A method for predicting piezoelectric properties of potassium sodium niobate-based ceramics based on machine learning

Abstract

The present invention discloses a method for predicting the piezoelectric properties of potassium sodium niobate-based ceramics based on machine learning, which belongs to the field of ceramic material performance prediction and includes the following steps: S1, data collection; S2, dividing the data set into a training set and a validation set according to a set ratio; S3, building an automated machine learning model; S4, inputting the training set into the automated machine learning model for training; S5, model evaluation; S6, obtaining the best automated machine learning model, and using the best automated machine learning model for prediction; S7, inputting the new formula data of potassium sodium niobate-based ceramics into the verified best automated machine learning model to predict piezoelectric properties. The present invention adopts the above-mentioned method for predicting the piezoelectric properties of potassium sodium niobate-based ceramics based on machine learning, constructs an alternative model through training data to predict materials with target properties, realizes efficient and accurate new material research technology, and promotes the application of machine learning in interdisciplinary research.

Description

Technical Field

[0001] The present invention relates to the technical field of ceramic material performance prediction, and in particular to a method for predicting the piezoelectric performance of potassium sodium niobate-based ceramics based on machine learning. Background Art

[0002] Piezoelectric ceramics are a type of functional ceramic material that can convert mechanical energy into electrical energy, so piezoelectric ceramics play an important role in scientific research. Therefore, they are widely used in various electronic devices, such as sensors, actuators, and buzzers.

[0003] Among them, lead-free piezoelectric ceramics are usually composed of components such as barium carbonate, sodium bismuth titanate and sodium potassium niobate. Since they do not contain volatile lead, they have better sustainability and are therefore of great concern for their environmental protection. This has led to the widespread use of devices based on lead-free piezoelectric ceramics in functional fields such as clinical treatment, ultrasonic technology and filtration systems.

[0004] Among lead-free piezoelectric ceramics, potassium sodium niobate ceramics are widely recognized for their excellent electrical properties, which are closely related to their manufacturing process. However, the following challenges still exist in the powder synthesis stage:

[0005] 1. Forming soft-hard agglomerates, etc.

[0006] 2. Although the forming process is simple, it lacks precision and requires secondary processing;

[0007] 3. The sintering temperature is high and the components are highly volatile, so the preparation method needs to be further improved. Summary of the invention

[0008] To solve the above problems, the present invention provides a method for predicting the piezoelectric properties of potassium sodium niobate-based ceramics based on machine learning. An alternative model is constructed through training data to predict materials with target properties, thereby realizing efficient and accurate new material research technology and promoting the application of machine learning in interdisciplinary research.

[0009] To achieve the above object, the present invention provides a method for predicting the piezoelectric properties of potassium sodium niobate-based ceramics based on machine learning, comprising the following steps:

[0010] S1. Data collection: Collect the experimental data of potassium sodium niobate-based ceramics that have been completed, and make a data set based on the collected experimental data of potassium sodium niobate-based ceramics;

[0011] S2, divide the data set into a training set and a validation set according to a set ratio, and preprocess the training set;

[0012] S3. Build an automated machine learning platform;

[0013] S4. Input the training set into the automated machine learning model for training, and save the trained automated machine learning model;

[0014] S5, model evaluation: input the validation set into the best automated machine learning model to determine whether the evaluation index meets the set requirements. If so, output the best automated machine learning model, otherwise return to step S4;

[0015] S6. Obtain the best automated machine learning model by calling the train() function in H2O AutoML and setting the maximum number of automated machine learning models and the random seed, use the best automated machine learning model to make predictions, and convert the piezoelectric performance prediction results into NumPy arrays;

[0016] S7. Input the new formula data of potassium sodium niobate-based ceramics into the verified optimal automated machine learning model to predict the piezoelectric performance and output the prediction results.

[0017] Preferably, the potassium sodium niobate-based ceramic experimental data described in step S1 includes comprehensive information on the elemental composition of potassium sodium niobate-based ceramics, and is distributed and calculated based on the potassium sodium niobate ceramic formula according to its molar proportion: the coefficient corresponding to each element is multiplied by the scalar value to obtain the molar proportion of each element, and the data are collected one by one to obtain a data set.

[0018] Preferably, in step S2, the ratio is set to 8:2.

[0019] Preferably, the evaluation index described in step S5 includes the root mean square error and the coefficient of determination , and the calculation formula is as follows:

[0020] ;

[0021] In the formula, represents the number of samples in the training set, Indicates the measured value, is the predicted value, Indicates the average of the measured values.

[0022] Preferably, in step S6, the fitting curve of the best automated machine learning model is output according to the coefficient of determination The feasibility of subsequent prediction of the size judgment data is determined. If feasible, the best model is used to predict the test data, and the prediction results are converted into NumPy arrays. At the same time, the feature column names are defined according to the feature names and data formats.

[0023] Preferably, step S7 specifically includes the following steps:

[0024] S71. Create a new NumPy array as prediction data;

[0025] S72, input the formula data of each element of the potassium sodium niobate-based ceramic to be predicted into the new NumPy array, and input the predicted element ratio independent variable The proportion of

[0026] S73, iterate 10 times for loop, in each iteration, the potassium sodium niobate-based ceramic ratio independent variable The value is increased by 0.01, and in this loop, the updated value is used Generate prediction data samples;

[0027] S74. Use the best automated machine learning model to predict the generated prediction data samples and output the prediction results.

[0028] The present invention has the following beneficial effects:

[0029] 1. Build the best machine learning model through training data (which is a surrogate model, which is an inference model. It is fitted to the data and the parameters are optimized through a cross-validation process. The model is applied to test data that is not used in model construction to evaluate the performance of the model and complete the prediction) to predict materials with target properties, realize efficient and accurate new material research technology, and promote the application of machine learning in interdisciplinary research;

[0030] 2. Apply efficient machine learning computing and prediction capabilities to the research of piezoelectric ceramics, use existing research results and data to identify patterns from observed data samples, and use these patterns to predict future or unobservable data, shortening the development cycle of lead-free potassium sodium niobate piezoelectric ceramics;

[0031] 3. Extract the characteristic value of the molar proportion of each element from the potassium sodium niobate ceramic formula. At the same time, by designing the relationship between the element combination and coefficient proportion of the input formula, the piezoelectric performance results can be directly output, thereby achieving the effect of efficient verification of new materials.

[0032] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is a flow chart of a method for predicting piezoelectric properties of potassium sodium niobate-based ceramics based on machine learning of the present invention;

[0034] Figure 2 A model framework diagram of a method for predicting piezoelectric properties of potassium sodium niobate-based ceramics based on machine learning according to the present invention;

[0035] Figure 3 This is a comparison chart of multi-model evaluation results of simulation experiments;

[0036] Figure 4 It is the fitting diagram of the simulation experiment of the present invention;

[0037] Figure 5 is (0.92-x)K of the simulation experiment of the present invention 0.48 Na 0.52 Comparison of the predicted and experimental results of the piezoelectric coefficients of the NbO3-xLiTaO3-0.08AgSbO3 ceramic components. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical scheme and advantages disclosed in the embodiments of the present invention clearer, the embodiments of the present invention are further described in detail in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the embodiments of the present invention and are not intended to limit the embodiments of the present invention. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application. Examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions.

[0039] It should be noted that the terms "include", "have" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products or devices.

[0040] Like reference numerals and letters denote similar items in the following drawings, and thus, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0041] like Figure 1 and Figure 2 As shown, a method for predicting the piezoelectric properties of potassium sodium niobate-based ceramics based on machine learning includes the following steps:

[0042] S1. Data collection: Collect the experimental data of potassium sodium niobate-based ceramics that have been completed, and make a data set based on the collected experimental data of potassium sodium niobate-based ceramics;

[0043] The potassium sodium niobate-based ceramic experimental data described in step S1 includes comprehensive information on the elemental composition of potassium sodium niobate-based ceramics, and is distributed and calculated based on the potassium sodium niobate ceramic formula according to its molar proportion: the coefficient corresponding to each element is multiplied by the scalar amount to obtain the molar proportion of each element, and the data are collected one by one to obtain a data set.

[0044] S2. Divide the data set into a training set and a validation set according to a set ratio, and preprocess the training set. The preprocessing process of this embodiment is as follows: Divide the original data set into a training set and a validation set. A common division ratio is to use a portion (80%) of the data set as a training set and the remaining portion (20%) as a validation set. Perform feature processing on the training set, including feature selection, feature extraction, or feature conversion;

[0045] S3. Build an automated machine learning platform. In this embodiment, the automated machine learning platform is built as follows: import the h2o library (which is an open source distributed machine learning platform for building and deploying machine learning models), import the H2OAutoML class from the h2o library (which provides automated machine learning functions), and use the H2OAutoML class to search and select the best model.

[0046] S4. Input the training set into the automated machine learning model for training, and save the trained automated machine learning model;

[0047] S5, model evaluation: input the validation set into the best automated machine learning model to determine whether the evaluation index meets the set requirements. If so, output the best automated machine learning model, otherwise return to step S4;

[0048] The evaluation index described in step S5 includes the root mean square error and the coefficient of determination , and the calculation formula is as follows:

[0049] ;

[0050] In the formula, represents the number of samples in the training set, Indicates the measured value, is the predicted value, Indicates the average of the measured values.

[0051] S6. Obtain the best automated machine learning model by calling the train() function in H2O AutoML and setting the maximum number of automated machine learning models and the random seed, use the best automated machine learning model to make predictions, and convert the piezoelectric performance prediction results into NumPy arrays;

[0052] The conversion method of this embodiment is as follows: use best_model to predict the feature column feature_cols in the test data set test_h2o, and save the prediction results in predictions; then extract the prediction results from predictions, convert them into numpy array form, and save them in y_pred; and assume that the column name of the prediction result is 'predict'. Finally, extract the value of the target variable 'd33 (pC / N)' from the test data set test_df, convert it into numpy array form, and save it in y_test.

[0053] In this embodiment, the maximum number is set to 6 and the random seed is set to 42;

[0054] In step S6, the fitting curve of the best automated machine learning model is output, according to the coefficient of determination The feasibility of subsequent prediction of size determination data, usually When the data feasibility is relatively effective, the optimal model is used to predict the test data if feasible, and the prediction results are converted into NumPy arrays. At the same time, the feature column names are defined according to the feature names and data formats.

[0055] S7. Input the new formula data of potassium sodium niobate-based ceramics into the verified optimal automated machine learning model to predict the piezoelectric performance and output the prediction results.

[0056] Step S7 specifically includes the following steps:

[0057] S71. Create a new NumPy array as prediction data;

[0058] S72, input the formula data of each element of the potassium sodium niobate-based ceramic to be predicted into the new NumPy array, and input the predicted element ratio independent variable The proportion of

[0059] S73, iterate 10 times for loop, in each iteration, the potassium sodium niobate-based ceramic ratio independent variable The value is increased by 0.01, and in this loop, the updated value is used Generate prediction data samples;

[0060] S74. Use the best automated machine learning model to predict the generated prediction data samples and output the prediction results.

[0061] Simulation experiment

[0062] like Figure 3-Figure 5 As shown, for example, a conventional solid phase reaction was used to synthesize (0.92- x )K0.48 Na 0.52 NbO3- x LiTaO3-0.08AgSbO3 ceramic material. Raw materials including K2CO3 (99%), Na2CO3 (99.8%), Nb2O5 (99.5%), Sb2O3 (99.99%), Li2CO3 (98%), Ag2O (99.7%) and Ta2O5 (99.99%) were accurately weighed; the powder was then ball-milled with ethanol as a dispersant for 24 hours; then, after drying, the ball-milled powder was calcined at 850°C for 6 hours; then, using 8wt% polyvinyl alcohol (PVA) as a binder, the powder was pressed into a disc with a diameter of 10 mm and a thickness of 1 mm at a pressure of 10MPa; after burning off the PVA, the disc was sintered at 1080-1110°C for 3 hours. To measure the electrical properties, the sintered samples were coated with silver paste and then heated at 600°C for 10 min. Finally, the samples were polarized in a silicone oil bath at 30 kV / cm at room temperature. The piezoelectric properties were then measured to be 100-300 (PC / N).

[0063] Therefore, the present invention adopts the above-mentioned potassium sodium niobate-based ceramic piezoelectric performance prediction method based on machine learning, constructs an alternative model through training data to predict materials with target properties, realizes efficient and accurate new material research technology, and promotes the application of machine learning in interdisciplinary research.

[0064] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for predicting the piezoelectric properties of potassium sodium niobate-based ceramics based on machine learning, characterized in that: The following steps are involved: S1. Data collection: Collect the experimental data of potassium sodium niobate-based ceramics that have been completed, and make a data set based on the collected experimental data of potassium sodium niobate-based ceramics; The potassium sodium niobate-based ceramic experimental data described in step S1 includes comprehensive information on the element composition of the potassium sodium niobate-based ceramic, and is distributed and calculated based on the potassium sodium niobate ceramic formula according to its molar proportion: the coefficient corresponding to each element is multiplied by the angular scalar to obtain the molar proportion of each element, and the data are collected one by one to obtain a data set; S2, divide the data set into a training set and a validation set according to a set ratio, and preprocess the training set; S3. Build an automated machine learning platform; S4. Input the training set into the automated machine learning model for training, and save the trained automated machine learning model; S5, model evaluation: input the validation set into the best automated machine learning model to determine whether the evaluation index meets the set requirements. If so, output the best automated machine learning model, otherwise return to step S4; S6. Obtain the best automated machine learning model by calling the train() function in H2O AutoML and setting the maximum number of automated machine learning models and the random seed, and use the best automated machine learning model for prediction; S7, inputting the new formula data of potassium sodium niobate-based ceramics into the verified optimal automated machine learning model to predict the piezoelectric performance and output the prediction results; Step S7 specifically includes the following steps: S71. Create a new NumPy array as prediction data; S72, input the formula data of each element of the potassium sodium niobate-based ceramic to be predicted into the new NumPy array, and input the predicted element ratio independent variable The proportion of S73, iterate 10 times for loop, in each iteration, the potassium sodium niobate-based ceramic ratio independent variable The value is increased by 0.01, and in this loop, the updated value is used Generate prediction data samples; S74. Use the best automated machine learning model to predict the generated prediction data samples and output the prediction results.

2. The method for predicting piezoelectric properties of potassium sodium niobate-based ceramics based on machine learning according to claim 1, characterized in that: In step S2, the ratio is set to 8:

2.

3. The method for predicting piezoelectric properties of potassium sodium niobate-based ceramics based on machine learning according to claim 1, characterized in that: The evaluation index described in step S5 includes the root mean square error and the coefficient of determination , and the calculation formula is as follows: ； In the formula, represents the number of samples in the training set, Indicates the measured value, is the predicted value, Indicates the average of the measured values.

4. The method for predicting piezoelectric properties of potassium sodium niobate-based ceramics based on machine learning according to claim 3, characterized in that: In step S6, the fitting curve of the best automated machine learning model is output, according to the coefficient of determination The size determines the feasibility of subsequent prediction of the data. If feasible, the best model is used to predict the test data, and the prediction results are converted into NumPy arrays. At the same time, the feature column names are defined according to the feature names and data formats.

Images