Cast iron new material bidirectional design model construction method, device, equipment and medium

Abstract

The invention discloses a cast iron new material bidirectional design model construction method and device, equipment and a medium, and relates to the technical field of machine learning. The method comprises the following steps: acquiring multi-scale multi-modal data in a cast iron process; performing bucket separation processing on the multi-scale multi-modal data to obtain a training data set and a test data set; constructing a plurality of neural network models based on the weights of the neuron nodes and the connected nodes; performing bidirectional training on each neural network model according to the training data set and the test data set to obtain a plurality of cast iron new material multi-modal models; and on the basis of an active optimization method, a target cast iron new material bidirectional design model is obtained from all the cast iron new material multi-modal models. Therefore, by utilizing the target cast iron new material bidirectional design model to predict the mechanical property of the cast iron new material, element components and process parameters are predicted, and an optimal balance point between the strength and the plasticity of the cast iron new material is sought, so that the problem of inversion of the strength and the plasticity is solved.

Description

Technical Field

[0001] The present invention relates to the field of machine learning, and in particular to a method, device, equipment and medium for constructing a bidirectional design model for a new cast iron material. Background Art

[0002] In 2023, my country's total casting output will reach 53.65 million tons, with a market size of 1.17 trillion yuan. The total output will remain basically stable, ranking first in the world for 24 consecutive years. Among them, cast iron is the most important material for castings, accounting for 72.4% of the total output. It is widely used in almost all industrial categories such as machine tools, construction machinery, wind power, etc., and is an industry-wide basic material. However, there is an obvious mismatch between the strength and plasticity of current new cast iron materials. High strength is often accompanied by low plasticity, which limits its application in high-end castings and complex stress environments. Therefore, the development of high-performance new cast iron materials with strong plasticity matching has great economic value and strategic significance. Summary of the invention

[0003] In view of this, the purpose of the present invention is to overcome the shortcomings of the prior art and provide a method, device, equipment and medium for constructing a bidirectional design model of a new cast iron material, which is used to construct a bidirectional design model of a target new cast iron material using multi-scale and multi-modal data. The model can be used to predict the mechanical properties of the new cast iron material, and can also predict the elemental composition and process parameters. Combined with the organizational morphology, the structure-activity relationship of "composition-process-organization-performance" is established to seek the optimal balance between the strength and plasticity of the new cast iron material, thereby solving the problem of strength-plasticity inversion.

[0004] The present invention provides the following technical solutions:

[0005] In a first aspect, the present invention provides a method for constructing a bidirectional design model for a new cast iron material, comprising:

[0006] Obtain multi-scale and multi-modal data of the entire process of cast iron preparation;

[0007] The multi-scale multi-modal data is bucketed to obtain a training data set and a test data set;

[0008] Construct various neural network models based on neuron nodes and connected node weights;

[0009] Bidirectionally training each of the neural network models according to the training data set and the test data set to obtain multiple multimodal models of new cast iron materials;

[0010] Based on the active optimization method, a bidirectional design model of the target new cast iron material is obtained from the multimodal models of each of the new cast iron materials.

[0011] In one embodiment, the multi-scale multi-modal data includes new cast iron material characteristic data in digital mode and tissue morphology data in image mode, and the new cast iron material characteristic data includes element composition data, process parameter data and mechanical property data.

[0012] In one embodiment, forward training and reverse training are performed on each of the neural network models according to the training data set and the test data set to obtain a plurality of multimodal models of the new cast iron material;

[0013] Among them, the output data of the forward training is mechanical property data, and the output data of the reverse training is element composition and process parameter data.

[0014] In one implementation, the multi-scale multi-modal data is preprocessed to obtain a multi-scale multi-modal target database;

[0015] The multi-scale multi-modal target database is bucketed to obtain the training data set and the test data set.

[0016] In one embodiment, based on the active optimization method, the model parameters of each of the multimodal models of the cast iron new material are optimized to obtain a plurality of candidate multimodal models;

[0017] Obtaining the prediction accuracy of each of the candidate multimodal models, and determining the highest prediction accuracy from the prediction accuracy;

[0018] The candidate multimodal model corresponding to the highest prediction accuracy is used as the bidirectional design model of the target cast iron new material.

[0019] In one embodiment, for each of the candidate multimodal models, multiple quantitative indicators are calculated according to the predicted value of the candidate multimodal model and its corresponding actual value;

[0020] The prediction accuracy is determined according to each of the quantitative indicators.

[0021] In one embodiment, acquiring new multi-scale multi-modal data;

[0022] Based on forward design and reverse design, the target cast iron new material bidirectional design model is iteratively trained according to the new multi-scale multimodal data until the prediction results of the target cast iron new material bidirectional design model all meet multiple indicator thresholds, thereby obtaining a final cast iron new material design model.

[0023] In a second aspect, the present invention provides a device for constructing a bidirectional design model for a new cast iron material, comprising:

[0024] Acquisition module, used to acquire multi-scale and multi-modal data in the cast iron process;

[0025] A processing module, used for performing bucket processing on the multi-scale multi-modal data to obtain a training data set and a test data set;

[0026] Building modules for building various neural network models based on neuron nodes and connected node weights;

[0027] A training module, used for performing bidirectional training on each of the neural network models according to the training data set and the test data set to obtain multiple multimodal models of the new cast iron material;

[0028] The optimization module is used to obtain a target cast iron new material bidirectional design model from each of the cast iron new material multimodal models based on an active optimization method.

[0029] In a third aspect, the present invention proposes a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the method for constructing a bidirectional design model for a new cast iron material as described in the first aspect is implemented.

[0030] In a fourth aspect, the present invention proposes a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the method for constructing a bidirectional design model for a new cast iron material as described in the first aspect.

[0031] The method, device, equipment and medium for constructing a bidirectional design model for new cast iron materials disclosed in the present invention obtain multi-scale multi-modal data of the entire process of cast iron preparation; bucket the multi-scale multi-modal data to obtain a training data set and a test data set; construct a variety of neural network models based on neuron nodes and connected node weights; perform bidirectional training on each of the neural network models according to the training data set and the test data set to obtain a plurality of multi-modal models of new cast iron materials; and obtain a target bidirectional design model of new cast iron materials from each of the multi-modal models of new cast iron materials based on an active optimization method. In this way, various types of neural network models are bidirectionally trained using multi-scale and multi-modal data in the cast iron process to obtain multiple multi-modal models of new cast iron materials. Then, based on the active optimization method, a target bidirectional design model of new cast iron materials that meets the requirements is found from multiple multi-modal models of new cast iron materials. The target bidirectional design model of new cast iron materials can be used to predict the mechanical properties of the new cast iron materials, and its elemental composition and process parameters can also be predicted. Combined with the organizational morphology, the structure-activity relationship of "composition-process-organization-performance" is established to seek the optimal balance between the strength and plasticity of the new cast iron materials, thereby solving the problem of strength-plasticity inversion. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solution of the present invention, the following will briefly introduce the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope of protection of the present invention. In each of the drawings, similar components are numbered similarly.

[0033] Figure 1 A schematic diagram of a process of constructing a bidirectional design model for a new cast iron material proposed in this embodiment is shown;

[0034] Figure 2 Another schematic diagram of the process of constructing a bidirectional design model for a new cast iron material proposed in this embodiment is shown;

[0035] Figure 3 Another schematic diagram of the process of constructing a bidirectional design model for a new cast iron material proposed in this embodiment is shown;

[0036] Figure 4 Another schematic diagram of the process of constructing a bidirectional design model for a new cast iron material proposed in this embodiment is shown;

[0037] Figure 5 A schematic diagram of a device for constructing a bidirectional design model for new cast iron materials proposed in this embodiment is shown.

[0038] Description of the accompanying drawings:

[0039] 500 - bidirectional design model building device for new cast iron material; 501 - acquisition module; 502 - processing module; 503 - building module; 504 - training module; 505 - optimization module. DETAILED DESCRIPTION

[0040] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0041] The components of the embodiments of the present invention generally described and shown in the drawings herein may be arranged and designed in a variety of different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0042] Hereinafter, the terms "including", "having" and their cognates, which may be used in various embodiments of the present invention, are intended only to indicate specific features, numbers, steps, operations, elements, components or combinations of the foregoing items, and should not be understood as first excluding the existence of one or more other features, numbers, steps, operations, elements, components or combinations of the foregoing items or adding the possibility of one or more features, numbers, steps, operations, elements, components or combinations of the foregoing items.

[0043] Furthermore, the terms “first”, “second”, “third”, etc. are merely used for distinguishing descriptions and are not to be understood as indicating or implying relative importance.

[0044] Unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meanings as those generally understood by those skilled in the art to which the various embodiments of the present invention belong. The terms (such as those defined in generally used dictionaries) will be interpreted as having the same meanings as the contextual meanings in the relevant technical field and will not be interpreted as having idealized meanings or overly formal meanings unless clearly defined in the various embodiments of the present invention.

[0045] Example 1

[0046] The disclosed embodiment provides a method for constructing a bidirectional design model for a new cast iron material, which is used to construct a bidirectional design model for a target new cast iron material using multi-scale and multi-modal data. The model can then be used to predict the mechanical properties of the new cast iron material, and its elemental composition and process parameters. Combined with the organizational morphology, a structure-activity relationship of "composition-process-organization-performance" is established to seek the optimal balance between the strength and plasticity of the new cast iron material, thereby solving the problem of strength-plasticity inversion.

[0047] See also Figure 1 A method for constructing a bidirectional design model for a new cast iron material includes steps S101 to S105, and each step is described in detail below.

[0048] Step S101, obtaining multi-scale and multi-modal data of the entire cast iron production process.

[0049] In this embodiment, multi-scale and multi-modal data in the cast iron process are extracted in batches and one-to-one by using detectors to detect elemental composition, monitor process flow, take metallographic images, and test mechanical properties.

[0050] Among them, the multi-scale and multi-modal data include the new cast iron material characteristic data in the digital mode and the organizational morphology data in the image mode. The new cast iron material characteristic data include element composition data, process parameter data and mechanical property data.

[0051] For the tissue morphology data in the image mode, multi-scale can be expressed as images taken at different angles, distances or resolutions; for the characteristic data of new cast iron materials in the digital mode, multi-scale can be expressed as data at different time scales.

[0052] Using multi-scale and multimodal data, the model learns how to identify and understand data features at different scales, and can also understand how these features change at different scales. This ability enables the model to show better robustness and adaptability when processing new data that has not been seen, especially data of different sizes and shapes. At the same time, multimodal data covers information from image modalities and digital modalities, which is used to learn the associations and complementarities between different modalities, thereby understanding the data more comprehensively. In this way, through cross-modal learning, the model can achieve better generalization capabilities in new fields and adapt to new tasks and new data more quickly.

[0053] It should be added that element composition data include the content of copper Cu, silicon Si, manganese Mn and other elements, process parameter data include molten iron temperature, pouring temperature, etc., mechanical property data include hardness, yield strength and elongation, etc. The main form of organizational morphology data is metallographic images before and after corrosion.

[0054] Step S102, bucketing the multi-scale multi-modal data to obtain a training data set and a test data set.

[0055] In this embodiment, continuous multi-scale multimodal data are divided into discrete intervals (i.e., "buckets"), and the multi-scale multimodal data are classified according to these intervals, so that the multi-scale multimodal data are divided into two parts: one part is used to train the model (training data set), and the other part is used to evaluate the model performance (test data set).

[0056] In a specific embodiment, step S102 includes: preprocessing the multi-scale multi-modal data to obtain a multi-scale multi-modal target database; and bucketing the multi-scale multi-modal target database to obtain the training data set and the test data set.

[0057] In this embodiment, since the data quality of the multi-scale multi-modal data directly obtained is not high enough to a certain extent, it is necessary to preprocess the multi-scale multi-modal data to construct a multi-scale multi-modal target database; and then the pre-processed multi-scale multi-modal target database is bucketed to obtain a training data set and a test data set.

[0058] Bucket processing can divide a large amount of data into multiple buckets (or intervals) according to certain rules or standards, and the data in each bucket has similar characteristics or attributes. In this way, the originally complex data set is simplified into multiple small data sets that are easier to understand and analyze. For training sets and test sets, this helps to reduce the complexity and amount of data processing and improve the efficiency of data processing. In addition, bucket processing can also help the model better process data with different characteristics and improve the generalization ability of the model.

[0059] Specifically, the digital data such as the elemental composition, process parameters and mechanical properties of cast iron materials and the image data such as organizational morphology are sorted and numbered respectively. Different cast iron materials have independent numbers, which correspond to their own elemental composition, process parameters, organizational morphology and mechanical properties, so as to establish a corresponding multi-scale and multi-modal initial database of the entire process of "composition-process-organization-performance" of cast iron materials.

[0060] Furthermore, for digital data such as elemental composition, process parameters and mechanical properties in the multi-scale multi-modal initial database, blank values ​​were deleted, blank rows were converted to NaN, and some outliers caused by human or machine errors were adjusted; for image data such as tissue morphology, the image pixels were adjusted to 1920*1080, the format was unified to jpg or png format, the clarity was 30-40 points, and the contrast was as high as possible.

[0061] Furthermore, normalization and standardization are performed on digital data such as elemental composition, process parameters and mechanical properties; for image data such as organizational morphology, features such as phase composition and phase content before and after corrosion are extracted, and different descriptors are selected to describe a series of features in the form of numerical values ​​or strings to obtain digital data on image data such as organizational morphology, and normalization is performed first, dividing each column of data by the maximum absolute value of the column of data at the same time, so as to scale the data to a smaller range, and then the normalized data is standardized to convert the data into a standard normal distribution.

[0062] Furthermore, digital data such as elemental composition, process parameters and mechanical properties are combined with digital data obtained from image data such as tissue morphology. The formats of both are unified into numbers rather than strings, so they can be integrated and aligned by number to obtain a multi-scale and multi-modal target database.

[0063] Furthermore, the preprocessed multi-scale multi-modal target database is bucketed, and the dimensionality reduction methods such as tsne are used to sequentially calculate the similarity in high-dimensional space, construct the distribution in low-dimensional space, and optimize the point position by using the distribution difference between high-dimensional and low-dimensional space. The multi-dimensional data is reduced to two dimensions and a scatter plot is drawn. Cast iron materials with high similarity will be closer in the scatter plot, and several scattered points with close distances to each other will form a cluster. According to the image, the cluster to which each numbered cast iron material belongs can be obtained, thereby realizing bucketing of the multi-scale multi-modal target database.

[0064] Furthermore, the data after bucketing is renumbered, and a portion of the data is randomly extracted from each bucket according to a certain ratio (such as the number of training data sets: the number of test data sets = 9:1) to obtain the training data set and the test data set.

[0065] Step S103, constructing multiple neural network models based on neuron nodes and connected node weights.

[0066] In this embodiment, the weights of different neuron nodes and connected nodes are set to construct a variety of neural network models. For example, a feedforward neural network with a simple multi-layer structure and a one-way flow of data from the input layer to the output layer, a convolutional neural network that automatically learns spatial hierarchical features through a convolutional layer, a recursive neural network that contains feedback connections and can process sequence data, etc., which are not limited in this embodiment. Among them, the neuron node is the basic unit of the neural network, and the weight of the connected node is used to measure the interaction, adjust the learning process and feature learning.

[0067] Step S104, bidirectionally training each of the neural network models according to the training data set and the test data set to obtain multiple multimodal models of the new cast iron material.

[0068] In this embodiment, each neural network model is bidirectionally trained according to the training data set and the test data set to obtain multiple multimodal models of cast iron new materials. At this time, each neural network model is bidirectionally trained with the data set, and the corresponding multimodal model of cast iron new materials can be used to forward predict cast iron mechanical properties data and reversely predict element composition and process parameter data of cast iron new materials.

[0069] In a specific embodiment, step S104 includes: performing forward training and reverse training on each of the neural network models according to the training data set and the test data set, respectively, to obtain multiple multimodal models of the new cast iron material; wherein the output data of the forward training is mechanical property data, and the output data of the reverse training is elemental composition and process parameter data.

[0070] In this embodiment, each neural network model is forward trained and reversely trained using a training data set and a test data set, respectively, so that each neural network model can learn the characteristics of tissue morphology data, elemental composition data, process parameter data and mechanical property data according to the training data set and the test data set, and during the training process, the output data of the forward training is designed to be mechanical property data, and the output data of the reverse training is designed to be elemental composition and process parameter data, thereby obtaining a corresponding multimodal model of new cast iron materials that can be used for both forward prediction of cast iron mechanical property data and reverse prediction of elemental composition and process parameter data of new cast iron materials.

[0071] Among them, the output data of forward training is mechanical property data, the purpose is to optimize the prediction of mechanical property data; while the output data of reverse training is element composition and process parameter data, the purpose is to optimize the prediction of element composition and process parameter data.

[0072] Step S105: Based on an active optimization method, a bidirectional design model of a target new cast iron material is obtained from each of the multimodal models of the new cast iron material.

[0073] In this embodiment, the active optimization method includes grid search (exhaustively enumerate all combinations in parameter space and select the optimal combination), gradient descent method (use gradient information to guide parameter update and gradually approach the result) and Newton method (calculate the second-order derivative of the function), etc. Using the active optimization method, the target cast iron new material bidirectional design model can be a cast iron new material multimodal model with the best model parameters.

[0074] See also Figure 2 In a specific embodiment, step S105 includes steps S1051 to S1053, and each step is described in detail below.

[0075] Step S1051, based on the active optimization method, optimizing the model parameters of each of the multimodal models of the new cast iron material to obtain a plurality of candidate multimodal models.

[0076] In this embodiment, the selected active optimization method is used to search for the optimal parameter combination in the parameter space of each cast iron new material multimodal model. During the search process, the effects of different parameter combinations are evaluated based on the performance evaluation indicators of the model (such as accuracy, recall rate, F1 score, etc.). Based on the parameter combinations with better performance found during the optimization process, multiple candidate multimodal models are generated.

[0077] Step S1052, obtaining the prediction accuracy of each of the candidate multimodal models, and determining the highest prediction accuracy from the prediction accuracy.

[0078] In this embodiment, the prediction accuracy of each candidate multimodal model is obtained, and the highest prediction accuracy is determined from the prediction accuracy rates.

[0079] In order to obtain accurate prediction accuracy, a data set independent of the training data set can be used for prediction. This data set should contain true labels so that it can be compared with the prediction results of the model. For each candidate multimodal model, the data set is used for prediction to obtain prediction results. The prediction accuracy of the candidate multimodal model is then determined based on the prediction results and true labels. Among them, the prediction results include the prediction results of the forward design and the prediction results of the reverse design. The prediction results of the forward design are mechanical performance parameters, and the prediction results of the reverse design are element composition parameters and process parameters.

[0080] In a specific embodiment, step S1052 includes: for each of the candidate multimodal models, calculating multiple quantitative indicators based on the predicted value of the candidate multimodal model and its corresponding actual value; and determining the prediction accuracy rate based on each of the quantitative indicators.

[0081] In this embodiment, for each candidate multimodal model, multiple quantitative indicators are calculated based on the predicted value of the candidate multimodal model and its corresponding actual value; the quantitative indicators between the candidate multimodal models are compared to determine the prediction accuracy of each candidate multimodal model.

[0082] Among them, multiple quantitative indicators include Pearson correlation coefficient, explained variance and mean absolute error, etc. Generally speaking, if the Pearson correlation coefficient is closer to 1, the explained variance is closer to 1, and the mean absolute error is smaller, the model prediction accuracy is higher.

[0083] Therefore, the quantitative indicators among the candidate multimodal models are comprehensively measured, and the prediction accuracy of the candidate multimodal model whose quantitative indicators are closer to the above indicator standards is selected as the highest prediction accuracy. In addition, the indicator threshold can also be used to assist in measuring the prediction accuracy.

[0084] For example, if the Pearson correlation coefficient of candidate multimodal model 1 is 0.95, the explained variance is 0.9, and the mean absolute error is 1.22, and the Pearson correlation coefficient of candidate multimodal model 2 is 0.85, the explained variance is 0.9, and the mean absolute error is 1.22, it can be seen that at this time, candidate multimodal model 1 and candidate multimodal model 2 each have a quantitative indicator due to each other. Therefore, the coefficient indicator threshold corresponding to the Pearson correlation coefficient can be set to 0.9 according to the prediction requirements, and the variance indicator threshold corresponding to the explained variance can be set to 0.9, so that it can be determined that the prediction accuracy of candidate multimodal model 1 is higher.

[0085] Step S1053: using the candidate multimodal model corresponding to the highest prediction accuracy as the target cast iron new material bidirectional design model.

[0086] In this embodiment, the candidate multimodal model corresponding to the highest prediction accuracy is used as the target cast iron new material bidirectional design model. At this time, the target cast iron new material bidirectional design model has the best prediction effect on mechanical property data and element composition and process parameter data among all candidate multimodal models. The determination process of the target cast iron new material bidirectional design model is as follows: Figure 3 shown.

[0087] In a specific embodiment, step S105 includes: obtaining new multi-scale multi-modal data; based on forward design and reverse design, iteratively training the target cast iron new material bidirectional design model according to the new multi-scale multi-modal data until the prediction results of the target cast iron new material bidirectional design model meet multiple indicator thresholds, thereby obtaining the final cast iron new material design model.

[0088] In this embodiment, after the bidirectional design model of the target cast iron new material is determined, the model can be verified through experiments and iteratively optimized.

[0089] First, new multi-scale multi-modal data are acquired and preprocessed. Furthermore, the target cast iron new material bidirectional design model is iteratively trained according to the preprocessed new multi-scale multi-modal data to obtain the final cast iron new material design model, so that the elemental composition, process parameters and microstructure of the cast iron new material can be input through the final cast iron new material design model to accurately predict its mechanical properties (forward design). Conversely, the elemental composition and process parameters of the material can be accurately inferred (reverse design) by inputting the target performance requirements, thereby realizing the intelligent bidirectional design of cast iron new materials.

[0090] For details, see Figure 4 , iterative optimization includes steps S401 to S406.

[0091] Step S401: import the new multi-scale multi-modal data as input data into the final cast iron new material design model for iterative training, and output multiple experimental results.

[0092] Step S402, calculating corresponding multiple quantitative indicators according to various experimental results, and judging whether all quantitative indicators meet the standards according to the indicator threshold.

[0093] Specifically, if there is one quantitative indicator among all the quantitative indicators that does not meet the indicator threshold, step S403 is executed; if all the quantitative indicators meet the indicator threshold, step S406 is executed.

[0094] Step S403, determining substandard quantitative indicators from all quantitative indicators according to the indicator threshold, and removing the input data corresponding to the substandard quantitative indicators in the input data to obtain high-quality data.

[0095] Step S404: import the high-quality data into the final cast iron new material design model for iterative training, and output multiple experimental results.

[0096] Step S405, importing the experimental results and high-quality data in step S404 as input data into the final cast iron new material design model for iterative training, and outputting multiple new experimental results.

[0097] Specifically, step S402 is executed based on the new experimental results until all quantitative indicators corresponding to the experimental results output according to the input data meet the standards, and then step S406 is executed.

[0098] Step S406, iteration stops.

[0099] It should be noted that the iterative optimization step can first be based on the forward design to conduct experiments. When the quantitative indicators corresponding to the experimental results in the forward design are all up to standard, then based on the reverse design, the forward design experimental results that have all met the standards and their corresponding input data are used as the input data of the reverse design, and imported into the final cast iron new material design model for iterative training, and steps S402 to S405 are executed repeatedly until the quantitative indicators corresponding to the experimental results in the reverse design are all up to standard. The final cast iron new material design model obtained at this time can be used for accurate bidirectional prediction of actual cast iron new materials.

[0100] It should be noted that the active optimization method can be used to further optimize the model parameters of the final cast iron new material design model, thereby further improving the model prediction accuracy.

[0101] The method for constructing a bidirectional design model for new cast iron materials proposed in this embodiment obtains multi-scale multi-modal data in the cast iron process; buckets the multi-scale multi-modal data to obtain a training data set and a test data set; constructs multiple neural network models based on neuron nodes and connected node weights; bidirectionally trains each of the neural network models according to the training data set and the test data set to obtain multiple multi-modal models of new cast iron materials; and obtains a target bidirectional design model of new cast iron materials from each of the multi-modal models of new cast iron materials based on an active optimization method. In this way, various types of neural network models are bidirectionally trained using multi-scale and multi-modal data in the cast iron process to obtain multiple multi-modal models of new cast iron materials. Then, based on the active optimization method, a target bidirectional design model of new cast iron materials that meets the requirements is found from multiple multi-modal models of new cast iron materials. The target bidirectional design model of new cast iron materials can be used to predict the mechanical properties of the new cast iron materials, and its elemental composition and process parameters can also be predicted. Combined with the organizational morphology, the structure-activity relationship of "composition-process-organization-performance" is established to seek the optimal balance between the strength and plasticity of the new cast iron materials, thereby solving the problem of strength-plasticity inversion.

[0102] Example 2

[0103] In addition, the present disclosure provides a new cast iron material bidirectional design model construction device 500, see Figure 5 , the device comprises:

[0104] An acquisition module 501 is used to acquire multi-scale and multi-modal data of the entire process of cast iron preparation;

[0105] A processing module 502 is used to perform bucket processing on the multi-scale multi-modal data to obtain a training data set and a test data set;

[0106] A construction module 503 is used to construct a variety of neural network models based on neuron nodes and connected node weights;

[0107] A training module 504 is used to perform bidirectional training on each of the neural network models according to the training data set and the test data set to obtain multiple multimodal models of the new cast iron material;

[0108] The optimization module 505 is used to obtain a target cast iron new material bidirectional design model from each of the cast iron new material multimodal models based on an active optimization method.

[0109] Optionally, the multi-scale multi-modal data include new cast iron material characteristic data in digital mode and tissue morphology data in image mode, and the new cast iron material characteristic data include element composition data, process parameter data and mechanical property data.

[0110] Optionally, the training module 504 is also used to perform forward training and reverse training on each of the neural network models according to the training data set and the test data set, respectively, to obtain multiple multimodal models of the new cast iron material; wherein the output data of the forward training is mechanical property data, and the output data of the reverse training is elemental composition and process parameter data.

[0111] Optionally, the processing module 502 is further used to preprocess the multi-scale multi-modal data to obtain a multi-scale multi-modal target database; and perform bucket processing on the multi-scale multi-modal target database to obtain the training data set and the test data set.

[0112] Optionally, the optimization module 505 is also used to optimize the model parameters of each of the multimodal models of the new cast iron material based on the active optimization method to obtain multiple candidate multimodal models; obtain the prediction accuracy of each of the candidate multimodal models; determine the highest prediction accuracy from each of the prediction accuracy rates; and use the candidate multimodal model corresponding to the highest prediction accuracy as the target cast iron new material bidirectional design model.

[0113] Optionally, the optimization module 505 is further used to calculate multiple quantitative indicators for each candidate multimodal model according to the predicted value of the candidate multimodal model and its corresponding actual value; and determine the prediction accuracy according to the indicator threshold and each quantitative indicator.

[0114] Optionally, the optimization module 505 is also used to obtain new multi-scale multi-modal data; based on forward design and reverse design, the target cast iron new material bidirectional design model is iteratively trained according to the new multi-scale multi-modal data until the prediction results of the target cast iron new material bidirectional design model all meet multiple indicator thresholds, thereby obtaining the final cast iron new material design model.

[0115] The device provided in the embodiment of the present disclosure can execute the steps of the method for constructing a bidirectional design model for new cast iron materials provided in Example 1, which will not be described again to avoid repetition.

[0116] The method and device for constructing a bidirectional design model for new cast iron materials proposed in this embodiment obtains multi-scale multi-modal data in the cast iron process; buckets the multi-scale multi-modal data to obtain a training data set and a test data set; constructs multiple neural network models based on neuron nodes and connected node weights; bidirectionally trains each of the neural network models according to the training data set and the test data set to obtain multiple multi-modal models of new cast iron materials; and based on an active optimization method, obtains a target bidirectional design model of new cast iron materials from each of the multi-modal models of new cast iron materials. In this way, various types of neural network models are bidirectionally trained using multi-scale and multi-modal data in the cast iron process to obtain multiple multi-modal models of new cast iron materials. Then, based on the active optimization method, a target bidirectional design model of new cast iron materials that meets the requirements is found from multiple multi-modal models of new cast iron materials. The target bidirectional design model of new cast iron materials can be used to predict the mechanical properties of the new cast iron materials, and its elemental composition and process parameters can also be predicted. Combined with the organizational morphology, the structure-activity relationship of "composition-process-organization-performance" is established to seek the optimal balance between the strength and plasticity of the new cast iron materials, thereby solving the problem of strength-plasticity inversion.

[0117] Example 3

[0118] In addition, an embodiment of the present disclosure provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the method for constructing a bidirectional design model for a new cast iron material described in Example 1 is implemented.

[0119] The device provided in the embodiment of the present disclosure can execute the steps of the method for constructing a bidirectional design model for new cast iron materials provided in Example 1, which will not be described again to avoid repetition.

[0120] Example 4

[0121] The embodiment of the present disclosure provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the method for constructing a bidirectional design model for a new cast iron material described in Embodiment 1 is implemented.

[0122] In this embodiment, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0123] The computer-readable storage medium provided in this embodiment can implement the method for constructing a bidirectional design model for new cast iron materials provided in Example 1, and will not be described in detail here to avoid repetition.

[0124] In all examples shown and described herein, any specific values ​​should be interpreted as merely exemplary and not as limiting, and thus other examples of the exemplary embodiments may have different values.

[0125] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0126] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the present invention. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.

Claims

1. A method for constructing a bidirectional design model for a new cast iron material, characterized in that: include: Obtain multi-scale and multi-modal data of the entire process of cast iron preparation; The multi-scale multi-modal data is bucketed to obtain a training data set and a test data set; Construct various neural network models based on neuron nodes and connected node weights; Bidirectionally training each of the neural network models according to the training data set and the test data set to obtain multiple multimodal models of new cast iron materials; Based on the active optimization method, a bidirectional design model of the target new cast iron material is obtained from the multimodal models of each of the new cast iron materials.

2. The method for constructing a bidirectional design model for new cast iron materials according to claim 1, characterized in that: The multi-scale multi-modal data include new cast iron material characteristic data in digital mode and tissue morphology data in image mode, and the new cast iron material characteristic data include element composition data, process parameter data and mechanical property data.

3. The method for constructing a bidirectional design model for new cast iron materials according to claim 1, characterized in that: The bidirectional training of each neural network model is performed according to the training data set and the test data set to obtain multiple multimodal models of new cast iron materials, including: According to the training data set and the test data set, each of the neural network models is forward trained and reversely trained to obtain a plurality of multimodal models of the new cast iron material; Among them, the output data of the forward training is mechanical property data, and the output data of the reverse training is element composition and process parameter data.

4. The method for constructing a bidirectional design model for new cast iron materials according to claim 1, characterized in that: The step of performing bucket processing on the multi-scale multi-modal data to obtain a training data set and a test data set includes: Preprocessing the multi-scale multi-modal data to obtain a multi-scale multi-modal target database; The multi-scale multi-modal target database is bucketed to obtain the training data set and the test data set.

5. The method for constructing a bidirectional design model for new cast iron materials according to claim 1, characterized in that: The active optimization method is used to obtain a bidirectional design model of a target new cast iron material from each of the multimodal models of the new cast iron material, including: Based on the active optimization method, the model parameters of each of the multimodal models of the cast iron new material are optimized to obtain a plurality of candidate multimodal models; Obtaining the prediction accuracy of each of the candidate multimodal models, and determining the highest prediction accuracy from the prediction accuracy; The candidate multimodal model corresponding to the highest prediction accuracy is used as the bidirectional design model of the target cast iron new material.

6. The method for constructing a bidirectional design model for new cast iron materials according to claim 5, characterized in that: The obtaining of the prediction accuracy of each of the candidate multimodal models comprises: For each of the candidate multimodal models, a plurality of quantitative indicators are calculated according to the predicted value of the candidate multimodal model and the corresponding actual value; The prediction accuracy is determined according to each of the quantitative indicators.

7. The method for constructing a bidirectional design model for new cast iron materials according to claim 3, characterized in that: After obtaining the target cast iron new material bidirectional design model from each of the cast iron new material multimodal models based on the active optimization method, the method includes: Acquire new multi-scale and multi-modal data; Based on forward design and reverse design, the target cast iron new material bidirectional design model is iteratively trained according to the new multi-scale multimodal data until the prediction results of the target cast iron new material bidirectional design model all meet multiple indicator thresholds, thereby obtaining a final cast iron new material design model.

8. A device for constructing a bidirectional design model for new cast iron materials, characterized in that: include: Acquisition module, used to acquire multi-scale and multi-modal data in the cast iron process; A processing module, used for performing bucket processing on the multi-scale multi-modal data to obtain a training data set and a test data set; Building modules for building various neural network models based on neuron nodes and connected node weights; A training module, used for performing bidirectional training on each of the neural network models according to the training data set and the test data set to obtain multiple multimodal models of the new cast iron material; The optimization module is used to obtain a target cast iron new material bidirectional design model from each of the cast iron new material multimodal models based on an active optimization method.

9. A computer device, characterized in that: It comprises a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the method for constructing a bidirectional design model for a new cast iron material as claimed in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: It stores a computer program, which, when executed by a processor, implements the method for constructing a bidirectional design model for a new cast iron material as described in any one of claims 1 to 7.

Images