Material property prediction method and apparatus

Abstract

The present disclosure provides a material property prediction method and device, which firstly acquires component data of a plurality of sample materials as training data, standardizes the training data, linearly interpolates the standardized training data to obtain a training data set, then trains a material property prediction model according to the training data set, inputs material data to be measured into the trained material property prediction model to obtain a material property prediction result, removes elements affecting the performance of the model by standardizing the training data, thereby improving the quality of the training data, and increases the size of the data set by using a linear interpolation method, which can be well applied to small sample material property prediction, and also uses a neural network composed of full connection layers to accurately predict the material property, and the method combining data expansion and neural networks is superior to the automatic machine learning method.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of chemical materials, and in particular, to a material property prediction method and device. BACKGROUND

[0002] Machine learning has achieved remarkable success in data-intensive applications, but in the case of a small number of training samples, the current machine learning method always fails to achieve the desired performance. In the field of materials science, machine learning methods have solved many problems in materials science, especially in predicting basic properties. More and more general models allow people to quickly screen materials in a wider design space.

[0003] The precondition for the success of the above applications is to have enough training examples. For example, Jonathan Schmidt et al. proved in the perovskite compound formation energy prediction experiment that when the training set is doubled, the prediction error is reduced by about 20%. However, the data set in the field of chemical reactions is usually limited, and it is possible that it does not support the accurate operation of the machine learning model. Therefore, there is a wide demand for small sample learning in the field of chemistry at present. SUMMARY

[0004] To solve at least one of the above technical problems, the present disclosure provides a material property prediction method and device.

[0005] The first aspect of the present disclosure provides a material property prediction method, comprising: obtaining component data of a plurality of sample materials as training data; standardizing the training data; performing linear interpolation on the standardized training data to obtain a training data set; training a material property prediction model according to the training data set; inputting the material data to be tested into the trained material property prediction model to obtain a material property prediction result.

[0006] According to one embodiment of the present disclosure, the standardization of the training data comprises: identifying all chemical element categories contained in the training data; filtering the chemical element categories to obtain a first element category set; and standardizing the training data of all element categories in the first element category set.

[0007] According to one embodiment of the present disclosure, filtering the chemical element categories to obtain a first element category set comprises: determining the number of data items containing each chemical element; and determining a first element category from the chemical element categories according to the number of data items.

[0008] According to one embodiment of the present disclosure, determining the first element category from the chemical element categories according to the number of data items comprises: determining a plurality of second element categories with the lowest number of data items from the chemical element categories; and screening the second element categories from all element categories, and taking the remaining element categories as the first element category.

[0009] According to one embodiment of the present disclosure, normalizing the training data of all element categories in the first element category set comprises: for each element category in the first element category set, determining a mass proportion value of the element category; determining a maximum mass proportion value of each element from the mass proportion values; and determining a ratio between each mass proportion value and the maximum mass proportion value as the normalized training data.

[0010] According to one embodiment of the present disclosure, linearly interpolating the normalized training data comprises: inserting a plurality of data between each two training data in a linear interpolation manner.

[0011] According to one embodiment of the present disclosure, the plurality of inserted data is four data.

[0012] According to one embodiment of the present disclosure, the material property is yield strength, elongation or fracture toughness.

[0013] According to one embodiment of the present disclosure, the material property prediction model adopts a deep neural network (DNN).

[0014] The second aspect of the present disclosure provides a material property prediction device, comprising: a memory storing execution instructions; and a processor executing the execution instructions stored in the memory, so that the processor executes the material property prediction method of any one of the above embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0015] The accompanying drawings illustrate exemplary embodiments of the present disclosure and together with the general description given above, explain the principles of the present disclosure, in which the drawings serve as an example to provide a further understanding of the present disclosure, and the drawings are included in the present specification and form a part of the present specification.

[0016] Figure 1 is a flowchart of a material property prediction method according to one embodiment of the present disclosure.

[0017] Figure 2 is a flowchart of normalizing training data according to one embodiment of the present disclosure.

[0018] Figure 3 is a flowchart of screening chemical element categories according to one embodiment of the present disclosure.

[0019] Figure 4 This is a statistical diagram showing the number of materials containing each chemical element in the training data.

[0020] Figure 5 This is a schematic diagram of a process for standardizing training data according to another embodiment of the present disclosure.

[0021] Figure 6 This is a statistical diagram illustrating the average element mass percentage before standardization, according to one embodiment of this disclosure.

[0022] Figure 7 It corresponds to Figure 6 A schematic diagram of the element mass distribution.

[0023] Figure 8 This is a statistical diagram showing the average element mass percentage after standardization according to one embodiment of the present disclosure.

[0024] Figure 9 It corresponds to Figure 8 A schematic diagram of the element mass distribution.

[0025] Figure 10 This is a schematic diagram of the distribution of the original training data.

[0026] Figure 11 This is a schematic diagram showing the distribution of the training data after interpolation.

[0027] Figure 12 This is the parameter table for the material property prediction model.

[0028] Figure 13 This is a performance comparison table of material property prediction models under different data preprocessing methods.

[0029] Figure 14 This is a performance comparison table of material property prediction models under different numbers of interpolations.

[0030] Figure 15 This is a performance comparison table between different models.

[0031] Figure 16 This is a schematic diagram of a material property prediction device employing a hardware implementation of a processing system according to one embodiment of the present disclosure. Detailed Implementation

[0032] The present disclosure will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the disclosure. Furthermore, it should be noted that, for ease of description, only the parts relevant to the present disclosure are shown in the accompanying drawings.

[0033] It should be noted that the embodiments and features in the present disclosure can be combined with each other in the case of no conflict. The technical solutions of the present disclosure will be described in detail below with reference to the drawings and in combination with the embodiments.

[0034] Unless otherwise specified, the exemplary embodiments / instances shown will be understood as providing exemplary features of various details that can implement the technical concepts of the present disclosure in practice. Therefore, unless otherwise specified, the features of various embodiments / instances can be additionally combined, separated, interchanged and / or rearranged without departing from the technical concepts of the present disclosure.

[0035] The terms used herein are for the purpose of describing specific embodiments and are not limiting. As used herein, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. Furthermore, the use of the term "comprise" and / or "include" and variations thereof, in the specification and claims, is intended to mean that the features, integers, steps, operations, components, components, and / or elements listed are present but not excluding the presence of one or more additional features, integers, steps, operations, components, components, and / or elements. It is also noted that, as used in this specification, the terms "substantially," "approximately," and other similar terms are used as terms of approximation and not as terms of degree, and as such, they are used to explain that the present disclosure is not intended to be limited to the precise values, calculations, and / or provided values that would be recognized by one of ordinary skill in the art.

[0036] The material property prediction method and device of the present disclosure are described below with reference to the accompanying drawings.

[0037] Figure 1 is a flowchart of a material property prediction method according to one embodiment of the present disclosure. Please refer to Figure 1 The material property prediction method of the present embodiment can include the following steps.

[0038] S100, obtaining component data of a plurality of sample materials as training data.

[0039] The material property to be predicted can be yield strength, elongation or fracture toughness. The yield strength is the yield limit of the metal material when the yield phenomenon occurs, that is, the stress resisting slight plastic deformation. The elongation refers to the percentage of the total deformation of the gauge length segment after the specimen is stretched to break to the original gauge length. The fracture toughness characterizes the ability of the material to resist crack propagation and is a quantitative indicator of the toughness of the material.

[0040] The training data can specifically use the MatBench v0.1 dataset. The MatBench v0.1 dataset contains mechanical property data of more than 800 steels, including yield strength data, elongation data, and fracture toughness data, etc., and takes components as material primitives instead of structures.

[0041] This embodiment takes the yield strength material attribute as an example for illustration. The yield strength attribute prediction is a regression task. The yield strength attribute prediction task can be achieved by using the matbench_v0.1 matbench_steels task contained in the MatBench v0.1 dataset. The matbench_v0.1 matbench_steels task contains 312 component data, part of which is used as training data and the other part is used as test data. These component data are used to predict the yield strength of steel from chemical composition, and each component data is composed of contained elements and the proportion of the mass of the corresponding element in the total mass.

[0042] S200, standardizing the training data.

[0043] The matbench_steels task of MatBench v0.1 only contains 312 data, which belongs to a small sample task. However, the data plays a crucial role in determining the performance of the machine learning model. There may be a large amount of incomplete, inconsistent, and even abnormal data in the original data. Therefore, by standardizing the training data, the quality of the data is improved, and the abnormality in the original data is eliminated, so that the data can adapt to the needs of the model and match the model.

[0044] Figure 2 is a flowchart of standardizing the training data according to one embodiment of the present disclosure. Referring to Figure 2 , step S200 can specifically include steps S210 to S230.

[0045] S210, identifying all chemical element species contained in the training data.

[0046] The training data in the above-mentioned 312 component data is identified by chemical element species, and it is determined that the training data contains 14 elements including Co (cobalt), Ni (nickel), W (tungsten), Nb (niobium), Cr (chromium), Al (aluminum), Fe (iron), C (carbon), V (vanadium), Ti (titanium), Mo (molybdenum), N (nitrogen), Mn (manganese), and Si (silicon) elements. A 14-dimensional vector composed of atomic mass ratio can be used to encode each data.

[0047] S220, screening the chemical element species to obtain a first element species set.

[0048] By screening the above-mentioned 14 elements, some elements that have a negative impact on the performance of the model are screened, thereby eliminating the abnormality in the original data, and adapting the data to the needs of the model and matching the model. After screening, the selected elements form a first element category set.

[0049] Figure 3 is a flowchart of screening chemical element categories according to one embodiment of the present disclosure. Referring to Figure 3 , step S220 can specifically include the following steps S221 and S222.

[0050] S221, for each chemical element, determining the number of data items containing the chemical element.

[0051] The number of data items of a chemical element is the number of materials containing the chemical element. Figure 4 is a statistical diagram of the number of materials containing each chemical element in the training data. Referring to Figure 4 , Element represents a chemical element, and numbers represent the number of materials containing the chemical element.

[0052] S222, determining a first element category set from the chemical element categories according to the number of data items.

[0053] Chemical element screening can be performed by analyzing the number of materials (component data) containing the chemical element, thereby obtaining the first element category set.

[0054] Exemplarily, a plurality of second element categories with the lowest number of data items can be determined from the chemical element categories, the second element categories are screened out from all element categories, and the remaining element categories are taken as the first element categories.

[0055] It can be seen from Figure 4 that among the chemical elements, the chemical element with the least number of materials containing the chemical element is the W element, followed by the N element, followed by the Ti element, followed by the C element, and so on. Thus, the k elements with the lowest number of data items are determined, k>1. Since the number of data items of the W and N elements in the plurality of elements with the lowest number of data items is significantly lower than that of other elements, the W and N elements can be screened out as second elements, at this time k=2, and the first element category set obtained contains Co, Ni, Nb, Cr, Al, Fe, C, V, Ti, Mo, Mn and Si elements.

[0056] Element screening is equivalent to data preprocessing, which eliminates the abnormality in the original data through preprocessing, and adapts the data to the needs of the model and matches the model.

[0057] It can be understood that the W element can also be screened out as the second element only, and k = 1. Or the W, N and Ti elements are all screened out as the second element, and k = 3. There are many specific screening methods. When k = 3, the second element includes Ti, W and N elements. The optional screening methods at this time include: only screening out the Ti element, only screening out the W element, only screening out the N element, only screening out the Ti and W elements, only screening out the Ti and N elements, only screening out the W and N elements, and screening out the Ti, W and N elements.

[0058] The specific value of n can be selected according to the difference between the number of data items. For example, for the W element with the lowest number of data items and the N element with the second lowest number of data items, the number of data items of both is less than 50, and the number of data items of the Ti element with the third lowest number of data items is about 200, which is significantly higher than the number of data items of the W and N elements. Therefore, only the W and N elements can be selected as the second element category.

[0059] S230, standardizing the training data according to all element categories in the first element category set.

[0060] Figure 5 is a flowchart of standardizing training data according to another embodiment of the present disclosure. Referring to Figure 5 , step S230 can specifically include steps S231 to S233.

[0061] S231, for each element category in the first element category set, determining the mass fraction value of the element category.

[0062] The first element category set contains the above-mentioned 14 elements. By analyzing the component data, the mass fraction data of each element in the component data is determined. In addition, the mass fraction average value can also be obtained according to the mass fraction value.

[0063] Figure 6 is a statistical diagram of the element mass fraction average value before standardization according to an embodiment of the present disclosure. Referring to Figure 6 , where Average represents the average value. The mass fraction average value of the Fe element is more than 0.7, and the mass fraction average values of other elements are smaller, and even the mass fraction average value of some elements (such as the Nb element) is close to zero.

[0064] Figure 7 is an element mass fraction distribution diagram corresponding to Figure 6 . Referring to Figure 7wherein, Qualityproportion represents the quality proportion. The proportion of the mass of the Fe element in the total mass of the material is mainly distributed between 0.6 and 0.8, and the proportion of the mass of other elements in the total mass of the material is mainly distributed between 0 and 0.2. It can be seen that the mass proportions of the elements in the first element category set are unevenly distributed. Therefore, data standardization is needed.

[0065] S232, determining the maximum mass proportion value of each element from the mass proportion values.

[0066] The data sample is usually multi-dimensional, and each dimension usually has a different distribution. When performing comprehensive analysis, the role of the dimension with a higher characteristic value level can be highlighted, and the role of the dimension with a lower characteristic value level can be relatively weakened. At this time, each element corresponds to a dimension. Therefore, for each element category in the first element category set, the corresponding maximum mass proportion value is determined. The maximum mass proportion value of each element category is the Y-axis value of the distribution point located at the highest position in Figure 7 .

[0067] S233, determining the ratio between each mass proportion value and the maximum mass proportion value as the standardized training data.

[0068] For example, the maximum mass proportion value of the Fe element is 0.86, and the mass proportion values corresponding to the Fe element include a first value 0.75 and a second value 0.66, then the standardized new first value = 0.75 / 0.86, and the standardized new second value = 0.66 / 0.86. The standardization of other mass proportion values of the Fe element is sequentially extended. The standardization of other elements in the first element category set is sequentially extended. The new value obtained after standardization is the new dimension value.

[0069] Figure 8 is a statistical diagram of the average element mass proportion after standardization according to one embodiment of the present disclosure. Figure 9 is an element mass proportion distribution diagram corresponding to Figure 8 . Referring to Figure 9 , wherein, Qualityproportion represents the quality proportion. Referring to Figure 8 and Figure 9 , the gap between the average mass proportions of the elements is reduced compared to before standardization, and the mass proportion distribution of each element is more uniform compared to before standardization. Through standardization processing, the characteristic values of different dimensions are in the same order of magnitude, and the accuracy of the model is improved.

[0070] S300: Linear interpolation is performed on the standardized training data to obtain the training dataset. Linear interpolation increases the dataset size. Specifically, step S300 may include the following steps: inserting multiple data points between every two training data points using linear interpolation.

[0071] Since the dataset provided in the matbench_v0.1 matbench_steels task only contains 312 samples, it belongs to the small dataset learning task. Figure 10 This is a schematic diagram illustrating the distribution of the original training data. (See also...) Figure 10 This shows all the sample points of the training data. Therefore, data augmentation can be used to increase the number of data samples in the dataset, thereby improving the training effect and performance of the model.

[0072] Specifically, four new data points can be inserted between every two training data points using linear interpolation. Figure 11 This is a schematic diagram showing the distribution of the training data after interpolation. (See also...) Figure 11 The number of samples was expanded from the original to 124,744, forming the training dataset. Understandably, the test set data from the 312 samples was not interpolated.

[0073] S400 trains the material property prediction model based on the training dataset.

[0074] Material property prediction models can employ deep neural networks (DNNs). Deep neural networks (DNNs) use deep structures within neural networks, consisting of input layers, output layers, and hidden layers. DNNs are widely used in the field of artificial intelligence.

[0075] The material property prediction model used in this step can include 10 fully connected layers. Figure 12 This is the parameter table for the material property prediction model. (See also...) Figure 12 This includes the number of neurons in each fully connected layer, the activation function used, and the number of parameters.

[0076] The interpolated training data is input into the material property prediction model to train the model and obtain a trained material property prediction model.

[0077] S500 inputs the data of the material to be tested into the pre-trained material property prediction model to obtain the material property prediction results.

[0078] The following is an explanation of the evaluation process for the material property prediction model.

[0079] In evaluating the material property prediction model, the model can be evaluated by using the mean absolute error (MAE). The mean absolute error (MAE) is the average of the absolute values of the deviations of all individual observations from the arithmetic mean. The mean absolute error can avoid the problem of mutual offset of errors, and thus can accurately reflect the size of the actual prediction error.

[0080] The calculation process of MAE is as follows:

[0081]

[0082] where y i is the i-th real value, is the i-th predicted value, and n is the number of data input into the model. By summing the absolute values of the differences between the n real values and the n predicted values and then taking the average, the mean absolute error MAE is obtained. In evaluating the model, five-fold nested cross-validation can be used, and the random seed is set to 18012019. Therefore, the final evaluation standard can be the average of the five scores MAE_mean.

[0083] Regarding data preprocessing, Figure 13 is a data table for comparing the performance of material property prediction models under different data preprocessing methods. See Figure 13 In the evaluation, the method of using dimensionality reduction to process data is involved in the comparison of preprocessing, including PCA, KPCA and LLE. The PCA (Principle Component Analysis) algorithm refers to using dimensionality reduction to convert multiple indicators into a few comprehensive indicators. The KPCA (Kernel Principle Component Analysis) algorithm refers to mapping sample data from the input space to a high-dimensional feature space through nonlinear transformation, and then extracting features in the high-dimensional feature space using PCA. The LLE (Locally Linear Embedding) algorithm maps signals in a high-dimensional space to a low-dimensional space while maintaining the properties of the original data. "None" represents no preprocessing operation. In addition, the seven screening methods contained in the three elements with the lowest number of data items (i.e. k = 3) are also involved in the comparison of preprocessing. Therefore, a total of 11 preprocessing methods are compared.

[0084] In the comparison, the MAE_mean value is used to compare the size of the MAE_mean value, which is the optimal result after reducing to different dimensions. The smaller the MAE_mean value, the better the model performance. From Figure 13It can be seen that the optimal result is the preprocessing method of "removing N and W elements", which shows that elements with significantly lower data item quantity will affect the model performance and accuracy. Therefore, filtering out elements that may affect the performance of the model can eliminate anomalies in the original data and make the data adapt to the model and match the needs of the model.

[0085] Regarding the data set, the data set in the material field may be smaller compared to other fields. The model trained based on a small amount of data may produce inaccurate results. Therefore, for the yield strength attribute prediction task with only 312 data, the embodiment selects linear interpolation for data augmentation.

[0086] Figure 14 is a performance comparison data table of material attribute prediction models under different interpolation numbers. Refer to Figure 14 It can be seen that the model performs best when the interpolation number between every two training data is 4. The reason is that when the interpolation number is small, information cannot be fully extracted and the coverage is not complete; when the interpolation number is large, too much reliance is placed on false information, and the recognition of true information is reduced. Data augmentation is often used in scenarios where data is insufficient or model parameters are large, which can effectively increase the size of the data set to improve model performance.

[0087] Regarding the model, the public available code library of MODNet, Automatminer, RF, CrabNet and Dummy can be used for yield strength attribute testing.

[0088] Figure 15 is a performance comparison data table between different models. Refer to Figure 15 DNNs represent deep neural networks, which are the models used in the embodiment. MODNet is the MO&DSD Operational / Development Network) model. AMMExpress v2020 corresponds to the Automatminer model, which can automatically train machine learning models and obtain results. RF-SCM / Magpie corresponds to the RF (Random Forests) model, which refers to a classifier model that uses multiple trees to train and predict samples. CrabNet (Compositionally Restricted Attention-Based network), Dummy are models that can be applied to the material field, where the Dummy algorithm forms a baseline by predicting the average value of the training set in the regression case.

[0089] Figure 15The performance indicators of the above models in this task are shown in the following table. The evaluation process of these models all use the same data set provided by Matbench, and the value of MAE_mean is based on the five-fold nested cross-validation proposed by Matbench. In this way, when comparing models, arbitrary bias can be minimized to achieve maximum fairness. It can be seen from the following table that Figure 15 It can be seen that the DNNs adopted by the present disclosure obtain the highest score of 91.432 on the task, and significantly improve the MAE_mean, improving the performance of yield strength attribute prediction. Data standardization improves data quality, and data augmentation increases data set size, thereby significantly improving overall performance. In addition, the model adopted by the present disclosure has only 13305 parameters, which is a significant reduction in the number of parameters compared to complex models (for example, CrabNet has 11987206 parameters), and the training time of the model adopted by the present disclosure is only 2.5 hours, greatly saving the time cost.

[0090] The material attribute prediction method according to the embodiment of the present disclosure removes elements affecting the performance of the model by standardizing the training data, thereby improving the quality of the training data, and increases the size of the data set by using a linear interpolation method, which can be well applied to small sample material attribute prediction. At the same time, a neural network composed of a full connection layer is used to accurately predict the material attribute, and the method of combining data augmentation and neural network is superior to the automatic machine learning method.

[0091] Figure 16 is a schematic diagram of a material attribute prediction device according to an embodiment of the present disclosure using a hardware implementation of a processing system. Referring to Figure 16 , the material attribute prediction device 1000 of the present embodiment includes a memory 1300 and a processor 1200. The memory 1300 stores execution instructions, and the processor 1200 executes the execution instructions stored in the memory 1300, so that the processor 1200 executes the material attribute prediction method of any of the above embodiments.

[0092] The device 1000 can include corresponding modules that perform one or more steps in the above flowcharts. Therefore, each step or several steps in the above flowcharts can be performed by corresponding modules, and the device can include one or more of these modules. The modules can be one or more hardware modules specially configured to perform the corresponding steps, or implemented by a processor configured to perform the corresponding steps, or stored in a computer readable medium for implementation by a processor, or implemented by some combination.

[0093] For example, the material property prediction apparatus 1000 can include a training data acquisition module 1002, a data standardization module 1004, a data interpolation module 1006, a prediction model training module 1008, and a material property prediction module 1010. The training data acquisition module 1002 is configured to acquire component data of a plurality of sample materials as training data. The data standardization module 1004 is configured to standardize the training data. The data interpolation module 1006 is configured to linearly interpolate the standardized training data to obtain a training data set. The prediction model training module 1008 is configured to train a material property prediction model according to the training data set. The material property prediction model can be a deep neural network (DNN). The material property prediction module 1010 is configured to input material data to be tested into the trained material property prediction model to obtain a material property prediction result. The material property can be yield strength, elongation, or fracture toughness.

[0094] The data standardization module 1004 can identify all chemical element categories contained in the training data, filter the chemical element categories to obtain a first element category set, and standardize the training data of all element categories in the first element category set. Specifically, the data standardization module 1004 can determine, for each chemical element, a number of data items containing the chemical element, and determine a first element category from the chemical element categories according to the number of data items. The data standardization module 1004 can determine a plurality of second element categories with the lowest number of data items from the chemical element categories, filter the second element categories from all element categories, and take the remaining element categories as the first element category. The data standardization module 1004 can determine, for each element category in the first element category set, a mass proportion value of the element category, determine a maximum mass proportion value of each element from the mass proportion values, and determine a ratio between each mass proportion value and the maximum mass proportion value as the standardized training data.

[0095] The data interpolation module 1006 can insert a plurality of data between each two training data in a linear interpolation manner. The inserted plurality of data can be four data.

[0096] The hardware structure can be implemented using a bus architecture. The bus architecture can include any number of interconnecting buses and bridges, depending on the specific application of the hardware and overall design constraints. The bus 1100 connects various circuits including one or more processors 1200, memory 1300, and / or hardware modules together. The bus 1100 can also connect various other circuits 1400 such as peripheral devices, voltage regulators, power management circuits, external antennas, and the like.

[0097] The bus 1100 can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, or the like. The bus can be divided into an address bus, a data bus, a control bus, and the like. For ease of representation, only one connection line is shown in the figure, but it does not mean that there is only one bus or only one type of bus.

[0098] It should be noted that details not disclosed in the material attribute prediction device 1000 provided in the embodiment can refer to the details disclosed in the material attribute prediction method of the above-mentioned embodiments of the present disclosure, which will not be described here.

[0099] The material attribute prediction device provided by the embodiment of the present disclosure removes elements affecting the performance of the model by standardizing the training data, thereby improving the quality of the training data, and increases the size of the data set by using the linear interpolation method, which can be well applied to small sample material attribute prediction. In addition, the neural network composed of the full connection layer is used to accurately predict the material attribute, and the method combining data augmentation and neural network is better than the automatic machine learning method.

[0100] Any process or method descriptions in flow charts or described elsewhere herein can be understood as representing one or more steps of a method or process, and the scope of the preferred embodiments of the present disclosure includes additional implementation in which one or more steps are performed in a different order, including an order that is not shown or discussed, and / or including additional steps not shown or discussed. The various methods and processes described above can be performed by a processor executing executable instructions stored in memory. For example, the method embodiments in the present disclosure can be implemented as a software program tangibly embodied in a machine-readable medium, such as a memory. In some embodiments, portions of the software program or all of the software program can be loaded and / or installed via the memory and / or communication interface. When the software program is loaded into the memory and executed by the processor, one or more steps of the above-described methods can be performed. Alternatively, in other embodiments, the processor can be configured to perform one of the above-described methods by any other suitable means, such as by means of firmware.

[0101] The logic and / or steps represented in the flowcharts and / or described herein can be embodied in any computer-readable medium for use by an instruction execution system, apparatus, or device, such as a computer-based system, processor- containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. In the context of this document, a "computer-readable medium" can be any means that can store such instructions, including a memory or a computer diskette, or can be any apparatus that can fetch the instructions from the instructions execution system, apparatus, or device and execute the instructions.

[0102] It should be understood that parts of the present disclosure can be realized by hardware, software, or a combination thereof. In the above-described embodiments, a plurality of steps or methods can be realized by software stored in a memory and executed by an appropriate instruction execution system. For example, if realized by hardware, and as in another embodiment, any one or a combination of the following technologies known in the art can be used: discrete logic circuitry having logic gates for implementing logic functions on data signals, application specific integrated circuits having appropriate combinations of logic gates, programmable gate arrays (PGA), field programmable gate arrays (FPGA), and the like.

[0103] Those skilled in the art of the present technology can understand that all or part of the steps of the above-described embodiment methods can be completed by programs instructing relevant hardware, and the programs can be stored in a readable storage medium. When executed, the programs include one or a combination of the steps of the method embodiments.

[0104] In addition, each functional unit in each embodiment of the present disclosure can be integrated into one processing module, or each unit can be physically present separately, or two or more units can be integrated into one module. The above-mentioned integrated module can be realized in the form of hardware or in the form of a software functional module. If the integrated module is realized in the form of a software functional module and sold or used as an independent product, it can also be stored in a readable storage medium. The storage medium can be a read-only memory, a magnetic disk or an optical disk, etc.

[0105] In the description of the present specification, the description referring to the terms "one embodiment / way", "some embodiments / ways", "specific examples", or "some examples" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment / way or example are included in at least one embodiment / way or example of the present disclosure. In the present specification, the illustrative expressions of the above terms are not necessarily the same embodiment / way or example. Also, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in any one or more embodiments / ways or examples. In addition, a person skilled in the art can combine and combine the different embodiments / ways or examples described in the present specification and the features of the different embodiments / ways or examples, without contradiction.

[0106] Furthermore, the terms "first", "second", etc. are used herein for descriptive purposes only and should not be construed as indicating or implying relative importance or an ordered ranking such that the technical features so designated should possess. Thus, features having a "first", "second", etc. designation can implicitly or explicitly include at least one of the features. In the description of the disclosure, the meaning of "a plurality" is at least two, for example, two, three, etc., unless otherwise specifically defined.

[0107] Those skilled in the art will understand that the above-described embodiments are merely for the purpose of clearly illustrating the present disclosure and are not intended to limit the scope of the present disclosure. Other changes or modifications can be made by those skilled in the art based on the above disclosure, and these changes or modifications are still within the scope of the present disclosure.

Claims

1. A method for predicting material properties, characterized in that, include: Obtain component data from multiple sample materials as training data; The training data is standardized; Linear interpolation is performed on the standardized training data to obtain the training dataset. The material property prediction model is trained based on the training dataset; Input the data of the material to be tested into the pre-trained material property prediction model to obtain the material property prediction results; The standardization of the training data includes: identifying all chemical element types contained in the training data; filtering the chemical element types to obtain a first element type set; and standardizing the training data of all element types in the first element type set. The process of filtering the chemical element types to obtain a first element type set includes: for each chemical element, determining the number of data items containing the chemical element; and determining the first element type from the chemical element types based on the number of data items. The process of determining the first element type from the chemical element types based on the number of data items includes: determining a plurality of second element types with the lowest number of data items from the chemical element types; filtering out the second element types from all element types, and taking the remaining element types as the first element types.

2. The method according to claim 1, characterized in that, The training data for all element types in the first element type set are standardized, including: For each element type in the first set of element types, determine the quality percentage of the element type. Determine the maximum mass percentage value for each element from the mass percentage values; The ratio between each mass percentage value and the maximum mass percentage value is determined as the standardized training data.

3. The method according to claim 1, characterized in that, Linear interpolation is performed on the standardized training data, including inserting multiple data points between every two training data points using linear interpolation.

4. The method according to claim 3, characterized in that, The inserted data consists of four data items.

5. The method according to claim 1 or 4, characterized in that, The material properties are yield strength, elongation, or fracture toughness.

6. The method according to claim 1, characterized in that, The material property prediction model employs a deep neural network (DNN).

7. A material property prediction device, characterized in that, include: The memory stores execution instructions; as well as A processor that executes execution instructions stored in the memory, causing the processor to perform the material property prediction method as described in any one of claims 1 to 6.

Images