阅读说明：本技术 一种利用机器学习辅助生长大尺寸lbo晶体的方法及其应用 (Method for growing large-size LBO crystal by machine learning assistance and application thereof ) 是由 胡章贵 邱海龙 齐小方 朱显超 吴以成 于 2020-05-22 设计创作，主要内容包括：本发明提供了一种利用机器学习辅助生长大尺寸LBO晶体的方法及其应用,所述方法包括以下步骤：步骤1,LBO单晶生长数据的采集与预处理,确定对应的影响因素和质量标签,建立数据集；步骤2,对步骤1得到的数据集中所有数据进行归一化处理,将处理后的数据集划分为训练集和测试集两部分；步骤3,利用训练集和机器学习算法,构造预测模型,并测试该预测模型在测试集上的可靠性,直至预测精度满足要求,得到对应的目标预测模型；步骤4,利用所述目标预测模型预测最优实验条件。本发明通过机器学习的分析与预测,探索合适的生长条件,减少试探性实验次数,降低LBO晶体生长成本,提高晶体质量。(The invention provides a method for growing a large-size LBO crystal by machine learning assistance and application thereof, wherein the method comprises the following steps: step 1, acquiring and preprocessing LBO single crystal growth data, determining corresponding influence factors and quality labels, and establishing a data set; step 2, performing normalization processing on all data in the data set obtained in the step 1, and dividing the processed data set into a training set and a test set; step 3, constructing a prediction model by utilizing a training set and a machine learning algorithm, and testing the reliability of the prediction model on a test set until the prediction precision meets the requirement to obtain a corresponding target prediction model; and 4, predicting optimal experimental conditions by using the target prediction model. According to the invention, through the analysis and prediction of machine learning, appropriate growth conditions are explored, the times of tentative experiments are reduced, the growth cost of the LBO crystal is reduced, and the crystal quality is improved.)

1. A method for growing a large-size LBO crystal with machine learning assistance, comprising the steps of:

step 1, acquiring and preprocessing LBO single crystal growth data, determining corresponding influence factors and quality labels, and establishing a data set;

step 2, performing normalization processing on all data in the data set obtained in the step 1, and dividing the processed data set into a training set and a test set;

step 3, constructing a prediction model by utilizing a training set and a machine learning algorithm, and testing the reliability of the prediction model on a test set until the prediction precision meets the requirement to obtain a corresponding target prediction model;

and 4, predicting optimal experimental conditions by using the target prediction model.

2. The method for growing LBO crystal with machine learning assistance as claimed in claim 1, wherein said method for growing LBO crystal with machine learning assistance further comprises the step of 5, cultivating LBO crystal or verifying the size of the cultivated LBO crystal is up to standard by using the optimum experimental conditions obtained in the experimental verification step 4.

3. The method for growing a large-sized LBO crystal with machine learning assistance as claimed in claim 1, wherein in step 1, the influencing factors are mainly the starting temperature and the ending temperature of the three temperature zones, the raw material ratio, the raw material concentration, the kind of the cosolvent or the reaction time.

4. The method for growing large-size LBO crystals using machine learning assistance as claimed in claim 1, wherein in said step 2, the ratio of said training set to said test set is calculated by convergence.

5. The method for growing large-size LBO crystals with machine learning assistance as claimed in claim 1, wherein in step 2, the division method of the training set and the test set is random division or Kennard-Stone division.

6. The method for growing a large-sized LBO crystal with machine learning assistance as claimed in claim 1, wherein in step 3, the machine learning algorithm is an artificial neural network algorithm.

7. The method for growing a large-sized LBO crystal with machine learning assistance as claimed in claim 1, wherein in said step 3, said evaluation index of reliability includes a correlation coefficient R and a root mean square error MSE.

8. The method for growing large-size LBO crystals with machine learning assistance as claimed in claim 7, wherein the correlation coefficient R is calculated as follows:

9. the method for growing large-size LBO crystals with machine learning assistance as claimed in claim 7 wherein the root mean square error MSE is as follows:

wherein n is the total number of samples, y_iAndfor the experimental and predicted values of the ith sample,is the average of all sample experimental values in the corresponding data set,the average of the predicted values for all samples in the corresponding data set.

10. Use of a method according to any one of claims 1 to 9 in LBO crystal cultivation or LBO crystal size verification.

Technical Field

The invention relates to the technical field of nonlinear optical crystal growth, in particular to a method for growing a large-size LBO (lithium triborate) crystal by machine learning assistance and application thereof.

Background

The nonlinear optical crystal is a laser frequency conversion core material, and has important application and indispensable functions in the fields of information, optical communication, national defense, national security and the like. Lithium triborate crystal (LBO) is an excellent nonlinear optical material. The crystal has a large enough nonlinear coefficient, can realize phase matching under a greenhouse, is not deliquescent, has stable chemical properties and moderate hardness, and is especially longer than other nonlinear crystals in terms of larger laser damage threshold, phase matching allowable angle and wide light transmission range. The method is generally applied to the field of solid laser frequency conversion. However, because the LBO single crystal growth has more influence factors and coupling is likely to exist, the difficulty in finding suitable conditions is high, and therefore, a laboratory is always required to spend a great deal of time and material cost to probe the suitable growth conditions of the single crystal material.

Disclosure of Invention

The invention aims to provide a method for growing a large-size LBO crystal by machine learning assistance, aiming at solving the problem that the exploration of conditions influencing the growth of the LBO single crystal is time-consuming and labor-consuming in the prior art.

It is another object of the invention to provide the use of said method in LBO crystal cultivation or LBO crystal size verification.

The technical scheme adopted for realizing the purpose of the invention is as follows:

a method for growing large-size LBO crystals with machine learning assistance, comprising the steps of:

step 1, acquiring and preprocessing LBO single crystal growth data, determining corresponding influence factors and quality labels, and establishing a data set;

step 2, performing normalization processing on all data in the data set obtained in the step 1, and dividing the processed data set into a training set and a test set;

step 3, constructing a prediction model by utilizing a training set and a machine learning algorithm, and testing the reliability of the prediction model on a test set until the prediction precision meets the requirement to obtain a corresponding target prediction model;

and 4, predicting optimal experimental conditions by using the target prediction model.

In the above technical solution, the method for growing a large-size LBO crystal with machine learning assistance further comprises a step 5 of verifying the optimal experimental conditions obtained in the step 4 by using an experiment to cultivate the LBO crystal or to verify whether the size of the grown LBO crystal reaches the standard.

In the above technical scheme, in the step 1, the influencing factors are mainly the starting temperature and the ending temperature of the three temperature zones, the raw material ratio, the raw material concentration, the cosolvent type or the reaction time.

In the above technical solution, in the step 2, the ratio of the training set to the test set is obtained by convergence calculation.

In the above technical solution, in the step 2, the method for dividing the training set and the test set is random division or Kennard-Stone division.

In the above technical solution, in the step 3, the machine learning algorithm is an artificial neural network algorithm.

In the above technical solution, in step 3, the evaluation index of reliability includes a correlation coefficient R and a root mean square error MSE, where the correlation coefficient R is calculated by the following formula:

the root mean square error MSE is as follows:

wherein n is the total number of samples, y_iAndfor the experimental and predicted values of the ith sample,is the average of all sample experimental values in the corresponding data set,the average of the predicted values for all samples in the corresponding data set.

In another aspect of the invention, the application of the method in LBO crystal cultivation or LBO crystal size verification is also included.

Compared with the prior art, the invention has the beneficial effects that:

1. the method analyzes the existing LBO crystal growth data of a laboratory through machine learning, predicts the successful growth possibility of the large-size and high-quality single crystal by utilizing the analysis result, and finds out the key factors of the LBO single crystal growth.

2. According to the invention, through the analysis and prediction of machine learning, appropriate growth conditions are explored, the times of tentative experiments are reduced, the growth cost of the LBO crystal is reduced, and the crystal quality is improved.

Drawings

FIG. 1 is a flow chart of the method of the present invention.

FIG. 2 is a distribution diagram of experimental and predicted crystal quality values for a test set in accordance with an embodiment of the present invention.

FIG. 3 is a graph of the artificial neural network classification recognition rate in accordance with an embodiment of the present invention.

In fig. 4, (a) is a picture of LBO single crystal obtained by the example of the present invention, and (b) is its weak absorption spectrum.

In FIG. 5, (a) shows a transmission spectrum of LBO single crystal obtained in an example of the present invention, and (b) shows a refractive index test.

Detailed Description

The invention is described in further detail below with reference to the figures and specific examples. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

Example 2

The research data of the embodiment mainly come from the experimental records of LBO single crystal growth by a fluxing agent method in a key laboratory of functional crystal materials in Tianjin, wherein the experimental records comprise the starting temperature and the ending temperature of three temperature zones of a crystal growth furnace, the growth time and crystal quality labels (the quality labels are 'qualified' and 'unqualified', the label that the crystal quality is more than 1Kg is 'qualified', and the labels that the rest are 'unqualified'). These data were derived from 652 samples, and 500 sets of data remained after removing invalid data (failed experiments due to power failure, operational errors, etc.) to form a data set.

The data were normalized according to their maximum and minimum values so that all data were in the (0, 1) range. In the embodiment, a random division method is used for dividing the whole data set into a training set and a test set, the proportion of the training set and the test set is determined by a convergence test, and the convergence test can refer to documents: qichong, chenqisong, zhanjuli, etc. a method for predicting strength parameters of filling materials by machine learning, china, CN 109523069 a, 2018.11.01; wang jian. feedforward neural network gradient learning algorithm convergence analysis [ D ]. university of great chain of workers, 2012.

In this example, the training set accounts for 80% of the total data set, and the test set accounts for 20% of the total data set.

And constructing a prediction model and model evaluation by using a training set through an artificial neural network algorithm. And testing the reliability of the prediction model on the test set until the prediction precision meets the requirement to obtain a corresponding target prediction model.

And adopting a correlation coefficient R and a root mean square error MSE as a standard for judging the prediction accuracy, wherein the calculation formula of the correlation coefficient R is as follows:

the root mean square error MSE is as follows:

wherein n is the total number of samples, y_iAndfor the experimental and predicted values of the ith sample,is the average of all sample experimental values in the corresponding data set,the average of the predicted values for all samples in the corresponding data set. The root mean square error reflects a measure of the degree of difference between the predicted value and the experimental value, and the lower the value, the closer to zero, the higher the prediction accuracy, and no fixed value. (values on the order of 10 are generally considered to be^-3I.e. meets the prediction accuracy requirements. ) The correlation coefficient is a quantity for researching the linear correlation degree between variables, the value of the correlation coefficient is between-1 and 1, the closer the value is to 0, the weaker the correlation degree is (equal to 0, linear independence), the more the value moves to 1 (positive correlation) or-1 (negative correlation), and the correlation is enhanced.

Growing LBO single crystal by a fluxing agent method, which relates to the cooperative change of a plurality of operation parameters, wherein an input layer selects the initial temperature and the ending temperature of three temperature zones of a crystal growing furnace and key operation parameters of growth time as input vectors; and adopting classification indexes representing the quality of the crystals, namely unqualified and qualified as output vectors, and realizing effective classification of the crystal quality by using an artificial neural network. The number of hidden layers and number of neurons in each layer of the neural network are determined using 10-fold Cross validation, particle swarm optimization and mean square error (see the documents R.R. Picard, R.D. Cook, Cross-validation of regression models, Journal of the American Statistical Association,79(1984) 575-583.). Finally, the structure of the neural network is determined to be 7-22-2, wherein the hidden layer is a single layer, and the number of the neurons is 22. The invention has 500 crystal growth data groups in total, 400 groups of data are randomly selected from the crystal growth data groups as training data to train the artificial neural network, and 100 groups of data are used as test data to test the classification capability of the network. The mean square error of the artificial neural network is calculated to be 0.00618, and the correlation coefficient of the crystal quality is 0.91, so that the artificial neural network model has good precision and reliability and can be used for predicting the crystal quality. In practical application, the number of hidden layers, the number of neurons and the discrimination criteria can be adjusted according to the change of the data set.

LBO crystal quality prediction: the example uses a trained neural network model to perform classification prediction on crystal quality in a test set. The ordinates 1 and 2 in FIG. 3 represent the crystal quality "off-spec" and "acceptable", respectively. The classification basis is as follows: the crystals with a mass greater than 1kg were marked as pass and the rest as fail. And the accuracy rate is the recognition rate, the predicted crystal quality is compared with the crystal quality obtained by the test centralized experiment, and the accuracy rate is obtained by dividing the total number of the samples which are predicted accurately by the total number of the test samples of the experiment. The distribution between the crystal quality predicted using the neural network on the test set and the experimental values is shown in fig. 2, and it can be seen that the predicted values and the experimental values agree well. The classification prediction accuracy of the two crystal qualities is calculated to be as high as 95.7% and 96.3%, as shown in figure 3. As can be seen from the artificial neural network classification prediction results, the quality classification algorithm based on the artificial neural network has higher accuracy, and the class to which the crystal quality belongs can be predicted quickly and accurately.

Optimal parameter verification experiment: and obtaining the converged optimal parameters through an optimal model (the specific method can refer to N.Dropka, M.Holana, Optimization of magnetic drive direct knowledge of silicon using specific neural networks and Gaussian process models, Journal of Crystal Growth,471(2017)53-61), randomly generating input values in the range of key operation parameters when the temperature change range of the Crystal Growth furnace temperature zone and the temperature change range of the Crystal Growth furnace temperature zone are 500-800 ℃, and the Crystal quality is predicted based on an artificial neural network, wherein the Crystal Growth process is carried out on the basis of the Growth parameters, and the Crystal quality is predicted according to the Crystal Growth furnace temperature range and the Growth time change range is 50-120 days. When the crystal growth quality is qualified, the corresponding parameters are the optimized parameters, and a group of crystal growth experiments are selected to verify the prediction effect of the model. The initial temperature is 725 ℃, the cooling rate is 0.2 ℃/day, the growth time is 65 days, the parameter is used as the experimental condition, and the LBO crystal is grown by a molten salt growth method. Obtaining LBO crystals with large size and high quality, wherein the mass of the LBO crystals reaches 3.870Kg, and the sizes are 240mm multiplied by 160mm multiplied by 110mm (shown in figure 4 (a)); the absorption of the crystal in both the c-axis and a-axis directions was less than 50ppm/cm (shown in FIG. 4 (b)), indicating that the light absorption of the crystal was small; as shown in FIG. 5 (a), the crystal transmittance is 250-1500 nm; the refractive index of the LBO crystal changes as shown in fig. 5 (b), when the crystal is irradiated by a 633nm light source, the overall refractive index of the crystal is relatively uniform (mostly in the blue region), and the refractive index of the edge part is changed (in the red and green regions), which indicates that the overall quality of the crystal is relatively good and only the edge quality is relatively poor. The model is proved to have a guiding function on LBO crystal growth. The key factors for obtaining the growth of the large-size LBO crystal by the embodiment are the initial temperature, the cooling rate and the growth time.

The foregoing is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, various modifications and decorations can be made without departing from the principle of the present invention, and these modifications and decorations should also be regarded as the protection scope of the present invention.