this range is used as a preset initial optimization window.

10. An electronic device, characterized in that, include: Processor, adapted to execute computer programs; and, A computer-readable storage medium storing a computer program that, when executed by the processor, implements the method of any one of claims 1 to 9.

Images