Microalloy steel multi-mode mechanical property prediction method and device and readable storage medium

Abstract

The invention provides a microalloy steel multi-mode mechanical property prediction method and device and a readable storage medium, and relates to the technical field of steel performance prediction. The method comprises the following steps: constructing and verifying a phase field model based on thermodynamics and phase field dynamics of a microalloyed steel system to obtain a phase field simulation microstructure image of microalloyed steel; alloy component data and heat treatment process data on a production line are collected, and phase field simulation microscopic structure images and mechanical property indexes corresponding to the alloy component data and the heat treatment process data are obtained; performing standardized preprocessing and sample construction on the phase field simulation microscopic structure image, the alloy component data, the heat treatment process data and the mechanical property indexes, and performing data set division; establishing a multi-modal deep learning model containing an image branch and a numerical branch; carrying out training, evaluation and hyper-parameter optimization based on the training set and the test set, and determining an optimal multi-modal deep learning model; and determining a mechanical property prediction result of the to-be-tested sample based on the optimal multi-modal deep learning model.

Description

Technical Field

[0001] This application relates to the field of steel performance prediction technology, and in particular to a method for predicting the multimodal mechanical properties of microalloyed steel, a computer device, and a readable storage medium. Background Technology

[0002] Microalloyed steel, as a key basic material for structural components and energy, transportation, and engineering machinery, possesses strength, toughness, weldability, and formability, directly affecting the lightweighting and safe and reliable service of equipment. In production practice, the rapid and reliable acquisition of the mechanical properties of microalloyed steel is a prerequisite for process design, quality release, and service life assessment. Therefore, accurately predicting the mechanical properties of microalloyed steel has significant engineering implications.

[0003] Existing data-driven methods mostly use chemical composition and process parameters as inputs to machine learning models, with a small number of works supplementing them with manual statistical features, such as grain size. However, microstructure, which is directly related to mechanical properties, is often not directly incorporated, especially microstructure images containing information on grain morphology, orientation, and defects. Due to this limitation, machine learning models are prone to problems such as poor cross-batch generalization, sensitivity to parameter perturbations, and the need for extensive manual feature engineering. At the same time, relying on long-term mechanical testing for label acquisition is costly and time-consuming, making it difficult to meet the dual requirements of prediction accuracy and timeliness for rapid R&D and process control. Summary of the Invention

[0004] In view of this, this application provides a method for predicting the multimodal mechanical properties of microalloyed steel, a computer device, and a readable storage medium, which solves the above-mentioned technical problems.

[0005] In a first aspect, embodiments of this application provide a method for predicting the multimodal mechanical properties of microalloyed steel, including: Step S1: Based on the thermodynamics and phase field dynamics of the microalloyed steel system, a phase field model is constructed, and the phase field model is verified by scanning and optical microstructure experiments. The phase field simulated microstructure image of the microalloyed steel is obtained through the verified phase field model. Step S2: Collect alloy composition data and heat treatment process data from the production line, and obtain phase field simulation microstructure images and mechanical property indicators corresponding to the alloy composition data and heat treatment process data. Step S3 involves standardizing and preprocessing the phase-field simulation microstructure images, alloy composition data, heat treatment process data, and mechanical property indicators, constructing samples, and dividing the dataset into training and testing sets. Step S4: Establish a multimodal deep learning model that includes image branches and numerical branches; Step S5: Based on the training set and the test set, train, evaluate and optimize the hyperparameters of the multimodal deep learning model to determine the optimal multimodal deep learning model; Step S6: Based on the optimal multimodal deep learning model, and according to the alloy composition data, heat treatment process data, and target phase field simulation microstructure image of the sample to be tested, output the predicted mechanical properties of the sample to be tested.

[0006] In a second aspect, embodiments of this application provide a computer device including a processor and a memory, the memory storing programs or instructions that can run on the processor, the programs or instructions implementing the steps of the method as described in the first aspect when executed by the processor.

[0007] Thirdly, embodiments of this application provide a readable storage medium on which a program or instructions are stored, which, when executed by a processor, implement the steps of the method as described in the first aspect.

[0008] The multimodal mechanical property prediction method, computer equipment, and readable storage medium for microalloyed steel according to embodiments of this application first construct a high-precision phase-field model based on thermodynamics and phase-field dynamics, generating simulated microstructure images that are highly consistent with experimental results, and verifying them through scanning and optical microstructure experiments. Using alloy composition, heat treatment process, and simulated phase-field microstructure images as input, after normalization preprocessing and sample construction, a deep learning model including numerical and image branches is employed. An attention mechanism is introduced in the image branch to achieve multi-instance aggregation. In the fusion layer, numerical features and image feature vectors are concatenated and learned to map to the target performance, achieving unified prediction of multiple indicators. This method can significantly improve the generalization and stability of samples from different batches and with different processes, reduce reliance on expensive experiments, and provide an efficient and feasible technical approach for alloy design, process optimization, and production quality control.

[0009] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0010] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A flowchart illustrating the method for predicting the multimodal mechanical properties of microalloyed steel according to an embodiment of this application is shown. Figure 2A schematic diagram of phase-field simulation microstructure images according to an embodiment of this application is shown; Figure 3 An architecture diagram of a multimodal deep learning model according to an embodiment of this application is shown; Figure 4 A structural block diagram of a computer device according to an embodiment of this application is shown. Detailed Implementation

[0011] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0012] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0013] The following description, in conjunction with the accompanying drawings, details the method for predicting the multimodal mechanical properties of microalloyed steel, the computer equipment, and the readable storage medium provided in this application through specific embodiments and application scenarios. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0014] This application provides a method for predicting the multimodal mechanical properties of microalloyed steel, such as... Figure 1 As shown, the method includes: Step S1: Based on the thermodynamics and phase field dynamics of the microalloyed steel system, a phase field model is constructed, and the phase field model is verified by scanning and optical microstructure experiments. The phase field simulated microstructure image of the microalloyed steel is obtained through the verified phase field model.

[0015] In this step, the phase-field model is constructed and validated. Specifically, a high-precision phase-field model is constructed based on thermodynamics and phase-field dynamics, and validated through scanning and optical microstructure experiments. The phase-field model is used to generate simulated microstructure images corresponding to the experimental gold. Through the phase-field model, microstructure images of the microalloyed steel are obtained, such as... Figure 2The image shows images corresponding to different initial austenite grain sizes at different temperatures (840℃, 800℃, 750℃, and 650℃, respectively).

[0016] It should be noted that the phase-field model is constructed based on thermodynamics and phase-field dynamics, and is used to generate simulated images that are highly consistent with the experimental metallographic structure, replacing the metallographic images obtained by experiments in related technologies, reducing data acquisition costs and improving sample coverage.

[0017] In one embodiment of this application, step S1 specifically includes: Step S11: Based on the thermodynamic properties of the microalloyed steel system, establish the phase-field governing equations describing the microstructure evolution of the microalloyed steel system, and introduce a multi-physics field coupled free energy functional to construct a model framework reflecting the phase transition process; set the thermodynamic parameters of the microalloyed steel system using a thermodynamic database, and set the kinetic parameters according to actual process conditions; numerically discretize the model framework using the finite difference method or the finite element method, and set initial and boundary conditions to complete the construction of the phase-field model; Step S12: Perform heat treatment experiments on the microalloyed steel sample under conditions consistent with the phase-field simulation to prepare metallographic samples. Obtain experimental microstructure images under different processes using scanning electron microscopy (SEM) and optical microscopy (OM). Quantitatively compare the grain morphology, size distribution, phase composition, and interface characteristics of the experimental microstructure images with the phase-field prediction results obtained by the phase-field model to evaluate the prediction accuracy of the phase-field model. Systematically correct the thermodynamic terms or kinetic parameters in the phase-field model based on the key differences between the experimental microstructure images and the phase-field prediction results. Step S13: Obtain the microstructure image of the microalloyed steel by predicting it using the modified phase-field model.

[0018] In this embodiment, step S11 involves constructing the phase-field model. Specifically, based on the thermodynamic properties of the microalloyed steel system, a phase-field governing equation describing its microstructure evolution is established. By introducing a multiphysics field coupled free energy functional including chemical free energy, interfacial energy, and elastic strain energy, a model framework reflecting the phase transition process is constructed. Thermodynamic parameters of the system are set using a thermodynamic database, and kinetic parameters such as diffusion coefficient, mobility, and gradient energy coefficient are set according to actual process conditions. The model is numerically discretized using the finite difference method or the finite element method, and reasonable initial and boundary conditions are set to complete the construction of the phase-field model.

[0019] Step S12 involves experimental verification. Specifically, the microalloyed steel sample undergoes heat treatment experiments consistent with the simulation conditions to prepare metallographic samples. Microstructure images under different processing conditions are then acquired using scanning electron microscopy (SEM) and optical microscopy (OM). The experimentally observed grain morphology, size distribution, phase composition, and interface characteristics are quantitatively compared with the phase-field simulation results to evaluate the accuracy of the model predictions. For key differences, the thermodynamic terms or kinetic parameters in the model are systematically corrected.

[0020] In this embodiment, a phase-field model for the microstructure transformation of microalloyed steel is constructed based on thermodynamics and phase-field kinetics, and thermodynamic and kinetic parameters such as interfacial energy and diffusion coefficient are set. Furthermore, to systematically verify the accuracy and reliability of the constructed phase-field model, the simulation results are compared quantitatively with actual scanning electron microscopy and optical microscopy microstructure characterization experiments from multiple perspectives, and the phase-field model is then corrected.

[0021] Step S2: Collect alloy composition data and heat treatment process data from the production line, and obtain phase-field simulation microstructure images and mechanical property indicators corresponding to the alloy composition data and heat treatment process data.

[0022] In this step, alloy composition data and heat treatment process data from the production line are collected, and corresponding phase-field simulation microstructure images and mechanical property indicators are acquired simultaneously to form a dataset for model training and evaluation.

[0023] In one embodiment of this application, step S2 specifically includes: Step S21, Data Acquisition: Collect alloy composition data and heat treatment process data of the sample from the production line, and simultaneously acquire phase field simulation microstructure images of the sample.

[0024] Step S22: Data Alignment: Map and align the alloy composition data, heat treatment process data, phase field simulation microstructure images, and mechanical property indicators of each sample using the sample ID, and unify the units of measurement.

[0025] In this embodiment, alloy composition data and heat treatment process data are collected from the production line. The alloy composition data is in mass fraction, such as C, Mn, Cr, Ni, Mo, V, Nb, Ti, B, etc. The heat treatment process parameters include austenitizing temperature / time, cooling method and cooling rate, tempering temperature / time, etc. Simultaneously, phase-field simulation microstructure images of the target steel grade are acquired, covering multiple heats and multiple heat treatment batches; and metadata such as simulation dimension, mesh size, simulation area size, and spatial resolution are recorded.

[0026] Furthermore, tag and metadata alignment is performed. Specifically, the composition, process, image, and mechanical properties of each sample are mapped and aligned using a unique sample ID. That is, a mapping relationship of "composition-process-image-mechanical properties" is established using a unified sample ID. Then, the units of measurement are standardized, such as the unit of strength MPa and the unit of temperature ℃, and data consistency checks are completed. Among them, the mechanical properties include one or more of yield strength, tensile strength, elongation after fracture, impact absorption energy, and brittle transition temperature.

[0027] Step S3 involves standardizing and preprocessing the phase-field simulation microstructure images, alloy composition data, heat treatment process data, and mechanical property indicators, and constructing samples, then dividing the dataset into training and testing sets.

[0028] In this step, data preprocessing and sample construction are performed. Specifically, the microscopic tissue images and composition-process-performance numerical data are standardized and samples are constructed. The dataset is then divided into training and test sets according to the 80 / 20 rule.

[0029] In one embodiment of this application, step S3 specifically includes: Step S31, Image preprocessing: The phase field simulation microscopic tissue image is subjected to descaler and annotation processing, abnormal region processing, scale unification processing and gray value standardization processing, and is cropped into multiple image sub-images according to a fixed size, and data augmentation is performed by rotation, mirroring or mild affine transformation. Step S32, Numerical preprocessing: The alloy composition data and the heat treatment process data are processed for missing values, outliers, and numerical standardization. Step S33, Sample Construction and Division: The processed data samples are randomly shuffled and divided into training set and test set according to a ratio of 80%:20%. The data samples include the phase field simulation microstructure image, the alloy composition data, the heat treatment process data, and the mechanical property indicators.

[0030] In this embodiment, the phase field simulation image undergoes descaler and annotation removal, abnormal region masking, scale unification, and grayscale value standardization. It is then cropped into multiple image sub-images at a fixed size, and data augmentation is performed using rotation, mirroring, and mild affine transformations. Exemplarily, this involves descaler and annotation removal, thermal / defect pixel masking, standardizing the original image to a divisible size of 1024×1024, and using a histogram-based grayscale value remapping method (such as adaptive piecewise linear stretching or histogram matching) to standardize the data distribution and unify the numerical ranges between different simulation results. The entire image is cropped into sub-images in 224×224 windows and augmented with random rotation (±15°), horizontal / vertical flipping, mild affine transformation (scaling 0.9–1.1, translation ≤5%), and grayscale value perturbation (±10%). During training, 32–64 sub-images are randomly sampled for each sample, and during testing, all sub-images are covered.

[0031] Missing value handling, outlier truncation, and standardization are performed on the composition and process data, while missing indicator variables are retained to enhance model robustness. For example, missing components / process data are filled with the median and a corresponding "missing indicator variable" is added; outliers are truncated to the 1st / 99th percentile; and all numerical features are standardized to zero mean and unit variance.

[0032] Furthermore, the processed data samples were divided into training and test sets at a ratio of 80%:20%. To facilitate early stopping and parameter tuning, an additional 10% was reserved in the training set as an internal validation set. Before the division, all samples were randomly shuffled, and a fixed random seed was used to ensure reproducibility.

[0033] Step S4: Establish a multimodal deep learning model that includes image branches and numerical branches.

[0034] In this step, a deep learning model is built. Specifically, a multimodal deep learning model is established, which includes image and numerical branches. An attention mechanism is introduced into the image branch to aggregate multiple instances, and the image feature vectors and numerical features are combined in a fusion layer to form the model input.

[0035] An attention mechanism is introduced into the image branch to aggregate multiple instances, and the image feature vectors and numerical features are combined in the fusion layer to form the model input.

[0036] In one embodiment of this application, step S4 specifically includes: Step S41, Image Branch: Input the image sub-image into the convolutional neural network encoder, extract the feature vector of each image sub-image, and perform multi-instance aggregation of the features of multiple image sub-images of the same sample based on the attention mechanism to obtain the sample-level image representation; Step S42, Numerical Branch: Input the processed alloy composition data and the heat treatment process data into a multilayer perceptron (MLP) to obtain a sample-level numerical representation; Step S43, Multimodal fusion: The image representation and the numerical representation are spliced ​​and fused. The resulting fused features are further integrated by a gated linear unit (GLU) and a shared feature layer, and finally connected to a multi-task regression head to achieve joint prediction of multiple mechanical performance indicators.

[0037] In this embodiment, such as Figure 3 As shown, for the image branch, the image sub-image (1×224×224) is input into the convolutional neural network encoder to obtain the image feature vector. In one embodiment, the convolutional neural network encoder adopts a layered structure, including three layered structures, one global average pooling layer, and one fully connected layer. The layered structure includes, in sequence, a convolutional layer, a GeLU activation function, a layer normalization layer, and a max pooling layer. Specifically, three layers are set up, each containing two 3×3 convolutional layers, a GeLU activation function, and a layer normalization layer (LayerNorm), followed by a 2×2 max pooling layer for downsampling. The convolutional branch can be combined with a random depth (DropPath) to suppress overfitting. The feature channels of the three layers are 32, 64, and 128, respectively. After the output of the last layered structure, global average pooling is performed to obtain a 128-dimensional vector, which is then projected onto a 256-dimensional sub-image through a fully connected layer.

[0038] Furthermore, attention is used to aggregate multiple instances of features from multiple subgraphs of the same sample, with attention weights arranged according to... Calculations are performed to obtain sample-level image representations. , h is the weight of the i-th image subgraph. i Let V be the feature vector of the i-th image sub-image, where V and w are both learnable parameters.

[0039] For the numerical branch, the standardized alloy composition and heat treatment process data are input into a multilayer perceptron. In one embodiment, the multilayer perceptron includes multiple fully connected layers, batch normalization, and ReLU activation functions, with the first two fully connected layers each receiving batch normalization and ReLU activation. The number of hidden units in each fully connected layer is 64 to 256, for example, 128, 128, or 64, and the dropout of each fully connected layer is 0.1 to 0.3, for example, applying a random deactivation of 0.2. This yields a sample-level numerical representation. .

[0040] Furthermore, multimodal fusion is performed. Specifically, [the process involves]... and The feature-level concatenation is performed in the fusion layer to form a 320-dimensional vector, and then weighted fusion is performed by gated linear units (GLU) to obtain a 256-dimensional fused representation. Then, two layers of shared full connection are connected, and finally, the mechanical properties of the selected target are predicted simultaneously using parallel linear regression output heads.

[0041] Step S5: Based on the training set and the test set, train, evaluate and optimize the hyperparameters of the multimodal deep learning model to determine the optimal multimodal deep learning model.

[0042] In this step, the deep learning model is trained, evaluated, and its hyperparameters are optimized. Specifically, the model is trained and its hyperparameters are optimized on the training set, and then evaluated on the test set using the coefficient of determination R0. 2 The model is evaluated using mean squared error (MSE) as an indicator to determine the optimal deep learning model.

[0043] In one embodiment of this application, step S5 specifically includes: Step S51, Training: The AdamW optimizer is used to train the model on the training set, the loss function is Huber loss, and multi-task joint optimization is performed using homoscedastic uncertainty weighting. Step S52, Evaluation and Hyperparameter Optimization: Using R 2 The model is evaluated on the test set with MSE as the core evaluation metric. If the prediction result does not reach the preset threshold, the learning rate, batch size, training epochs, regularization strength, and number of hidden units are searched and optimized based on the training set until the prediction result reaches the preset threshold.

[0044] In this embodiment, the AdamW optimizer is used for end-to-end training with a weight decay of 0.05. The Huber loss function is used. Group learning rates are set for different layers: 1e-4 for the convolutional encoder and attention aggregation layer, and 5e-4 for the fusion and sharing layer. After a linear warm-up in the first 5 epochs, cosine annealing scheduling is switched. The system is organized into minimum batches of "samples" (8–16 samples per batch, with 32–64 subgraphs randomly sampled per sample). The Huber loss function (δ=1.0) is used. Homoscedastic uncertainty weighting is introduced when predicting multiple mechanical indices simultaneously to achieve multi-task joint optimization. The total number of training epochs is 100–150, and an early stopping strategy (patience value 15) is used on the internal validation set.

[0045] On the test set, the coefficient of determination R 2 The mean squared error (MSE) is used as the core indicator to evaluate model performance; when the indicator does not reach a preset threshold, such as R... 2When the value is ≥0.85, grid or Bayesian search is performed on hyperparameters such as learning rate, batch size, number of training rounds, regularization strength (weight decay / Dropout), and number of hidden units based on the training set. The optimal combination is selected on the internal validation set and retested on the test set to confirm it. Finally, the optimal model parameters and random seed are solidified.

[0046] Step S6: Based on the optimal multimodal deep learning model, and according to the alloy composition data, heat treatment process data, and target phase field simulation microstructure image of the sample to be tested, output the predicted mechanical properties of the sample to be tested.

[0047] In this step, performance prediction is performed. Specifically, based on the optimal deep learning model, and according to the alloy composition, experimental conditions, and microstructure images of the sample under test, the predicted mechanical properties of the sample are output.

[0048] In step S6, the sample to be tested is preprocessed according to the specifications defined in step S3, and then input into the optimal deep learning model to output the sample-level prediction results of one or more selected target mechanical properties, which include yield strength, tensile strength, elongation after fracture, impact absorption energy, and brittle transition temperature.

[0049] In one embodiment of this application, step S6 specifically includes: Step S61: Perform image preprocessing and numerical preprocessing on the alloy composition data, heat treatment process data, and target phase field simulation microstructure image of the sample to be tested. Step S62: Input the preprocessed alloy composition data, heat treatment process data, and target phase field simulation microstructure image into the optimal multimodal deep learning model, and output the mechanical property prediction results of the sample to be tested. The mechanical property prediction results include at least one of the following: yield strength, tensile strength, elongation after fracture, impact absorption energy, and brittle transition temperature.

[0050] In this embodiment, the microscopic tissue images obtained from phase-field simulation, along with the composition and process data, are preprocessed using the same workflow as in steps S31 and S32. Image cropping and intensity normalization are consistent with those used in the training phase, and no random enhancement is applied during the testing phase. The preprocessed multi-patch images and numerical vectors are input into the optimal multimodal model determined in step S5. Forward inference is performed in the order of subgraph encoding—attention aggregation—fusion—shared layer—regression head, outputting the sample-level predicted value of the selected target mechanical properties. Simultaneously, the sample ID, timestamp, model version, and prediction results are recorded for subsequent traceability and verification.

[0051] This application aims to address the shortcomings of existing prediction methods based on chemical composition and process parameters, such as insufficient utilization of microstructure information, weak batch-to-batch generalization ability, and reliance on numerous destructive tests. It proposes a multimodal mechanical property prediction method for microalloyed steel based on phase-field simulation and an attention mechanism. This method integrates alloy composition, heat treatment process values, and phase-field simulation microstructure images into the model input. Through standardized data construction and a multimodal fusion network, an attention mechanism is introduced to achieve intelligent aggregation of image sub-images, enabling high-precision prediction of the target mechanical properties of microalloyed steel. This meets the efficiency and reliability requirements of process design, quality release, and service evaluation.

[0052] Furthermore, while related multimodal methods incorporate microscopic images, they still suffer from the following shortcomings: they rely on experimental metallographic images, resulting in high costs, long cycles, and difficulty in covering the entire process space; their feature fusion methods are simple, such as dot multiplication or stitching, failing to fully consider the correlation between local image features and global semantics; and they lack an attention mechanism, making it impossible to differentiate and weight image sub-images, thus limiting the model's ability to capture key microscopic features. The multimodal mechanical property prediction method for microalloyed steel based on phase-field simulation and an attention mechanism proposed in this application does not rely on experimental metallographic images but instead uses phase-field simulated microscopic images and a multimodal deep learning model. This method fully considers the correlation between local image features and global semantics and introduces an attention mechanism to differentiate and weight image sub-images, enhancing the model's ability to capture key microscopic features.

[0053] This application also provides a computer device, such as... Figure 4 As shown, the computer device 400 includes a processor 401 and a memory 402. The memory 402 stores a program or instruction that can run on the processor 401. When the program or instruction is executed by the processor 401, it implements the various steps of the above-described embodiment of the multimodal mechanical property prediction method for microalloyed steel and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0054] The memory 402 can be used to store software programs and various data. The memory 402 may primarily include a first storage area for storing programs or instructions and a second storage area for storing data. The first storage area may store the operating system, application programs or instructions required for at least one function (such as sound playback, image playback, etc.). Furthermore, the memory 402 may include volatile memory or non-volatile memory, or both. The non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory can be random access memory (RAM), static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct memory bus RAM (DRRAM). The memory 402 in this embodiment includes, but is not limited to, these and any other suitable types of memory.

[0055] Processor 401 may include one or more processing units; optionally, processor 401 integrates an application processor and a modem processor, wherein the application processor mainly handles operations involving the operating system, user interface, and applications, and the modem processor mainly handles wireless communication signals, such as a baseband processor. It is understood that the aforementioned modem processor may also not be integrated into processor 401.

[0056] This application also provides a readable storage medium storing a program or instructions. When the program or instructions are executed by a processor, they implement the various processes of the above-described microalloyed steel multimodal mechanical property prediction method embodiment and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0057] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0058] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A method for predicting the multimodal mechanical properties of microalloyed steel, characterized in that, include: Step S1: Based on the thermodynamics and phase field dynamics of the microalloyed steel system, a phase field model is constructed, and the phase field model is verified by scanning and optical microstructure experiments. The phase field simulated microstructure image of the microalloyed steel is obtained through the verified phase field model. Step S2: Collect alloy composition data and heat treatment process data from the production line, and obtain phase field simulation microstructure images and mechanical property indicators corresponding to the alloy composition data and heat treatment process data. Step S3 involves standardizing and preprocessing the phase-field simulation microstructure images, alloy composition data, heat treatment process data, and mechanical property indicators, constructing samples, and dividing the dataset into training and testing sets. Step S4: Establish a multimodal deep learning model that includes image branches and numerical branches; Step S5: Based on the training set and the test set, train, evaluate and optimize the hyperparameters of the multimodal deep learning model to determine the optimal multimodal deep learning model; Step S6: Based on the optimal multimodal deep learning model, and according to the alloy composition data, heat treatment process data, and target phase field simulation microstructure image of the sample to be tested, output the predicted mechanical properties of the sample to be tested.

2. The method for predicting the multimodal mechanical properties of microalloyed steel according to claim 1, characterized in that, Step S3 specifically includes: Step S31, Image preprocessing: The phase field simulation microscopic tissue image is subjected to descaler and annotation processing, abnormal region processing, scale unification processing and gray value standardization processing, and is cropped into multiple image sub-images according to a fixed size, and data augmentation is performed by rotation, mirroring or mild affine transformation. Step S32, Numerical preprocessing: The alloy composition data and the heat treatment process data are processed for missing values, outliers, and numerical standardization. Step S33, Sample Construction and Division: The processed data samples are randomly shuffled and divided into training set and test set according to a ratio of 80%:20%. The data samples include the phase field simulation microstructure image, the alloy composition data, the heat treatment process data, and the mechanical property indicators.

3. The method for predicting the multimodal mechanical properties of microalloyed steel according to claim 2, characterized in that, Step S4 specifically includes: Step S41, Image Branch: Input the image sub-image into the convolutional neural network encoder, extract the feature vector of each image sub-image, and perform multi-instance aggregation of the features of multiple image sub-images of the same sample based on the attention mechanism to obtain the sample-level image representation; Step S42, numerical branch: Input the processed alloy composition data and the heat treatment process data into the multilayer sensor to obtain a sample-level numerical representation; Step S43, Multimodal fusion: The image representation and the numerical representation are spliced ​​and fused. The resulting fused features are further integrated by a gated linear unit and a shared feature layer, and finally connected to a multi-task regression head to achieve joint prediction of multiple mechanical performance indicators.

4. The method for predicting the multimodal mechanical properties of microalloyed steel according to claim 3, characterized in that, The convolutional neural network encoder for the image branch includes three hierarchical structures, one global average pooling layer, and one fully connected layer. The hierarchical structures sequentially include a convolutional layer, a GeLU activation function, a layer normalization, and a max pooling layer. The weights of the multi-instance aggregation are calculated as follows: ,in, h is the weight of the i-th image subgraph. i Let V be the feature vector of the i-th image sub-image, where V and w are both learnable parameters; The numerical branch of the multilayer perceptron includes multiple fully connected layers, as well as batch normalization and ReLU activation functions. The number of hidden units in each fully connected layer is 64 to 256, and the Dropout of each fully connected layer is 0.1 to 0.

3.

5. The method for predicting the multimodal mechanical properties of microalloyed steel according to claim 1, characterized in that, Step S5 specifically includes: Step S51, Training: The AdamW optimizer is used to train the model on the training set, the loss function is Huber loss, and multi-task joint optimization is performed using homoscedastic uncertainty weighting. Step S52, Evaluation and Hyperparameter Optimization: Using R 2 The model is evaluated on the test set with MSE as the core evaluation metric. If the prediction result does not reach the preset threshold, the learning rate, batch size, training epochs, regularization strength, and number of hidden units are searched and optimized based on the training set until the prediction result reaches the preset threshold.

6. The method for predicting the multimodal mechanical properties of microalloyed steel according to claim 1, characterized in that, Step S6 specifically includes: Step S61: Perform image preprocessing and numerical preprocessing on the alloy composition data, heat treatment process data, and target phase field simulation microstructure image of the sample to be tested. Step S62: Input the preprocessed alloy composition data, heat treatment process data, and target phase field simulation microstructure image into the optimal multimodal deep learning model, and output the mechanical property prediction results of the sample to be tested. The mechanical property prediction results include at least one of the following: yield strength, tensile strength, elongation after fracture, impact absorption energy, and brittle transition temperature.

7. The method for predicting the multimodal mechanical properties of microalloyed steel according to claim 1, characterized in that, Step S1 specifically includes: Step S11: Based on the thermodynamic properties of the microalloyed steel system, establish the phase-field governing equations describing the microstructure evolution of the microalloyed steel system, and introduce a multi-physics field coupled free energy functional to construct a model framework reflecting the phase transition process; set the thermodynamic parameters of the microalloyed steel system using a thermodynamic database, and set the kinetic parameters according to actual process conditions; numerically discretize the model framework using the finite difference method or the finite element method, and set initial and boundary conditions to complete the construction of the phase-field model; Step S12: Perform heat treatment experiments on the microalloyed steel sample under conditions consistent with the phase-field simulation to prepare metallographic samples. Obtain experimental microstructure images under different processes using scanning electron microscopy and optical microscopy. Quantitatively compare the grain morphology, size distribution, phase composition, and interface characteristics of the experimental microstructure images with the phase-field prediction results obtained by the phase-field model to evaluate the prediction accuracy of the phase-field model. Systematically correct the thermodynamic terms or kinetic parameters in the phase-field model based on the key differences between the experimental microstructure images and the phase-field prediction results. Step S13: Obtain the microstructure image of the microalloyed steel by predicting it using the modified phase-field model.

8. The method for predicting the multimodal mechanical properties of microalloyed steel according to claim 1, characterized in that, Step S2 specifically includes: Step S21, Data Acquisition: Collect alloy composition data and heat treatment process data of the sample from the production line, and simultaneously acquire phase field simulation microstructure images of the sample. Step S22: Data Alignment: Map and align the alloy composition data, heat treatment process data, phase field simulation microstructure images, and mechanical property indicators of each sample using the sample ID, and unify the units of measurement.

9. A computer device, characterized in that, It includes a processor and a memory, the memory storing a program or instructions that run on the processor, the program or instructions being executed by the processor to implement the steps of the method for predicting the multimodal mechanical properties of microalloyed steel as described in any one of claims 1 to 8.

10. A readable storage medium having a program or instructions stored thereon, characterized in that, When the program or instructions are executed by the processor, they implement the steps of the method for predicting the multimodal mechanical properties of microalloyed steel as described in any one of claims 1 to 8.

Images