Grey wolf optimization based method for carbide morphology control of wear layer of high chromium cast iron

Abstract

This invention discloses a method for controlling the morphology of carbide carbides in the wear layer of high-chromium cast iron based on gray wolf optimization, comprising: generating a target index system; obtaining a preprocessed historical manufacturing dataset of high-chromium cast iron; constructing a compositional thermodynamic inversion model to output an initial set of manufacturable process paths; defining a multi-objective optimization function set and setting a brittleness threshold and a carbide network degree threshold as infeasible solution criteria; using quantum state encoding to initialize the initial set of manufacturable process paths into a population, constructing a quantum state gray wolf population; generating a candidate morphology control solution set within the framework of the quantum adaptive gray wolf optimization algorithm; constructing a service-state morphology evolution proxy model and outputting a service evaluation result set; repeating the steps until convergence to obtain a Pareto optimal morphology control solution set; and selecting the final process scheme set from the Pareto optimal morphology control solution set. This invention reduces the risk of performance drift between batches and adapts to the fine control requirements under complex service conditions.

Description

Technical Field

[0001] This invention relates to the field of metal technology, and in particular to a method for controlling the morphology of carbides in the wear layer of high-chromium cast iron based on gray wolf optimization. Background Technology

[0002] High-chromium cast iron is widely used as a key component material in heavy-load wear conditions such as mining, metallurgy, and cement due to its excellent wear resistance and comprehensive mechanical properties. The size, volume fraction, connectivity, and spatial distribution of carbides in its wear layer play a decisive role in wear resistance, crack resistance, and service life.

[0003] In multi-component high-chromium cast iron systems, existing process designs typically rely on limited experimental experience or lookup methods based on static phase diagrams. Traditional methods struggle to reliably back-engineer specific target morphologies proposed by engineers to a manufacturable and repeatable composition-temperature-time process path. When faced with dynamic processes such as non-equilibrium solidification, tempering hardening, and secondary precipitation in service, they cannot simultaneously ensure the consistency and feasibility of morphology control. Product performance is prone to fluctuations between batches, and the reproducibility of morphology control and service stability are difficult to guarantee.

[0004] In industrial scenarios that require multiple performance objectives such as wear resistance, spalling resistance, and crack resistance, existing optimization methods often employ single-objective or fixed-weight multi-objective trade-offs, which can easily lead to local optima and result in significant performance compromises. This makes it difficult to effectively avoid brittle failure caused by coarse carbide connectivity while simultaneously ensuring both wear resistance and toughness. Summary of the Invention

[0005] One objective of this invention is to propose a method for controlling the carbide morphology of high-chromium cast iron wear layer based on gray wolf optimization. This invention reduces the risk of performance drift between batches and meets the needs of fine control under complex service conditions.

[0006] A method for controlling the carbide morphology of high-chromium cast iron wear layer based on gray wolf optimization according to an embodiment of the present invention includes: A multi-dimensional index system oriented towards the morphology and service performance of target carbides is constructed, and a target index system is generated. Collect historical manufacturing datasets of high-chromium cast iron and perform unified preprocessing to obtain preprocessed historical manufacturing datasets of high-chromium cast iron. Based on a multi-component phase diagram database and a non-equilibrium solidification kinetics model, a composition thermodynamic inversion model is constructed. The target index system and the pre-processed historical manufacturing dataset of high-chromium cast iron are input into the composition thermodynamic inversion model, and the initial set of manufacturable process paths is output. Define a set of multi-objective optimization functions, and set the brittleness threshold and the carbide network degree threshold as infeasible solution criteria; A quantum state encoding method is used to initialize the initial set of manufacturable process paths, constructing a quantum state gray wolf population, and combining it with an adaptive step size strategy to form a quantum adaptive gray wolf optimization algorithm framework. In the quantum adaptive gray wolf optimization algorithm framework, a multi-objective optimization function set is used as the driving force to iteratively optimize the quantum state gray wolf population and generate a candidate morphology control solution set. A service-state morphology evolution proxy model is constructed. The candidate morphology regulation solution set is input into the service-state morphology evolution proxy model to evaluate the stability of secondary carbide precipitation and morphology evolution during service, and output the service evaluation result set. The service evaluation result set is fed back to the quantum adaptive gray wolf optimization algorithm framework to correct the search direction of the quantum state gray wolf population. The steps are repeated until convergence to obtain the Pareto optimal morphology control solution set. The final set of process schemes is obtained by screening within the Pareto optimal morphology control solution set.

[0007] Optionally, the construction of a multi-dimensional index system oriented towards the morphology and service performance of the target carbide includes: Construct a target vector of carbide morphology indices, which consists of carbide size distribution index, carbide aspect ratio index, carbide connectivity index, carbide orientation dispersion index, and carbide volume fraction index. Construct a service performance target vector composed of wear rate index, spalling probability index, and crack initiation threshold index; The target vectors for carbide morphology and service performance are normalized respectively, and each index value is mapped to a target index system vector with uniform dimensions. The normalized target index system vector is used as the target index system.

[0008] Optionally, the high-chromium cast iron historical manufacturing dataset includes alloy composition, process parameters, and service morphology observation data, and the unified preprocessing includes missing value completion, outlier removal, and normalization.

[0009] Optionally, the construction of the composition thermodynamic inversion model includes: Based on a multi-element phase diagram database, an equilibrium phase stability mapping matrix is ​​established. Based on the non-equilibrium solidification kinetics model, a non-equilibrium solidification mapping matrix is ​​constructed. The equilibrium phase stability mapping matrix and the non-equilibrium solidification mapping matrix are fused to construct the composition thermodynamic inversion kernel function, and the predicted target index system vector is output. The target error function is constructed using the Euclidean norm of the difference between the predicted target indicator system vector and the normalized target indicator system vector. The inverse problem is solved by using the objective error function as the optimization objective and the preprocessed high-chromium cast iron historical manufacturing dataset as the constraint condition, and the initial set of manufacturable process paths is obtained.

[0010] Optionally, the multi-objective optimization function set includes: The objective function for wear resistance is defined as the wear rate index as a function of the alloy composition vector, solidification cooling rate, and heat treatment temperature-time curve. Based on the crack initiation threshold index, the reciprocal of the crack initiation threshold index is calculated, and the reciprocal is used as the toughness objective function of the alloy composition vector, solidification cooling rate, and heat treatment temperature-time curve. The cost objective function is obtained by summing the products of the mass fraction of all elements and their corresponding unit proportion cost coefficients, adding the product of the unit cooling rate cost coefficient and the solidification cooling rate, adding the unit temperature-time integral cost coefficient and the integral of the heat treatment temperature-time curve over the total heat treatment time, and normalizing with the benchmark cost constant. A solution is considered feasible only when the spalling probability index is less than or equal to the spalling probability index threshold, the carbide connectivity index is less than or equal to the carbide connectivity index threshold, and the crack initiation threshold index is greater than or equal to the crack initiation threshold index threshold. If any inequality is not met, the solution is considered infeasible. A multi-objective optimization function set is constructed, which is composed of wear resistance objective function, toughness objective function and cost objective function weighted together. The feasible region is defined as the combination set of alloy composition vector, solidification cooling rate and heat treatment temperature-time curve that satisfy the manufacturability boundary constraint set and infeasibility solution criterion.

[0011] Optionally, the construction of the quantum-state gray wolf population includes: A quantum state encoding space is established within the parameter range of the initial set of manufacturable process paths, and a corresponding qubit pair is set for each variable to be optimized; Based on the parameter boundaries in the initial set of manufacturable process paths, the mass fraction range, solidification cooling rate range, and heat treatment temperature-time curve set of each element are read, and the mapping relationship from quantum state to parameter space is defined. Set the number of individual quantum state gray wolves, and initialize all qubit pairs by uniformly sampling in the interval [0,π / 2] using angle parameters, so that the complex amplitude of each qubit pair of all individual quantum state gray wolves is composed of the cosine and sine values ​​of the angle, forming a set of quantum registers; For each individual gray wolf in a quantum state, the initial decision vector is obtained by decoding the mapping relationship from the quantum state to the parameter space; The composition thermodynamic inversion kernel function is used to perform forward calculation on each initial decision vector to obtain the prediction target index system vector corresponding to each quantum state gray wolf individual; Each initial decision vector is screened using infeasibility criteria and manufacturability boundary constraints to determine whether each quantum state gray wolf individual meets the feasibility requirements. If a quantum state gray wolf individual does not meet the feasibility requirements, it is re-initialized until the feasibility requirements are met, so that all initial quantum state gray wolf populations are located in the feasible region. Based on the normalized target index system vector and the predicted target index system vector of each quantum state gray wolf individual, the initial target deviation metric of each quantum state gray wolf individual is calculated, and the initial value of the adaptive step size is calculated by dividing the maximum step size by one and adding the product of the step size sensitivity coefficient and the mean of the initial target deviation. The update operator of the quantum adaptive gray wolf optimization algorithm framework is defined. It generates two random vectors and constructs a dimensionless coefficient vector with an adaptive step size and a linear combination of the random vectors. The complex amplitude of each qubit pair is updated with a rotation angle increment.

[0012] Optionally, the iterative optimization of the quantum-state gray wolf population includes: Taking the initial quantum state gray wolf population as the starting point of the iteration, for the initial decision vector of each individual quantum state gray wolf, the prediction target index system vector of the individual is calculated sequentially using the compositional thermodynamic inversion kernel function; Based on the difference between the predicted target index system vector and the normalized target index system vector of each quantum state gray wolf individual, a multi-target deviation vector is obtained. Each component of the multi-target deviation vector is normalized to obtain a normalized multi-target error vector. Based on the normalized multi-objective error vector of each individual quantum-state gray wolf, the Pareto non-dominated sorting method is used to sort the current quantum-state gray wolf population: If there are two quantum-state gray wolf individuals, denoted as the first quantum-state gray wolf individual and the second quantum-state gray wolf individual, if the normalized error vector of the first quantum-state gray wolf individual is less than or equal to that of the second quantum-state gray wolf individual in all index dimensions, and the normalized error vector in at least one index dimension is strictly less than that of the second quantum-state gray wolf individual, then it is determined that the first quantum-state gray wolf individual dominates the second quantum-state gray wolf individual. After each round of iteration, all quantum-state gray wolf individuals that are not dominated by other quantum-state gray wolf individuals are collected into a set of non-dominant front individuals. The individual with the smallest multi-objective deviation metric between the non-dominated frontier individual prediction target indicator system vector and the normalized target indicator system vector is taken as the leader gray wolf individual group. The optimal solution, the second-best solution, and the third-best solution are selected respectively and denoted as the first leader gray wolf individual, the second leader gray wolf individual, and the third leader gray wolf individual.

[0013] For each individual gray wolf in the current population, excluding the leader gray wolf, its quantum state is updated using a rotational update rule. For each updated individual gray wolf in quantum state, a new round of decision vector is decoded based on the mapping method from quantum state to parameter space. The new round of decision vector is then recalculated using the composition thermodynamic inversion kernel function to predict the target index system vector, error vector, and multi-target deviation metric. Repeat the above steps for a specified number of iterations or until the multi-objective deviation metric values ​​of all quantum-state gray wolf individuals converge in the dimension of the target indicator system. Finally, merge all non-dominated frontier individuals in each iteration as a candidate morphology regulation solution set.

[0014] Optionally, the candidate morphology regulation solution set can be input into the service-state morphology evolution surrogate model to evaluate the stability of secondary carbide precipitation and morphology evolution during service, including: A proxy model for the evolution of service morphology was constructed using a coupled structure of multilayer perceptron and recurrent neural network. Each candidate topology control solution in the candidate topology control solution set is represented as an input parameter triplet. The input parameter triplet is combined with the preset service stress-temperature field boundary conditions to form an input tensor. Each input tensor is fed into the trained service morphology evolution proxy model, and the predicted carbide morphology evolution trajectory is output. By setting a stability criterion for morphological evolution and constructing a service stability error function, the service stability error function values ​​of candidate morphological control solutions are obtained. The service stability error function value of each candidate morphology control solution is compared with a preset stability threshold. When the service stability error function value is less than or equal to the preset stability threshold, the corresponding candidate morphology control solution is determined to be morphologically stable during service and is output to the service evaluation result set. When the service stability error function value is greater than the preset stability threshold, the corresponding candidate morphology control solution is determined to not meet the morphology stability requirements during service and is removed. All candidate morphology control solutions that satisfy the service stability criterion are recorded sequentially as the service evaluation result set.

[0015] Optionally, the repeated steps until convergence to obtain the Pareto optimal topography control solution set include: The service stability error function value corresponding to each candidate morphology control solution in the service evaluation result set is combined with its predicted target index system vector to construct a set of enhanced target index system vectors. The enhanced target indicator system vector set is used as the new evaluation basis to reconstruct the normalized enhanced target indicator system vector. The enhancement deviation vector of each quantum-state gray wolf individual is calculated by normalizing the enhancement target index system vector. The square of the enhancement deviation vector is multiplied by the weight factor by weighted summation. The sum of the squared differences of all weighted summations is used to obtain the enhancement target function value of the quantum-state gray wolf individual. The rotation angle weight coefficient of each individual gray wolf in the quantum state update process is dynamically adjusted based on the enhancement objective function value of each individual gray wolf in the quantum state. The recalculated rotation angle increment is applied to the update of the qubit pairs to obtain the corrected quantum state gray wolf population. A new round of optimization iteration is completed through qubit pair decoding, prediction, sorting and Pareto non-dominated screening process. In the new round of optimization iteration, the service stability error function value is used as the convergence monitoring index, and a global convergence criterion is set: if the difference between the maximum and minimum values ​​of the service stability error function values ​​of all Pareto non-dominated front individuals is less than the set stability threshold within a specified number of consecutive rounds, the algorithm is judged to have converged, and the iterative process of the quantum adaptive gray wolf optimization algorithm is terminated. After the convergence condition is met, the parameter triplet corresponding to all Pareto non-dominated front individuals in the last iteration is taken as the Pareto optimal morphology control solution set.

[0016] Optionally, the filtering within the Pareto optimal morphology control solution set includes: For each Pareto optimal topography control solution, a repeatability score is calculated. A comprehensive cost score is calculated for each Pareto optimal topography control solution; A dual-objective scoring function is constructed based on manufacturing repeatability score and cost comprehensive score. All Pareto optimal morphology control solutions are sorted, and the top K solutions with the highest scores are selected to form the final process scheme set.

[0017] The beneficial effects of this invention are: This invention is based on a deep coupling mechanism of quantum adaptive gray wolf optimization and compositional thermodynamic inversion. It integrates the ternary constraint of "carbide morphology-composition-process" into the optimization process. By constructing a multi-component phase diagram and non-equilibrium solidification kinetic mapping, integrating thermodynamic equilibrium and non-equilibrium multi-scale information, and using the target error drive of the inversion kernel function, it can back-derive manufacturable and repeatable high-chromium cast iron alloy formulation window, solidification cooling path and heat treatment parameters according to actual engineering needs. It significantly improves the controllability and cross-batch stability of multi-dimensional indicators such as target carbide size distribution, connectivity and volume fraction, reduces the risk of performance drift between batches, and adapts to the fine control requirements under complex service conditions.

[0018] This invention introduces a gray wolf optimization algorithm based on quantum state encoding and an adaptive step size mechanism. By constructing objective functions for wear resistance, toughness, and cost, as well as manufacturing feasibility boundaries, brittleness thresholds, and connectivity thresholds, and combining Pareto non-dominated sorting, it solves the problems of existing methods that are prone to getting trapped in local optima due to fixed weights or single objectives, performance trade-offs, and high risk of brittle fracture. The improved quantum gray wolf optimization algorithm can efficiently search for multi-objective optimal solutions in the global-local space, achieving simultaneous improvement in wear resistance and toughness. It effectively suppresses the probability of spalling and early failure caused by coarse connectivity and networking of eutectic carbides, and significantly improves the overall reliability and economy of the process solution. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a method for controlling the carbide morphology of high-chromium cast iron wear layer based on gray wolf optimization proposed in this invention; Figure 2 The flowchart shows the quantum adaptive gray wolf optimization algorithm for a method to control the carbide morphology of high-chromium cast iron wear layer based on gray wolf optimization proposed in this invention. Detailed Implementation

[0020] Example 1: Reference Figures 1-2 A method for controlling the carbide morphology of high-chromium cast iron wear layer based on gray wolf optimization includes: A multi-dimensional index system oriented towards the morphology and service performance of target carbides is constructed, and a target index system is generated. In this embodiment, a multi-dimensional index system oriented towards the morphology and service performance of the target carbide is constructed, including: Construct a target vector of carbide morphology indices, which consists of carbide size distribution index, carbide aspect ratio index, carbide connectivity index, carbide orientation dispersion index, and carbide volume fraction index. The carbide size distribution index describes the target mean and target variance of carbide particle size; the carbide aspect ratio index describes the ratio between the major axis and the minor axis of the carbide; the carbide connectivity index describes the spatial proportion when carbides form a network connection structure; the carbide orientation dispersion index describes the variance of the distribution direction of carbides in the wear layer; and the carbide volume fraction index describes the ratio between the total volume of carbides and the total volume of the wear layer.

[0021] Construct a service performance target vector composed of wear rate index, spalling probability index, and crack initiation threshold index; The service performance target vector is used to characterize the wear resistance, spalling resistance and crack resistance that high-chromium cast iron is expected to achieve during service.

[0022] The target vectors for carbide morphology and service performance are normalized respectively, and each index value is mapped to a target index system vector with uniform dimensions. The normalized target index system vector is used as the target index system.

[0023] The normalized target index system vector is composed of the normalized carbide size distribution index, carbide aspect ratio index, carbide connectivity index, carbide orientation dispersion index, carbide volume fraction index, wear rate index, spalling probability index, and crack initiation threshold index.

[0024] A historical manufacturing dataset of high-chromium cast iron was collected and preprocessed to obtain a preprocessed historical manufacturing dataset of high-chromium cast iron. The historical manufacturing dataset of high-chromium cast iron includes alloy composition, process parameters and service morphology observation data. The unified preprocessing includes missing value completion, outlier removal and normalization.

[0025] Based on a multi-component phase diagram database and a non-equilibrium solidification kinetics model, a composition thermodynamic inversion model is constructed. The target index system and the pre-processed historical manufacturing dataset of high-chromium cast iron are input into the composition thermodynamic inversion model, and the initial set of manufacturable process paths is output. In this embodiment, the composition thermodynamic inversion model is constructed, including: Based on a multi-element phase diagram database, an equilibrium phase stability mapping matrix is ​​established. In Example 1, the full combination composition range of relevant elements in high-chromium cast iron is enumerated based on a multi-component phase diagram database. Under all possible combinations of alloy composition vector and temperature range for each combination, the multi-component phase diagram database is called to calculate the equilibrium phase for the corresponding combination of alloy composition and temperature conditions. The equilibrium phase fractions of each metal phase under each combination of alloy composition vector and temperature conditions are recorded sequentially in the equilibrium phase stability mapping matrix. Each data in the equilibrium phase stability mapping matrix consists of an alloy composition vector, a temperature range, and the corresponding equilibrium fraction of each phase. All data are obtained by traversing all combinations of composition vectors and temperature ranges, completely covering the target composition range and service temperature range of high-chromium cast iron.

[0026] Based on the non-equilibrium solidification kinetics model, a non-equilibrium solidification mapping matrix is ​​constructed. In Example 1, each alloy composition vector in the historical manufacturing dataset of high-chromium cast iron is used as an input variable. Combined with preset solidification cooling rate and time parameters, the microstructure transformation process under each cooling rate and time condition is simulated sequentially according to the non-equilibrium solidification kinetics model. The changing trend of the main phase fraction at each time point and under each cooling rate condition is recorded, forming a mapping relationship with alloy composition vector, solidification cooling rate and time as multi-dimensional inputs and main phase fraction as output. The phase fraction represents the volume fraction ratio of the microstructure at each stage.

[0027] The equilibrium phase stability mapping matrix and the non-equilibrium solidification mapping matrix are fused to construct the composition thermodynamic inversion kernel function, and the predicted target index system vector is output. In Example 1, based on the selected alloy composition vector within a specified temperature range, the corresponding phase fraction data are retrieved using the equilibrium phase stability mapping matrix. Under the same alloy composition vector and corresponding cooling rate and time conditions, the phase fraction change data during the solidification process are obtained using the non-equilibrium solidification mapping matrix. For the same alloy composition vector and cooling path parameters, the phase fraction data output by the equilibrium phase stability mapping matrix and the phase fraction change data output by the non-equilibrium solidification mapping matrix are weighted and fused accordingly to achieve complementary integration of multi-scale information under thermodynamic equilibrium and non-equilibrium solidification states. The weighted and fused phase fraction information is input into the composition thermodynamic inversion kernel function. Based on the given alloy composition vector, solidification cooling rate, and heat treatment temperature-time curve, the composition thermodynamic inversion kernel function outputs a predicted target index system vector. Each term of the predicted target index system vector is an index value obtained after multi-source phase fraction fusion and normalization. Each index value represents the comprehensive prediction result of carbide morphology and service performance under alloy composition and process parameters.

[0028] The target error function is constructed using the Euclidean norm of the difference between the predicted target indicator system vector and the normalized target indicator system vector. The inverse problem is solved by using the objective error function as the optimization objective and the preprocessed high-chromium cast iron historical manufacturing dataset as the constraint condition, and the initial set of manufacturable process paths is obtained.

[0029] In Example 1, the target error function is used as the optimization objective. The Euclidean norm between the predicted target index system vector and the normalized target index system vector is used as the evaluation criterion for optimization. Combined with the preprocessed historical manufacturing dataset of high-chromium cast iron, the manufacturability boundary constraint set is determined. The manufacturability boundary constraint set includes the allowable range of mass fraction of each element, the maximum allowable solidification cooling rate, and the highest and lowest heat treatment temperature and time interval that the equipment can achieve. Under the combined action of the target error function and the manufacturability boundary constraint set, the search is conducted in all feasible parameter combination spaces to minimize the target error function. A set of parameter combinations that minimizes the error between the predicted target index system vector and the normalized target index system vector and that all parameters satisfy the manufacturability boundary constraint set is obtained. All parameter combinations that meet the conditions are recorded in sequence to obtain the initial set of manufacturable process paths.

[0030] Define a set of multi-objective optimization functions, and set the brittleness threshold and the carbide network degree threshold as infeasible solution criteria; In this embodiment, the multi-objective optimization function set includes: The objective function for wear resistance is defined as the wear rate index as a function of the alloy composition vector, solidification cooling rate, and heat treatment temperature-time curve. In Example 1, the wear resistance objective function takes the alloy composition vector, solidification cooling rate and heat treatment temperature-time curve as inputs, obtains the predicted target index system vector through the composition thermodynamic inversion kernel function, and extracts the wear rate index in the vector as the output of the wear resistance objective function.

[0031] Based on the crack initiation threshold index, the reciprocal of the crack initiation threshold index is calculated, and the reciprocal is used as the toughness objective function of the alloy composition vector, solidification cooling rate, and heat treatment temperature-time curve. The cost objective function is obtained by summing the products of the mass fraction of all elements and their corresponding unit proportion cost coefficients, adding the product of the unit cooling rate cost coefficient and the solidification cooling rate, adding the unit temperature-time integral cost coefficient and the integral of the heat treatment temperature-time curve over the total heat treatment time, and normalizing with the benchmark cost constant. A solution is considered feasible only when the spalling probability index is less than or equal to the spalling probability index threshold, the carbide connectivity index is less than or equal to the carbide connectivity index threshold, and the crack initiation threshold index is greater than or equal to the crack initiation threshold index threshold. If any inequality is not met, the solution is considered infeasible. A multi-objective optimization function set is constructed, which is composed of wear resistance objective function, toughness objective function and cost objective function weighted together. The feasible region is defined as the combination set of alloy composition vector, solidification cooling rate and heat treatment temperature-time curve that satisfy the manufacturability boundary constraint set and infeasibility solution criterion.

[0032] The manufacturability boundary constraints set includes the allowable range of mass fraction for each element, the maximum allowable solidification cooling rate, and the highest and lowest heat treatment temperature and time range that the equipment can achieve.

[0033] A quantum state encoding method is used to initialize the initial set of manufacturable process paths, constructing a quantum state gray wolf population, and combining it with an adaptive step size strategy to form a quantum adaptive gray wolf optimization algorithm framework. In this embodiment, constructing a quantum-state gray wolf population includes: A quantum state encoding space is established within the parameter range of the initial set of manufacturable process paths, and a corresponding qubit pair is set for each variable to be optimized; Each pair of qubits is represented by a pair of complex amplitude parameters, and the sum of the squares of the moduli of the pair of complex amplitude parameters is equal to 1.

[0034] Based on the parameter boundaries in the initial set of manufacturable process paths, the mass fraction range, solidification cooling rate range, and heat treatment temperature-time curve set of each element are read, and the mapping relationship from quantum state to parameter space is defined. In Example 1, the minimum and maximum allowable mass fractions of each alloy component are extracted based on the parameter boundaries in the initial set of manufacturable process paths, and are denoted as the upper and lower limits of the mass fraction interval for each component. The minimum and maximum allowable values ​​of the solidification cooling rate are also extracted and denoted as the upper and lower limits of the solidification cooling rate interval. All feasible heat treatment temperature-time curves are compiled to form a set of heat treatment temperature-time curves. For each parameter of each individual in the quantum-state gray wolf population, the square of the modulus of the first complex amplitude of the qubit pair corresponding to the parameter is taken as a dimensionless value. The dimensionless value is mapped in the linear interval to the actual value interval of the parameter, specifically: For alloy components, the dimensionless value is multiplied by the span of the mass fraction range and then added to the lower limit of the mass fraction to obtain the actual mass fraction value of the component. For the solidification cooling rate, the dimensionless value is multiplied by the span of the solidification cooling rate range and then added to the lower limit to obtain the actual solidification cooling rate value. For the heat treatment temperature-time curve, multiply the dimensionless value by the number of curves in the heat treatment temperature-time curve set minus one, round it to the nearest integer, and then select one curve from the set to obtain the actual heat treatment temperature-time curve value.

[0035] Set the number of individual quantum state gray wolves, and initialize all qubit pairs by uniformly sampling in the interval [0,π / 2] using angle parameters, so that the complex amplitude of each qubit pair of all individual quantum state gray wolves is composed of the cosine and sine values ​​of the angle, forming a set of quantum registers; Each set of quantum registers corresponds to a single individual gray wolf in a quantum state; For each individual gray wolf in a quantum state, the initial decision vector is obtained by decoding the mapping relationship from the quantum state to the parameter space; The initial decision vector includes a set of alloy composition components, a solidification cooling rate value, and a heat treatment temperature-time curve value.

[0036] The composition thermodynamic inversion kernel function is used to perform forward calculation on each initial decision vector to obtain the prediction target index system vector corresponding to each quantum state gray wolf individual; Each initial decision vector is screened using infeasibility criteria and manufacturability boundary constraints to determine whether each quantum state gray wolf individual meets the feasibility requirements. If a quantum state gray wolf individual does not meet the feasibility requirements, it is re-initialized until the feasibility requirements are met, so that all initial quantum state gray wolf populations are located in the feasible region. Based on the normalized target index system vector and the predicted target index system vector of each quantum state gray wolf individual, the initial target deviation metric of each quantum state gray wolf individual is calculated, and the initial value of the adaptive step size is calculated by dividing the maximum step size by one and adding the product of the step size sensitivity coefficient and the mean of the initial target deviation. The initial target deviation metric is equal to the sum of the squares of the differences between the corresponding components of the individual's predicted target indicator system vector and the normalized target indicator system vector.

[0037] The update operator of the quantum adaptive gray wolf optimization algorithm framework is defined. It generates two random vectors and constructs a dimensionless coefficient vector with an adaptive step size and a linear combination of the random vectors. The complex amplitude of each qubit pair is updated with a rotation angle increment.

[0038] The rotation angle increment is equal to the product of the rotation gain coefficient, the adaptive step size, and the dimensionless direction weight coefficient. The dimensionless direction weight coefficient is dynamically set based on the quantum state gray wolf leader individual information.

[0039] The updated complex amplitude pair is obtained by calculating the rotation angle increment and the original complex amplitude pair using a two-dimensional rotation matrix.

[0040] The quantum register set, initial decoding population, adaptive step size initial value, and update operator are used together as the initialization output of the quantum adaptive gray wolf optimization algorithm framework.

[0041] In the quantum adaptive gray wolf optimization algorithm framework, a multi-objective optimization function set is used as the driving force to iteratively optimize the quantum state gray wolf population and generate a candidate morphology control solution set. In this embodiment, iterative optimization of the quantum state gray wolf population includes: Taking the initial quantum state gray wolf population as the starting point of the iteration, for the initial decision vector of each individual quantum state gray wolf, the prediction target index system vector of the individual is calculated sequentially using the compositional thermodynamic inversion kernel function; Based on the difference between the predicted target index system vector and the normalized target index system vector of each quantum state gray wolf individual, a multi-target deviation vector is obtained. Each component of the multi-target deviation vector is normalized to obtain a normalized multi-target error vector. Based on the normalized multi-objective error vector of each individual quantum-state gray wolf, the Pareto non-dominated sorting method is used to sort the current quantum-state gray wolf population: If there are two quantum-state gray wolf individuals, denoted as the first quantum-state gray wolf individual and the second quantum-state gray wolf individual, if the normalized error vector of the first quantum-state gray wolf individual is less than or equal to that of the second quantum-state gray wolf individual in all index dimensions, and the normalized error vector in at least one index dimension is strictly less than that of the second quantum-state gray wolf individual, then it is determined that the first quantum-state gray wolf individual dominates the second quantum-state gray wolf individual. After each round of iteration, all quantum-state gray wolf individuals that are not dominated by other quantum-state gray wolf individuals are collected into a set of non-dominant front individuals. The individual with the smallest multi-objective deviation metric between the non-dominated frontier individual prediction target indicator system vector and the normalized target indicator system vector is taken as the leader gray wolf individual group. The optimal solution, the second-best solution, and the third-best solution are selected respectively and denoted as the first leader gray wolf individual, the second leader gray wolf individual, and the third leader gray wolf individual.

[0042] For each individual gray wolf in the current population, excluding the leader gray wolf, its quantum state is updated using a rotational update rule. Each parameter to be optimized is uniquely represented by a pair of complex amplitude parameters. The complex amplitude parameter pair can be equivalently understood as a two-dimensional vector. The rotation update process is as follows: calculate the rotation angle increment of the parameter. The rotation angle increment is equal to the product of the rotation gain coefficient, the current iteration step size, and the dimensionless direction weight coefficient. Perform a two-dimensional rotation on the complex amplitude parameter pair using the rotation angle increment to obtain the updated complex amplitude parameter pair. All parameter updates maintain the same mapping relationship with the original parameter space.

[0043] For each updated individual gray wolf in quantum state, a new round of decision vector is decoded based on the mapping method from quantum state to parameter space. The new round of decision vector is then recalculated using the composition thermodynamic inversion kernel function to predict the target index system vector, error vector, and multi-target deviation metric. Repeat the above steps for a specified number of iterations or until the multi-objective deviation metric values ​​of all quantum-state gray wolf individuals converge in the dimension of the target indicator system. Finally, merge all non-dominated frontier individuals in each iteration as a candidate morphology regulation solution set.

[0044] A service-state morphology evolution proxy model is constructed. The candidate morphology regulation solution set is input into the service-state morphology evolution proxy model to evaluate the stability of secondary carbide precipitation and morphology evolution during service, and output the service evaluation result set. In this embodiment, the candidate morphology regulation solution set is input into the service-state morphology evolution proxy model to evaluate the stability of secondary carbide precipitation and morphology evolution during service, including: A proxy model for the evolution of service morphology was constructed using a coupled structure of multilayer perceptron and recurrent neural network. A training dataset is constructed based on historical service morphology evolution data. This data includes time-series data of carbide morphology indicators, service stress field data, service temperature field data, and wear mode data collected within a specific service cycle. The historical service morphology evolution data is processed in a unified format to form a service state input tensor set and a corresponding morphology response tensor set. The service state input tensor set represents the service condition input information of each sample at each time point, including alloy composition vector, service temperature, external load, and service time. The morphology response tensor set represents the carbide morphology response information of each sample at each time point, including particle size, volume fraction, and connectivity morphology evolution indicators.

[0045] The service status input tensor set is used as the input to the service status morphology evolution surrogate model, and the corresponding morphology response tensor set is used as the output of the service status morphology evolution surrogate model. The service status morphology evolution surrogate model is trained using supervised learning. By continuously adjusting the internal parameters of the service status morphology evolution surrogate model, the mean square error between the morphology response tensor set output by the service status morphology evolution surrogate model and the actual morphology response tensor set in the historical service status morphology evolution data is minimized. The mean square error is equal to the sum of the squares of the differences between the service status morphology evolution surrogate model output and the actual observed data at all samples and all time points, and then the average value is taken. The training is iterated until convergence, and the service status morphology evolution surrogate model with the best fitting ability is obtained.

[0046] Each candidate topology control solution in the candidate topology control solution set is represented as an input parameter triplet. The input parameter triplet is combined with the preset service stress-temperature field boundary conditions to form an input tensor. The input parameter ternary set includes the alloy composition vector, solidification cooling rate, and heat treatment temperature-time curve.

[0047] Each input tensor is fed into the trained service morphology evolution proxy model, and the predicted carbide morphology evolution trajectory is output. The predicted carbide morphology evolution trajectory includes the predicted particle size, predicted volume fraction, and predicted connectivity at each point in time throughout the entire service life.

[0048] By setting a stability criterion for morphological evolution and constructing a service stability error function, the service stability error function values ​​of candidate morphological control solutions are obtained. The service stability error function is equal to the sum of the absolute values ​​of the differences between the predicted particle size and the factory particle size, the absolute values ​​of the differences between the predicted volume fraction and the factory volume fraction, and the absolute values ​​of the differences between the predicted connectivity and the factory connectivity at each time point throughout the entire service cycle, multiplied by their respective weighting coefficients. The average of the weighted sums at all time points is then calculated to obtain the service stability error function value of the candidate morphology control solution.

[0049] The service stability error function value of each candidate morphology control solution is compared with a preset stability threshold. When the service stability error function value is less than or equal to the preset stability threshold, the corresponding candidate morphology control solution is determined to be morphologically stable during service and is output to the service evaluation result set. When the service stability error function value is greater than the preset stability threshold, the corresponding candidate morphology control solution is determined to not meet the morphology stability requirements during service and is removed. All candidate morphology control solutions that satisfy the service stability criterion are recorded sequentially as the service evaluation result set.

[0050] S8. Feed the service evaluation result set back to the quantum adaptive gray wolf optimization algorithm framework to correct the search direction of the quantum state gray wolf population, and repeat the steps until convergence to obtain the Pareto optimal morphology control solution set; In this embodiment, the steps are repeated until convergence to obtain the Pareto optimal topography control solution set, including: The service stability error function value corresponding to each candidate morphology control solution in the service evaluation result set is combined with its predicted target index system vector to construct a set of enhanced target index system vectors. The enhanced target indicator system vector is based on the original predicted target indicator system vector, with the addition of the service stability error function value as a new dimension, and combined into a multi-dimensional vector. The multi-dimensional vector includes, in order, carbide size distribution index, carbide aspect ratio index, carbide connectivity index, carbide orientation dispersion index, carbide volume fraction index, wear rate index, spalling probability index, crack initiation threshold index, and service stability error function value.

[0051] The enhanced target indicator system vector set is used as the new evaluation basis to reconstruct the normalized enhanced target indicator system vector. The enhancement deviation vector of each quantum-state gray wolf individual is calculated by normalizing the enhancement target index system vector. The square of the enhancement deviation vector is multiplied by the weight factor by weighted summation. The sum of the squared differences of all weighted summations is used to obtain the enhancement target function value of the quantum-state gray wolf individual. The reinforcement bias vector is a multidimensional vector composed of the differences between the reinforcement target index system vector of the quantum state gray wolf individual and the normalized reinforcement target index system vector in each index dimension.

[0052] The rotation angle weight coefficient of each individual gray wolf in the quantum state update process is dynamically adjusted based on the enhancement objective function value of each individual gray wolf in the quantum state. Specifically, the rotation gain coefficient is multiplied by the current iteration step size and the ratio of the quantum state gray wolf individual enhancement objective function value to the average value of the current population enhancement objective function value to obtain the rotation angle increment. Individuals with larger quantum state gray wolf individual enhancement objective function values ​​obtain larger update angles.

[0053] The recalculated rotation angle increment is applied to the update of the qubit pairs to obtain the corrected quantum state gray wolf population. A new round of optimization iteration is completed through qubit pair decoding, prediction, sorting and Pareto non-dominated screening process. In the new round of optimization iteration, the service stability error function value is used as the convergence monitoring index, and a global convergence criterion is set: if the difference between the maximum and minimum values ​​of the service stability error function values ​​of all Pareto non-dominated front individuals is less than the set stability threshold within a specified number of consecutive rounds, the algorithm is judged to have converged, and the iterative process of the quantum adaptive gray wolf optimization algorithm is terminated. After the convergence condition is met, the parameter triplet corresponding to all Pareto non-dominated front individuals in the last iteration is taken as the Pareto optimal morphology control solution set.

[0054] The final set of process schemes is obtained by screening within the Pareto optimal morphology control solution set.

[0055] In this embodiment, the selection within the Pareto optimal morphology control solution set includes: For each Pareto optimal topography control solution, a repeatability score is calculated. Manufacturing repeatability scores are obtained by comprehensively evaluating the following factors: a) Manufacturing stability coefficient of each element in the alloy composition vector; b) The matching coefficient between the solidification cooling rate and the cooling capacity of the production equipment; c) Temperature control accuracy index of the heat treatment temperature-time curve; d) Consistency score of heat treatment process timing across different batches; The manufacturing repeatability score is calculated by weighted averaging the above scores. The higher the manufacturing repeatability score, the better the repeatability of achieving a consistent morphology under actual process conditions.

[0056] A comprehensive cost score is calculated for each Pareto optimal topography control solution; The overall cost score is calculated based on the following parameters: a) The sum of the products of the unit mass price and mass fraction of each element in the alloy composition vector; b) The product of solidification cooling rate and unit cooling energy consumption coefficient; c) The heat treatment energy consumption integral required for the heat treatment temperature-time curve; d) Energy consumption, material consumption, and resource consumption of auxiliary treatments in surface strengthening process parameters; The sum of the above cost factors is normalized to the benchmark production cost, and its reciprocal is taken as the overall cost score. The higher the overall cost score, the better the economic efficiency of the solution.

[0057] A dual-objective scoring function is constructed based on manufacturing repeatability score and cost comprehensive score. All Pareto optimal morphology control solutions are sorted, and the top K solutions with the highest scores are selected to form the final process scheme set.

[0058] The final process solution set includes: The high-chromium cast iron alloy formulation window consists of the mass fraction of each element and its controllable deviation range of the alloy composition vector corresponding to the Pareto optimal morphology control solution, and is represented as the recommended main element and trace modifier element types, addition ratios and control precision requirements. Solidification cooling path parameters: including recommended cooling rate range, cooling method, and suggestions for controlling cooling rate before, during, and after solidification; Heat treatment curve parameters: including austenitizing temperature, holding time, tempering stage temperature range, time distribution, and cooling rate range for each stage; Surface strengthening process parameters: including whether laser hardening, shot peening, and plasma treatment technology combinations are applicable, and the energy input, effective area, application sequence, and execution window for each process.

[0059] Example 2: During the development of a new high-chromium cast iron wear-resistant part, the implementation team collected data from the most recent 50 batches of production samples. They planned to use the method of this invention to achieve coordinated and precise control of carbide morphology and service performance, and compare the actual performance with traditional processes. The specific process is as follows: The team defined the engineering objectives for this batch of products, including: the average particle size of carbides is controlled at 3.5~5.0μm, with a variance of no more than 1.1μm²; the connectivity is no higher than 0.17; the volume fraction is no less than 22%; the average wear rate during service life is less than 1.4mg / cm²·h, and the peeling probability is less than 4%.

[0060] The team extracted 500 training samples of high-chromium cast iron formulation, process, and service performance from a historical database. In Example 2, the data for sample S101 is as follows: Alloy composition (mass fraction, %): C 2.56, Cr 22.8, Mo 0.60, V 0.12, rare earth 0.15, Mn 0.62, Si 1.00; cooling rate: 22 K / s; heat treatment parameters: austenitizing temperature 1003℃, holding time 2.1h, tempering temperature 320℃, tempering time 3.5h; particle size 4.09μm, variance 0.92μm², connectivity 0.15, volume fraction 22.7%, wear rate 1.28 mg / cm²·h, spalling probability 3.1%; Based on batch samples, the implementation team adopted normalization processing, mapping indicators such as particle size, connectivity, volume fraction, wear rate, and peeling probability to the [0,1] interval. In Example 2, the target indicator vector of sample S101 after normalization is: [0.41, 0.37, 0.53, 0.32, 0.19].

[0061] In the historical data, missing items were removed, and the units of all elements were standardized. For example, all cooling rates were expressed in K / s, heat treatment time was standardized in hours, and temperature was standardized in °C. The original cooling rate of some sample S109 was in °C / min, which was converted to 21.5 K / s after conversion. The data was then standardized and entered into the analysis library.

[0062] The team used the CALPHAD database and their self-developed non-equilibrium solidification simulation model to calculate the phase fraction changes for each combination of alloy formulations and process paths in all historical data. In Example 2, for an alloy with C 2.60%, Cr 22.9%, Mo 0.62%, V 0.13%, and rare earth elements 0.16%, at a cooling rate of 24 K / s, an austenitizing temperature of 1008℃, a holding time of 2.0 h, a tempering temperature of 330℃, and a tempering time of 3.2 h, the simulation output was as follows: Eutectic carbide volume fraction: 23.0%; average primary carbide particle size after rapid cooling: 4.25 μm; secondary precipitation increase during service life: 0.42 μm; An inversion mapping kernel function is generated, and a total of 115 sets of manufacturable process path parameters are output in batches as the subsequent optimization population.

[0063] For each process path, the following are calculated: wear rate objective function, such as the predicted wear rate of 1.19 mg / cm²·h for process S117; toughness objective function (measured as the reciprocal of the crack initiation threshold), with a corresponding value of 0.00101 MPa⁻¹; cost objective function, such as the overall production cost of S117 / standard baseline = 1.07; connectivity criterion: the predicted connectivity is 0.13, which is lower than 0.17, and is therefore considered feasible.

[0064] Within the parameter space, all cooling rates are mapped to qubit pairs. The initial sample, such as S128 quantum encoding and decoding, yields a cooling rate of 23.4 K / s, a C content of 2.53%, and a Cr content of 22.7%, with a corresponding initial target index deviation of 0.18.

[0065] After 150 iterations of optimizing a quantum gray wolf population (set to 100 individuals), the Pareto non-dominated front converges to 10 solutions. Some data are as follows: S210: Particle size 4.01μm, connectivity 0.13, volume fraction 23.5%, wear rate 1.17mg / cm²·h, spalling probability 2.6%, cost score 0.93.

[0066] S215: Particle size 4.12μm, connectivity 0.14, volume fraction 22.9%, wear rate 1.21mg / cm²·h, spalling probability 2.8%, cost score 0.91.

[0067] Using a pre-trained service morphology evolution proxy model, inputting the triplet parameters of S210 and S215, a service stress of 600 MPa, and a service period of 60 days, the morphology evolution is dynamically predicted: S210: During its service life, the particle size increased by 0.34 μm, the connectivity increased by 0.02, and the volume fraction decreased by 0.6%. S215: Particle size increased by 0.40 μm, connectivity increased by 0.03, and volume fraction decreased by 0.7%; All values ​​are below the set threshold, indicating stable morphology.

[0068] The service stability error function values ​​were 0.017 and 0.020, respectively. After normalization, these values ​​were incorporated into the target index system to adjust the optimization direction. After 20 iterations, the Pareto front no longer fluctuated, and the final solution set was obtained.

[0069] Based on manufacturing repeatability and cost scores, three final process schemes were output, with S210 being the optimal one: formula C 2.55%, Cr 22.8%, Mo 0.60%, V 0.13%, rare earth 0.15%; cooling rate 23.7K / s; austenitizing temperature 1004℃, holding time 2.2h, tempering temperature 325℃, tempering time 3.3h, and surface strengthening using laser quenching with an energy density of 9.0J / mm².

[0070] The implementation team simultaneously applied the optimized process of this invention with a traditional empirical formula (C 2.68%, Cr 23.1%, Mo 0.71%, natural cooling, austenitization at 1010℃ for 2.0 h, tempering at 350℃ for 2.0 h, no surface strengthening) to production trials. Each scheme produced 30 units, and the results were collected after an actual wear test period of 60 days, as shown in Table 1 below. Table 1. Comparison of data between the S210 scheme of the present invention and traditional empirical schemes. Average particle size (pm) 4.01 6.10 Interconnectedness 0.13 0.25 Volume fraction (%) 23.5 20.0 Wear rate (mg / cm2-h) 1.17 1.89 Probability of spalling (%) 2.6 13.8 Service crack initiation threshold (MPa) 951 893 Manufacturing repeatability score 0.94 0.69 Cost score 0.92 0.74 Furthermore, in 50 training samples, the product using the solution of this invention had a significantly lower standard deviation of performance fluctuation during service than the traditional solution, with an average fluctuation of 3.1%, compared to 16.1% for the traditional solution.

[0071] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for controlling the carbide morphology of high-chromium cast iron wear layer based on gray wolf optimization, characterized in that, include: A multi-dimensional index system oriented towards the morphology and service performance of target carbides is constructed, and a target index system is generated. Collect historical manufacturing datasets of high-chromium cast iron and perform unified preprocessing to obtain preprocessed historical manufacturing datasets of high-chromium cast iron. Based on a multi-component phase diagram database and a non-equilibrium solidification kinetics model, a composition thermodynamic inversion model is constructed. The target index system and the pre-processed historical manufacturing dataset of high-chromium cast iron are input into the composition thermodynamic inversion model, and the initial set of manufacturable process paths is output. Define a set of multi-objective optimization functions, and set the brittleness threshold and the carbide network degree threshold as infeasible solution criteria; A quantum state encoding method is used to initialize the initial set of manufacturable process paths, constructing a quantum state gray wolf population, and combining it with an adaptive step size strategy to form a quantum adaptive gray wolf optimization algorithm framework. In the quantum adaptive gray wolf optimization algorithm framework, a multi-objective optimization function set is used as the driving force to iteratively optimize the quantum state gray wolf population and generate a candidate morphology control solution set. A service-state morphology evolution proxy model is constructed. The candidate morphology regulation solution set is input into the service-state morphology evolution proxy model to evaluate the stability of secondary carbide precipitation and morphology evolution during service, and output the service evaluation result set. The service evaluation result set is fed back to the quantum adaptive gray wolf optimization algorithm framework to correct the search direction of the quantum state gray wolf population. The steps are repeated until convergence to obtain the Pareto optimal morphology control solution set. The final set of process schemes is obtained by screening within the Pareto optimal morphology control solution set.

2. The method for controlling the carbide morphology of high-chromium cast iron wear layer based on gray wolf optimization according to claim 1, characterized in that, The construction of a multi-dimensional index system oriented towards target carbide morphology and service performance includes: Construct a target vector of carbide morphology indices, which consists of carbide size distribution index, carbide aspect ratio index, carbide connectivity index, carbide orientation dispersion index, and carbide volume fraction index. Construct a service performance target vector composed of wear rate index, spalling probability index, and crack initiation threshold index; The target vectors for carbide morphology and service performance are normalized respectively, and each index value is mapped to a target index system vector with uniform dimensions. The normalized target index system vector is used as the target index system.

3. The method for controlling the carbide morphology of high-chromium cast iron wear layer based on gray wolf optimization according to claim 1, characterized in that, The high-chromium cast iron historical manufacturing dataset includes alloy composition, process parameters, and service morphology observation data. The unified preprocessing includes missing value completion, outlier removal, and normalization.

4. The method for controlling the carbide morphology of high-chromium cast iron wear layer based on gray wolf optimization according to claim 1, characterized in that, The construction of the composition thermodynamic inversion model includes: Based on a multi-element phase diagram database, an equilibrium phase stability mapping matrix is ​​established. Based on the non-equilibrium solidification kinetics model, a non-equilibrium solidification mapping matrix is ​​constructed. The equilibrium phase stability mapping matrix and the non-equilibrium solidification mapping matrix are fused to construct the composition thermodynamic inversion kernel function, and the predicted target index system vector is output. The target error function is constructed using the Euclidean norm of the difference between the predicted target indicator system vector and the normalized target indicator system vector. The inverse problem is solved by using the objective error function as the optimization objective and the preprocessed high-chromium cast iron historical manufacturing dataset as the constraint condition, and the initial set of manufacturable process paths is obtained.

5. The method for controlling the carbide morphology of high-chromium cast iron wear layer based on gray wolf optimization according to claim 1, characterized in that, The multi-objective optimization function set includes: The objective function for wear resistance is defined as the wear rate index as a function of the alloy composition vector, solidification cooling rate, and heat treatment temperature-time curve. Based on the crack initiation threshold index, the reciprocal of the crack initiation threshold index is calculated, and the reciprocal is used as the toughness objective function of the alloy composition vector, solidification cooling rate, and heat treatment temperature-time curve. The cost objective function is obtained by summing the products of the mass fraction of all elements and their corresponding unit proportion cost coefficients, adding the product of the unit cooling rate cost coefficient and the solidification cooling rate, adding the unit temperature-time integral cost coefficient and the integral of the heat treatment temperature-time curve over the total heat treatment time, and normalizing with the benchmark cost constant. A solution is considered feasible only when the spalling probability index is less than or equal to the spalling probability index threshold, the carbide connectivity index is less than or equal to the carbide connectivity index threshold, and the crack initiation threshold index is greater than or equal to the crack initiation threshold index threshold. If any inequality is not met, the solution is considered infeasible. A multi-objective optimization function set is constructed, which is composed of wear resistance objective function, toughness objective function and cost objective function weighted together. The feasible region is defined as the combination set of alloy composition vector, solidification cooling rate and heat treatment temperature-time curve that satisfy the manufacturability boundary constraint set and infeasibility solution criterion.

6. The method for controlling the carbide morphology of high-chromium cast iron wear layer based on gray wolf optimization according to claim 1, characterized in that, The construction of the quantum state gray wolf population includes: A quantum state encoding space is established within the parameter range of the initial set of manufacturable process paths, and a corresponding qubit pair is set for each variable to be optimized; Based on the parameter boundaries in the initial set of manufacturable process paths, the mass fraction range, solidification cooling rate range, and heat treatment temperature-time curve set of each element are read, and the mapping relationship from quantum state to parameter space is defined. Set the number of individual quantum state gray wolves, and initialize all qubit pairs by uniformly sampling in the interval [0,π / 2] using angle parameters, so that the complex amplitude of each qubit pair of all individual quantum state gray wolves is composed of the cosine and sine values ​​of the angle, forming a set of quantum registers; For each individual gray wolf in a quantum state, the initial decision vector is obtained by decoding the mapping relationship from the quantum state to the parameter space; The composition thermodynamic inversion kernel function is used to perform forward calculation on each initial decision vector to obtain the prediction target index system vector corresponding to each quantum state gray wolf individual; Each initial decision vector is screened using infeasibility criteria and manufacturability boundary constraints to determine whether each quantum state gray wolf individual meets the feasibility requirements. If a quantum state gray wolf individual does not meet the feasibility requirements, it is re-initialized until the feasibility requirements are met, so that all initial quantum state gray wolf populations are located in the feasible region. Based on the normalized target index system vector and the predicted target index system vector of each quantum state gray wolf individual, the initial target deviation metric of each quantum state gray wolf individual is calculated, and the initial value of the adaptive step size is calculated by dividing the maximum step size by one and adding the product of the step size sensitivity coefficient and the mean of the initial target deviation. The update operator of the quantum adaptive gray wolf optimization algorithm framework is defined. It generates two random vectors and constructs a dimensionless coefficient vector with an adaptive step size and a linear combination of the random vectors. The complex amplitude of each qubit pair is updated with a rotation angle increment.

7. The method for controlling the carbide morphology of high-chromium cast iron wear layer based on gray wolf optimization according to claim 1, characterized in that, The iterative optimization of the quantum state gray wolf population includes: Taking the initial quantum state gray wolf population as the starting point of the iteration, for the initial decision vector of each individual quantum state gray wolf, the prediction target index system vector of the individual is calculated sequentially using the compositional thermodynamic inversion kernel function; Based on the difference between the predicted target index system vector and the normalized target index system vector of each quantum state gray wolf individual, a multi-target deviation vector is obtained. Each component of the multi-target deviation vector is normalized to obtain a normalized multi-target error vector. Based on the normalized multi-objective error vector of each individual quantum-state gray wolf, the Pareto non-dominated sorting method is used to sort the current quantum-state gray wolf population: If there are two quantum-state gray wolf individuals, denoted as the first quantum-state gray wolf individual and the second quantum-state gray wolf individual, if the normalized error vector of the first quantum-state gray wolf individual is less than or equal to that of the second quantum-state gray wolf individual in all index dimensions, and the normalized error vector in at least one index dimension is strictly less than that of the second quantum-state gray wolf individual, then it is determined that the first quantum-state gray wolf individual dominates the second quantum-state gray wolf individual. After each round of iteration, all quantum-state gray wolf individuals that are not dominated by other quantum-state gray wolf individuals are collected into a set of non-dominated front individuals. The individual with the smallest multi-objective deviation metric between the non-dominated frontier individual prediction target indicator system vector and the normalized target indicator system vector is taken as the leader gray wolf individual group. The optimal solution, the second-best solution and the third-best solution are selected respectively and denoted as the first leader gray wolf individual, the second leader gray wolf individual and the third leader gray wolf individual. For each individual gray wolf in the current population, excluding the leader gray wolf, its quantum state is updated using a rotational update rule. For each updated individual gray wolf in quantum state, a new round of decision vector is decoded based on the mapping method from quantum state to parameter space. The new round of decision vector is then recalculated using the composition thermodynamic inversion kernel function to predict the target index system vector, error vector, and multi-target deviation metric. Repeat the above steps for a specified number of iterations or until the multi-objective deviation metric values ​​of all quantum-state gray wolf individuals converge in the dimension of the target indicator system. Finally, merge all non-dominated frontier individuals in each round of iteration as a candidate morphology regulation solution set.

8. The method for controlling the carbide morphology of high-chromium cast iron wear layer based on gray wolf optimization according to claim 1, characterized in that, The candidate morphology regulation solution set is input into the service-state morphology evolution proxy model to evaluate the stability of secondary carbide precipitation and morphology evolution during service, including: A proxy model for the evolution of service morphology was constructed using a coupled structure of multilayer perceptron and recurrent neural network. Each candidate topology control solution in the candidate topology control solution set is represented as an input parameter triplet. The input parameter triplet is combined with the preset service stress-temperature field boundary conditions to form an input tensor. Each input tensor is fed into the trained service morphology evolution proxy model, and the predicted carbide morphology evolution trajectory is output. By setting a stability criterion for morphological evolution and constructing a service stability error function, the service stability error function values ​​of candidate morphological control solutions are obtained. The service stability error function value of each candidate morphology control solution is compared with a preset stability threshold. When the service stability error function value is less than or equal to the preset stability threshold, the corresponding candidate morphology control solution is determined to be morphologically stable during service and is output to the service evaluation result set. When the service stability error function value is greater than the preset stability threshold, the corresponding candidate morphology control solution is determined to not meet the morphology stability requirements during service and is removed. All candidate morphology control solutions that satisfy the service stability criterion are recorded sequentially as the service evaluation result set.

9. The method for controlling the carbide morphology of high-chromium cast iron wear layer based on gray wolf optimization according to claim 1, characterized in that, The repeated steps until convergence to obtain the Pareto optimal topography control solution set include: The service stability error function value corresponding to each candidate morphology control solution in the service evaluation result set is combined with its predicted target index system vector to construct a set of enhanced target index system vectors. The enhanced target indicator system vector set is used as the new evaluation basis to reconstruct the normalized enhanced target indicator system vector. The enhancement deviation vector of each quantum-state gray wolf individual is calculated by normalizing the enhancement target index system vector. The square of the enhancement deviation vector is multiplied by the weight factor by weighted summation. The sum of the squared differences of all weighted summations is used to obtain the enhancement target function value of the quantum-state gray wolf individual. The rotation angle weight coefficient of each individual gray wolf in the quantum state update process is dynamically adjusted based on the enhancement objective function value of each individual gray wolf in the quantum state. The recalculated rotation angle increment is applied to the update of the qubit pairs to obtain the corrected quantum state gray wolf population. A new round of optimization iteration is completed through qubit pair decoding, prediction, sorting and Pareto non-dominated screening process. In the new round of optimization iteration, the service stability error function value is used as the convergence monitoring index, and a global convergence criterion is set: if the difference between the maximum and minimum values ​​of the service stability error function values ​​of all Pareto non-dominated front individuals is less than the set stability threshold within a specified number of consecutive rounds, the algorithm is judged to have converged, and the iterative process of the quantum adaptive gray wolf optimization algorithm is terminated. After the convergence condition is met, the parameter triplet corresponding to all Pareto non-dominated front individuals in the last iteration is taken as the Pareto optimal morphology control solution set.

10. The method for controlling the carbide morphology of high-chromium cast iron wear layer based on gray wolf optimization according to claim 1, characterized in that, The selection within the Pareto optimal morphology control solution set includes: For each Pareto optimal topography control solution, a repeatability score is calculated. A comprehensive cost score is calculated for each Pareto optimal topography control solution; A dual-objective scoring function is constructed based on manufacturing repeatability score and cost comprehensive score. All Pareto optimal morphology control solutions are sorted, and the top K solutions with the highest scores are selected to form the final process scheme set.

Images