Method and device for predicting evolution trend of material performance, and storage medium

Abstract

The application discloses a material performance evolution trend prediction method and device and a storage medium, relates to the technical field of material science, and mainly aims at improving the determination efficiency and accuracy of the material performance evolution trend. The method comprises the following steps: in response to a material performance evolution trend prediction signal of a target material in an aging process, obtaining aging environment information of the target material and historical material performance of a previous aging time point corresponding to a current aging time point; taking the current aging time point and a plurality of future aging time points as to-be-predicted time points, inputting the historical material performance and the aging environment information into a preset performance prediction model to gradually perform material performance recursive prediction of each to-be-predicted time point, and sequentially obtaining material performance of each to-be-predicted time point; and determining the material performance evolution trend of the target material in the aging process based on the historical material performance and the material performance of each to-be-predicted time point.

Description

Technical Field

[0001] This invention relates to the field of materials science and technology, and in particular to a method, apparatus and storage medium for predicting the evolution trend of material properties. Background Technology

[0002] Elastic-plastic materials are widely used in various fields such as construction, aerospace, and machinery manufacturing due to their excellent properties. However, these materials undergo different degradation stages during use, including the elastic stage, yielding stage, strengthening stage, and ultimately necking and fracture. Therefore, studying the changing trends of material properties is crucial for material selection and product design.

[0003] Currently, material properties and their changing trends are typically estimated manually. However, this manual estimation method is time-consuming and labor-intensive, and is also affected by subjective human factors, making it impossible to accurately determine the future properties of materials, let alone the evolution trend of their properties. Summary of the Invention

[0004] This invention provides a method, apparatus, and storage medium for predicting the evolution trend of material properties, which mainly improves the efficiency and accuracy of determining the evolution trend of material properties.

[0005] According to a first aspect of the present invention, a method for predicting the evolution trend of material properties is provided, comprising:

[0006] In response to the predicted trend signal of material property evolution of the target material during the aging process, the aging environment information of the target material and the historical material properties of the previous aging time point corresponding to the current aging time point are obtained.

[0007] The current aging time point and multiple future aging time points are used as the time points to be predicted. The historical material properties and the aging environment information are input into the preset performance prediction model to recursively predict the material properties of each time point to be predicted, and the material properties of each time point to be predicted are obtained in sequence.

[0008] Based on the historical material properties and the material properties at each predicted time point, the material property evolution trend of the target material during the aging process is determined.

[0009] Optionally, before inputting the historical material properties and the aging environment information into a preset performance prediction model to recursively predict the material properties at each of the predicted time points, and sequentially obtaining the material properties at each of the predicted time points, the method further includes:

[0010] A preset initial performance prediction model is constructed, and a sample dataset is obtained, wherein the sample dataset includes sample aging environment information of sample materials with material performance labels at different aging time points;

[0011] The sample dataset is divided into training data and test data. The preset initial performance prediction model is trained using the training data, and the trained preset initial performance prediction model is tested using the test data. Based on the test results, the preset performance prediction model is determined.

[0012] Optionally, obtaining the sample dataset includes:

[0013] Aging tests are conducted on different sample materials under different aging environments. During the aging tests, the performance of each sample material at different aging time points is determined. The method for determining the performance of each sample material at different aging time points includes:

[0014] The stress-strain curves of each sample material at different aging time points are determined, and the stress-strain curves of each sample material in the elastic stage are linearly fitted to obtain a fitted line. Based on the fitted line, the first performance parameters of each sample material at different aging time points in the elastic stage are determined.

[0015] The fitted line is translated according to a preset translation condition to obtain an auxiliary line. Based on the stress-strain curve and the auxiliary line, the second performance parameters of each sample material in the elastic stage at different aging time points are determined.

[0016] Based on the stress-strain curve of each sample material in the plastic stage, a strain range is determined, and based on the strain range, two data points are selected on the stress-strain curve in the plastic stage. Based on the stress-strain value of each data point, a third performance parameter of each sample material in the plastic stage at different aging time points is determined. The first performance parameter, the first performance parameter, and the first performance parameter are collectively referred to as the sample material performance of each sample material at different aging time points.

[0017] Optionally, determining the preset performance prediction model based on the test results includes:

[0018] Based on the test data, determine the prediction performance evaluation parameters of the trained preset initial performance prediction model;

[0019] Based on the prediction performance evaluation parameters, it is determined whether the prediction performance of the trained preset initial performance prediction model meets the preset performance conditions. If it does not meet the conditions, the method for adjusting the trained preset initial performance prediction model is determined based on the prediction performance evaluation parameters, and the trained preset initial performance prediction model is adjusted based on the adjustment method to obtain the preset performance prediction model.

[0020] Optionally, determining the method for adjusting the trained preset initial performance prediction model based on the prediction performance evaluation parameters includes:

[0021] Determine whether the predicted performance evaluation parameter is greater than a preset parameter threshold. If it is, determine that the method for adjusting the trained preset initial performance prediction model is a structural adjustment method.

[0022] Based on the aforementioned structural adjustment method, the trained preset initial performance prediction model is adjusted to obtain the preset performance prediction model, including:

[0023] Based on the expected prediction performance of the preset performance prediction model to be constructed and the prediction complexity of the material property evolution trend, the expected number of network layers and the expected number of neurons of the preset initial performance prediction model after training are determined.

[0024] Based on the number of network layers and the expected number of neurons, the number of network layers and the number of neurons in the trained preset initial performance prediction model are adjusted, and the adjusted preset initial performance prediction model is used as the preset performance prediction model.

[0025] Optionally, after determining whether the predictive performance evaluation parameter is greater than a preset parameter threshold, the method further includes:

[0026] If the prediction performance evaluation parameter is less than or equal to the preset parameter threshold, then the method for adjusting the trained preset initial performance prediction model is determined to be the hyperparameter adjustment method.

[0027] Based on the hyperparameter adjustment method, the trained preset initial performance prediction model is adjusted to obtain the preset performance prediction model, including:

[0028] The trained preset initial performance prediction model is determined as the model to be adjusted, and at least one target hyperparameter term that needs to be adjusted and its corresponding parameter adjustment range are determined.

[0029] Determine the loss function corresponding to the model to be adjusted, and fit a surrogate model of the loss function using a Gaussian process;

[0030] Based on the parameter adjustment range, multiple candidate hyperparameter combinations are generated. Based on each candidate hyperparameter combination, the prediction uncertainty coefficient of the surrogate model is determined. Based on the prediction uncertainty coefficient and the preset acquisition function, the acquisition function value corresponding to each candidate hyperparameter combination is determined.

[0031] Based on the obtained function value, a target hyperparameter combination is selected in each of the candidate hyperparameter combinations, and the prediction performance of the model to be adjusted is verified based on the target hyperparameter combination and the verification data. If the prediction performance meets the preset performance conditions, the hyperparameters of the model to be adjusted are adjusted based on the target hyperparameter combination.

[0032] Optionally, before inputting the historical material properties and the aging environment information into a preset performance prediction model to recursively predict the material properties at each of the predicted time points, and sequentially obtaining the material properties at each of the predicted time points, the method further includes:

[0033] The abnormal performance data in the historical material properties and the abnormal environmental data in the aging environment information are determined respectively. The abnormal performance data is removed to obtain the cleaned historical material properties, and the abnormal environmental data is removed to obtain the cleaned aging environment information.

[0034] The historical material properties and the aging environment information after cleaning are normalized to obtain the normalized historical material properties and the normalized aging environment information.

[0035] According to a second aspect of the present invention, a device for predicting the evolution trend of material properties is provided, comprising:

[0036] The acquisition unit is used to acquire the aging environment information of the target material and the historical material properties of the previous aging time point corresponding to the current aging time point in response to the prediction signal of the material property evolution trend of the target material during the aging process.

[0037] The performance prediction unit is used to take the current aging time point and multiple future aging time points as the time points to be predicted, input the historical material performance and the aging environment information into the preset performance prediction model, and recursively predict the material performance of each time point to be predicted in turn, so as to obtain the material performance of each time point to be predicted in sequence.

[0038] The determining unit is used to determine the material property evolution trend of the target material during the aging process based on the historical material properties and the material properties at each of the predicted time points.

[0039] According to a third aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements a method for predicting the evolution trend of the above-mentioned material properties.

[0040] According to a fourth aspect of the present invention, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement a method for predicting the evolution trend of the above-mentioned material properties.

[0041] According to the present invention, a method, apparatus, and storage medium for predicting the evolution trend of material properties provide a method that, compared with the current method of manually estimating material properties and their changing trends, uses a preset performance prediction model to recursively predict the material properties at the current time point and multiple future time points based on the aging environment information of the target material and the historical material properties at the previous aging time point. That is, the material properties at each time point are predicted based on the material properties and aging environment at the previous time point. Finally, based on the material properties at each time point, the performance evolution trend of the target material during the aging process is determined. Thus, based on historical data, the preset performance prediction model can predict the future performance of the material, thereby quickly and effectively capturing the long-term trend of material properties changing over time, avoiding manual intervention, and improving the prediction efficiency and accuracy of material property evolution trends. Attached Figure Description

[0042] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0043] Figure 1 A flowchart of a method for predicting the evolution trend of material properties provided by an embodiment of the present invention is shown;

[0044] Figure 2 This diagram illustrates a centralized time window division of a sample dataset according to an embodiment of the present invention.

[0045] Figure 3 A flowchart of another method for predicting the evolution trend of material properties provided by an embodiment of the present invention is shown;

[0046] Figure 4 This diagram illustrates the acquisition of material properties of sample materials with different aging times, as provided in an embodiment of the present invention.

[0047] Figure 5 A schematic diagram of the structure of a device for predicting the evolution trend of material properties provided in an embodiment of the present invention is shown;

[0048] Figure 6A schematic diagram of another material property evolution trend prediction device provided in an embodiment of the present invention is shown;

[0049] Figure 7 A schematic diagram of the physical structure of a computer device provided in an embodiment of the present invention is shown. Detailed Implementation

[0050] The present invention will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the present application can be combined with each other.

[0051] Currently, estimating material properties and trends manually is time-consuming and labor-intensive. Furthermore, it is affected by subjective human factors and cannot accurately determine material properties, let alone accurately determine the evolution trend of material properties.

[0052] To address the above problems, embodiments of the present invention provide a method for predicting the evolution trend of material properties, such as... Figure 1 As shown, the method includes:

[0053] 101. In response to the predicted trend signal of material property evolution during the aging process of the target material, obtain the aging environment information of the target material and the historical material properties of the previous aging time point corresponding to the current aging time point.

[0054] The target material can be any material, such as elastoplastic material, rubber material, metal material, or composite material; the aging environment information refers to the environmental information such as temperature, humidity, light, and gas during the aging process of the target material; the material properties include the Young's modulus, yield strength, and tangent modulus of the target material.

[0055] In order to predict the performance evolution trend of the target material during the aging process, in accordance with the embodiments of the present invention, it is first necessary to determine the aging environment information of the target material and the historical material performance at the previous aging time point corresponding to the current aging time point.

[0056] 102. Using the current aging time point and multiple future aging time points as the time points to be predicted, input the historical material properties and aging environment information into the preset performance prediction model to recursively predict the material properties at each time point to be predicted, and obtain the material properties at each time point to be predicted in sequence.

[0057] In this embodiment of the invention, to improve the prediction accuracy of material properties, a preset performance prediction model needs to be trained and constructed using a sample dataset. During training, the data in the sample dataset can be divided into multiple time windows according to different time intervals, with each window containing data from several consecutive time points. For example, if a sample material in the dataset has data from 10 degradation time points, two time points can be selected as a window, and the performance parameters (such as Young's modulus, yield strength, and tangent modulus) at the last time point of each time window are designated as the label for that window. Figure 2 As shown, a schematic diagram of the time window division is presented, in which, Figure 2 The aging load in the model represents the aging environment information of the sample. Then, the pre-built performance prediction model is used to predict the material properties at each time point to be predicted. When using the pre-built performance prediction model to predict material properties, the model is based on the material properties and aging environment at the previous aging time point to predict the material properties at the next time point, thus recursively predicting the material properties at each aging time point. For example, following a chronological order, the time points are divided into previous time point a, current time point b, time point c, time point d, time point e, time point f, time point g, and so on. The aging environment information and the historical material performance corresponding to the previous time point a are input into a preset performance prediction model. The prediction process of the preset performance prediction model is as follows: First, based on the aging environment information and the historical material performance corresponding to the previous time point a, the preset performance prediction model predicts the material performance corresponding to the current time point b. Then, based on the aging environment information and the material performance corresponding to the current time point b, it predicts the material performance corresponding to the time point c. Next, based on the aging environment information and the material performance corresponding to the time point c, it predicts the material performance corresponding to the time point d. Finally, based on the aging environment information and the material performance corresponding to the time point d, it predicts the material performance corresponding to the time point e. This process is repeated sequentially, using the predicted value of the previous time point as the input for the next time point, progressively advancing to achieve multi-step recursive prediction, thereby obtaining the material performance at each aging time point. The embodiments of the present invention can achieve performance prediction of the target material throughout the entire aging cycle by using a preset performance prediction model. That is, based on historical data, the preset performance prediction model can predict the future performance of the material, thereby quickly and effectively capturing the long-term trend of material performance changes over time.

[0058] 103. Based on historical material properties and material properties at each point in time to be predicted, determine the evolution trend of material properties of the target material during the aging process.

[0059] Specifically, after predicting the material properties of the target material at various future aging time points using a preset performance prediction model, the evolution trend of the target material's properties throughout the aging process can be determined based on the material properties at historical aging time points, the material properties at the current aging time point, and the material properties at various future aging time points. For example, charts (such as line graphs, scatter plots, etc.) can be used to visually display the performance evolution trend. Thus, this embodiment of the invention, based on historical material properties and using a preset performance prediction model to predict the material's future performance, can quickly and effectively capture the long-term trend of material properties changing over time, avoiding manual intervention, thereby improving the prediction efficiency and accuracy of material performance evolution trends.

[0060] According to the present invention, a method for predicting the evolution trend of material properties is provided. Compared with the current method of determining the performance of a material at a certain stage by manually observing the surface color of the material, the present invention is based on the aging environment information of the target material and the historical material performance at the previous aging time point. It uses a preset performance prediction model to recursively predict the material performance at the current time point and multiple future time points. That is, the material performance at each time point is predicted based on the material performance at the previous time point and the aging environment. Finally, based on the material performance at each time point, the performance evolution trend of the target material in the aging process is determined. Thus, the model can quickly realize the prediction of material performance and evolution trend at multiple aging time points, avoiding manual intervention, thereby improving the prediction efficiency and accuracy of the material performance evolution trend.

[0061] Furthermore, to better illustrate the above data classification process, and as a refinement and extension of the above embodiments, this invention provides another method for predicting the evolution trend of material properties, such as... Figure 3 As shown, the method includes:

[0062] 201. Construct a preset initial performance prediction model and obtain a sample dataset, which includes sample aging environment information of sample materials with material performance labels at different aging time points.

[0063] The material performance label can include properties such as Young's modulus, yield strength, and tangent modulus of the sample material at different aging time points; the sample aging environment information includes environmental information such as temperature, humidity, light, and gas.

[0064] In this embodiment of the invention, to train and construct a preset performance prediction model, it is first necessary to obtain a sample dataset. Based on this, step 201 specifically includes: conducting aging tests on different sample materials under different aging environments; during the aging tests, determining the performance of each sample material at different aging time points; wherein the method for determining the performance of each sample material at different aging time points includes: determining the stress-strain curve of each sample material at different aging time points; performing linear fitting on the stress-strain curve of each sample material in the elastic stage to obtain a fitting line; and based on the fitting line, determining the first performance parameter of each sample material at different aging time points in the elastic stage. The fitted line is translated according to preset translation conditions to obtain an auxiliary line. Based on the stress-strain curve and the auxiliary line, a second performance parameter of each sample material in the elastic stage at different aging time points is determined. Based on the stress-strain curve of each sample material in the plastic stage, a strain range is determined. Based on the strain range, two data points are selected on the stress-strain curve in the plastic stage. Based on the stress-strain value of each data point, a third performance parameter of each sample material in the plastic stage at different aging time points is determined. The first performance parameter, the first performance parameter, and the first performance parameter are collectively referred to as the sample material performance of each sample material at different aging time points.

[0065] The first performance parameter is Young's modulus, the second performance parameter is yield strength, and the third performance parameter is tangent modulus. Specifically, different sample materials can be of different types or from different production batches. The same sample material includes multiple sample materials from the same production batch, and the material properties of sample materials from the same production batch are completely identical. Taking sample materials from the same production batch as an example, aging tests are performed on sample materials from the same batch under the same initial conditions for different aging times, and relevant parameters are recorded, including aging environment, aging time, and the number of aging samples. Figure 4As shown, during the aging test, stress-strain load curves of each sample material are obtained at different aging time points. Standard tensile and indentation testing methods are used to determine the stress-strain load curves of the sample material at different strain rates. Based on the characteristics of the stress-strain curves, they are roughly divided into two stages: the elastic stage and the plastic stage. In the elastic stage, a linear fit is performed on the stress-strain curve to obtain fitted line 1. The slope of fitted line 1 represents Young's modulus. This fitted line is then shifted to a point where the strain rate on the horizontal axis is 0.002, forming a second auxiliary line. The intersection of the stress-strain curve and this auxiliary line is defined as the yield point, and the stress value at this point is the yield strength. For the plastic stage, the stress-strain curve is fitted again to obtain fitted line 2. The slope of fitted line 2 represents the tangent modulus. For samples from the same batch and under the same aging conditions, the average value, variance, and other statistical indicators of their performance parameters are statistically analyzed to create a performance parameter dataset with statistical characteristics. Therefore, following the above method, the Young's modulus, yield strength, tangent modulus, and other performance properties of different sample materials at different aging time points can be determined. Finally, the aging environment information, different aging time points, and the corresponding sample material properties were used as the sample dataset. The sample material properties served as the label information.

[0066] 202. Divide the sample dataset into training data and test data. Use the training data to train the preset initial performance prediction model, and use the test data to test the trained preset initial performance prediction model. Based on the test results, determine the preset performance prediction model.

[0067] In this embodiment of the invention, after determining the sample dataset, the sample dataset is divided into training data and test data. The aging environment and the sample material performance at the previous time point in the training data are used as input data, and the material performance at each time point after the previous time point is used as output data. This is used to train a preset initial performance prediction model. After training, the trained preset initial performance prediction model is tested using test data. During the testing process, if the test results are not ideal, the preset initial performance prediction model needs to be adjusted to meet the actual prediction requirements. Based on this, the method includes: determining the prediction performance evaluation parameters of the trained preset initial performance prediction model based on the test data; determining whether the prediction performance of the trained preset initial performance prediction model meets the preset performance conditions based on the prediction performance evaluation parameters; if not, determining the adjustment method for the trained preset initial performance prediction model based on the prediction performance evaluation parameters, and adjusting the trained preset initial performance prediction model based on the adjustment method to obtain the preset performance prediction model.

[0068] Specifically, when testing the trained preset initial performance prediction model using test data, the aging environment and historical performance parameters at the initial time point in the test data are input into the trained preset initial performance prediction model. The model outputs predicted performance parameters at multiple time points after the initial time point. Based on the differences between the actual performance parameters and predicted performance parameters at multiple time points after the initial time point, such as mean squared error and mean absolute error, the predicted performance evaluation parameters of the trained preset initial performance prediction model are determined. If the predicted performance evaluation parameters are less than or equal to a preset threshold, the trained preset initial performance prediction model is determined to meet the preset performance conditions. At this time, the trained preset initial performance prediction model is used as the preset performance evaluation model. If the prediction performance evaluation parameter is greater than a preset threshold (the preset threshold is set according to actual needs), then it is determined that the trained preset initial performance prediction model does not meet the preset performance conditions (wherein, the preset performance conditions are set according to actual needs). In this case, the trained preset initial performance prediction model needs to be adjusted. Before adjustment, the adjustment method of the model needs to be determined according to the prediction performance evaluation parameter. Based on this, the method includes: determining whether the prediction performance evaluation parameter is greater than the preset parameter threshold; if it is greater, then determining that the adjustment method of the trained preset initial performance prediction model is a structural adjustment method; adjusting the trained preset initial performance prediction model according to the structural adjustment method to obtain the preset performance prediction model, including: determining the expected number of network layers and the expected number of neurons of the trained preset initial performance prediction model based on the expected prediction performance of the preset performance prediction model to be constructed and the prediction complexity of the material performance evolution trend; adjusting the number of network layers and the expected number of neurons of the trained preset initial performance prediction model based on the number of network layers and the expected number of neurons, and using the adjusted preset initial performance prediction model as the preset performance prediction model.

[0069] Among them, the preset parameter threshold is set according to actual needs; expected prediction performance refers to the expected accuracy and expected prediction rate of the model; prediction complexity refers to the data scale on which the prediction is based, the difficulty of the prediction task, and the amount of resources required for the prediction.

[0070] Specifically, when adjusting the model structure, if higher prediction accuracy is required, deeper networks and more neurons may be needed to capture more complex features. If the task requires a fast response, the number of network layers and neurons should be moderate to avoid excessive computation time. Meanwhile, prediction complexity involves multiple aspects, including data complexity, task difficulty, and model generalization ability. The number and type of data features, as well as the correlation between features, all affect prediction complexity. High-dimensional data, nonlinear relationships, or noisy data typically increase prediction complexity. Simple linear or near-linear problems can use shallower networks and fewer neurons; while complex nonlinear problems may require deeper networks and more neurons. However, excessively deep networks and too many neurons can lead to overfitting and reduced generalization ability. Therefore, it is necessary to comprehensively consider prediction complexity, expected prediction performance, and generalization ability to determine the expected number of network layers and neurons. For example, based on the complexity of the problem and the scale of the data, a reasonable number of network layers and neurons can be initially set. For simple problems, shallower networks (e.g., 2-3 layers) and fewer neurons can be selected; for complex problems, deeper networks (e.g., 5-10 layers or more) and more neurons can be attempted. Finally, based on the expected number of network layers and neurons, the number of network layers and neurons in the trained preset initial performance prediction model are adjusted to obtain the adjusted preset initial performance prediction model. To further improve model performance, the adjusted preset initial performance prediction model can be tested using test data, and the adjusted preset initial performance prediction model that meets the test conditions is ultimately adopted as the preset performance prediction model. This embodiment of the invention, by adjusting the model structure, enables the model to maintain high performance while minimizing computational resource and time consumption.

[0071] Furthermore, if the prediction performance evaluation parameter is less than or equal to a preset parameter threshold, hyperparameter adjustment is required for the trained preset initial performance prediction model. Based on this, the method includes: determining the trained preset initial performance prediction model as the model to be adjusted; determining at least one target hyperparameter term to be adjusted and its corresponding parameter adjustment range; determining the loss function corresponding to the model to be adjusted, and fitting a surrogate model of the loss function using a Gaussian process; generating multiple candidate hyperparameter combinations based on the parameter adjustment range; determining the prediction uncertainty coefficient of the surrogate model based on each candidate hyperparameter combination; determining the acquisition function value corresponding to each candidate hyperparameter combination based on the prediction uncertainty coefficient and a preset acquisition function; selecting a target hyperparameter combination from each candidate hyperparameter combination based on the acquisition function value; verifying the prediction performance of the model to be adjusted based on the target hyperparameter combination and validation data; if the prediction performance meets a preset performance condition, then hyperparameter adjustment is performed on the model to be adjusted based on the target hyperparameter combination.

[0072] Hyperparameters can include learning rate, batch size, regularization coefficient, etc. Specifically, the loss function corresponding to the model to be tuned is first determined. This loss function measures the difference between the model's predicted values ​​and the actual values, and is a core indicator in the optimization process. For example, in regression problems, we might choose mean squared error (MSE) as the loss function; in classification problems, we might choose cross-entropy loss. Next, we use a Gaussian process (GP) to fit a surrogate model to the loss function. By constructing this surrogate model, we can quickly evaluate the performance of different hyperparameter combinations without actually training the model. Then, based on the parameter adjustment range, multiple candidate hyperparameter combinations are generated. For each candidate combination, we predict its loss function value using the surrogate model and calculate the prediction uncertainty coefficient. This uncertainty coefficient reflects the reliability of the prediction result: the higher the uncertainty coefficient, the less reliable the prediction result. Further, based on the prediction uncertainty coefficient and a preset acquisition function, i.e., by directly feeding the uncertainty coefficient into the preset acquisition function, the acquisition function value is obtained. The acquisition function is used to balance exploration and utilization: exploring regions with high prediction uncertainty while also utilizing regions that are known to perform well. Common acquisition functions include Expected Improvement (EI) and Upper Confidence Bound (UCB). Based on the acquisition function value, one or more candidate hyperparameter combinations are selected as the target hyperparameter combination. Then, validation data is used to train the model and evaluate its predictive performance. If the predictive performance meets preset performance conditions (such as accuracy, AUC (Area Under the Curve), etc., reaching a certain threshold), an effective hyperparameter combination is considered found. Finally, the hyperparameters of the model to be adjusted are based on the found target hyperparameter combination to obtain the predicted model with the preset performance. This adjusted model will have more accurate and faster predictive performance and can play a greater role in practical applications. It should be noted that in each iteration, the surrogate model is updated based on new observation data, and the acquisition function value is recalculated to select the next target hyperparameter combination. This process continues until a stopping condition is met (such as reaching a preset number of iterations, or the predictive performance no longer significantly improving).

[0073] 203. In response to the predicted signal of the material property evolution trend of the target material during the aging process, obtain the aging environment information of the target material and the historical material properties of the previous aging time point corresponding to the current aging time point.

[0074] 204. Using the current aging time point and multiple future aging time points as the time points to be predicted, input the historical material properties and aging environment information into the preset performance prediction model to recursively predict the material properties at each time point to be predicted, and obtain the material properties at each time point to be predicted in sequence.

[0075] In this embodiment of the invention, after obtaining aging environment information and historical material properties, in order to improve the prediction accuracy of material property evolution trends, it is also necessary to perform data preprocessing on the aging environment information and historical material properties. Based on this, the method includes: determining the performance abnormal data in the historical material properties and the environmental abnormal data in the aging environment information, respectively, removing the performance abnormal data to obtain the cleaned historical material properties, and removing the environmental abnormal data to obtain the cleaned aging environment information; and performing normalization processing on the cleaned historical material properties and the cleaned aging environment information to obtain the normalized historical material properties and the normalized aging environment information.

[0076] Specifically, the previous aging time point can be multiple historical aging time points. Taking aging environment information as an example, abnormal data in the aging environment information is determined as follows: The data density within a preset neighborhood corresponding to the aging environment information is determined, and based on the data density, core data points and non-core data points are determined in the aging environment information; any core data point among the core data points is taken as a target core data point, and the remaining core data points are clustered using the target core data point as the cluster center to obtain core data points under different cluster categories; any non-core data point among the non-core data points is taken as a target non-core data point, and the reference core data point closest to the non-core data point is determined among the core data points under each cluster category, and it is determined whether the distance between the non-core data point and the reference core data point is less than a preset distance threshold; if the distance between the non-core data point and the reference core data point is less than the preset distance threshold, the non-core data point is assigned to the cluster category to which the reference core data point belongs; otherwise, the non-core data point is determined as abnormal data.

[0077] Specifically, the preset neighborhood and the preset distance threshold are set according to actual needs. Data density can specifically refer to the amount of data. Specifically, if the data density within the preset neighborhood corresponding to a certain aging environment is greater than the preset density threshold, then that aging environment is identified as a core data point. Conversely, if the data density within the preset neighborhood corresponding to a certain aging environment is less than or equal to the preset density threshold, then that aging environment is identified as a non-core data point. Then, each core data point is used as a cluster center to cluster the remaining core data points, resulting in different cluster categories. Next, it is determined whether each non-core data point can be clustered into any of the above cluster categories. Finally, non-core points that cannot be clustered into any of the above cluster categories are identified as outliers. Then, the outliers in the aging environment are deleted, resulting in cleaned aging environment information. Thus, following the above method, the cleaned historical material properties can be determined. This embodiment of the invention improves data quality by removing outliers, thereby increasing the accuracy of predicting material performance change trends. Meanwhile, in high-dimensional data spaces, the distribution of data points can become very complex. Therefore, the neighborhood density-based method of this invention identifies anomalies by examining the density distribution of data points in their neighborhoods, which can improve the accuracy of identifying abnormal data.

[0078] Furthermore, the historical material properties and aging environment information after cleaning are normalized to ensure that all information is on a similar order of magnitude, thereby enabling the model to quickly find the optimal solution during prediction, thus improving the model's prediction efficiency.

[0079] Furthermore, the historical material properties and aging environment information after cleaning and normalization are input into the preset performance prediction model to recursively predict the material properties at each future time point.

[0080] 205. Based on historical material properties and material properties at each point in time to be predicted, determine the material property evolution trend of the target material during the aging process.

[0081] Specifically, the historical aging time point and each predicted time point are used as the horizontal axis, and the material properties corresponding to each time point are used as the vertical axis, so as to determine the material property evolution trend of the target material throughout the aging process.

[0082] According to another method for predicting the evolution trend of material properties provided by the present invention, compared with the current method of determining the performance of a material at a certain stage by manually observing the surface color of the material, the present invention is based on the aging environment information of the target material and the historical material performance at the previous aging time point. It uses a preset performance prediction model to recursively predict the material performance at the current time point and multiple future time points. That is, the material performance at each time point is predicted based on the material performance at the previous time point and the aging environment. Finally, based on the material performance at each time point, the performance evolution trend of the target material in the aging process is determined. Thus, the model can quickly realize the prediction of material performance and evolution trend at multiple aging time points, avoiding manual intervention, thereby improving the prediction efficiency and accuracy of the material performance evolution trend.

[0083] Furthermore, as Figure 1 In specific implementation, embodiments of the present invention provide a device for predicting the evolution trend of material properties, such as... Figure 5 As shown, the device includes: an acquisition unit 31, a performance prediction unit 32, and a determination unit 33.

[0084] The acquisition unit 31 can be used to acquire the aging environment information of the target material and the historical material properties of the previous aging time point corresponding to the current aging time point in response to the prediction signal of the material property evolution trend of the target material during the aging process.

[0085] The performance prediction unit 32 can be used to take the current aging time point and multiple future aging time points as the time points to be predicted, input the historical material performance and the aging environment information into the preset performance prediction model, and recursively predict the material performance of each time point to be predicted in turn, so as to obtain the material performance of each time point to be predicted in sequence.

[0086] The determining unit 33 can be used to determine the material performance evolution trend of the target material during the aging process based on the historical material performance and the material performance at each predicted time point.

[0087] In specific application scenarios, in order to train and build a preset performance prediction model, such as Figure 6 As shown, the device also includes a construction unit 34.

[0088] The construction unit 34 can be used to construct a preset initial performance prediction model and obtain a sample dataset, wherein the sample dataset includes sample aging environment information of sample materials with material performance labels at different aging time points; the sample dataset is divided into training data and test data, the preset initial performance prediction model is trained using the training data, and the trained preset initial performance prediction model is tested using the test data, and the preset performance prediction model is determined based on the test results.

[0089] In specific application scenarios, to obtain sample datasets, the construction unit 34 can be used to conduct aging tests on different sample materials under different aging environments. During the aging test, the performance of each sample material at different aging time points is determined. The method for determining the performance of each sample material at different aging time points includes: determining the stress-strain curve of each sample material at different aging time points; performing linear fitting on the stress-strain curve of each sample material in the elastic stage to obtain a fitted line; and based on the fitted line, determining the first performance parameter of each sample material at different aging time points in the elastic stage; and adjusting the fitted line... The sample material is translated according to preset translation conditions to obtain an auxiliary line. Based on the stress-strain curve and the auxiliary line, the second performance parameter of each sample material in the elastic stage at different aging time points is determined. Based on the stress-strain curve of each sample material in the plastic stage, the strain range is determined. Based on the strain range, two data points are selected on the stress-strain curve of the plastic stage. Based on the stress-strain value of each data point, the third performance parameter of each sample material in the plastic stage at different aging time points is determined. The first performance parameter, the first performance parameter, and the first performance parameter are collectively referred to as the sample material performance of each sample material at different aging time points.

[0090] In specific application scenarios, in order to determine the preset performance prediction model, the construction unit 34 includes a determination module 341 and an adjustment module 342.

[0091] The determining module 341 can be used to determine the predictive performance evaluation parameters of the trained preset initial performance prediction model based on the test data.

[0092] The adjustment module 342 can be used to determine whether the prediction performance of the trained preset initial performance prediction model meets the preset performance conditions based on the prediction performance evaluation parameters. If it does not meet the conditions, the module determines the adjustment method for the trained preset initial performance prediction model based on the prediction performance evaluation parameters, and adjusts the trained preset initial performance prediction model based on the adjustment method to obtain the preset performance prediction model.

[0093] In specific application scenarios, in order to adjust the model, the adjustment module 342 can be used to determine whether the prediction performance evaluation parameter is greater than a preset parameter threshold. If it is greater, the adjustment method for the trained preset initial performance prediction model is determined to be a structural adjustment method. Based on the structural adjustment method, the trained preset initial performance prediction model is adjusted to obtain the preset performance prediction model, including: determining the expected number of network layers and the expected number of neurons of the trained preset initial performance prediction model based on the expected prediction performance of the preset performance prediction model to be constructed and the prediction complexity of the material performance evolution trend; adjusting the number of network layers and the expected number of neurons of the trained preset initial performance prediction model based on the number of network layers and the expected number of neurons, and using the adjusted preset initial performance prediction model as the preset performance prediction model.

[0094] In specific application scenarios, in order to adjust the model hyperparameters, the adjustment module 342 can be specifically used to determine that the method of adjusting the trained preset initial performance prediction model is a hyperparameter adjustment method if the prediction performance evaluation parameter is less than or equal to the preset parameter threshold; and to adjust the trained preset initial performance prediction model based on the hyperparameter adjustment method to obtain the preset performance prediction model, including: determining the trained preset initial performance prediction model as the model to be adjusted; determining at least one target hyperparameter term that needs to be adjusted in the model to be adjusted and its corresponding parameter adjustment range; and determining the loss function corresponding to the model to be adjusted. The model is fitted with a Gaussian process to fit the loss function; multiple candidate hyperparameter combinations are generated based on the parameter adjustment range; the prediction uncertainty coefficient of the surrogate model is determined based on each candidate hyperparameter combination; and the acquisition function value corresponding to each candidate hyperparameter combination is determined based on the prediction uncertainty coefficient and a preset acquisition function; a target hyperparameter combination is selected from each candidate hyperparameter combination based on the acquisition function value; and the prediction performance of the model to be adjusted is verified based on the target hyperparameter combination and validation data. If the prediction performance meets the preset performance conditions, the hyperparameters of the model to be adjusted are adjusted based on the target hyperparameter combination.

[0095] In specific application scenarios, in order to preprocess historical material properties and aging environment information, the device further includes a preprocessing unit 35.

[0096] The preprocessing unit 35 can be used to determine the abnormal performance data in the historical material properties and the abnormal environmental data in the aging environment information, respectively, and remove the abnormal performance data to obtain the cleaned historical material properties, and remove the abnormal environmental data to obtain the cleaned aging environment information; and perform normalization processing on the cleaned historical material properties and the cleaned aging environment information respectively to obtain the normalized historical material properties and the normalized aging environment information.

[0097] It should be noted that other corresponding descriptions of the functional modules involved in the material property evolution trend prediction device provided in this embodiment of the invention can be found in [reference]. Figure 1 The corresponding description of the method shown will not be repeated here.

[0098] Based on the above, Figure 1 Accordingly, this embodiment of the invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, performs the following steps: responding to a prediction signal of the material performance evolution trend of a target material during aging, acquiring aging environment information of the target material and historical material performance at the previous aging time point corresponding to the current aging time point; using the current aging time point and multiple future aging time points as prediction time points, inputting the historical material performance and the aging environment information into a preset performance prediction model to recursively predict the material performance at each prediction time point, thereby obtaining the material performance at each prediction time point in sequence; and determining the material performance evolution trend of the target material during aging based on the historical material performance and the material performance at each prediction time point.

[0099] Based on the above, Figure 1 The method shown and as Figure 5 The embodiment of the device shown in the invention also provides a physical structure diagram of a computer device, such as... Figure 7As shown, the computer device includes: a processor 41, a memory 42, and a computer program stored in the memory 42 and executable on the processor. Both the memory 42 and the processor 41 are mounted on a bus 43. When the processor 41 executes the program, it performs the following steps: In response to a prediction signal indicating the evolution trend of the target material's properties during the aging process, it acquires the aging environment information of the target material and the historical material properties of the previous aging time point corresponding to the current aging time point; it uses the current aging time point and multiple future aging time points as prediction time points, inputs the historical material properties and the aging environment information into a preset performance prediction model, and recursively predicts the material properties of each prediction time point sequentially to obtain the material properties of each prediction time point; based on the historical material properties and the material properties of each prediction time point, it determines the evolution trend of the target material's properties during the aging process.

[0100] Through the technical solution of this invention, the present invention recursively predicts the material performance at the current time point and multiple future time points by using a preset performance prediction model based on the aging environment information of the target material and the historical material performance at the previous aging time point. That is, the material performance at each time point is predicted based on the material performance at the previous time point and the aging environment. Finally, based on the material performance at each time point, the performance evolution trend of the target material during the aging process is determined. Thus, the model can quickly realize the prediction of material performance and evolution trend at multiple aging time points, avoiding manual intervention, thereby improving the prediction efficiency and accuracy of material performance evolution trend.

[0101] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0102] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting the evolution trend of material properties, characterized in that, include: In response to the predicted trend signal of material property evolution of the target material during the aging process, the aging environment information of the target material and the historical material properties of the previous aging time point corresponding to the current aging time point are obtained. The current aging time point and multiple future aging time points are used as the time points to be predicted. The historical material properties and the aging environment information are input into the preset performance prediction model to recursively predict the material properties of each time point to be predicted, and the material properties of each time point to be predicted are obtained in sequence. Based on the historical material properties and the material properties at each predicted time point, the material property evolution trend of the target material during the aging process is determined.

2. The method according to claim 1, characterized in that, Before inputting the historical material properties and the aging environment information into a preset performance prediction model to recursively predict the material properties at each of the predicted time points, and sequentially obtaining the material properties at each of the predicted time points, the method further includes: Construct a preset initial performance prediction model and obtain a sample dataset, wherein the sample dataset includes sample aging environment information of sample materials with material performance labels at different aging time points; The sample dataset is divided into training data and test data. The preset initial performance prediction model is trained using the training data, and the trained preset initial performance prediction model is tested using the test data. Based on the test results, the preset performance prediction model is determined.

3. The method according to claim 2, characterized in that, The acquisition of the sample dataset includes: Aging tests are conducted on different sample materials under different aging environments. During the aging tests, the performance of each sample material at different aging time points is determined. The method for determining the performance of each sample material at different aging time points includes: The stress-strain curves of each sample material at different aging time points are determined, and the stress-strain curves of each sample material in the elastic stage are linearly fitted to obtain a fitted line. Based on the fitted line, the first performance parameters of each sample material at different aging time points in the elastic stage are determined. The fitted line is translated according to a preset translation condition to obtain an auxiliary line. Based on the stress-strain curve and the auxiliary line, the second performance parameters of each sample material in the elastic stage at different aging time points are determined. Based on the stress-strain curve of each sample material in the plastic stage, a strain range is determined, and based on the strain range, two data points are selected on the stress-strain curve in the plastic stage. Based on the stress-strain value of each data point, a third performance parameter of each sample material in the plastic stage at different aging time points is determined. The first performance parameter, the first performance parameter, and the first performance parameter are collectively referred to as the sample material performance of each sample material at different aging time points.

4. The method according to claim 2, characterized in that, The step of determining the preset performance prediction model based on the test results includes: Based on the test data, determine the prediction performance evaluation parameters of the trained preset initial performance prediction model; Based on the prediction performance evaluation parameters, it is determined whether the prediction performance of the trained preset initial performance prediction model meets the preset performance conditions. If it does not meet the conditions, the method for adjusting the trained preset initial performance prediction model is determined based on the prediction performance evaluation parameters, and the trained preset initial performance prediction model is adjusted based on the adjustment method to obtain the preset performance prediction model.

5. The method according to claim 4, characterized in that, The method for determining the adjustment of the trained preset initial performance prediction model based on the prediction performance evaluation parameters includes: Determine whether the predicted performance evaluation parameter is greater than a preset parameter threshold. If it is, determine that the method for adjusting the trained preset initial performance prediction model is a structural adjustment method. Based on the aforementioned structural adjustment method, the trained preset initial performance prediction model is adjusted to obtain the preset performance prediction model, including: Based on the expected prediction performance of the preset performance prediction model to be constructed and the prediction complexity of the material property evolution trend, the expected number of network layers and the expected number of neurons of the preset initial performance prediction model after training are determined. Based on the number of network layers and the expected number of neurons, the number of network layers and the number of neurons in the trained preset initial performance prediction model are adjusted, and the adjusted preset initial performance prediction model is used as the preset performance prediction model.

6. The method according to claim 5, characterized in that, After determining whether the predicted performance evaluation parameter is greater than a preset parameter threshold, the method further includes: If the prediction performance evaluation parameter is less than or equal to the preset parameter threshold, then the method for adjusting the trained preset initial performance prediction model is determined to be the hyperparameter adjustment method. Based on the hyperparameter adjustment method, the trained preset initial performance prediction model is adjusted to obtain the preset performance prediction model, including: The trained preset initial performance prediction model is determined as the model to be adjusted, and at least one target hyperparameter term that needs to be adjusted and its corresponding parameter adjustment range are determined. Determine the loss function corresponding to the model to be adjusted, and fit a surrogate model of the loss function using a Gaussian process; Based on the parameter adjustment range, multiple candidate hyperparameter combinations are generated. Based on each candidate hyperparameter combination, the prediction uncertainty coefficient of the surrogate model is determined. Based on the prediction uncertainty coefficient and the preset acquisition function, the acquisition function value corresponding to each candidate hyperparameter combination is determined. Based on the obtained function value, a target hyperparameter combination is selected in each of the candidate hyperparameter combinations, and the prediction performance of the model to be adjusted is verified based on the target hyperparameter combination and the verification data. If the prediction performance meets the preset performance conditions, the hyperparameters of the model to be adjusted are adjusted based on the target hyperparameter combination.

7. The method according to claim 1, characterized in that, Before inputting the historical material properties and the aging environment information into a preset performance prediction model to recursively predict the material properties at each of the predicted time points, and sequentially obtaining the material properties at each of the predicted time points, the method further includes: The abnormal performance data in the historical material properties and the abnormal environmental data in the aging environment information are determined respectively. The abnormal performance data is removed to obtain the cleaned historical material properties, and the abnormal environmental data is removed to obtain the cleaned aging environment information. The historical material properties and the aging environment information after cleaning are normalized to obtain the normalized historical material properties and the normalized aging environment information.

8. A device for predicting the evolution trend of material properties, characterized in that, include: The acquisition unit is used to acquire the aging environment information of the target material and the historical material properties of the previous aging time point corresponding to the current aging time point in response to the prediction signal of the material property evolution trend of the target material during the aging process. The performance prediction unit is used to take the current aging time point and multiple future aging time points as the time points to be predicted, input the historical material performance and the aging environment information into the preset performance prediction model, and recursively predict the material performance of each time point to be predicted in turn, so as to obtain the material performance of each time point to be predicted in sequence. The determining unit is used to determine the material property evolution trend of the target material during the aging process based on the historical material properties and the material properties at each of the predicted time points.

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

10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images