Material property design method, device, equipment and medium based on CPFEM

Abstract

The application discloses a material performance design method and device based on CPFEM, equipment and medium, the method comprises the following steps: based on the key material parameters of the material to be analyzed, generating a structured parameter set of representative volume element models with different microstructure characteristics through a parameterization script; driving CPFEM to calculate different structured parameter sets in parallel through a high-performance computing platform to obtain calculation results including macroscopic mechanical response; extracting the key performance indicators of each representative volume element model from the calculation results; constructing a mapping relationship between the structured parameter set and the key performance indicators, so as to predict material performance and perform material reverse design based on the mapping relationship. Thus, the model parameter set covering different microstructure characteristics is generated through the parameterization script, and the high-performance computing platform is used to improve the research and development efficiency, explore a wide design space, determine a better performance combination, and realize the reverse design of the material through the mapping relationship.

Description

Technical Field

[0001] This application relates to the field of materials science and technology, and in particular to a material performance design method, apparatus, equipment and medium based on CPFEM. Background Technology

[0002] New materials research and development is the core driving force for the development of high-end manufacturing. Traditional materials research and development relies heavily on the "trial and error method," that is, repeatedly preparing samples and conducting performance tests to screen formulas and processes. This process is time-consuming and costly, and has become a bottleneck restricting the innovation of new materials.

[0003] Specifically, the trial-and-error method has the following drawbacks: First, it relies on manual operation, resulting in low R&D efficiency; second, it lacks systematicity, with existing research focusing on the performance reproduction and mechanism analysis of known structures, failing to explore the vast microstructure design space to discover better performance combinations; third, the large amount of data generated during the R&D process has not been effectively integrated and mined, making it impossible to achieve reverse design of materials.

[0004] Therefore, how to improve the efficiency of materials research and development, accurately and quickly discover the optimal combination of materials performance, and realize the reverse design of materials are problems that urgently need to be solved by those skilled in the art. Summary of the Invention

[0005] In view of this, one aspect of this application provides a material property design method based on CPFEM, the method comprising: Obtain the key material parameters of the material to be analyzed; Based on the key material parameters, a set of structured parameters for representative volume element models with different microstructural features is generated through a parameterized script. The CPFEM simulation software is driven by a high-performance computing platform to perform parallel calculations on different sets of structured parameters, and the calculation results, including the macroscopic mechanical response of each representative volume element model, are obtained. From the calculation results, extract the key performance indicators of each representative volume element model; A mapping relationship is constructed between the structured parameter set and the key performance indicators so as to perform material performance prediction and material reverse design based on the mapping relationship.

[0006] Optionally, the key material parameters include at least chemical composition, preparation process, service performance, and microstructure; Based on the aforementioned key material parameters, a structured parameter set for representative volumetric element models with different microstructural features is generated using a parameterization script, including: A microstructure image of the material to be analyzed is obtained; and the microstructure image is analyzed using an image processing algorithm to obtain the statistical distribution parameters of the microstructure features. The key material parameters and the statistical distribution parameters are transformed into a set of quantified parameters; the set of quantified parameters is a set of parameters that can be processed by a calculation script and includes at least a range of values, a distribution form, and a unit. The structured parameter set is generated using the parameterization script based on the quantization parameter set.

[0007] Optionally, the structured parameter set includes at least the geometric topology configuration, material constitutive parameters, boundary loading conditions, and simulation task metadata of the corresponding representative volume element model; The CPFEM simulation software is driven by a high-performance computing platform to perform parallel computations on different sets of structured parameters, including: Based on the simulation task metadata, a simulation input file including modeling instructions is generated for the structured parameter set corresponding to each representative volume element model; The computational resources required for the simulation of each representative volume element model are estimated to obtain resource parameters; the resource parameters include at least the number of CPU cores, computation time, and memory size. Based on the resource parameters, computing resources are allocated to each of the simulation input files, and the CPFEM simulation software is driven to perform calculations in parallel.

[0008] Optionally, the computational resources required for the simulation of each representative volumetric element model are estimated to obtain resource parameters, including: From the structured parameter set, the intrinsic features of each representative volume element model are extracted; the intrinsic features include mesh generation parameters, material constitutive parameters, loading solution features, and derived feature parameters; Based on the inherent characteristics, the computational resources required for the simulation of each representative volume element model are estimated using a pre-constructed estimation model, thereby obtaining the resource parameters. The prediction model includes a random forest model for predicting the computation time and a linear regression model for predicting the memory size.

[0009] Optionally, the CPFEM-based material property design method further includes: During the parallel computing process, task status information is acquired; When the task status information is in a specified state, the log information of the corresponding task is extracted; the specified state includes at least one of failure state, timeout state, and node failure. Based on the log information, the fault type is determined; the fault type includes a first type of fault caused by insufficient resources and a second type of fault caused by an error in the structured parameter set; When the fault type is the first type, the simulation input file corresponding to the fault task is parsed, the new resource parameters of the fault task are determined, and the calculation is retried. When the fault type is the second type, the simulation input file corresponding to the fault task is marked as an invalid task.

[0010] Optionally, the key performance indicators include at least yield strength, tensile strength, and fracture strain; From the calculation results, key performance indicators for each representative volumetric element model are extracted, including: Extract the initial stress-strain curve from the calculation results; The initial stress-strain curve is cleaned and standardized to obtain the target stress-strain curve; Based on preset rules, the key performance indicators are determined from the target stress-strain curve; The preset rules include: for curves with a yield point, the stress value corresponding to the first point in the curve where the slope is less than a threshold is taken as the yield strength; for curves without a yield point, the yield strength is determined by the offset method; the maximum stress value on the curve is taken as the tensile strength; and the strain value corresponding to the termination point on the curve is taken as the fracture strain.

[0011] Optionally, a mapping relationship is constructed between the structured parameter set and the key performance indicators to enable material performance prediction and reverse material design based on the mapping relationship, including: Using the structured parameter set as input and the key performance indicators as output, a structural performance database for reverse design of the material is constructed. Based on the structural performance database, a correlation map of structural performance is generated; and using the structural performance database as a training dataset, a performance prediction model for predicting the material performance is trained; wherein, the correlation map includes at least one of a two-dimensional scatter plot, a parallel coordinate plot, and a three-dimensional scatter plot, and the correlation map is used to analyze the microstructural parameters corresponding to the expected material performance of the material.

[0012] Another aspect of this application provides a CPFEM-based material property design apparatus, the apparatus comprising: The key parameter acquisition module is used to acquire the key material parameters of the material to be analyzed. The parameter set generation module is used to generate a structured parameter set of representative volume element models with different microstructural features based on the key material parameters and through a parameterized script. The parallel computing module is used to drive the CPFEM simulation software to perform parallel calculations on different sets of structured parameters through a high-performance computing platform, so as to obtain calculation results including the macroscopic mechanical response of each representative volume element model; The performance index extraction module is used to extract key performance indicators of each representative volume element model from the calculation results. The mapping relationship construction module is used to construct the mapping relationship between the structured parameter set and the key performance indicators, so as to perform material performance prediction and material reverse design based on the mapping relationship.

[0013] Another aspect of this application provides an electronic device including a memory and a processor, wherein the memory stores a computer program executable on the processor, and the processor executes the computer program to implement the steps of the CPFEM-based material property design method.

[0014] Another aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the CPFEM-based material property design method.

[0015] The material performance design method, apparatus, equipment, and medium based on CPFEM provided in this application have the following beneficial effects: First, by automatically generating model parameter sets covering different microstructural features through parametric scripts, and using a high-performance computing platform to drive CPFEM software for parallel computation, a fully automated process from model generation to performance extraction is achieved, improving R&D efficiency and avoiding the tediousness and time-consuming nature of manual operations. Second, by performing parallel simulations on multiple microstructural designs, a broad design space that is difficult to cover by traditional methods is explored, thereby efficiently selecting better performance combinations. Furthermore, by constructing a quantitative mapping relationship between microstructural parameters and key performance indicators, not only is rapid prediction of material performance achieved, but also reverse optimization of microstructural parameters based on target performance is supported, realizing reverse design of materials. Attached Figure Description

[0016] Figure 1 A schematic flowchart illustrating a material property design method based on CPFEM provided in this application embodiment; Figure 2 This is a flowchart illustrating a material property design method based on CPFEM according to another embodiment of this application; Figure 3 A two-dimensional scatter plot illustrating the relationship between grain size and yield strength, provided as an embodiment of this application; Figure 4(a) is a schematic diagram of the interaction between grain size and phase fraction provided in an embodiment of this application; Figure 4(b) is a schematic diagram of another interaction between grain size and phase fraction provided in the embodiments of this application; Figure 5 This is a schematic diagram of a material property design device based on CPFEM provided in an embodiment of this application; Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0017] The attached diagram is labeled as follows: 50 is the key parameter acquisition module, 51 is the parameter set generation module, 52 is the parallel computing module, 53 is the performance index extraction module, 54 is the mapping relationship construction module, 60 is the memory, 61 is the processor, 62 is the display screen, 63 is the input / output interface, 64 is the communication interface, 65 is the power supply, 66 is the communication bus, 601 is the computer program, 602 is the operating system, and 603 is the data. Detailed Implementation

[0018] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0019] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0020] Figure 1 This is a flowchart illustrating a material property design method based on CPFEM provided in an embodiment of this application, as shown below. Figure 1 As shown, the method includes: S10: Obtain the key material parameters of the material to be analyzed; In materials research and development and performance optimization, it is necessary to analyze the microstructure-performance relationship of specific material systems (such as duplex steel, aluminum alloys, etc.). To this end, in specific embodiments, key material parameters of the material to be analyzed are obtained. Key material parameters may include, but are not limited to, chemical composition (such as the content of alloying elements), preparation process (such as heat treatment temperature and cooling rate), service performance target (such as target tensile strength), and known microstructure characteristics (such as known physical metallurgical principles and empirical laws).

[0021] It should be noted that the key material parameters can be obtained through user input, reading from a database, or importing experimental data; this application does not limit the method. It is worth noting that the key material parameters provide the physical basis and constraint boundaries for the subsequent definition of the microstructure design space.

[0022] S11: Based on key material parameters, a structured parameter set of representative volume element models with different microstructural features is generated through parameterized scripts; After obtaining the key material parameters, a series of structured parameter sets for representative volume element (RVE) models with different combinations of microstructure parameters are systematically generated using parametric scripts. In other words, multiple RVE models are generated, each with different microstructure characteristics and corresponding to different structured parameter sets. These microstructure characteristics include, but are not limited to, phase fraction, grain size and distribution, grain orientation texture, and second-phase morphology.

[0023] Specifically, in a specific embodiment, the parameterization script reads information such as the parameter range, distribution form, and constraints (e.g., the sum of phase fractions is 1) from the key material parameters, and automatically generates hundreds or thousands of RVE model parameter configurations covering the design space using the Voronoi mosaic algorithm or other geometric modeling methods.

[0024] In one alternative embodiment, the structured parameter set can be stored in a machine-readable format such as CSV, with each row corresponding to a unique RVE model, containing a complete description of the RVE model's geometric topological parameters, material constitutive parameters, boundary loading conditions, etc.

[0025] S12: Through a high-performance computing platform, the CPFEM simulation software is driven to perform parallel calculations on different structured parameter sets to obtain calculation results including the macroscopic mechanical response of each representative volume element model; Furthermore, in an alternative embodiment, a high-performance computing platform (such as a supercomputing cluster deployed with Slurm or PBS job scheduling system) is used to drive the CPFEM simulation software to perform high-throughput parallel computation on the structured parameter set generated in step S11.

[0026] Specifically, each line of the structured parameter set is parsed to automatically generate the simulation input file for the corresponding RVE model (such as the .inp file in Abaqus), and the simulation input file is submitted to the computing cluster in batches through a job array or task queue according to the estimated computing resource requirements (number of CPU cores, memory, runtime, etc.) of each RVE model.

[0027] During execution, each simulation model independently occupies allocated computing resources and runs in parallel to simulate its macroscopic mechanical response under uniaxial tension, compression, or shear loading conditions. After all simulation tasks are completed, the system collects the output result files of each model (such as stress-strain curves, field variable outputs, etc.) to form a computational aggregate.

[0028] S13: Extract the key performance indicators of each representative volume element model from the calculation results; S14: Construct a mapping relationship between a structured parameter set and key performance indicators to enable material performance prediction and reverse design based on the mapping relationship.

[0029] After completing the large-scale simulation, key performance indicators are further extracted from the calculation results of each RVE model. At this point, the structured parameter set obtained in step S11 and the key performance indicators obtained in step S13 are combined to construct the structure-performance mapping relationship of different RVE models. Subsequently, material performance prediction and material reverse design are performed based on this mapping relationship.

[0030] Specifically, it can be understood that, based on the structure-performance mapping relationship, users can predict the properties of a material based on its known microstructure. Simultaneously, this mapping relationship can also be used to reverse-engineer materials with the desired properties.

[0031] Therefore, the CPFEM-based material performance design method provided in this application automatically generates model parameter sets covering different microstructural features through parametric scripts, and uses a high-performance computing platform to drive CPFEM software for parallel computation, realizing a fully automated process from model generation to performance extraction, improving R&D efficiency and avoiding the tediousness and time-consuming nature of manual operations. Secondly, by performing parallel simulations on multiple microstructural designs, a vast design space that is difficult to cover by traditional methods is explored, thereby efficiently selecting better performance combinations. Furthermore, by constructing a quantitative mapping relationship between microstructural parameters and key performance indicators, not only is rapid prediction of material performance achieved, but it also supports reverse optimization of microstructural parameters based on target performance, realizing reverse design of materials.

[0032] In one optional embodiment, the key material parameters include at least chemical composition, preparation process, service performance, and microstructure. Specifically, the key material parameters include the material's chemical composition and phase diagram characteristics, preparation and processing characteristics, macroscopic performance targets and service environment, and known physical metallurgical principles and empirical laws.

[0033] In one alternative embodiment, when acquiring key material parameters, reverse engineering can be performed from the performance target to identify which microscopic physical mechanisms dominate this macroscopic performance. The fabrication and processing technology can also be examined to determine which microstructural parameters can be effectively controlled by the process. Furthermore, known principles of physical metallurgy can be used to identify parameters that have the most significant impact on performance and may interact with each other. Finally, in a specific embodiment, the selected key material parameters are converted into quantitative parameters that can be processed by computer scripts.

[0034] It should be noted that the process of determining key material parameters is a systematic analysis that integrates performance requirements, process knowledge, physical principles and engineering judgments. The aim is to transform the vague concept of "material design" into a clear, quantifiable, and computable multi-dimensional parameter space, providing a foundation for digital screening and design.

[0035] Figure 2 This is a flowchart illustrating a material property design method based on CPFEM according to another embodiment of this application. As an optional embodiment based on the above embodiment, such as... Figure 2 As shown, based on key material parameters, a structured parameter set for representative volumetric element models with different microstructural characteristics is generated through a parameterization script, including: S20: Obtain microstructure images of the material to be analyzed; and analyze the microstructure images using image processing algorithms to obtain statistical distribution parameters of the microstructure features; In a specific embodiment of generating a structured parameter set, a microstructure image of the material to be analyzed is first obtained, and an image processing algorithm is used to analyze the microstructure image to obtain the statistical distribution parameters of the microstructure features.

[0036] It should be noted that the microscopic tissue images can be images acquired by scanning electron microscope (SEM) or experimental images acquired by electron backscatter diffraction (EBSD), and this application does not limit them.

[0037] Furthermore, it should be noted that statistical distribution parameters include, but are not limited to, phase fraction, grain size distribution, grain orientation texture, second-phase morphology parameters, and grain boundary characteristics. Phase fraction refers to the volume fraction of each phase and its spatial distribution statistics (e.g., the ratio of the two phases in a martensitic / ferrite dual-phase steel). Grain size distribution refers to the distribution of the equivalent circle diameter of the grains, log-normal distribution parameters (mean, standard deviation), or grain area distribution. Grain orientation texture refers to the pole figure, inverse pole figure, or orientation distribution function (ODF), as well as the volume fraction and dispersion of the texture components. Second-phase morphology parameters refer to the size distribution, aspect ratio, shape factor (e.g., spherical, acicular), and spatial distribution uniformity of the second phase (e.g., whether it is distributed along grain boundaries). Grain boundary characteristics refer to the ratio of large-angle / small-angle grain boundaries, grain boundary curvature, etc.

[0038] In specific embodiments, these statistical distribution parameters can be directly used in parameterization scripts to generate RVE model clusters with the same statistical characteristics but different microstructural details.

[0039] Specifically, statistical distribution parameters can provide quantitative input for geometric modeling of the RVE model. Specifically, based on grain size distribution, when generating a polycrystalline RVE model using the Voronoi mosaic algorithm, the size of each grain can be controlled to conform to the actual log-normal distribution, rather than assuming all grains are the same size. Furthermore, based on the statistics of phase fractions, the volume proportions of different phases can be precisely allocated in the RVE model to ensure that the macroscopic phase proportions are consistent with the actual material. Additionally, based on the second-phase morphology parameters, second-phase particles with specific sizes and shapes can be randomly embedded in the matrix, ensuring their distribution conforms to the statistical laws observed in actual data.

[0040] Statistical distribution parameters can construct a more realistic design space. Specifically, they provide not only the average value but also the distribution range and variation information. This allows parametric script design to systematically explore minute fluctuations or extreme cases surrounding the microstructure of real materials. For example, while keeping the average grain size constant, changing the standard deviation of the size distribution can be used to study its impact on performance. In specific embodiments, this design space based on real statistics is more physically meaningful than parameter ranges based solely on empirical assumptions, and the selected high-performance microstructures are easier to realize in actual processes.

[0041] In specific embodiments, statistical distribution parameters can also improve the confidence and generalization ability of simulation results. Specifically, since the statistical characteristics of microstructure are one of the key factors determining macroscopic performance, RVE models generated using real statistical distribution parameters have mechanical responses that more closely resemble the actual behavior of real materials. In the subsequent mapping relationship construction, the machine learning model trained on the structure-performance database built based on these RVE models can more accurately predict the performance of real materials, rather than being applicable only to idealized models.

[0042] Furthermore, statistical distribution parameters can achieve a closed loop from experience-based design to data-driven design. Specifically, statistical distribution parameters obtained from image processing can be used to generate initial RVE models or as target parameters for material reverse engineering. For example, when a performance prediction model indicates that a certain ideal performance requires a specific grain size distribution, it can guide the adjustment of process parameters, and this distribution is a quantifiable and measurable indicator from actual images. Examples will be provided below for easier understanding.

[0043] For example, the statistical result of grain size distribution extracted from EBSD images is the average grain diameter. Standard deviation Furthermore, the grain size follows a log-normal distribution. In the parametric design script, the grain size is set as a random variable with a probability density function of... ,in, .

[0044] Then, for each RVE model to be generated, the script randomly samples a set of grain sizes from the distribution and drives the Voronoi algorithm to generate grains of the corresponding sizes, thereby obtaining a series of statistically equivalent but different RVE models with different specific geometric configurations.

[0045] S21: Transform key material parameters and statistical distribution parameters into a set of quantified parameters; the set of quantified parameters is a set of parameters that can be processed by a calculation script and includes at least the range of values, distribution form, and unit. Furthermore, key material parameters (chemical composition, process parameters, etc.) are combined with statistical distribution parameters extracted from the images to transform them into a quantized parameter set. This quantized parameter set is a set of parameters that can be processed by the computational script and includes at least the range of values, distribution form, and units. For example, the martensite volume fraction ranges from 0.2 to 0.5 and is uniformly distributed. The ferrite average grain size is 2-10 μm and follows a log-normal distribution with a standard deviation of 0.3.

[0046] During the conversion process, it is necessary to determine the definition of each parameter. For example, it can be a continuous type with a phase fraction of 0.2-0.5, or a discrete type with a second phase morphology (spherical / acicular / lamellar). Furthermore, it is necessary to determine the value range of each parameter and the constraint relationships between different parameters (e.g., martensite fraction + ferrite fraction + retained austenite fraction = 1). By writing the above information into the input template of the parameterization script, it can be converted into a quantization parameter set.

[0047] Table 1 is a schematic table of a set of quantization parameters provided in the embodiments of this application. For ease of understanding, the following description will be based on Table 1.

[0048] Table 1 is a schematic table of a set of quantization parameters.

[0049] Table 1 provides an example of DP steel. In this example, the constraint is that the martensite carbon content and martensite fraction must satisfy the phase diagram constraint (when the carbon content is too high, the martensite fraction is limited by the carbon distribution).

[0050] S22: Generate a structured parameter set based on the quantization parameter set using a parameterized script.

[0051] Finally, a structured parameter set (CSV file) corresponding to a microstructure design point in each row is generated in batches based on the quantized parameter set using a parameterization script (such as a Python script). In one implementation, if experimental images are lacking, statistical distribution parameters can also be defined based on empirical values ​​from literature or theoretical assumptions; this application does not limit this approach.

[0052] Therefore, by extracting statistical distribution parameters from real microstructures using image processing algorithms, the generated RVE model is statistically equivalent to the real material, thereby ensuring that the simulation results can reproduce real deformation behavior and improving the credibility and physical authenticity of the test.

[0053] In one alternative embodiment, the structured parameter set includes at least the geometric topology of the corresponding representative volumetric element model, material constitutive parameters, boundary loading conditions, and simulation task metadata.

[0054] The geometric topological configuration includes, but is not limited to, phase fraction, grain size and distribution, grain orientation texture, and second phase morphology. In a specific embodiment, the geometric topological configuration of the RVE model directly determines the structure of the RVE model, i.e., the microstructure of the material.

[0055] Material constitutive parameters include, but are not limited to, crystal plasticity model parameters (describing the deformation behavior of a crystal in a specific slip system), initial critical decomposition shear stress (the initial stress required to initiate slip), hardening parameters (how the slip system's ability to resist further slip changes with increasing deformation, specifically including self-hardening and latent hardening coefficients), rate sensitivity index (the sensitivity of the material's deformation rate to force), rate sensitivity index (the stiffness matrix describing the elastic deformation of the material), and phase attribution (identifying which phase each of the above sets of material parameters corresponds to). In specific embodiments, the material constitutive parameters assign mechanical properties to each phase, i.e., how they deform under stress.

[0056] Boundary loading conditions include, but are not limited to, the loading method (specifically, uniaxial tension, compression, shear, or complex multiaxial loading), the loading direction (the relative relationship between the loading direction and the material texture), the strain rate (i.e., the loading speed), and the total strain / load step (how long or how many steps the calculation requires to run). In specific embodiments, boundary loading conditions define how the mechanical simulation is performed.

[0057] Simulation task metadata includes, but is not limited to, a unique model identifier (Model ID) and its corresponding parameter group. The Model ID can be a unique number, such as Job_0001 or Job_0002, which can link all parameters, input files, and output results together. The Parameter Group is used to identify which batch the RVE model belongs to, facilitating data management.

[0058] In a specific embodiment, the structured parameter set is stored in the form of a CSV file. In the CSV file, each column corresponds to one of the above parameters, and each row corresponds to a unique RVE model.

[0059] Based on the above embodiments, as an optional embodiment, a high-performance computing platform is used to drive the CPFEM simulation software to perform parallel calculations on different structured parameter sets, including: Based on the simulation task metadata, generate simulation input files including modeling instructions for the structured parameter sets corresponding to each representative volume element model; Estimate the computational resources required for simulation of each representative volume element model to obtain resource parameters; resource parameters should include at least the number of CPU cores, computation time, and memory size. Based on the resource parameters, computing resources are allocated to each simulation input file, and the CPFEM simulation software is driven to perform calculations in parallel.

[0060] In a specific embodiment, based on the simulation task metadata, the task generator script reads each line of the structured parameter set and fills the specific geometric and material parameters into the general template file through text replacement or parameter injection techniques, thereby generating a simulation input file containing complete modeling instructions for each RVE model.

[0061] For example, a dedicated simulation input file (such as Job_0001.inp) is generated for the simulation task metadata Job_0001, and specific modeling instructions such as "grain size 5μm, second phase volume fraction 0.2..." are written into the simulation input file. At the same time, an index is established to ensure that the generated Job_0001.inp file and the Job_0001 line in the CSV file form a strict and traceable correspondence.

[0062] Furthermore, thousands of tasks with the same software environment requirements but different simulation input files can be packaged into an array job and submitted. The job scheduling system (such as Slurm, PBS) will automatically estimate the computing resources required for the simulation for each array index and allocate computing resources according to the resource parameters, thereby achieving high-throughput parallel computing.

[0063] For example, in the submission script, the command "#SBATCH--array=1-1000" defines a job array containing 1000 independent subtasks, thereby reducing the tediousness of writing and submitting 1000 independent jobs. At the same time, the system will assign a unique array job ID to this array and assign a unique task index to each subtask in the array, such as an index setting from 1 to 1000.

[0064] When allocating resources, resource requests defined in the script (such as the number of CPU cores, memory size, runtime, etc.) are independently allocated to each subtask in the array.

[0065] For example, in one alternative embodiment, the following is defined in the script: #SBATCH --cpus-per-task=4 #SBATCH --mem=8G #SBATCH --time=02:00:00 This means that the simulated RVE model with task index 1 will have exclusive access to 4 CPU cores and 8GB of memory, running for a maximum of 2 hours. Similarly, the RVE model with task index 2 also has completely independent and equivalent resources. This task-level parallelism mode, compared to internally partitioning a large cluster containing 4000 cores, reduces resource fragmentation and waiting time, and improves the scheduler's response speed.

[0066] Furthermore, based on resource parameters, computational resources are allocated to each simulation input file, driving the CPFEM simulation software to perform parallel calculations. In a specific embodiment, when using task indexes to precisely execute the simulation, as an optional implementation, direct mapping based on filenames can be used. Specifically, when the simulation input files are named according to task indexes, such as model_1.inp, model_2.inp,..., model_1000.inp, then only one line of command, my_solvermodel_${SLURM_ARRAY_TASK_ID}.inp, is needed in the script to automatically load and calculate model_500.inp for the 500th task.

[0067] In another alternative embodiment, dynamic mapping based on index files can also be used. Specifically, a master manifest file `index.idx` is written to the path of the simulation input file or control parameters, with each line corresponding to an RVE model. The script uses the command `sed -n ${SLURM_ARRAY_TASK_ID} pindex.idx` to dynamically extract the content of the Nth line, and then loads the corresponding file to start the calculation. This method is more flexible and does not depend on a fixed filename format. The specific index execution method is not limited in this application.

[0068] In the above parallel execution process, as an optional embodiment, in order to avoid the impact of submitting a massive number of tasks at once on the scheduling system, and also in order to fairly share cluster resources, this application embodiment provides a parallelism management method. Specifically, the % operator can be used to control the number of tasks running at the same time.

[0069] Specifically, a parallelism limit is added to the submission command, using the % operator to control the number of tasks running simultaneously. For example, the control command is #SBATCH--array=1-1000%50. This means that although 1000 tasks are submitted at once, it guarantees that at any given time, only a maximum of 50 (the specified number) tasks are running concurrently. When a task finishes, the scheduler automatically starts the next one from the queue until all tasks are completed. This sliding window execution method ensures high throughput while avoiding resource contention and system overload.

[0070] Based on the above embodiments, as an optional embodiment, the computational resources required for simulation of each representative volume element model are estimated to obtain resource parameters, including: The intrinsic features of each representative volume element model are extracted from the structured parameter set; the intrinsic features include mesh generation parameters, material constitutive parameters, loading and solution features, and derived feature parameters. Based on the inherent characteristics, the computational resources required for the simulation of each representative volume element model are estimated through a pre-constructed prediction model, and the resource parameters are obtained. The prediction models include a random forest model for predicting computation time and a linear regression model for predicting memory size.

[0071] In a specific implementation, a lightweight manager is developed to maintain a queue of tasks awaiting computation. The manager dynamically retrieves tasks from the queue, submits them based on the current idle status of cluster nodes, and monitors the task status. Failed tasks can be automatically re-added to the queue for retry. The algorithm can request appropriate computing resources (number of CPU cores, memory size, computation time) for each RVE model based on its estimated computational and memory requirements, thereby improving the overall utilization of the cluster.

[0072] It should be noted that in the early stages of system operation, heuristic rules can be used for conservative estimation. Specifically, as computing tasks accumulate, we will associate the collected actual consumption data (real CPU time, peak memory, runtime) with the corresponding structured parameter set to build a training dataset, and then train a machine learning model to obtain a prediction model for estimating the required computing resources.

[0073] Based on the above, it can be understood that the resource consumption of each RVE model is determined by its intrinsic characteristics. Therefore, when calculating the resource parameters required by the RVE model, the intrinsic characteristics of each RVE model are first extracted from the structured parameter set. In specific embodiments, the intrinsic characteristics include, but are not limited to, mesh generation parameters, material constitutive parameters, loading and solving characteristics, and derived characteristic parameters.

[0074] The meshing parameters include, but are not limited to, the total number of elements (NElem), the total number of nodes (NNode), the element type (such as C3D8, C3D20, with higher-order elements requiring more computation), and mesh quality indicators (such as minimum Jacobian, as distorted meshes may lead to convergence difficulties and increase computation time).

[0075] Material constitutive parameters include, but are not limited to, the type of RVE model (e.g., phenomenology-based vs. physics-based), the number of active slip systems (NSlip, such as 12 slip systems for FCC materials), the complexity of the hardening model (e.g., whether latent hardening, back stress, etc. are considered), and the number of material parameters (which affects the computational cost of each iteration).

[0076] Loading solution features includes, but is not limited to, the total number of increment steps (NSteps, which depends on the preset total strain and initial increment step size), the degree of nonlinearity (large deformation, contact, etc.), and output requirements (whether to output detailed state variables for each element, which will affect I / O and memory).

[0077] The derived feature parameters include, but are not limited to, the total number of degrees of freedom (NDOF = NNode * degrees of freedom of each node, which directly affects the size of the stiffness matrix), the total number of material points (NMatPoints = NElem * number of material points at each integration point, which directly affects the number of material constitutive calculations) and the model complexity index. The model complexity index can be obtained by weighted combination of the above features, Complexity = NElem * NSlip * log(NSteps).

[0078] Furthermore, when estimating resources using a pre-built prediction model, in one optional embodiment, the random forest model can be used to predict computation time because it can handle nonlinear relationships, provides good interpretability of feature importance, and is less prone to overfitting. Additionally, memory estimation is also a regression problem, but memory has a strong linear relationship with degrees of freedom (NDOF). Therefore, in a specific embodiment, linear regression can be used as the prediction model.

[0079] It should be noted that, in an alternative embodiment, cross-validation can be used to evaluate model accuracy. Specifically, as new tasks are completed, the model is periodically and incrementally trained to adapt it to new types of microstructure parameters.

[0080] Therefore, the CPFEM-based material performance design method provided in this application uses a machine learning model to make personalized predictions of the resource requirements for each simulation task, avoiding resource waste or task failure due to insufficient resources caused by traditional unified resource requests, thereby improving the overall utilization rate of the cluster and the success rate of tasks.

[0081] In an optional embodiment, the CPFEM-based material property design method further includes: During parallel computing, task status information is acquired; When the task status information is in a specified state, extract the log information of the corresponding task; the specified state includes at least one of failure state, timeout state, and node failure. Based on the log information, determine the fault type; the fault types include the first type of fault caused by insufficient resources and the second type of fault caused by an error in the structured parameter set; When the fault type is the first type, the simulation input file corresponding to the fault task is parsed, the new resource parameters of the fault task are determined, and the calculation is retried. When the fault type is type 2, the simulation input file corresponding to the fault task is marked as an invalid task.

[0082] In a specific implementation of parallel computing, a daemon process (such as monitor.py) is developed to periodically scan the status of all submitted jobs (running, completed, failed, timed out). When a task is detected to be in a specified state such as failed, timed out, or node failure, the error log file of the corresponding task is extracted.

[0083] Furthermore, the fault type is determined based on keywords in the log information (such as "convergencenotachieved" corresponding to model parameter errors, "Nodeisdown" corresponding to node failure, and "OUTOFMEMORY" corresponding to insufficient resources).

[0084] In one alternative embodiment, the fault types can be divided into two categories. The first category is faults caused by insufficient resources (such as memory overflow, timeout), and the second category is faults caused by errors in the structured parameter set itself (such as non-convergence caused by mesh malformation).

[0085] When the fault type is the first type, the simulation input file corresponding to the fault task is parsed, the fault node name or insufficient resource information is extracted from the log, and the task is re-added to the queue for retry after modifying the resource parameters (such as increasing memory requests by 10%, adding fault nodes to the node exclusion list, and increasing timeout by 20%). The number of retries shall not exceed the preset threshold (such as 3 times).

[0086] When the fault type is type 2, the simulation input file corresponding to the task is directly marked as an invalid task, the reason for failure is recorded (such as "mesh distortion caused non-convergence"), and it is excluded from the subsequent data processing flow and no longer participates in performance index extraction and database construction.

[0087] It is worth noting that, in an optional embodiment, for tasks that are unresponsive for an extended period of time, a timeout mechanism is implemented to automatically terminate and release resources, and the task is handled as a Type I fault.

[0088] Therefore, the CPFEM-based material property design method provided in this application achieves intelligent fault-tolerant processing for large-scale simulation tasks through automated status monitoring, log parsing, and fault classification. This isolates tasks that inevitably fail due to unreasonable physical parameters, avoiding resource waste. Simultaneously, it automatically retryes temporary resource issues, improving the robustness and completion rate of large-scale high-throughput computing.

[0089] In one optional embodiment, the key performance indicators include at least yield strength, tensile strength, and fracture strain; From the calculation results, key performance indicators for each representative volume element model were extracted, including: Extract the initial stress-strain curve from the calculation results; The initial stress-strain curve is cleaned and standardized to obtain the target stress-strain curve. Based on preset rules, key performance indicators are determined from the target stress-strain curve; The preset rules include: for curves with a yield point, the stress value corresponding to the first point in the curve where the slope is less than a threshold is taken as the yield strength; for curves without a yield point, the yield strength is determined by the offset method; the maximum stress value on the curve is taken as the tensile strength; and the strain value corresponding to the termination point on the curve is taken as the fracture strain.

[0090] In a specific embodiment, the original engineering stress-strain curves and / or actual stress-strain curve data are automatically extracted from the calculation result file (such as an ODB or DAT file). Furthermore, the initial curves are cleaned and standardized. Specifically, a moving average filter or a Savitzky-Golay filter can be used to smooth the curves to eliminate numerical noise. In addition, all data are uniformly converted to the International System of Units (SI) (stress in MPa, strain dimensionless).

[0091] Finally, key performance indicators are determined based on pre-defined materials science rules. Specifically, for materials with a clear yield point (such as low-carbon steel), the stress-strain curve data points are iterated through to calculate the derivative (slope) of stress with respect to strain. When the slope The first drop to near zero (i.e., , When a preset threshold (e.g., 0.001 × elastic modulus) is reached, the stress value corresponding to that point is taken as the yield strength. If a stress decrease occurs ( If the stress peak value before the decrease is taken as the upper yield strength, then the upper yield strength is taken as the upper yield strength.

[0092] For materials without a clear yield point (such as aluminum alloys and high-strength steel), the offset method is used to determine the specified plastic elongation strength. That is, the stress corresponding to 0.2% of the plastic strain is taken as the specified plastic elongation strength (usually denoted as σ). Specifically, first determine the elastic modulus E (by performing a linear fit within the elastic segment), then construct a straight line parallel to the elastic segment and offset by 0.2% of the strain. The equation of the line is... Furthermore, the intersection point of this straight line and the actual stress-strain curve is found, and the stress at the intersection point is accurately calculated using linear interpolation as the yield strength. .

[0093] In another alternative embodiment, the offset line can be solved simultaneously with the actual stress-strain curve (engineering stress-strain or actual stress-strain). Specifically, the stress difference is calculated. As strain increases, The point where the value changes from negative to positive, crossing the zero point, is the intersection point. The stress value corresponding to this intersection point is accurately calculated using linear interpolation. This stress value is the yield strength. .

[0094] For tensile strength, the maximum stress value on the entire stress-strain curve can be taken. In an optional embodiment, if local noise exists, several points before and after the maximum value can be taken, and parabolic fitting can be performed to find the extreme value.

[0095] For fracture strain, the strain value corresponding to the end point of the curve can be taken. In a specific embodiment, if damage evolution is introduced into the simulation, the cumulative equivalent plastic strain when the damage variable reaches 1 is taken. For calculations that terminate prematurely, they are marked as not actually fractured and the termination strain is recorded.

[0096] Therefore, the CPFEM-based material performance design method provided in this application transforms the standardized testing criteria of materials science into automatically executable mathematical algorithms, realizing high-throughput and automated extraction of performance indicators from large-scale simulation results, ensuring data consistency and repeatability, and avoiding subjective differences and inefficiencies in manual processing.

[0097] In one alternative embodiment, a mapping relationship is constructed between a structured parameter set and key performance indicators to enable material performance prediction and reverse material design based on the mapping relationship, including: Using a structured parameter set as input and key performance indicators as output, a structural performance database for reverse design of materials is constructed. Based on the structural performance database, a correlation map of structural performance is generated; and using the structural performance database as the training dataset, a performance prediction model for predicting material performance is trained; wherein, the correlation map includes at least one of two-dimensional scatter plot, parallel coordinate plot and three-dimensional scatter plot, and the correlation map is used to analyze the microstructural parameters corresponding to the expected material performance of the material.

[0098] In a specific embodiment, firstly, a structure-performance database for material reverse engineering is constructed, using a structured parameter set (including microstructural parameters such as phase fraction and grain size) as input features and extracted key performance indicators (yield strength, tensile strength, and fracture strain) as output labels. In the structure-performance database, each row corresponds to an RVE model, recording its complete input parameters and output performance.

[0099] Furthermore, two knowledge mining operations are performed based on the structure-performance database. The first is generating a structure-performance correlation graph. In a specific embodiment, constructing the performance-microstructure correlation graph is the core step in the knowledge mining process. Essentially, it maps the high-dimensional structure-performance data space onto an intuitively understandable two-dimensional or three-dimensional graph, thereby revealing how microstructure parameters affect macroscopic mechanical properties. This allows for intuitive analysis of the microstructure parameters corresponding to the expected material properties, facilitating rapid identification of optimization directions.

[0100] The correlation graph can include, but is not limited to, two-dimensional scatter plots, parallel coordinate plots, three-dimensional scatter plots or surface plots, heatmaps, radar plots, and box plots. Table 2 is a schematic table of relevant information for a correlation graph provided in an embodiment of this application. A visualized correlation graph is shown in the table below: Table 2 is a schematic table of relevant information for a type of association map.

[0101] Figure 3 Figure 4(a) is a two-dimensional scatter plot of the relationship between grain size and yield strength provided in an embodiment of this application. Figure 4(b) is a thermal diagram of the interaction between grain size and phase fraction provided in an embodiment of this application.

[0102] As shown in Table 2, the two-dimensional scatter plot uses a certain microstructural parameter as the X-axis and a certain performance as the Y-axis, with the third dimension parameter encoded by color. In an optional embodiment, such as... Figure 3 As shown, two-dimensional scatter plots of materials Ti-6Al-4V and 300M martensitic high-strength steel were generated. The yield strength of the two materials increased with decreasing grain size, but the degree of increase was different, indicating that fine grain strengthening and phase fraction strengthening have a superimposed effect.

[0103] As shown in Table 2, the heatmap can illustrate the impact of the interaction between two parameters on performance. In an optional embodiment, as shown in Figures 4(a) and 4(b), heatmaps were generated using Ti-6Al-4V and 300M martensitic high-strength steel as examples. This heatmap allows determination of the range of grain size and phase fraction values ​​corresponding to the darkest color region, thereby enabling the analysis and determination of the optimal design domain.

[0104] Furthermore, referring to Table 2, row plots can simultaneously display the relationship between multiple input parameters and multiple output performance parameters, facilitating rapid screening of high-performance regions. Three-dimensional scatter plots or surface plots can provide a three-dimensional view of the combined effect of two parameters on performance. Radar charts are used to compare the multidimensional performance of a few microstructures. Box plots are used to compare the distribution differences in performance across different categories.

[0105] The second aspect of knowledge mining involves using a structure-performance database as the training dataset to train a performance prediction model for predicting material properties. Specifically, machine learning algorithms such as random forests, support vector regression, and neural networks can be employed to establish a regression model with microstructure parameters as input and performance indicators as output.

[0106] After training, the model can quickly predict the mechanical properties of a given microstructure. Combined with optimization algorithms such as genetic algorithms or Bayesian optimization, it can also achieve reverse design. Specifically, it searches backwards for the optimal combination of microstructure parameters, using the target performance as a constraint.

[0107] Thus, by using two paths—association graphs and machine learning models—the big data generated by high-throughput simulations is transformed into intuitive and readable knowledge graphs and rapid prediction tools. This supports both user experience analysis and data-driven automatic optimization design, achieving a leap from "simulation computation" to "intelligent design."

[0108] To facilitate understanding by those skilled in the art, the following will use DP steel as an example to illustrate the process of this application from parametric design to knowledge mining (i.e., establishing mapping relationships). Specifically, it includes parametric design, high-throughput computing, data extraction, and knowledge mining.

[0109] In a specific embodiment of parametric design, the key material parameters for DP steel are determined as follows: martensite volume fraction (range 0.2-0.5), average ferrite grain size (range 2-10 μm), and average martensite island size (range 1-5 μm). A parametric script is written to generate 1000 parameter combinations within the aforementioned parameter space using the Latin hypercube sampling method, and the results are output as a structured parameter set file in CSV format. Each line represents an independent DP steel microstructure design and has a unique RVE model ID.

[0110] In a specific implementation of high-throughput computing, a task generator script is deployed on a high-performance computing cluster equipped with the Slurm job scheduling system. This script reads a structured parameter set in CSV format and, based on a generic crystal plastic finite element input file template, generates a unique simulation input file (e.g., Job_0001.inp) for each row of parameters. Subsequently, an array of 1000 simulation tasks is submitted via the #SBATCH --array=1-1000 command, and these tasks are executed in parallel.

[0111] In a specific implementation of data extraction, after all task calculations are completed, an automated post-processing script is started. This script traverses the output database files of all tasks and extracts the macroscopic stress-strain curve data for each model. Based on preset rules, the yield strength is calculated using the 0.2% offset method, the maximum stress point on the curve is taken as the tensile strength, and the strain at the calculation termination point is taken as the fracture strain.

[0112] In a specific implementation of knowledge mining, the extracted performance metrics are integrated with the input microstructural parameters into a database. A two-dimensional scatter plot is generated using a visualization library (such as Python's Matplotlib), for example, with ferrite grain size as the X-axis and yield strength as the Y-axis, and the color intensity of the dots representing the martensite volume fraction.

[0113] Thus, a "microstructure-performance" spectrum of DP steel can be successfully constructed, which intuitively reveals the superimposed effect of fine grain strengthening and phase strengthening on strength. For example, the spectrum shows that as the grain size decreases, the yield strength generally increases, and at the same grain size, the model with a higher martensite fraction also has higher strength, providing a clear optimization direction for the design of high-performance DP steel.

[0114] In the above embodiments, the material performance design method based on CPFEM has been described in detail. This application also provides an embodiment of the material performance design device based on CPFEM.

[0115] Figure 5 This is a schematic diagram of a material property design device based on CPFEM provided in an embodiment of this application, as shown below. Figure 5 As shown, the device includes: Key parameter acquisition module 50 is used to acquire key material parameters of the material to be analyzed. The parameter set generation module 51 is used to generate a structured parameter set of representative volume element models with different microstructural features based on key material parameters and through parameterized scripts. Parallel computing module 52 is used to drive the CPFEM simulation software to perform parallel calculations on different structured parameter sets through a high-performance computing platform, and obtain calculation results including the macroscopic mechanical response of each representative volume element model. The performance index extraction module 53 is used to extract key performance indicators of each representative volume element model from the calculation results. The mapping relationship construction module 54 is used to construct the mapping relationship between the structured parameter set and the key performance indicators, so as to perform material performance prediction and material reverse design based on the mapping relationship.

[0116] Furthermore, the CPFEM-based material property design apparatus provided in this application embodiment also includes: The image analysis module is used to acquire microstructure images of the material to be analyzed; and to analyze the microstructure images through image processing algorithms to obtain statistical distribution parameters of microstructural features; key material parameters include at least chemical composition, preparation process, service performance and microstructure; The parameter quantization module is used to transform key material parameters and statistical distribution parameters into a quantified parameter set; the quantified parameter set is a parameter that can be processed by the calculation script and includes at least the value range, distribution form, and unit; The parameter generation submodule is used to generate a structured parameter set based on the quantized parameter set using a parameterization script.

[0117] The simulation file generation module is used to generate simulation input files, including modeling instructions, for the structured parameter sets corresponding to each representative volume element model based on the simulation task metadata. The structured parameter sets include at least the geometric topology, material constitutive parameters, boundary loading conditions, and simulation task metadata of the corresponding representative volume element model. The computational resource estimation module is used to estimate the computational resources required for the simulation of each representative volumetric element model and obtain resource parameters; the resource parameters include at least the number of CPU cores, computation time and memory size. The simulation driver module is used to allocate computing resources to each simulation input file according to resource parameters and drive the CPFEM simulation software to perform calculations in parallel.

[0118] The intrinsic feature extraction module is used to extract the intrinsic features of each representative volume element model from the structured parameter set; the intrinsic features include mesh generation parameters, material constitutive parameters, loading and solution features, and derived feature parameters; The resource estimation submodule is used to estimate the computational resources required for simulation of each representative volume element model based on its inherent characteristics and through a pre-built estimation model, thereby obtaining resource parameters. The estimation model includes a random forest model for estimating computation time and a linear regression model for estimating memory size.

[0119] The status information acquisition module is used to acquire task status information during parallel computing. The log information extraction module is used to extract the log information of the corresponding task when the task status information is in a specified state; the specified state includes at least one of failure state, timeout state, and node failure. The fault type determination module is used to determine the fault type based on log information; the fault types include the first type of fault caused by insufficient resources and the second type of fault caused by errors in the structured parameter set; The retry module is used to parse the simulation input file corresponding to the faulty task, determine the new resource parameters of the faulty task, and perform calculation retry when the fault type is the first type. The marking module is used to mark the simulation input file corresponding to the fault task as an invalid task when the fault type is the second type.

[0120] The curve extraction module is used to extract the initial stress-strain curve from the calculation results; the preprocessing module is used to clean and standardize the initial stress-strain curve to obtain the target stress-strain curve. The index determination module is used to determine key performance indicators from the target stress-strain curve based on preset rules. The key performance indicators include at least yield strength, tensile strength, and fracture strain. The preset rules include: for curves with a yield point, the stress value corresponding to the first point on the curve where the slope is less than a threshold is taken as the yield strength; for curves without a yield point, the yield strength is determined by the offset method; the maximum stress value on the curve is taken as the tensile strength; and the strain value corresponding to the termination point on the curve is taken as the fracture strain.

[0121] The database construction module is used to build a structural performance database for reverse engineering of materials, taking a structured parameter set as input and key performance indicators as output. The graph generation module is used to generate correlation graphs of structural properties based on the structural performance database. The correlation graphs include at least one of two-dimensional scatter plots, parallel coordinate plots, and three-dimensional scatter plots. The correlation graphs are used to analyze the microstructural parameters corresponding to the expected material properties of the material.

[0122] The model training module is used to train a performance prediction model for predicting material properties using a structural performance database as the training dataset.

[0123] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application, such as... Figure 6 As shown, the electronic device includes: a memory 60 for storing computer programs; The processor 61 is used to execute computer programs to implement the steps of the CPFEM-based material property design method as described in the above embodiments.

[0124] The electronic devices provided in this embodiment may include, but are not limited to, laptops or desktop computers.

[0125] The processor 61 may include one or more processing cores, such as a quad-core processor or an octa-core processor. The processor 61 may be implemented using at least one of the following hardware forms: Digital Signal Processor (DSP), Field-Programmable Gate Array (FPGA), and Programmable Logic Array (PLA). The processor 61 may also include a main processor and a coprocessor. The main processor, also known as the Central Processing Unit (CPU), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, the processor 61 may integrate a Graphics Processing Unit (GPU), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, the processor 61 may also include an Artificial Intelligence (AI) processor, which is used to handle computational operations related to machine learning.

[0126] The memory 60 may include one or more computer-readable storage media, which may be non-transitory. The memory 60 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In this embodiment, the memory 60 is used to store at least the following computer program 601, which, after being loaded and executed by the processor 61, is capable of implementing the relevant steps of the CPFEM-based material property design method disclosed in any of the foregoing embodiments. In addition, the resources stored in the memory 60 may also include an operating system 602 and data 603, and the storage method may be temporary or permanent storage. The operating system 602 may include Windows, Unix, Linux, etc. The data 603 may include, but is not limited to, relevant data involved in the CPFEM-based material property design method.

[0127] In some embodiments, the electronic device may further include a display screen 62, an input / output interface 63, a communication interface 64, a power supply 65, and a communication bus 66.

[0128] Those skilled in the art will understand that Figure 6 The structures shown do not constitute a limitation on electronic devices and may include more or fewer components than those shown.

[0129] The electronic device provided in this application includes a memory and a processor. When the processor executes the program stored in the memory, it can implement the material property design method based on CPFEM in the above embodiments.

[0130] It should be noted that although the operations are depicted in a specific order in the accompanying drawings, this should not be construed as requiring these operations to be performed in the specific order shown or sequentially, or requiring all illustrated operations to be performed to achieve the desired result. In some cases, multitasking and parallel processing may be advantageous. Furthermore, the separation of various system modules and components in the above embodiments should not be construed as requiring such separation in all embodiments, and it should be understood that the described program components and systems can generally be integrated together in a single software product or packaged into multiple software products.

Claims

1. A material property design method based on CPFEM, characterized in that, The method includes: Obtain the key material parameters of the material to be analyzed; Based on the key material parameters, a set of structured parameters for representative volume element models with different microstructural features is generated through a parameterized script. The CPFEM simulation software is driven by a high-performance computing platform to perform parallel calculations on different sets of structured parameters, and the calculation results, including the macroscopic mechanical response of each representative volume element model, are obtained. From the calculation results, extract the key performance indicators of each representative volume element model; A mapping relationship is constructed between the structured parameter set and the key performance indicators so as to perform material performance prediction and material reverse design based on the mapping relationship.

2. The material property design method based on CPFEM as described in claim 1, characterized in that, The key material parameters include at least chemical composition, preparation process, service performance, and microstructure; Based on the aforementioned key material parameters, a structured parameter set for representative volumetric element models with different microstructural features is generated using a parameterization script, including: A microstructure image of the material to be analyzed is obtained; and the microstructure image is analyzed using an image processing algorithm to obtain the statistical distribution parameters of the microstructure features. The key material parameters and the statistical distribution parameters are transformed into a set of quantified parameters; the set of quantified parameters is a set of parameters that can be processed by a calculation script and includes at least a range of values, a distribution form, and a unit. The structured parameter set is generated using the parameterization script based on the quantization parameter set.

3. The material property design method based on CPFEM as described in claim 1, characterized in that, The structured parameter set includes at least the geometric topology, material constitutive parameters, boundary loading conditions, and simulation task metadata of the corresponding representative volume element model; The CPFEM simulation software is driven by a high-performance computing platform to perform parallel computations on different sets of structured parameters, including: Based on the simulation task metadata, a simulation input file including modeling instructions is generated for the structured parameter set corresponding to each representative volume element model; The computational resources required for the simulation of each representative volume element model are estimated to obtain resource parameters; the resource parameters include at least the number of CPU cores, computation time, and memory size. Based on the resource parameters, computing resources are allocated to each of the simulation input files, and the CPFEM simulation software is driven to perform calculations in parallel.

4. The material property design method based on CPFEM as described in claim 3, characterized in that, Estimate the computational resources required for simulation of each representative volumetric element model to obtain resource parameters, including: From the structured parameter set, the intrinsic features of each representative volume element model are extracted; the intrinsic features include mesh generation parameters, material constitutive parameters, loading solution features, and derived feature parameters; Based on the inherent characteristics, the computational resources required for the simulation of each representative volume element model are estimated using a pre-constructed estimation model, thereby obtaining the resource parameters. The prediction model includes a random forest model for predicting the computation time and a linear regression model for predicting the memory size.

5. The material property design method based on CPFEM as described in claim 3, characterized in that, The method further includes: During the parallel computing process, task status information is acquired; When the task status information is in a specified state, the log information of the corresponding task is extracted; the specified state includes at least one of failure state, timeout state, and node failure. Based on the log information, the fault type is determined; the fault type includes a first type of fault caused by insufficient resources and a second type of fault caused by an error in the structured parameter set; When the fault type is the first type, the simulation input file corresponding to the fault task is parsed, the new resource parameters of the fault task are determined, and the calculation is retried. When the fault type is the second type, the simulation input file corresponding to the fault task is marked as an invalid task.

6. The material property design method based on CPFEM as described in claim 1, characterized in that, The key performance indicators include at least yield strength, tensile strength, and fracture strain; From the calculation results, key performance indicators for each representative volumetric element model are extracted, including: Extract the initial stress-strain curve from the calculation results; The initial stress-strain curve is cleaned and standardized to obtain the target stress-strain curve; Based on preset rules, the key performance indicators are determined from the target stress-strain curve; The preset rules include: for curves with a yield point, the stress value corresponding to the first point in the curve where the slope is less than a threshold is taken as the yield strength; for curves without a yield point, the yield strength is determined by the offset method; the maximum stress value on the curve is taken as the tensile strength; and the strain value corresponding to the termination point on the curve is taken as the fracture strain.

7. The material property design method based on CPFEM as described in claim 1, characterized in that, Constructing a mapping relationship between the structured parameter set and the key performance indicators to enable material performance prediction and reverse design based on the mapping relationship includes: Using the structured parameter set as input and the key performance indicators as output, a structural performance database for reverse design of the material is constructed. Based on the structural performance database, a correlation map of structural performance is generated; and using the structural performance database as a training dataset, a performance prediction model for predicting the material performance is trained; wherein, the correlation map includes at least one of a two-dimensional scatter plot, a parallel coordinate plot, and a three-dimensional scatter plot, and the correlation map is used to analyze the microstructural parameters corresponding to the expected material performance of the material.

8. A material property design device based on CPFEM, characterized in that, The device includes: The key parameter acquisition module is used to acquire the key material parameters of the material to be analyzed. The parameter set generation module is used to generate a structured parameter set of representative volume element models with different microstructural features based on the key material parameters and through a parameterized script. The parallel computing module is used to drive the CPFEM simulation software to perform parallel calculations on different sets of structured parameters through a high-performance computing platform, so as to obtain calculation results including the macroscopic mechanical response of each representative volume element model; The performance index extraction module is used to extract key performance indicators of each representative volume element model from the calculation results. The mapping relationship construction module is used to construct the mapping relationship between the structured parameter set and the key performance indicators, so as to perform material performance prediction and material reverse design based on the mapping relationship.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the CPFEM-based material property design method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the CPFEM-based material property design method according to any one of claims 1 to 7.

Images