An artificial intelligence generation method and system for a ferroelectric material composition-temperature phase diagram

Abstract

The application relates to an artificial intelligence generation method and system of a ferroelectric material component-temperature phase diagram, and the method comprises the following steps: acquiring a chemical formula of a ferroelectric material with a doping variable, and setting the chemical formula into a string form; setting the change range and step length of the doping proportion and temperature; processing chemical elements and the doping proportion in the chemical formula of the ferroelectric material with the doping variable in the string form to obtain a chemical formula numerical vector list; forming a temperature list according to the set change range and step length of the temperature, and performing normalization processing to obtain a normalized temperature list; based on the chemical formula numerical vector list and the normalized temperature list, performing merging and inputting into a pre-trained crystal phase transition artificial intelligence model to output a crystal structure type; and based on the crystal structure type, a component-temperature phase diagram of the chemical formula of the ferroelectric material with the doping variable under different doping proportions and different temperature conditions is drawn. Compared with the prior art, the application has the advantages of automatically generating a component-temperature phase diagram.

Description

Technical Field

[0001] This invention relates to the field of new materials technology, and in particular to an artificial intelligence method and system for generating composition-temperature phase diagrams of ferroelectric materials. Background Technology

[0002] With the rapid development of cutting-edge fields such as next-generation information technology, new energy, and integrated circuits, the demand for electronic functional materials has increased dramatically. Ferroelectric materials, as fundamental electronic functional materials, are widely used in piezoelectric transducers, infrared sensors, and energy conversion. Traditional typical ferroelectric materials, such as barium titanate and lead titanate, are no longer sufficient to meet the diverse performance requirements of these advanced fields. Therefore, finding feasible solutions for the accurate and rapid development of novel high-performance ferroelectric materials is a crucial technology that urgently needs to be addressed. Researchers have discovered that ferroelectric materials located at phase boundaries—the intersections of multiple crystal structures—possess a variety of excellent electrical properties in the composition-temperature phase diagram of ferroelectric materials. Taking calcium-zirconium co-doped barium titanate as an example, at room temperature, the doping ratio at cubic, tetragonal, and trigonal phase interfaces generally improves the piezoelectric coefficient, dielectric constant, and other electrical properties by more than one order of magnitude compared to pure barium titanate. Designing doped ferroelectric materials based on phase interfaces with multiple crystal structures has become a guiding principle for the rapid development of ferroelectric materials in this field. Therefore, accurately and rapidly constructing the composition-temperature phase diagram of ferroelectric materials allows for the direct identification of multiphase interfaces and the direct synthesis of ferroelectric materials with high electrical properties. However, traditional methods for constructing composition-temperature phase diagrams typically involve designing experiments, synthesizing large quantities of materials with different doping ratios, and characterizing the crystal structure and phase transition temperature of each material to create the composition-temperature phase diagram and identify multiphase interfaces. For ferroelectric materials with diverse doping ratios and types, traditional methods cannot achieve rapid phase diagram construction and interface identification.

[0003] Patent application CN115410654A discloses a machine learning method for predicting the ferroelectric-paraelectric phase transition temperature of molecular ferroelectrics. The method collects the chemical formulas of molecular ferroelectrics and their experimental values ​​for ferroelectric-paraelectric phase transition temperatures from literature. It then uses literature, first-principles calculations, and topological structures to calculate feature parameters describing the sample characteristics. The dimensionality of the features is rapidly reduced, and a subset search strategy is used to embed a learner to obtain the optimal subset of features. Using the ferroelectric-paraelectric phase transition temperature as the target variable and the optimal features as independent variables, a quantitative prediction model for the ferroelectric-paraelectric phase transition temperature of molecular ferroelectrics is established using a support vector machine regression algorithm. New molecular ferroelectric samples are collected, their feature parameters are obtained, and the ferroelectric-paraelectric phase transition temperature is predicted based on the quantitative prediction model. However, this method requires first-principles calculations, topological structures, and feature parameters of the samples as input data. Furthermore, it does not involve predicting crystal structure types, generating composition-temperature phase diagrams for ferroelectric materials, or automatically generating such diagrams. Summary of the Invention

[0004] The purpose of this invention is to provide an artificial intelligence generation method and system for ferroelectric material composition-temperature phase diagrams that enables automated generation.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] An artificial intelligence method for generating composition-temperature phase diagrams of ferroelectric materials includes the following steps:

[0007] Obtain the chemical formula of the doped ferroelectric material and set it as a string, wherein the chemical formula of the doped ferroelectric material includes chemical elements and corresponding doping ratios, and the doping variable is included in the doping ratio;

[0008] Based on the chemical formula of doped variable ferroelectric materials in string form, the range and step size of the doping ratio and temperature variation are set;

[0009] Based on the set range and step size of the doping ratio, the chemical elements and doping ratios in the string-form chemical formula of the doped variable ferroelectric material are processed to obtain a list of chemical formula numerical vectors.

[0010] Based on the set temperature variation range and step size, a temperature list is generated and normalized to obtain a normalized temperature list.

[0011] Based on the list of chemical formula numerical vectors and the list of normalized temperatures, they are merged and used as input data into a pre-trained artificial intelligence model for crystal phase transformation, which outputs the crystal structure type.

[0012] Based on the crystal structure type, the composition-temperature phase diagrams of the chemical formula of the doped variable ferroelectric material under different doping ratios and temperature conditions were plotted.

[0013] Furthermore, the chemical formula of the doped variable ferroelectric material in string form uses symbols to distinguish chemical elements and doping ratios, specifically:

[0014] The chemical element is placed in the symbol, the doping ratio is placed on the side of the corresponding chemical element, or both the chemical element and the doping ratio are placed in different symbols, where the doping variable is included in the doping ratio.

[0015] Furthermore, the step of obtaining the list of chemical formula numerical vectors includes:

[0016] Based on the string-format chemical formula of the doped variable ferroelectric material, chemical formulas with different doping ratios are generated in batches according to the set range and step size of the doping ratio, forming a list of chemical formulas with different doping ratios. The number of formulas generated in batches is:

[0017]

[0018] In the formula, N1 is the number of chemical formulas generated in batches, R1 and R2 are the upper and lower limits of the doping ratio, respectively, and ΔR is the step size;

[0019] Based on the list of chemical formulas with different doping ratios, the doping ratios were normalized using a logarithmic normalization method to obtain the logarithmically normalized doping ratios:

[0020] By embedding the chemical elements and logarithmically normalized doping ratios of the chemical formulas in the periodic table, an initial list of numerical vectors for the chemical formulas is obtained. The expression for embedding is as follows:

[0021]

[0022] In the formula, v[j] is the numerical vector of the j-th initialization formula, and m′ i E represents the log-normalized doping ratio of the i-th chemical element. j Let ε be the j-th chemical formula containing n elements. i A list of elements in the periodic table;

[0023] Based on the initial list of chemical formula numerical vectors, principal component analysis is used to reduce the dimensionality, resulting in the final list of chemical formula numerical vectors. The dimensionality reduction is expressed as follows:

[0024] z p =v j U k

[0025] In the formula, z p v is the numerical vector of the chemical formula after dimensionality reduction. j For the j-th initialization formula numerical vector, U k It is a low-dimensional basis matrix.

[0026] Furthermore, the step of obtaining the log-normalized doping ratio includes:

[0027] Based on the list of chemical formulas with different doping ratios, the doping ratios are normalized, and the normalization expression is as follows:

[0028]

[0029] In the formula, m * i Let m be the normalized doping ratio coefficient for the i-th chemical element. i Let be the doping ratio of the i-th chemical element;

[0030] The normalized doping ratio is logarithmically transformed to obtain the logarithmically normalized doping ratio, which is expressed as follows:

[0031] m′ i =log(1+m) * i )

[0032] In the formula, m′ i is the log-normalized doping ratio of the i-th chemical element.

[0033] Furthermore, the step of obtaining the normalized temperature list includes:

[0034] Based on the set temperature variation range and step size, each temperature point is generated in batches to form a temperature list. Each temperature point represents the total number of calculations for each chemical formula numerical vector in the chemical formula numerical vector list. The calculation expression is:

[0035]

[0036] In the formula, N2 is the total number of calculations for each chemical formula numerical vector at different temperature points, T1 and T2 are the upper and lower limits of temperature, respectively, and ΔT is the step size;

[0037] Based on the temperature list, a normalized temperature list is obtained by normalizing the temperature using the maximum-minimum normalization algorithm. The expression for the maximum-minimum normalization algorithm is as follows:

[0038]

[0039] In the formula, T′ iFor the i-th normalized temperature point, T i Let T be the i-th initial temperature. min T max These are the minimum and maximum temperatures, respectively.

[0040] Furthermore, the crystal phase transformation artificial intelligence model includes an input layer, a hidden layer, and an output layer. The input layer is used to input input data obtained by merging the chemical formula numerical vector list and the normalized temperature list. The hidden layer is used to extract deep-level implicit features from the input data. The output layer is used to perform classification prediction based on the extracted deep-level implicit features and output the predicted crystal structure type.

[0041] Furthermore, the hidden layer has multiple layers, and each hidden layer adds a batch normalization operation and a discard operation. The execution steps of the hidden layer include:

[0042] In the first hidden layer, the input data is processed to obtain the initial output vector h of the first hidden layer. 1 :

[0043] h 1 =ReLU(W (1) Q+b (1) )

[0044] In the formula, W (1) Here, Q is the weight matrix of the first hidden layer, and b is the input vector. (1) This is the bias vector of the first hidden layer;

[0045] For the initial output vector h of the first layer 1 Perform batch normalization to obtain the initial output vector after batch normalization.

[0046]

[0047] In the formula, γ, β, and ∈ are training parameters, and μ batch and These are the mean and variance of the current batch;

[0048] Based on the batch-normalized initial output vector Perform a discard operation to obtain the final output vector of the first hidden layer, and use it as the input of the next hidden layer. Repeat the above steps until the processing of all hidden layers is completed.

[0049] Furthermore, the final output vector of the first hidden layer is represented as:

[0050]

[0051] In the formula, is the final output vector of the first hidden layer, and m is the mask.

[0052] Further, the step of outputting the crystal structure type includes:

[0053] Based on the list of chemical formula numerical vectors and the list of normalized temperatures, the total number of executions of the pre-trained AI model for crystal phase transformation is calculated. The expression for calculating the total number of executions is as follows:

[0054] N = N1 × N2

[0055] In the formula, N is the total number of executions, N1 is the number of chemical formula numerical vectors in the chemical formula numerical vector list, that is, the number of chemical formulas with different doping ratios, and N2 is the number of temperature points in the normalized temperature list, that is, the total number of calculations for each chemical formula numerical vector at different temperature points.

[0056] The chemical formula numerical vectors in the chemical formula numerical vector list and the normalized temperatures in the normalized temperature list are merged and used as input data for one execution. The pre-trained crystal phase transformation artificial intelligence model is then used to process the data and output the crystal structure type for this execution. This step is repeated N-1 times to obtain the crystal structure types of all chemical formula numerical vectors at each temperature point.

[0057] This invention also provides an artificial intelligence generation system for composition-temperature phase diagrams of ferroelectric materials, comprising:

[0058] Pre-trained AI module: used to acquire a dataset of phase transformation of electrical materials, train a multilayer artificial neural network model, perform hyperparameter optimization and cross-validation during training, and obtain a pre-trained phase transformation AI model;

[0059] The phase diagram automatic generation module is used to read the chemical formula, doping ratio, and temperature variation range and step size of the doped variable ferroelectric material, perform standardization processing, cyclically call the pre-trained crystal phase transformation artificial intelligence model for processing, output the crystal structure type, and draw the composition-temperature phase diagram of the doped variable ferroelectric material under different doping ratios and different temperature conditions according to the crystal structure type. The chemical formula of the doped variable ferroelectric material includes chemical elements and corresponding doping ratios, and the doping variable is included in the doping ratio.

[0060] Compared with the prior art, the present invention has the following beneficial effects:

[0061] (1) This invention generates chemical formulas with different doping ratios by setting the doping ratio and temperature of the chemical formula of the doped ferroelectric material. By merging the chemical formulas with different temperature points, the crystal structure type is predicted using an artificial intelligence model of crystal phase transformation. Thus, the composition-temperature phase diagrams under different doping ratios and temperature conditions can be drawn. This invention only requires inputting the chemical formula of the ferroelectric material, without any additional information such as material properties, structure, or atoms. It not only realizes the automated generation of the composition-temperature phase diagram of the ferroelectric material, but also reduces the generation difficulty.

[0062] (2) By using a pre-trained artificial intelligence model for crystal phase transformation, the model does not need to undergo unnecessary parameter tuning and other optimization processes when predicting the crystal structure of ferroelectric materials, thus improving execution efficiency.

[0063] (3) Compared with traditional experimental methods for constructing component-temperature phase diagrams, the present invention uses an artificial intelligence-based automatic generation method to significantly shorten the time required to generate a complete component-temperature phase diagram and reduce research and development costs. Attached Figure Description

[0064] Figure 1 This is a schematic diagram of the artificial intelligence generation method of the present invention;

[0065] Figure 2 This is a schematic diagram of the artificial intelligence generation system of the present invention;

[0066] Figure 3 This is a schematic diagram of the data processing flow for the chemical formula of the ferroelectric material of the present invention;

[0067] Figure 4 This represents the accuracy of the pre-trained phase transition artificial intelligence model in this invention on the training and validation sets.

[0068] Figure 5 The calcium-zirconium-doped barium titanate ferroelectric material (1-x)Ba(Zr) produced in this invention 0.2 Ti 0.8 O3-xBa 0.7 Ca 0.3 The composition-temperature phase diagram of TiO3;

[0069] Figure 6 The tin-doped barium titanate ferroelectric material BaSn generated in this invention x Ti 1-x O3 composition-temperature phase diagram;

[0070] Figure 7 The lead zirconate titanate ferroelectric material PbZr generated in this invention 1-x Ti x Composition-temperature phase diagram of O3. Detailed Implementation

[0071] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0072] Example 1

[0073] This embodiment provides an artificial intelligence method for generating composition-temperature phase diagrams of ferroelectric materials, such as... Figure 1 As shown, the method includes the following steps:

[0074] (1) Natural language processing was used to obtain ferroelectric material phase transformation data from existing literature, and the collected data was used to construct a ferroelectric material phase transformation dataset.

[0075] This ferroelectric material phase transformation dataset is a batch acquisition of ferroelectric material phase transformation information from accessible scientific literature databases using natural language processing or manual acquisition methods. The dataset includes data such as chemical formula, temperature, crystal structure, document identification code, journal name, publication year, and their source information. The phase transformation dataset is a linked list data structure, but it can also be an array, tree, graph, or other data structures.

[0076] (2) A multi-layer artificial neural network intelligent model was built, trained with crystal structure transformation data of ferroelectric materials, and obtained through cross-validation to obtain an artificial intelligence model of crystal phase transformation.

[0077] The artificial neural network module contains a pre-trained multilayer artificial neural network model, which is trained using a ferroelectric material phase transformation dataset.

[0078] A multi-layer artificial neural network model consists of one or more input layers, fully connected layers, and an output layer. The input layer receives processed data from the original data, with the proportions of chemical elements in the original chemical formula normalized. For a given chemical formula, E represents the proportions of chemical elements. d (i = 1, 2, 3…) represents a chemical formula containing n elements, with m… d (i = 1, 2, 3…) represents the doping ratio of each element. The normalized doping ratio coefficient m is calculated based on the following formula. * i (i = 1, 2, 3…),

[0079]

[0080] Then, by performing a logarithmic transformation on the proportionality coefficient according to the following formula, the logarithmically normalized doping ratio m′ can be obtained. i (i = 1, 2, 3…),

[0081] m′d =log(1+m) * i (2)

[0082] Then, based on the following formula, list the elements and proportions in the chemical formula using elements from the periodic table ε. i (i = 1, 2, 3...) are used for embedding representation.

[0083]

[0084] The numerical representation of the chemical formula is obtained, and the original chemical formula is converted into a new vector v.

[0085] The vector v is calculated according to its dimensions using the principal component analysis algorithm, and the result is obtained through the low-dimensional basis matrix U. k The input vector z is reduced to a specified p-dimensionality using the following formula. p ;

[0086] z p =v j U k (4)

[0087] Temperature T in the original input data i The normalized temperature T′ is obtained by calculating using the Min-Max normalization algorithm, as shown in the following formula:

[0088]

[0089] Input vector z p The normalized temperature T′ is combined with the normalized temperature T′ to form a new vector, which constitutes the input vector Q of the first layer of the neural network, with a dimension of d = p + 1;

[0090] Each layer of the artificial neural network extracts deep-seated implicit features from the input data. Using the weight matrix W and the bias vector b, the feature h extracted by the neuron in the s-th layer is calculated according to the following formula.

[0091] h s =ReLU(W (s) Q+b (s) (6)

[0092] Preferably, each layer of the artificial neural network includes one or more neurons, activation functions, regularizers, batch normalizers, and dropouts, which can improve the stability and generalization ability of the model; the hyperparameters in each layer of the artificial neural network are obtained through grid search to obtain the hyperparameters with the highest accuracy; the constructed artificial neural network module is a multilayer perceptron model for classification tasks, or it can be a model for other types of tasks; the last layer of the artificial neural network uses an activation function to output the predicted class probability.

[0093] The output layer is the Lth layer. The output vector o of the output layer is calculated according to the following formula, and the activation function is used to transform the output vector into the numerical data corresponding to the crystal structure type.

[0094] o = W (L) h L-1 +b (L) (7)

[0095] The activation function used is the Softmax function, and the total number of crystal structure types is C. The probability of the predicted crystal structure numerical type is calculated using the following formula: Take the maximum probability The crystal structure corresponding to the current input vector;

[0096]

[0097] The accuracy of the artificial neural network was evaluated using metrics such as loss function and accuracy. Cross-validation was performed on the training and validation sets, which were obtained by dividing the crystal phase transformation dataset acquired in step (1) according to a certain ratio. As the training rounds increased, the artificial neural network continuously optimized the weight matrix W and the bias coefficient b. The accuracy of the optimized crystal phase transformation artificial intelligence model was greater than 90%.

[0098] (3) Set the chemical formula of the doped ferroelectric material to be generated phase diagram.

[0099] The data format of the chemical formula is set as a string of one or more elements such as "element1[proportion1]element2[proportion2]". This method enables computer programs to read the elements and proportions in the chemical formula. This structured string can also be a combination of other symbols, such as using parentheses, commas, periods, semicolons and other symbols to distinguish chemical elements and proportions. It can also be a form in which elements are placed in symbols, or a form in which elements and proportions are placed in different symbols.

[0100] Doping variable refers to the doping of a chemical element into a matrix material. The doping ratio x is used as a doping variable and is contained in the string of doping ratio. The string containing the doping variable can be converted into a computer-recognizable form and mathematical operations can be performed.

[0101] (4) Set the upper and lower limits of the doping ratio to E1 and E2, and the step size to ΔR. Set the upper and lower limits of the temperature change to T1 and T2, and the step size to ΔT.

[0102] The range of doping ratio variation is the upper and lower limits of the composition axis in the proposed phase diagram; the range of variation of the specified temperature conditions is between 0 and 2000 K.

[0103] (5) Calculate the proportionality coefficients in the specified ferroelectric material chemical formula based on the set doping ratio range and step size, and calculate the total number N1 of the chemical formula using the following formula.

[0104]

[0105] After the calculation is completed, N1 chemical formulas are generated in batches. The numerical representation of the chemical formulas is obtained by using the embedding representation method, and the data dimension is optimized.

[0106] The generated chemical formulas are represented in an embedded manner using the periodic table of elements, converting strings of chemical elements and proportions into numerical lists to achieve numerical representation of the chemical formulas; and employing a logarithmic normalization algorithm to enhance the influence of trace doping elements.

[0107] Preferably, the doping ratio m of each element in the chemical formula is first determined according to formula (1). i (i = 1, 2, 3…) are normalized to obtain the normalized doping ratio coefficient m / i (i=1,2,3…), and then according to formula (2), the proportionality coefficient is logarithmically transformed to obtain the logarithmically normalized doping ratio m′. i (i = 1, 2, 3…);

[0108] Principal component analysis (PCA) and neural network-based encoders can be used to reduce the dimensionality of the original data, or other algorithms can be employed. Optimizing data dimensionality processing can reduce the dimensionality of the original data and improve the feature representation of the data.

[0109] (6) Based on the set temperature range and step size, generate each temperature point in batches to form a temperature list, and calculate the total number of calculations N2 required for each chemical formula at different temperature points using the following formula:

[0110]

[0111] The temperature is normalized by Min-Max, and the normalized temperature list T′ (i=1,2,3…) is obtained according to formula (5).

[0112] (7) Combine the dimension-reduced chemical formula numerical representation and the normalized temperature list into a new numerical list as input data.

[0113] The input data is fed into the pre-trained crystal phase transformation artificial intelligence model one by one. The model can either display the output results one by one or display all the results after execution. The output of the pre-trained crystal phase transformation artificial intelligence model is a numerical representation of the crystal structure type, which is converted into a string of crystal structure by a crystal structure type converter.

[0114] (8) Calculate the total number of times N, the artificial intelligence model for crystal phase transformation, needs to be invoked, according to the following formula.

[0115] N = N1 × N2 (11)

[0116] Within a range of N times, the artificial intelligence model for crystal phase transformation is called one by one, and the input data is read one by one, with the corresponding output being the crystal structure at different doping ratios and temperatures.

[0117] (9) After executing N times, based on the output results, draw the composition-temperature phase diagram of the current material under different doping ratios and different temperature conditions.

[0118] This embodiment verifies the experiment using the method described above. The crystal phase transition AI model used in the experiment is a multilayer perceptron model for classification tasks. Seven fully connected layers are set up, serving as one input layer, five hidden layers, and one output layer. The data received at the input layer is the processed form of the original data. Following the method described in this embodiment, after reading the chemical formula, the chemical formula is normalized, embedded, and dimensionality reduced to obtain a 20-dimensional chemical formula embedding vector and a normalized 1-dimensional temperature vector. In this embodiment, the crystal phase transition dataset contains more than 10,000 data entries. Some chemical formulas in the crystal phase transition dataset are processed as follows... Figure 3 As shown.

[0119] By merging the 20-dimensional chemical formula embedding vector Q and the normalized 1-dimensional temperature vector, the input vector in the input layer becomes 21-dimensional, in the form of:

[0120] Q = [Q1 Q2 …Q] 20 temp] T (12)

[0121] In the first hidden layer, the ReLU activation function is specified; therefore, the output vector h of the first layer is... 1 for

[0122] h 1 =ReLU(W (1) Q+b (1) (13)W (1) Let b be the weight matrix of the first layer. (1) This is the bias vector for the first layer. In the neurons of this layer, batch normalization and dropout operations are added to improve the model's generalization ability. Therefore, the output vector h... 1 Scaling and translation are performed using the following formulas:

[0123]

[0124] Where μ batch and These represent the mean and variance of the current batch, while γ, β, and ∈ are trainable parameters. The batch-normalized output vector is obtained through calculation. In the discard operation, the mask m is calculated.

[0125]

[0126] The output vector obtained after batch normalization and discarding operations is obtained. This vector is also the input data for the next layer of the neural network. In the subsequent second, third, fourth, and fifth hidden layers, the input vector is used to calculate the output vector using formulas (13) to (15). In the fifth hidden layer, the output vector is obtained as follows: The last layer of neurons is the output layer, calculated according to formulas (7) to (8). The output crystal structure is obtained.

[0127] By increasing the number of training rounds, dividing the training set and validation set for cross-validation, and continuously optimizing the weight matrix, bias vector and other hyperparameters in each layer of neurons, a crystal phase transition artificial intelligence model with an accuracy of more than 90% on both the training set and the validation set was finally obtained. Figure 4 This refers to the accuracy of the phase transformation artificial intelligence model built in this embodiment in the training and validation sets as training rounds progress.

[0128] Set the chemical formulas of the ferroelectric materials with doped variables to be used to generate the phase diagram. In the computer program, input the chemical formulas of three ferroelectric materials, as shown in Table 1. These three materials are:

[0129] (1) The material numbered 1 is a calcium zirconium doped barium titanate ferroelectric material:

[0130] The variable chemical formula of material No. 1 is set as "(1-x)Ba(Zr"). 0.2 Ti 0.8 O3-xBa 0.7 Ca 0.3 After standardization, TiO3 is obtained in the following form: Ba[1-0.3x]Zr[0.2-0.2x]Ti[0.8+0.2x]O[3]Ca[0.3x], where x is a variable to adjust the doping ratio.

[0131] The upper and lower limits of this doping ratio range are set to 0.2 and 0.5, respectively.

[0132] Set the scaling step size to 0.002.

[0133] The upper and lower limits of the temperature variation range are set to 120K and 440K, respectively.

[0134] Set the temperature change step size to 2K.

[0135] Table 1. Parameters for automatically generating three different material composition-temperature phase diagrams.

[0136]

[0137] A computer program is executed to generate a list of chemical formulas with different doping ratios in batches by reading the chemical formulas, doping ranges, and step size information with doping variables. This material contains 150 chemical formulas with different doping ratios. Based on the generated chemical formulas, the doping ratios are normalized using a logarithmic normalization method. Then, the elements and doping ratios are embedded using the periodic table. Principal component analysis (PCA) is used to reduce the dimensionality of the embedded representation of the chemical formulas, resulting in a 20-dimensional numerical representation of the chemical formulas.

[0138] A temperature list is generated, and the temperature list is normalized using the Min-Max normalization method.

[0139] The numerical representation of the 20-dimensional chemical formula and the temperature list after Min-Max normalization are combined as input data.

[0140] According to formulas (9) to (11), the total number of times the pre-trained crystal phase transformation artificial intelligence model needs to be called is 24,000. For each row of data in the input data, the crystal phase transformation artificial intelligence model is called one by one to output the crystal structure at different doping ratios and temperatures. After 24,000 executions, the processing of all input data is completed.

[0141] Based on the output results, composition-temperature phase diagrams were plotted for calcium zirconium-doped barium titanate ferroelectric materials in the doping ratio ranges of 0.2–0.5 and 120–440 K. The results are as follows. Figure 5 As shown. Comparing with the phase diagrams constructed experimentally by Acosta, Matias (2014) et al., it can be seen that the phase boundary change trends are consistent in the phase diagrams automatically generated in this embodiment of the invention.

[0142] (2) Material number 2 is tin-doped barium titanate:

[0143] The chemical formula of material No. 2 is set to "BaSn". x Ti 1-x After standardization, the chemical formula “O3” is converted into the following form: “Ba[1]Sn[x]Ti[1-x]O[3]”, where x is a variable for adjusting the doping ratio.

[0144] The upper and lower limits of this doping ratio range are set to 0 and 0.3, respectively.

[0145] Set the scaling step size to 0.002.

[0146] The upper and lower limits of the temperature variation range are set to 150K and 440K, respectively.

[0147] Set the temperature change step size to 2K.

[0148] Using the same processing method as for material No. 1, according to formulas (9) to (11), the total number of times the pre-trained artificial intelligence model for crystal phase transformation needs to be called is calculated to be 21750. After each execution is completed, the composition-temperature phase diagram of the tin-doped barium titanate ferroelectric material is plotted in the doping ratio range of 0-0.3 and 150-440K. The results are as follows. Figure 6 As shown. The phase boundary variation trend in this phase diagram is consistent with the experimental results of Liu, Wenfeng (2017) et al.

[0149] (3) The material numbered 3 is lead zirconate titanate:

[0150] The chemical formula of material No. 3 is set to "PbZr". 1-x Ti x After standardization, the chemical formula “O3” is converted into the following form: “Pb[1]Zr[1-x]Ti[x]O[3]”, where x is a variable for adjusting the doping ratio.

[0151] The upper and lower limits of this doping ratio range are set to 0.35 and 0.46, respectively.

[0152] Set the scaling step size to 0.001.

[0153] The upper and lower limits of the temperature variation range are set to 500K and 690K, respectively.

[0154] Set the temperature change step size to 2K.

[0155] Using the same processing method as for material No. 1, according to formulas (9) to (11), the total number of times the pre-trained artificial intelligence model for crystal phase transformation needs to be called is calculated to be 10450. After each execution is completed, the composition-temperature phase diagram of the material is plotted in the doping ratio range of 0.35 to 0.46 and 500 to 690 K, as shown below. Figure 7 As shown. The obtained trend of phase boundary changes is consistent with the experimental results published by A. Bouzid et al. (2003).

[0156] In this embodiment, the intelligent generation method for ferroelectric material composition-temperature phase diagrams described above was used to automatically generate composition-temperature phase diagrams for three different materials. Table 1 also lists that in this embodiment, the time required to construct the composition-temperature phase diagrams for the three different materials was less than 1000 seconds. Compared with the time consumed by traditional experimental methods (usually 1-2 years), the above method significantly shortens the research and development time and improves research and development efficiency. Using high-performance graphics cards and processors can further shorten the execution time of computer programs.

[0157] Example 2

[0158] This embodiment provides an artificial intelligence system for generating composition-temperature phase diagrams of ferroelectric materials, such as... Figure 2 As shown, it includes:

[0159] Pre-trained AI module: used to acquire a dataset of phase transformation of electrical materials, train a multilayer artificial neural network model, perform hyperparameter optimization and cross-validation during training, and obtain a pre-trained phase transformation AI model;

[0160] The phase diagram automatic generation module is used to read the chemical formula, doping ratio, and temperature variation range and step size of the doped variable ferroelectric material, perform standardization processing, cyclically call the pre-trained crystal phase transformation artificial intelligence model for processing, output the crystal structure type, and draw the composition-temperature phase diagram of the doped variable ferroelectric material under different doping ratios and different temperature conditions according to the crystal structure type. The chemical formula of the doped variable ferroelectric material includes chemical elements and corresponding doping ratios, and the doping variable is included in the doping ratio.

[0161] The rest are as in Example 1.

[0162] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0163] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention can be implemented using various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0164] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0165] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0166] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0167] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0168] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. An artificial intelligence method for generating composition-temperature phase diagrams of ferroelectric materials, characterized in that, Includes the following steps: Obtain the chemical formula of the doped ferroelectric material and set it as a string, wherein the chemical formula of the doped ferroelectric material includes chemical elements and corresponding doping ratios, and the doping variable is contained in the doping ratio. Based on the chemical formula of doped variable ferroelectric materials in string form, the range and step size of the doping ratio and temperature variation are set; Based on the set range and step size of the doping ratio, the chemical elements and doping ratios in the string-form chemical formula of the doped variable ferroelectric material are processed to obtain a list of chemical formula numerical vectors. Based on the set temperature variation range and step size, a temperature list is generated and normalized to obtain a normalized temperature list. The chemical formula numerical vector list and the normalized temperature list are merged and used as input data to a pre-trained crystal phase transformation artificial intelligence model, which outputs the crystal structure type. The crystal phase transformation artificial intelligence model includes an input layer, a hidden layer, and an output layer. The input layer is used to input the input data obtained by merging the chemical formula numerical vector list and the normalized temperature list. The hidden layer is used to extract deep-level implicit features from the input data. The output layer is used to perform classification and prediction based on the extracted deep-level implicit features and outputs the predicted crystal structure type. Based on the crystal structure type, the composition-temperature phase diagrams of the chemical formula of the doped variable ferroelectric material under different doping ratios and temperature conditions were plotted.

2. The artificial intelligence generation method for the composition-temperature phase diagram of ferroelectric materials according to claim 1, characterized in that, The chemical formula of the doped variable ferroelectric material in string form uses symbols to distinguish chemical elements and doping ratios, specifically: The chemical element is placed in the symbol, the doping ratio is placed on the side of the corresponding chemical element, or both the chemical element and the doping ratio are placed in different symbols, where the doping variable is included in the doping ratio.

3. The artificial intelligence generation method for the composition-temperature phase diagram of ferroelectric materials according to claim 1, characterized in that, The steps for obtaining the list of chemical formula numerical vectors include: Based on the string-format chemical formula of the doped variable ferroelectric material, chemical formulas with different doping ratios are generated in batches according to the set range and step size of the doping ratio, forming a list of chemical formulas with different doping ratios. The number of formulas generated in batches is: In the formula, The number of chemical formulas generated in batches. and The upper and lower limits of the doping ratio, respectively. Step size; Based on the list of chemical formulas with different doping ratios, the doping ratios were normalized using a logarithmic normalization method to obtain the logarithmically normalized doping ratios: By embedding the chemical elements and logarithmically normalized doping ratios of the chemical formulas in the periodic table, an initial list of numerical vectors for the chemical formulas is obtained. The expression for embedding is as follows: In the formula, For the first j An initialization formula numerical vector, For the first i The log-normalized doping ratio of each chemical element For the first j Includes Chemical formulas of the elements A list of elements in the periodic table. ; Based on the initial list of chemical formula numerical vectors, principal component analysis is used to reduce the dimensionality, resulting in the final list of chemical formula numerical vectors. The dimensionality reduction is expressed as follows: In the formula, This is the numerical vector of the chemical formula after dimensionality reduction. For the first j An initialization formula numerical vector, It is a low-dimensional basis matrix.

4. The artificial intelligence generation method for the composition-temperature phase diagram of ferroelectric materials according to claim 3, characterized in that, The step of obtaining the log-normalized doping ratio includes: Based on the list of chemical formulas with different doping ratios, the doping ratios are normalized, and the normalization expression is as follows: In the formula, For the first i The normalized doping ratio coefficient of each chemical element. For the first i The doping ratio of each chemical element; The normalized doping ratio is logarithmically transformed to obtain the logarithmically normalized doping ratio, which is expressed as follows: In the formula, For the first i The log-normalized doping ratio of each chemical element.

5. The artificial intelligence generation method for the composition-temperature phase diagram of ferroelectric materials according to claim 1, characterized in that, The step of obtaining the normalized temperature list includes: Based on the set temperature variation range and step size, each temperature point is generated in batches to form a temperature list. Each temperature point represents the total number of calculations for each chemical formula numerical vector in the chemical formula numerical vector list. The calculation expression is: In the formula, The total number of calculations for each chemical formula numerical vector at different temperature points. and These are the upper and lower limits of the temperature, respectively. Step size; Based on the temperature list, a normalized temperature list is obtained by normalizing the temperature using the maximum-minimum normalization algorithm. The expression for the maximum-minimum normalization algorithm is as follows: In the formula, For the first i A normalized temperature point, For the first i The original temperature, , These are the minimum and maximum temperatures, respectively.

6. The artificial intelligence generation method for the composition-temperature phase diagram of ferroelectric materials according to claim 1, characterized in that, The hidden layer has multiple layers, and each hidden layer adds batch normalization and discard operations. The execution steps of the hidden layer include: In the first hidden layer, the input data is processed to obtain the initial output vector of the first hidden layer. : In the formula, This is the weight matrix of the first hidden layer. For the input vector, This is the bias vector of the first hidden layer; The initial output vector of the first layer Perform batch normalization to obtain the initial output vector after batch normalization. : In the formula, , and These are training parameters. and These are the mean and variance of the current batch; Based on the batch-normalized initial output vector Perform a discard operation to obtain the final output vector of the first hidden layer, and use it as the input of the next hidden layer. Repeat the above steps until the processing of all hidden layers is completed.

7. The artificial intelligence generation method for the composition-temperature phase diagram of ferroelectric materials according to claim 6, characterized in that, The final output vector of the first hidden layer is represented as: In the formula, This is the final output vector of the first hidden layer. For masking.

8. The artificial intelligence generation method for the composition-temperature phase diagram of ferroelectric materials according to claim 1, characterized in that, The steps for determining the output crystal structure type include: Based on the list of chemical formula numerical vectors and the list of normalized temperatures, the total number of executions of the pre-trained AI model for crystal phase transformation is calculated. The expression for calculating the total number of executions is as follows: In the formula, For the total number of executions, This represents the number of chemical formula numerical vectors in the list of chemical formula numerical vectors, that is, the number of chemical formulas with different doping ratios. This represents the number of temperature points in the normalized temperature list, which is also the total number of calculations for each chemical formula's numerical vector at different temperature points. The chemical formula numerical vectors from the chemical formula numerical vector list and the normalized temperatures from the normalized temperature list are merged and used as input data for one execution. This data is then processed using the pre-trained crystal phase transformation AI model to output the crystal structure type for this execution. This step is repeated before further execution. N -1 times, to obtain the crystal structure type of all chemical formula numerical vectors at each temperature point.

9. An artificial intelligence generation system for composition-temperature phase diagrams of ferroelectric materials, characterized in that, The method for generating an artificial intelligence-based composition-temperature phase diagram of a ferroelectric material according to any one of claims 1-8 includes: Pre-trained AI module: used to acquire a dataset of phase transformation of electrical materials, train a multilayer artificial neural network model, perform hyperparameter optimization and cross-validation during training, and obtain a pre-trained phase transformation AI model; The phase diagram automatic generation module is used to read the chemical formula, doping ratio, and temperature variation range and step size of the doped variable ferroelectric material, perform standardization processing, cyclically call the pre-trained crystal phase transformation artificial intelligence model for processing, output the crystal structure type, and draw the composition-temperature phase diagram of the doped variable ferroelectric material under different doping ratios and different temperature conditions according to the crystal structure type. The chemical formula of the doped variable ferroelectric material includes chemical elements and corresponding doping ratios, and the doping variable is included in the doping ratio.

Images