Material performance prediction method and device based on structural model and computer equipment
By acquiring and integrating two-dimensional structural model images of materials and using deep learning neural networks for performance prediction, the problems of low accuracy and low efficiency in new material performance prediction methods have been solved, achieving efficient and accurate new material performance prediction and improving R&D efficiency.
Patent Information
- Application Number
- CN202511101405.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-07
- Publication Date
- 2025-11-14
AI Technical Summary
Existing methods for predicting the properties of new materials suffer from low accuracy and low efficiency, making it difficult to balance the accuracy and efficiency of the prediction results.
By acquiring a three-dimensional structural model of the material to be predicted, a two-dimensional structural model image is obtained by projecting it along the crystal axis. This image is then integrated into a two-dimensional structural model image using image fusion technology. The area of the characteristic groups is adjusted as input feature parameters, and finally, the image is input into a pre-trained deep learning neural network for performance prediction.
It enables efficient and accurate prediction of the properties of new materials, improves the efficiency of new material research and development, reduces experimental trial and error costs, and can accurately predict the properties of different crystal planes of the same material.
Smart Images
Figure CN120954590A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of artificial intelligence technology and new materials technology, specifically to a method, apparatus and computer equipment for predicting material properties based on structural models. Background Technology
[0002] The research and development of new materials mainly relies on experimental trial and error to continuously optimize performance, which leads to high research and development costs and long development time. The general approach to solving this problem is to use artificial intelligence technology to assist in the research and design of new materials. However, existing artificial intelligence-based material performance prediction methods are difficult to balance the accuracy and efficiency of prediction results.
[0003] Crystal axes, as imaginary straight lines passing through the center of a crystal, are used to establish the coordinate system of the crystal and characterize its spatial orientation. The concept of crystal axes can also be applied to the crystalline regions of semi-crystalline polymer materials. According to crystallographic theory, crystal axes include three axes: a, b, and c, encompassing
[100] ,
[010] ,
[001] , and [...]. 00]、[0 0]、[00 The projection of the six axes along the crystal axis to obtain molecular images of a specific crystal plane can intuitively reflect the specific internal pore structure and defect distribution of the molecule on that crystal plane, and can effectively predict molecular properties.
[0004] Therefore, in order to better research and develop new materials, it is necessary to quickly and accurately predict the properties of new materials. This invention proposes a material property prediction method that can efficiently and accurately predict the properties of new materials simply by reading and identifying two-dimensional images of the crystalline molecular structure models of the constituent materials, thus solving the technical problems of low accuracy and low efficiency in new material property prediction methods. Summary of the Invention
[0005] The purpose of this invention is to provide a material property prediction method, apparatus, and computer equipment based on structural models, so as to at least partially solve the technical problems of low accuracy and low efficiency in new material property prediction methods.
[0006] To achieve the above objectives, the present invention provides the following technical solution:
[0007] This invention provides a method for predicting material properties based on a structural model, the method comprising:
[0008] Obtain a three-dimensional structural model of the material to be predicted, wherein the three-dimensional structural model is a crystalline molecular structure model with micro- and nano-sized pores;
[0009] The three-dimensional structural model is projected along the crystal axis to obtain a two-dimensional structural model image, wherein the crystal axis includes
[100] ,
[010] ,
[001] , and [...]. 00]、[0 0]、[00 Six axes;
[0010] At least two two-dimensional structure model images projected along different crystal axes are fused into a unified two-dimensional structure model image using image fusion technology.
[0011] The total area of atoms contained in the characteristic groups in the integrated image of the two-dimensional structure model is adjusted to the optimal value of the area input feature parameter to obtain the parameter-tuned integrated image of the two-dimensional structure model. The optimal value of the area input feature parameter is obtained by training in a pre-built performance prediction model.
[0012] The integrated image of the parameter-tuned two-dimensional structural model is input into a pre-trained performance prediction model to obtain the performance prediction result of the material to be predicted.
[0013] The performance prediction model is obtained by training a pre-built deep learning neural network with an integrated image of the two-dimensional structure model after material parameter tuning and performance labels. The integrated image of the two-dimensional structure model is a crystalline molecular structure model image with micro- and nano-sized pores. The performance labels include a first label for performance meeting the standard and a second label for performance not meeting the standard.
[0014] In some embodiments, the performance prediction model is obtained by training a pre-built deep learning neural network, specifically including:
[0015] Obtain a vast number of 3D structural models and corresponding material property labels from materials databases;
[0016] Projecting all three-dimensional structural models along the crystal axis yields a two-dimensional structural model image;
[0017] At least two two-dimensional structural model images of the same material, projected onto different crystal axis directions, are fused into a unified two-dimensional structural model image using image fusion technology.
[0018] The characteristic groups in the integrated image of the two-dimensional structure model are annotated with images, and the area occupied by all the annotated atoms in the integrated image of the two-dimensional structure model is recorded as the area input feature parameter.
[0019] The massive amount of two-dimensional structural models are used to integrate images and corresponding material property labels to construct a sample set, which is then divided into a training set, a validation set, and a test set.
[0020] Based on the two-dimensional structure model, the images are integrated and the deep learning neural network is used to classify and model the training set and the validation set. The resulting classification model is then trained to obtain a performance prediction model.
[0021] The performance prediction model is evaluated using the test set based on the evaluation metrics.
[0022] The process of training the classification model to obtain a performance prediction model includes:
[0023] The classification model is trained on the training set, and its hyperparameters are tuned on the validation set. The area input feature parameters of the integrated images of all two-dimensional structure models in the sample set are adjusted simultaneously based on the cross-entropy loss of the validation set until the change in the cross-entropy loss of the validation set is less than 2% for at least three consecutive training epochs. The optimal value of the area input feature parameters is then obtained to obtain the performance prediction model.
[0024] In some embodiments, the three-dimensional structural model of the material to be predicted is a crystallographic information file (CIF).
[0025] In some embodiments, the characteristic groups are chemical groups that have hydrophilic or hydrophobic properties, including: carboxyl groups, hydroxyl groups, aldehyde groups, amide groups, amino groups, nitro groups, halogen atoms, and hydrocarbon groups.
[0026] In some embodiments, image annotation processing is performed on all the atoms contained in the characteristic groups of the integrated image of the two-dimensional structure model, specifically including:
[0027] Identify and label the atoms contained in all characteristic groups in the integrated image of the two-dimensional structural model, either manually or automatically. The labels are in the form of dots, lines, boxes, polygons, or text labels.
[0028] In some embodiments, the evaluation metrics include accuracy, the harmonic mean (F1) of precision and recall, the area under the receiver operating characteristic curve (AUC), and the confusion matrix.
[0029] The present invention also provides a material property prediction device based on a structural model, the device comprising:
[0030] The three-dimensional structure model acquisition unit is used to acquire the three-dimensional structure model of the material to be predicted from the material database. The three-dimensional structure model is a crystalline molecular structure model with micro- and nano-sized pores.
[0031] A two-dimensional image projection generation unit is used to project the three-dimensional structural model along the crystal axis to obtain a two-dimensional structural model image;
[0032] The image fusion unit is used to fuse two-dimensional structural model images obtained by projecting them onto different crystal axis directions into a two-dimensional structural model integrated image using image fusion technology.
[0033] The area input feature parameter adjustment unit is used to adjust the area input feature parameters to obtain a two-dimensional structural model integrated image with the area input feature parameters at their optimal values.
[0034] The material property prediction structure output unit is used to output the performance prediction results of the material to be predicted.
[0035] The present invention also provides a computer device, 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 the steps of the method described above.
[0036] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described above.
[0037] This invention provides a method, apparatus, and computer device for predicting material properties based on structural models. Its advantage lies in its ability to efficiently and accurately predict the properties of new materials simply by reading and recognizing two-dimensional images of the crystalline molecular structure models that make up the material. Since two-dimensional molecular structure model images of specific crystal planes obtained by projecting along the crystal axis can intuitively reflect the specific internal pore structure and defect distribution of molecules on that crystal plane, the material's properties can be accurately and effectively predicted. When developing new materials, the method, apparatus, and computer device provided by this invention can predict the material's properties based on its crystallographic information files, thereby improving the efficiency of new material development and reducing the trial-and-error costs caused by experiments. Furthermore, the method provided by this invention can also further predict the properties of different crystal planes of the same material based on two-dimensional images obtained by projecting along different crystal axes. This invention partially solves the technical problems of low accuracy and low efficiency in new material property prediction methods. Attached Figure Description
[0038] To clearly illustrate the embodiments of the present invention, a brief description of the embodiments will be provided below. Obviously, the following figures are merely examples; those skilled in the art can derive other embodiments based on the figures provided by the present invention without any creative effort.
[0039] The structures, proportions, sizes, etc. illustrated in this specification are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed herein, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.
[0040] Figure 1 One of the flowcharts for a material property prediction method based on a structural model provided by the present invention;
[0041] Figure 2 The second flowchart of a material property prediction method based on a structural model provided by the present invention;
[0042] Figure 3 Learning curves for the cross-entropy loss function on the training and validation sets;
[0043] Figure 4 A flowchart illustrating how to use the material property prediction model;
[0044] Figure 5 This is a schematic diagram of the confusion matrix for the test set.
[0045] Figure 6 A structural block diagram of a material property prediction device based on a structural model provided by the present invention;
[0046] Figure 7 A structural block diagram of a computer device provided by the present invention. Detailed Implementation
[0047] The technical solutions in the embodiments of the present invention will be fully described below with reference to the accompanying drawings and specific embodiments. Obviously, the following embodiments are only some embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of the present invention.
[0048] Please see Figure 1 , Figure 1 This is one of the flowcharts for a material property prediction method based on a structural model provided by the present invention.
[0049] In one specific embodiment, the present invention provides a material property prediction method based on a structural model, comprising the following steps:
[0050] S110: Obtain a three-dimensional structural model of the material to be predicted, wherein the three-dimensional structural model is a crystalline molecular structure model with micro- and nano-sized pores;
[0051] S120: Project the three-dimensional structural model along the crystal axis direction to obtain a two-dimensional structural model image, wherein the crystal axis direction includes
[100] ,
[010] ,
[001] , [...]. 00]、[0 0]、[00 Six axes;
[0052] S130: At least two two-dimensional structure model images obtained by projecting them in different crystal axis directions are fused into a two-dimensional structure model integrated image using image fusion technology;
[0053] S140: Adjust the total area of atoms contained in the characteristic groups in the integrated image of the two-dimensional structure model to the optimal value of the area input feature parameter to obtain the parameter-tuned integrated image of the two-dimensional structure model. The optimal value of the area input feature parameter is obtained by training in a pre-built performance prediction model.
[0054] S150: The integrated image of the parameter-tuned two-dimensional structural model is input into the pre-trained performance prediction model to obtain the performance prediction result of the material to be predicted;
[0055] The performance prediction model is obtained by training a pre-built deep learning neural network with an integrated image of the two-dimensional structure model after material parameter tuning and performance labels. The integrated image of the two-dimensional structure model is a crystalline molecular structure model image with micro- and nano-sized pores. The performance labels include a first label for performance meeting the standard and a second label for performance not meeting the standard.
[0056] In some embodiments, such as Figure 2 As shown, the performance prediction model is obtained by training a pre-built deep learning neural network. The specific steps include:
[0057] S210: Obtain a vast number of 3D structural models and corresponding material property labels from materials databases;
[0058] S220: Project all three-dimensional structural models along the crystal axis to obtain a two-dimensional structural model image;
[0059] S230: At least two two-dimensional structural model images of the same material projected in different crystal axis directions are fused into a two-dimensional structural model integrated image using image fusion technology;
[0060] S240: Perform image annotation processing on all atoms contained in the characteristic groups in the integrated image of the two-dimensional structure model, and record the area occupied by all the image-annotated atoms in the integrated image of the two-dimensional structure model as the area input feature parameter;
[0061] S250: Utilize the massive amount of two-dimensional structural models to integrate images and corresponding material property labels to construct a sample set, which is then divided into a training set, a validation set, and a test set;
[0062] S260: Based on the two-dimensional structure model, integrate the image and use the deep learning neural network to classify and model the training set and validation set, and train the obtained classification model to obtain a performance prediction model;
[0063] S270: Evaluate the performance prediction model using the test set based on the evaluation metrics.
[0064] In some embodiments, the obtained classification model is trained to obtain a performance prediction model, specifically including:
[0065] The classification model is trained on the training set, and its hyperparameters are tuned on the validation set. The area input feature parameters of the integrated images of all two-dimensional structure models in the sample set are adjusted simultaneously based on the cross-entropy loss of the validation set until the change in the cross-entropy loss of the validation set is less than 2% for at least three consecutive training epochs. The optimal value of the area input feature parameters is then obtained to obtain the performance prediction model.
[0066] In some embodiments, the three-dimensional structural model of the material to be predicted is a crystallographic information file (CIF).
[0067] In some embodiments, the characteristic groups are chemical groups that have hydrophilic or hydrophobic properties, including: carboxyl groups, hydroxyl groups, aldehyde groups, amide groups, amino groups, nitro groups, halogen atoms, and hydrocarbon groups.
[0068] In some embodiments, image annotation processing is performed on all the atoms contained in the characteristic groups of the integrated image of the two-dimensional structure model, specifically including:
[0069] The atoms contained in all characteristic groups are identified and marked in the integrated image of the two-dimensional structural model by means of manual or automated methods. The markings can be dots, lines, boxes, polygons, or text labels.
[0070] In some embodiments, the evaluation metrics include accuracy, the harmonic mean (F1) of precision and recall, the area under the receiver operating characteristic curve (AUC), and the confusion matrix.
[0071] In some embodiments, the deep learning neural network algorithm used in the performance prediction model is a Convolutional Neural Network (CNN), which is suitable for image classification problems. The model only needs to acquire a two-dimensional image of the crystalline molecular structure model of the material, fuse the image, adjust the parameters, and train it according to the above method to efficiently and accurately predict whether the performance of the new material meets the standards. The model has high prediction accuracy and good performance. The evaluation results are shown in Table 1.
[0072] Table 1. Evaluation index results based on material property prediction from the test set.
[0073] Actual performance meets data volume requirements The actual data volume that does not meet the performance standards accuracy Accuracy Recall rate F1 AUC 261 557 0.913 0.832 0.912 0.870 0.809
[0074] Among them, accuracy refers to the percentage of correctly predicted samples out of all predicted samples; precision refers to the percentage of samples that are actually true out of all samples that the system predicts as true; recall refers to the percentage of samples that are actually true out of all samples that are predicted as true; F1 is the harmonic mean of precision and recall; and AUC is the area under the receiver operating characteristic curve (ROC curve).
[0075] According to the specific embodiments described above, the method provided by this invention predicts material properties based on convolutional neural networks. Compared with traditional experimental methods, it has a wider prediction range, higher efficiency, simpler operation, and more comprehensive evaluation metrics. The evaluation metrics for the machine learning classification model include: accuracy, the harmonic mean (F1) of precision and recall, the area under the receiver operating characteristic curve (AUC), and the confusion matrix. The confusion matrix of the test set is as follows: Figure 5 As shown.
[0076] In a specific application scenario, a performance prediction model is first constructed using Convolutional Neural Networks (CNNs). The training samples include integrated images of all two-dimensional structural models from the training and validation sets, along with all corresponding material performance data. During model training, all integrated images of the two-dimensional structural models are processed according to the methods described in all the specific implementation methods above, and then input into the CNN to construct a classification model. The classification model is trained using a deep learning algorithm: the sample set is divided into 10 equal parts, with one part used as the test set and the remaining 9 parts used as training data, and evaluated using 10-fold cross-validation. The training data is then divided into 9 equal parts, with one part used as the validation set and the remaining 8 parts used as the training set. The training set is input into the model for training, and after 10 training epochs, the model training learning curves for the training and validation sets are as follows. Figure 3 As shown, Figure 3The curve represents the relationship between the cross-entropy loss function and the number of training epochs. The test set is input into the trained performance prediction model to evaluate its performance. The performance prediction model is then encapsulated and loaded into the material performance prediction device provided by this invention.
[0077] In some embodiments, such as Figure 4 As shown, the specific process of using the performance prediction model includes:
[0078] S310: Encapsulate and integrate the trained and tested performance prediction model into a module;
[0079] S320: Input the material to be predicted, adjust the area, input the optimal value of the feature parameters, integrate the two-dimensional structural model image, and run the program in the module;
[0080] S330: Outputs the prediction results of whether the material properties meet or do not meet the standards.
[0081] In addition to the methods described above, this invention also provides a material property prediction device based on a structural model, such as... Figure 6 As shown, the device includes:
[0082] The three-dimensional structure model acquisition unit 101 is used to acquire the three-dimensional structure model of the material to be predicted from the material database. The three-dimensional structure model is a crystalline molecular structure model with micro- and nano-scale pore sizes.
[0083] The two-dimensional image projection generation unit 102 is used to project the three-dimensional structure model along the crystal axis to obtain a two-dimensional structure model image;
[0084] The image fusion unit 103 is used to fuse two-dimensional structure model images obtained by projecting them onto different crystal axis directions into a two-dimensional structure model integrated image using image fusion technology.
[0085] The area input feature parameter adjustment unit 104 is used to adjust the area input feature parameters to obtain a two-dimensional structural model integrated image with the area input feature parameters at their optimal values.
[0086] The material property prediction structure output unit 105 is used to output the performance prediction results of the material to be predicted.
[0087] In one embodiment, the present invention also provides a computer device for predicting material properties based on a structural model. This computer device may be a server, and its internal structure diagram may be as follows: Figure 7As shown, the computer device includes a processor, memory, and network interface connected via a system bus. The processor provides computing and control capabilities; the memory includes a storage medium and internal memory; the storage medium includes programs, an operating system, and a database. The storage medium can be memory, such as volatile or non-volatile memory, or both. The internal memory provides an environment for the operating system and programs running in the storage medium. The processor of the computer device can be an integrated circuit chip, a general-purpose processor, a digital signal processor, an application-specific integrated circuit (ASIC), or other logic device. The processor reads information from the storage medium and, in conjunction with its hardware, completes the steps of the above method embodiments. The database of the computer device stores static and dynamic information data. The network interface of the computer device is used for communication with external terminals via a network connection. When the program of the computer device is executed by the processor, it can implement the steps of the above method embodiments.
[0088] Those skilled in the art will understand that Figure 7 The structure shown is merely a partial structural block diagram related to the present invention and does not constitute a limitation on the computer device to which the present invention is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.
[0089] Corresponding to the above embodiments, the present invention also provides a computer-readable storage medium, including a volatile storage medium and a non-volatile storage medium, wherein a computer program is stored thereon, and the computer program, when executed by a processor, implements the steps of the method described above.
[0090] For those skilled in the art, based on the ideas of the embodiments of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.
Claims
1. A method for predicting material properties based on structural models, characterized in that, The method includes: Obtain a three-dimensional structural model of the material to be predicted, wherein the three-dimensional structural model is a crystalline molecular structure model with micro- and nano-sized pores; The three-dimensional structural model is projected along the crystal axis to obtain a two-dimensional structural model image, wherein the crystal axis includes [100], [010], [001], and [...]. 00]、[0 0]、[00 Six axes; At least two two-dimensional structure model images obtained by projecting them onto different crystal axis directions are fused into a two-dimensional structure model integrated image using image fusion technology; The total area of atoms contained in the characteristic groups in the integrated image of the two-dimensional structure model is adjusted to the optimal value of the area input feature parameter to obtain the parameter-tuned integrated image of the two-dimensional structure model. The optimal value of the area input feature parameter is obtained by training in a pre-built performance prediction model. The integrated image of the parameter-tuned two-dimensional structural model is input into a pre-trained performance prediction model to obtain the performance prediction result of the material to be predicted. The performance prediction model is obtained by training a pre-built deep learning neural network with an integrated image of the two-dimensional structure model after material parameter tuning and performance labels. The integrated image of the two-dimensional structure model is a crystalline molecular structure model image with micro- and nano-sized pores. The performance labels include a first label for performance meeting the standard and a second label for performance not meeting the standard.
2. The material property prediction method based on a structural model according to claim 1, characterized in that, The performance prediction model is obtained by training a pre-built deep learning neural network, specifically including: Obtain a vast number of 3D structural models and corresponding material property labels from materials databases; Projecting all three-dimensional structural models along the crystal axis yields a two-dimensional structural model image; At least two two-dimensional structural model images of the same material, projected onto different crystal axis directions, are fused into a unified two-dimensional structural model image using image fusion technology. The characteristic groups in the integrated image of the two-dimensional structure model are annotated with images, and the area occupied by all the annotated atoms in the integrated image of the two-dimensional structure model is recorded as the area input feature parameter. The massive amount of two-dimensional structural models are used to integrate images and corresponding material property labels to construct a sample set, which is then divided into a training set, a validation set, and a test set. Based on the two-dimensional structure model, the images are integrated and the deep learning neural network is used to classify and model the training set and the validation set. The resulting classification model is then trained to obtain a performance prediction model. The performance prediction model is evaluated using the test set based on the evaluation metrics. The evaluation metrics include accuracy, harmonic mean (F1) of precision and recall, area under the receiver operating characteristic curve (AUC), and confusion matrix. The process of training the obtained classification model to obtain a performance prediction model includes: The classification model is trained on the training set, and its hyperparameters are tuned on the validation set. The area input feature parameters of the integrated images of all two-dimensional structure models in the sample set are adjusted simultaneously based on the cross-entropy loss of the validation set until the change in the cross-entropy loss of the validation set is less than 2% for at least three consecutive training epochs. The optimal value of the area input feature parameters is then obtained to obtain the performance prediction model.
3. The material property prediction method based on a structural model according to claim 1, characterized in that, The three-dimensional structural model of the material to be predicted is obtained as a crystallographic information file (CIF).
4. The material property prediction method based on a structural model according to claim 1, characterized in that, The characteristic groups are chemical groups that have hydrophilic or hydrophobic properties, including: carboxyl groups, hydroxyl groups, aldehyde groups, amide groups, amino groups, nitro groups, halogen atoms, and hydrocarbon groups.
5. The material property prediction method based on a structural model according to claim 1, characterized in that, The two-dimensional structure model is integrated into the image, and all the atoms contained in the characteristic groups are annotated. Specifically, this includes: Identify and label the atoms contained in all characteristic groups in the integrated image of the two-dimensional structural model, either manually or automatically. The labels are in the form of dots, lines, boxes, polygons, or text labels.
6. A material property prediction device based on a structural model, characterized in that, The device includes: The three-dimensional structure model acquisition unit is used to acquire the three-dimensional structure model of the material to be predicted from the material database. The three-dimensional structure model is a crystalline molecular structure model with micro- and nano-sized pores. A two-dimensional image projection generation unit is used to project the three-dimensional structural model along the crystal axis to obtain a two-dimensional structural model image; The image fusion unit is used to fuse two-dimensional structural model images obtained by projecting them along different crystal axis directions into a two-dimensional structural model integrated image using image fusion technology. The area input feature parameter adjustment unit is used to adjust the area input feature parameters to obtain a two-dimensional structural model integrated image with the area input feature parameters at their optimal values. The material property prediction structure output unit is used to output the performance prediction results of the material to be predicted.
7. A computer device comprising a memory, a central processing unit (CPU), and a computer program stored in the memory and executable on the CPU, characterized in that, When the processor executes the program, it implements the steps of the method as described in any one of claims 1 to 5.
8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the central processing unit, it implements the steps of the method as described in any one of claims 1 to 5.