Photocuring material performance prediction method based on deep learning

Abstract

The invention provides a photocuring material performance prediction method based on deep learning, and relates to the technical field of photocuring 3D printing, and the method comprises the steps: collecting the multi-modal data of a photocuring material in real time; performing data cleaning, de-noising and standardization processing on the acquired multi-modal data, fusing the processed multi-modal data, and extracting key index values and printing process parameters of a material formula; constructing an initial deep learning neural network model by using key index values in the extracted multi-modal data; printing process parameters are used for training the initial neural network model, gradient descent is adopted for optimizing the model weight, performance evaluation and optimization are conducted on the model through cross validation, and a material performance prediction model is obtained after verification; and predicting the performance of a to-be-predicted photocuring material based on the material performance prediction model. By comprehensively analyzing the formula, illumination conditions and printing parameters of the photocuring material, the performance of the material under different conditions is accurately predicted.

Description

Technical Field

[0001] The present invention relates to the technical field of photocuring 3D printing, and in particular to a method for predicting photocuring material properties based on deep learning. Background Art

[0002] Photocuring 3D printing technology generates three-dimensional objects by curing liquid resin materials through light, and is widely used in the fields of medicine, aerospace, and automobile manufacturing. The performance of the material directly affects the quality of the printed parts, and traditional performance testing methods require a lot of experiments, which is time-consuming and costly. Under complex material formulas and printing conditions, how to efficiently and accurately predict material properties has become a key challenge in the research and development of photocuring materials.

[0003] However, existing deep learning-based material performance prediction methods are still in their infancy and are usually limited to the prediction of a single performance, failing to fully consider the various performance manifestations of materials and the complex nonlinear relationships between different factors. At the same time, how to extract key features from complex multimodal data and how to optimize the model's hyperparameters to improve the accuracy and stability of predictions are still technical challenges currently faced. Therefore, an innovative, multi-level prediction model is urgently needed to improve the accuracy and reliability of photocurable material performance predictions, thereby accelerating the research and development of materials and promoting the further application and development of 3D printing technology. Summary of the invention

[0004] In order to solve the above problems, this paper proposes a method for predicting the performance of photocurable materials based on deep learning. By comprehensively analyzing the formulation, lighting conditions and printing parameters of photocurable materials, the performance of materials under different conditions can be accurately predicted. This method aims to improve the efficiency of material development, reduce experimental costs, and promote the innovation and application of photocurable materials.

[0005] The present invention is implemented by the following technical solutions:

[0006] In a first aspect, the present invention provides a method for predicting properties of photocurable materials based on deep learning, the method comprising the following steps:

[0007] Real-time collection of multimodal data of photocurable materials, including recipe data, lighting condition data, printing parameter data, and material performance data of photocurable materials;

[0008] Perform data cleaning, denoising and standardization on the collected multimodal data, fuse the processed multimodal data, and extract key indicator values ​​of material formula and printing process parameters;

[0009] Using the key indicator values ​​from the extracted multimodal data, an initial deep learning neural network model is constructed;

[0010] The initial neural network model is trained using printing process parameters, the model weights are optimized using gradient descent, the model performance is evaluated and tuned through cross-validation, and the material performance prediction model is obtained after verification;

[0011] The performance of the photocurable material to be predicted is predicted based on the material performance prediction model.

[0012] As a further solution of the present invention, when collecting multimodal data of photocurable materials in real time, printing parameter data including printing speed, layer thickness, temperature and pressure are obtained from the 3D printer, and the chemical composition and concentration formula data of the material are recorded in real time using a high performance liquid chromatograph; a light intensity sensor is used to measure the wavelength, intensity and irradiation time of the light source in real time as lighting condition data, and material performance data of the hardness and elastic modulus of the material are monitored in real time through material testing equipment to obtain multimodal data of the photocurable material.

[0013] As a further solution of the present invention, when fusing the processed multimodal data, the following steps are included:

[0014] Organize the cleaned, denoised and standardized multimodal data into a multimodal matrix;

[0015] Use the linear algebra library to calculate the eigenvalues ​​and eigenvectors of the covariance matrix, where the eigenvalues ​​and eigenvectors satisfy:

[0016] Cv i =λ i v i

[0017] In the formula, v i is the i-th eigenvector; i is the i-th eigenvalue;

[0018] The first k principal components are selected according to the size of the eigenvalues, and the data are mapped to a new subspace using the selected principal components to obtain the fused data after dimensionality reduction.

[0019] As a further solution of the present invention, the multimodal data is organized into a multimodal matrix, and the multimodal matrix X has a dimension of n×m, where n is the number of samples and m is the total number of features of each sample.

[0020] As a further solution of the present invention, when calculating the eigenvalues ​​and eigenvectors of the covariance matrix, constructing the covariance matrix includes the following steps:

[0021] Calculate the mean vector μ of the multimodal matrix X;

[0022] Centralize each sample data by subtracting the mean vector;

[0023] Calculate the covariance matrix Σ of the centered data matrix; the calculation formula of the covariance matrix Σ is:

[0024]

[0025] Where X is the m-dimensional data matrix; X T is the transposed matrix of X, X T X is the m-by-m covariance matrix.

[0026] As a further solution of the present invention, when calculating the eigenvalues ​​and eigenvectors of the covariance matrix, the NumPy linear algebra library is used to calculate the eigenvalues ​​and eigenvectors of the covariance matrix, the eigenvalues ​​are sorted, and the eigenvectors corresponding to the first k largest eigenvalues ​​are selected; the transformation matrix W is constructed using the selected k principal components, and the original data X is projected into the new subspace to obtain the reduced-dimensional data.

[0027] As a further solution of the present invention, extracting key indicator values ​​and printing process parameters of the material formula includes the following steps:

[0028] Extract the first k principal components from the fusion data after dimension reduction as key indicators;

[0029] Identify process parameters for material performance through data regression analysis;

[0030] Use principal components to map the data; extract key indicator values ​​and printing process parameters.

[0031] As a further aspect of the present invention, the extracted printing process parameters include printing speed, layer thickness, printing temperature and pressure.

[0032] As a further solution of the present invention, using the key indicator values ​​in the extracted multimodal data to construct an initial deep learning neural network model includes the following steps:

[0033] Using key indicator values ​​extracted from multimodal data as input data;

[0034] Construct a neural network with multiple hidden layers and determine the activation function to use for each layer;

[0035] Define and optimize the loss function of the task, use gradient descent to optimize the network parameters, and obtain the constructed initial deep learning neural network model.

[0036] As a further solution of the present invention, when constructing the initial deep learning neural network model, it also includes:

[0037] Use the key indicator values ​​obtained after fusion as the input data set;

[0038] Divide the dataset into training, validation and test sets;

[0039] The input layer, hidden layer and output layer of the network structure are set, wherein the input layer is determined according to the number of key indicator values ​​extracted by fusion; each layer of the hidden layer is set with a different number of neurons; the number of neurons and activation function of the output layer are determined according to the type of prediction task;

[0040] Initialize the weights of the network parameters through Xavier;

[0041] Each hidden layer consists of fully connected neurons, the output layer uses a linear activation function, cross entropy loss is used as the loss function, the model is trained using gradient descent, and the validation set is used to evaluate the model performance.

[0042] As a further solution of the present invention, when gradient descent is used to optimize the model weights, a loss function is defined to measure the error between the model prediction value and the actual value, where y is assumed to be i is the true value of the i-th sample, is the predicted value of the model, then the mean square error of the loss function is:

[0043]

[0044] Among them, θ is the parameter weight of the model, and n is the number of samples;

[0045] When calculating the gradient, each parameter θ j , calculate the loss function L(θ) about θ j The partial derivative of :

[0046]

[0047] Represents the rate of change of the loss function with respect to the parameters.

[0048] As a further solution of the present invention, when the model parameters are updated using gradient descent, if the learning rate is η, the update rule is:

[0049]

[0050] Among them, the learning rate η is used to control the step size of each update.

[0051] The present invention also includes a computer device, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor executes the photocurable material performance prediction method based on deep learning.

[0052] The present invention also includes a computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable the computer to execute the photocurable material performance prediction method based on deep learning.

[0053] Compared with the prior art, the photocurable material performance prediction method based on deep learning provided by the present invention has the following beneficial effects:

[0054] 1. Improved prediction accuracy. The present invention can capture the full range of effects of material formula, lighting conditions and printing process on material properties by real-time collection and fusion of multimodal data of photocurable materials. Compared with traditional single-modal data, the use of multimodal data can more comprehensively reflect the complexity of materials and improve the accuracy of the model.

[0055] 2. Enhanced model interpretability by extracting key indicators. The present invention extracts key indicator values ​​in material formula and printing process through fusion technology from multimodal data after data cleaning, denoising and standardization. These key indicators not only improve the efficiency of the model, but also enhance the interpretability of the model, allowing users to understand which parameters have a greater impact on material performance.

[0056] 3. Training deep learning models improves generalization capabilities. The present invention uses deep learning neural networks to construct material performance prediction models and uses large-scale data for training. By introducing a multi-layer network structure, the model can capture complex nonlinear relationships and has stronger prediction capabilities. At the same time, cross-validation is used for tuning to ensure the generalization ability of the model, that is, the performance on unseen data is still reliable.

[0057] 4. Gradient descent optimization ensures rapid convergence. The present invention uses the gradient descent method to optimize the model weights, which can ensure that the model converges quickly within a limited number of training times and reduce training time. The optimization of learning rate and weight update helps to avoid the model from falling into local optimality and improve the final prediction accuracy. The model is also evaluated through k-fold cross-validation, which effectively avoids the overfitting phenomenon of the model on a single data set. The verification results of each fold balance the impact of different data sets on model performance, and further through hyperparameter tuning, it is ensured that the final model has good robustness.

[0058] 5. Accurately predict material properties, shorten the R&D cycle, and improve product stability. Based on the final material property prediction model, the present invention can quickly predict the performance of the photocurable material to be predicted, greatly reducing the number of experiments and debugging. Through this data-driven prediction method, the material R&D cycle can be effectively shortened and the development cost can be reduced. The prediction model continuously provides performance prediction results at different stages of product development to ensure product consistency and stability. In mass production, the material properties predicted by this method can ensure that the quality of each batch of products meets the expected standards.

[0059] In summary, the present invention successfully achieves efficient and accurate prediction of the performance of photocurable materials through multimodal data fusion, construction and optimization of deep learning neural networks, and cross-validation evaluation and tuning. It not only improves the accuracy and robustness of the prediction model, but also provides effective support for the rapid research and development and production of materials.

[0060] These and other aspects of the present invention will be more clearly understood in the following description of the embodiments. It should be understood that the above general description and the following detailed description are only exemplary and explanatory and cannot limit the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or related technologies, the following briefly introduces the drawings required for use in the exemplary embodiments or related technical descriptions. The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation to the present invention. In the drawings:

[0062] Figure 1 This is a flow chart of a method for predicting properties of photocurable materials based on deep learning according to an embodiment of the present invention.

[0063] Figure 2 This is a flowchart of fusing processed multimodal data in a method for predicting properties of photocurable materials based on deep learning according to an embodiment of the present invention.

[0064] Figure 3 This is a flowchart of constructing an initial deep learning neural network model in the method for predicting properties of photocurable materials based on deep learning in an embodiment of the present invention. DETAILED DESCRIPTION

[0065] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0066] In some of the processes described in the specification and claims of the present invention and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., do not represent the order of precedence, and do not limit the "first" and "second" to be different types.

[0067] The following will be combined with the drawings in the exemplary embodiments of the present invention to clearly and completely describe the technical solutions in the exemplary embodiments of the present invention. Obviously, the exemplary embodiments described are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0068] Since the existing material property prediction methods based on deep learning are still in their infancy, they are usually limited to the prediction of a single property and cannot fully consider the various performance manifestations of the material and the complex nonlinear relationships between different factors. At the same time, how to extract key features from complex multimodal data and how to optimize the hyperparameters of the model to improve the accuracy and stability of the prediction are still the technical challenges currently faced. The present invention provides a method for predicting the performance of photocurable materials based on deep learning. By comprehensively analyzing the formulation, lighting conditions and printing parameters of photocurable materials, the performance of materials under different conditions can be accurately predicted. This method aims to improve the efficiency of material development, reduce experimental costs, and promote the innovation and application of photocurable materials.

[0069] The technical solution of the present invention is further described below in conjunction with specific embodiments:

[0070] See also Figure 1 As shown, Figure 1 A flowchart of a method for predicting photocurable material properties based on deep learning provided by the present invention. A method for predicting photocurable material properties based on deep learning provided in one embodiment of the present invention comprises the following steps:

[0071] Step S10: collecting multimodal data of the photocurable material in real time, including recipe data, illumination condition data, printing parameter data, and material performance data of the photocurable material.

[0072] In this step, when collecting multimodal data of the photocurable material in real time, printing parameter data including printing speed, layer thickness, temperature and pressure are obtained from the 3D printer, and the chemical composition and concentration formula data of the material are recorded in real time using a high performance liquid chromatograph; a light intensity sensor is used to measure the wavelength, intensity and irradiation time of the light source in real time as lighting condition data, and material performance data of the hardness and elastic modulus of the material are monitored in real time through material testing equipment to obtain multimodal data of the photocurable material.

[0073] Step S20: performing data cleaning, denoising, and standardization processing on the collected multimodal data, fusing the processed multimodal data, and extracting key indicator values ​​and printing process parameters of the material formula.

[0074] In this step, see Figure 2 As shown, when fusing the processed multimodal data, the following steps are included:

[0075] Step S201, organizing the cleaned, denoised and standardized multimodal data into a multimodal matrix;

[0076] Step S202: Use the linear algebra library to calculate the eigenvalues ​​and eigenvectors of the covariance matrix, where the eigenvalues ​​and eigenvectors satisfy:

[0077] Cv i =λ i v i

[0078] In the formula, v i is the i-th eigenvector; i is the i-th eigenvalue;

[0079] Step S203: select the first k principal components according to the size of the eigenvalues, and use the selected principal components to map the data to a new subspace to obtain fused data after dimensionality reduction.

[0080] Among them, multimodal data is organized into a multimodal matrix, and the dimension of the multimodal matrix X is n×m, where n is the number of samples and m is the total number of features of each sample. Assume that there are 100 samples (n=100) and each sample contains 10 features (m=10). Therefore, the dimension of the multimodal matrix is ​​100×10.

[0081] In this embodiment, when calculating the eigenvalues ​​and eigenvectors of the covariance matrix, constructing the covariance matrix includes the following steps:

[0082] Calculate the mean vector μ of the multimodal matrix X;

[0083] Centralize each sample data by subtracting the mean vector;

[0084] Calculate the covariance matrix Σ of the centered data matrix; the calculation formula of the covariance matrix Σ is:

[0085]

[0086] Where X is the m-dimensional data matrix; X T is the transposed matrix of X, X T X is the m-by-m covariance matrix.

[0087] Among them, when calculating the eigenvalues ​​and eigenvectors of the covariance matrix, the NumPy linear algebra library is used to calculate the eigenvalues ​​and eigenvectors of the covariance matrix, sort the eigenvalues, and select the eigenvectors corresponding to the first k largest eigenvalues; the selected k principal components are used to construct the transformation matrix W, and the original data X is projected into the new subspace to obtain the reduced-dimensional data.

[0088] In this embodiment, extracting key indicator values ​​and printing process parameters of the material formula includes the following steps:

[0089] Extract the first k principal components from the fusion data after dimension reduction as key indicators;

[0090] Identify process parameters for material performance through data regression analysis;

[0091] Use principal components to map the data; extract key indicator values ​​and printing process parameters.

[0092] Among them, the extracted printing process parameters include printing speed, layer thickness, printing temperature and pressure.

[0093] Step S30: Use the key indicator values ​​in the extracted multimodal data to build an initial deep learning neural network model.

[0094] In this step, see Figure 3 As shown, using the key indicator values ​​extracted from the multimodal data, an initial deep learning neural network model is constructed, including the following steps:

[0095] Step S301: using key indicator values ​​extracted from multimodal data as input data;

[0096] Step S302: construct a neural network including multiple hidden layers, and determine the activation function used for each layer;

[0097] Step S303: define and optimize the loss function of the task, use gradient descent to optimize the network parameters, and obtain the constructed initial deep learning neural network model.

[0098] Exemplarily, it is assumed that the key indicator values ​​of the material have been extracted, including printing speed, layer thickness, printing temperature and pressure, and the data matrix X obtained by fusion processing of these data is used for model construction. The input data is: the matrix X of the key indicator values ​​is used as the input data of the neural network. Assuming that the present invention has 1000 samples, each sample contains 8 key indicators (printing speed, layer thickness), so the dimension of the input data matrix X is 1000×8.

[0099] Among them, when building the initial deep learning neural network model, it also includes:

[0100] Use the key indicator values ​​obtained after fusion as the input data set;

[0101] Divide the dataset into training, validation and test sets;

[0102] The input layer, hidden layer and output layer of the network structure are set, wherein the input layer is determined according to the number of key indicator values ​​extracted by fusion; each layer of the hidden layer is set with a different number of neurons; the number of neurons and activation function of the output layer are determined according to the type of prediction task;

[0103] Initialize the weights of the network parameters through Xavier;

[0104] Each hidden layer consists of fully connected neurons, the output layer uses a linear activation function, cross entropy loss is used as the loss function, the model is trained using gradient descent, and the validation set is used to evaluate the model performance.

[0105] Step S40: Use the printing process parameters to train the initial neural network model, use gradient descent to optimize the model weights, evaluate and tune the model performance through cross-validation, and obtain the material performance prediction model after verification.

[0106] In this step, when using gradient descent to optimize the model weights, define the loss function that measures the error between the model prediction value and the actual value, where y is set i is the true value of the i-th sample, is the predicted value of the model, then the mean square error of the loss function is:

[0107]

[0108] Among them, θ is the parameter weight of the model, and n is the number of samples;

[0109] When calculating the gradient, each parameter θ j , calculate the loss function L(θ) about θ j The partial derivative of :

[0110]

[0111] Represents the rate of change of the loss function with respect to the parameters.

[0112] Among them, when using gradient descent to update model parameters, if the learning rate is η, the update rule is:

[0113]

[0114] Among them, the learning rate η is used to control the step size of each update.

[0115] Step S50: predicting the performance of the photocurable material to be predicted based on the material performance prediction model.

[0116] Among them, if the number of neurons in the input layer is equal to the number of key indicator values, both are 8 neurons; the first hidden layer of the hidden layer has 64 neurons, and the second hidden layer has 32 neurons, and each layer uses the ReLU activation function; the output layer has one neuron and uses a linear activation function; the weights of the network parameters are initialized using the Xavier initialization method, the neural network is trained using the training set data, and the model weights are optimized using the gradient descent method. The learning rate is set to 0.01 for parameter update, and cross-validation is used to evaluate the model performance. The data set is divided into 5 folds, and different training sets and validation sets are used in each fold for model training and performance evaluation. The trained material performance prediction model is used to predict the performance of new light-curing material samples. Enter the key indicator values ​​of the new light-curing material, and the model will output the predicted material performance indicators.

[0117] The present invention can capture the full range of effects of material formula, lighting conditions and printing process on material properties by real-time collection and fusion of multimodal data of photocurable materials. Compared with traditional single-modal data, the use of multimodal data can more comprehensively reflect the complexity of materials and improve the accuracy of the model. Through data cleaning, denoising and standardization of multimodal data, key indicator values ​​in material formula and printing process are extracted through fusion technology. These key indicators not only improve the efficiency of the model, but also enhance the interpretability of the model, so that users can understand which parameters have a greater impact on material properties.

[0118] The present invention adopts deep learning neural network to construct material property prediction model, and uses large-scale data for training. By introducing a multi-layer network structure, the model can capture complex nonlinear relationships and has stronger prediction capabilities. At the same time, cross-validation is used for tuning to ensure the generalization ability of the model, that is, the performance on unseen data is still reliable. The gradient descent method is used to optimize the model weights, which can ensure that the model converges quickly within a limited number of training times and reduce training time. The optimization of learning rate and weight update helps to avoid the model from falling into local optimality and improve the final prediction accuracy. The model is also evaluated by k-fold cross-validation, which effectively avoids the overfitting phenomenon of the model on a single data set. The verification results of each fold balance the impact of different data sets on model performance, and further through hyperparameter tuning, it is ensured that the final model has good robustness.

[0119] Based on the final material performance prediction model, the present invention can quickly predict the performance of the photocurable material to be predicted, greatly reducing the number of experiments and debugging. Through this data-driven prediction method, the material research and development cycle can be effectively shortened and the development cost can be reduced. The prediction model continuously provides performance prediction results at different stages of product development to ensure the consistency and stability of the product. In mass production, the material performance predicted by this method can ensure that the quality of each batch of products meets the expected standards.

[0120] In summary, the present invention successfully achieves efficient and accurate prediction of the performance of photocurable materials through multimodal data fusion, construction and optimization of deep learning neural networks, and cross-validation evaluation and tuning. It not only improves the accuracy and robustness of the prediction model, but also provides effective support for the rapid research and development and production of materials.

[0121] It should be understood that, although described in a certain order, these steps are not necessarily performed in sequence in the above order. Unless there is clear explanation in this article, the execution of these steps does not have strict order restriction, and these steps can be performed in other orders. Moreover, a part of the steps of the present embodiment may include a plurality of steps or a plurality of stages, and these steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a part of the steps or stages in other steps or other steps.

[0122] In one embodiment, a computer device is also provided in an embodiment of the present invention, comprising at least one processor and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor performs the steps of the method for predicting the properties of photocurable materials based on deep learning.

[0123] In one embodiment, the present invention further provides a computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable the computer to execute the steps of the method for predicting properties of photocurable materials based on deep learning.

[0124] A person skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be realized by instructing the relevant hardware through a computer program represented by computer instructions, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory.

[0125] Non-volatile memory may include read-only memory, magnetic tape, floppy disk, flash memory or optical storage, etc. Volatile memory may include random access memory or external cache memory. As an illustration and not limitation, RAM may be in various forms, such as static random access memory or dynamic random access memory, etc.

[0126] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting properties of photocurable materials based on deep learning, characterized in that: The method comprises the following steps: Real-time collection of multimodal data of photocurable materials, including recipe data, lighting condition data, printing parameter data, and material performance data of photocurable materials; Perform data cleaning, denoising and standardization on the collected multimodal data, fuse the processed multimodal data, and extract key indicator values ​​of material formula and printing process parameters; Using the key indicator values ​​from the extracted multimodal data, an initial deep learning neural network model is constructed; The initial neural network model is trained using printing process parameters, the model weights are optimized using gradient descent, the model performance is evaluated and tuned through cross-validation, and the material performance prediction model is obtained after verification; The performance of the photocurable material to be predicted is predicted based on the material performance prediction model.

2. The method for predicting properties of photocurable materials based on deep learning according to claim 1, characterized in that: When collecting multimodal data of photocurable materials in real time, printing parameter data including printing speed, layer thickness, temperature and pressure are obtained from the 3D printer, and the formula data of the chemical composition and concentration of the materials are recorded in real time using a high-performance liquid chromatograph; A light intensity sensor is used to measure the wavelength, intensity and irradiation time of the light source in real time as the lighting condition data, and the material performance data of the hardness and elastic modulus of the material are monitored in real time through material testing equipment to obtain multimodal data of the photocurable material.

3. The method for predicting properties of photocurable materials based on deep learning according to claim 2, characterized in that: When fusing the processed multimodal data, the following steps are included: Organize the cleaned, denoised and standardized multimodal data into a multimodal matrix; Use the linear algebra library to calculate the eigenvalues ​​and eigenvectors of the covariance matrix, where the eigenvalues ​​and eigenvectors satisfy: C1i=λivi In the formula, v i is the i-th eigenvector; i is the i-th eigenvalue; The first k principal components are selected according to the size of the eigenvalues, and the data are mapped to a new subspace using the selected principal components to obtain the fused data after dimensionality reduction.

4. The method for predicting properties of photocurable materials based on deep learning according to claim 3, characterized in that: Multimodal data are organized into a multimodal matrix X, and the dimension of the multimodal matrix X is n×m, where n is the number of samples and m is the total number of features for each sample.

5. The method for predicting properties of photocurable materials based on deep learning according to claim 4, characterized in that: When calculating the eigenvalues ​​and eigenvectors of the covariance matrix, constructing the covariance matrix includes the following steps: Calculate the mean vector μ of the multimodal matrix X; Centralize each sample data by subtracting the mean vector; Calculate the covariance matrix Σ of the centered data matrix; the calculation formula of the covariance matrix Σ is: Where X is the m-dimensional data matrix; X T is the transposed matrix of X, X T X is the m-by-m covariance matrix.

6. The method for predicting properties of photocurable materials based on deep learning according to claim 5, characterized in that: When calculating the eigenvalues ​​and eigenvectors of the covariance matrix, the NumPy linear algebra library is used to calculate the eigenvalues ​​and eigenvectors of the covariance matrix, sort the eigenvalues, and select the eigenvectors corresponding to the first k largest eigenvalues; the selected k principal components are used to construct the transformation matrix W, and the original data X is projected into the new subspace to obtain the reduced-dimensional data.

7. The method for predicting properties of photocurable materials based on deep learning according to claim 6, characterized in that: Extracting the key indicator values ​​of the material formula and the printing process parameters includes the following steps: Extract the first k principal components from the fusion data after dimension reduction as key indicators; Identify process parameters for material performance through data regression analysis; Use principal components to map the data; extract key indicator values ​​and printing process parameters. As a further aspect of the present invention, the extracted printing process parameters include printing speed, layer thickness, printing temperature and pressure.

8. The method for predicting properties of photocurable materials based on deep learning according to claim 7, characterized in that: Using the key indicator values ​​from the extracted multimodal data, an initial deep learning neural network model is constructed, including the following steps: Using key indicator values ​​extracted from multimodal data as input data; Construct a neural network with multiple hidden layers and determine the activation function to use for each layer; Define and optimize the loss function of the task, use gradient descent to optimize the network parameters, and obtain the constructed initial deep learning neural network model.

9. The method for predicting properties of photocurable materials based on deep learning according to claim 8, characterized in that: When building the initial deep learning neural network model, also include: Use the key indicator values ​​obtained after fusion as the input data set; Divide the dataset into training, validation and test sets; The input layer, hidden layer and output layer of the network structure are set, wherein the input layer is determined according to the number of key indicator values ​​extracted by fusion; each layer of the hidden layer is set with a different number of neurons; the number of neurons and activation function of the output layer are determined according to the type of prediction task; Initialize the weights of the network parameters through Xavier; Each hidden layer consists of fully connected neurons, the output layer uses a linear activation function, cross entropy loss is used as the loss function, the model is trained using gradient descent, and the validation set is used to evaluate the model performance.

10. The method for predicting properties of photocurable materials based on deep learning according to claim 9, characterized in that: When using gradient descent to optimize model weights, define a loss function that measures the error between the model's predicted value and the actual value, where y is assumed to be i is the true value of the i-th sample, is the predicted value of the model, then the mean square error of the loss function is: Among them, θ is the parameter weight of the model, and n is the number of samples; When calculating the gradient, each parameter θ j , calculate the loss function L(θ) about θ j The partial derivative of : Represents the rate of change of the loss function with respect to the parameters.

Images