Prediction method for damage growth characteristic of fused quartz element under laser irradiation

Abstract

A fused quartz element damage growth characteristic prediction method based on deep learning comprises the following steps: using a pre-prepared damage point array, using a microscopic imaging system to shoot the area of each damage point of the damage point array to calculate the equivalent diameter of the damage point, covering the damage point array with a laser spot, and calculating the damage growth characteristic of the fused quartz element; and recording laser irradiation parameter energy, pulse width and local flux of a damage point. And then shooting the damage array after laser irradiation by using a microscopic imaging system, and calculating the equivalent diameter of each damage point. And finally, sorting the data, substituting the data into a deep learning decision tree algorithm, training a fused quartz damage growth model, and determining an optimal model according to a model damage function. According to the method, the damage growth rule is learned and summarized from the fused quartz damage growth data through the decision tree algorithm, the method is used for analyzing and predicting the damage growth characteristics, few training samples are needed, and the damage growth characteristics of the damage points can be analyzed and predicted in batches.

Description

Technical Field

[0001] The present invention relates to the technical field of optical component damage prediction, and in particular to a method for predicting damage growth characteristics of a fused quartz component based on a deep learning decision tree algorithm. Background Art

[0002] As an optical material with excellent performance, fused quartz is widely used in various lasers, especially in the ultraviolet band. However, due to the increase in photon energy in the ultraviolet band, initial damage will occur on the front and back surfaces of fused quartz optics, and a large number of initial damage points will experience damage growth under subsequent laser irradiation.

[0003] The initial damage characteristics and damage growth characteristics of fused quartz are two main research focuses, among which the accurate prediction of the damage growth characteristics of fused quartz components is of great significance for the stable operation of laser devices. At present, the damage growth characteristics of the surface of fused quartz components are affected by many factors, and the damage growth characteristics cannot be directly measured, including the following: the internal chemical composition and material structure of a single damage point are different, resulting in different damage growth characteristics; multiple laser irradiations at the damage point show exponential growth and linear growth, and even a mixture of linear growth and exponential growth; the size of each damage point has different effects on the damage growth threshold, making it impossible to predict the probability of damage growth at the damage point; the internal composition, material structure, laser energy absorption factor and other factors of the damage point are difficult to quantify, and each damage point is subject to different conditions such as laser energy, pulse width, local flux, etc., which affect the damage growth. The existing damage growth characteristics are based on a statistical probability model (see prior art 1, Probability of growth of small damage sites on the exit surface of fused silica optics, 2012 Vol. 20, No. 12 OPTICS EXPRESS), which predicts the damage growth characteristics by observing the growth behavior of a large number of damage points under ideal conditions and fitting a mathematical model based on statistical data.

[0004] However, the statistical probability model is difficult to accurately reflect the growth of each damage point under actual conditions; it is also difficult to adapt to the batch prediction of damage characteristics with a large number of damage points, a large range of damage sizes, and uneven distribution of damage point sizes. Therefore, it is necessary to find a more effective, more accurate, and more universal method to predict the surface damage growth characteristics of fused silica components. Summary of the invention

[0005] In order to overcome the above-mentioned deficiencies of the prior art, the present invention proposes a method for predicting the damage growth characteristics of a fused quartz component. A damage growth prediction model is trained using a small sample data set, which can realize the prediction of the growth characteristics of damage points on the surface of a fused quartz component.

[0006] The technical solution of the present invention is as follows:

[0007] A method for predicting damage growth characteristics of a fused quartz component is characterized in that it comprises the following steps:

[0008] Step S1. A method for obtaining sample data of damage growth characteristics of a fused silica component;

[0009] Step S1.1: converge the laser beam and irradiate the fused quartz element, and prepare a damage point array by adjusting the laser energy, each array containing at least one marking point, and number the damage points that are not marking points;

[0010] Step S1.2: Use a microscopic imaging system to record the initial equivalent area S of the array damage point i , and calculate the initial equivalent diameter D of the damage point i , the formula is as follows:

[0011]

[0012] Where i = 1, 2, 3, ..., n, i represents the i-th damage point;

[0013] Step S1.3: Adjust the position of the fused quartz element, cover the prepared damage point array with a centimeter-level flat-top laser spot, and record the laser irradiation parameters, including laser energy, pulse width, near-field light intensity distribution of the laser spot, etc. According to the relative spatial position relationship between the damage point and the light intensity distribution, combined with the laser energy, calculate the local laser irradiation flux of each damage point;

[0014] Step S1.4 Use the microscopic imaging system to record the equivalent area S of each damage point in the array after irradiation. i ', calculate the equivalent diameter D after the damage point grows i ';

[0015]

[0016] Step S1.5: taking the initial equivalent diameter, laser energy, pulse width, local laser irradiation flux, etc. of each damage point in the array as prediction input features, and taking the equivalent diameter of each damage point after growth as prediction output feature;

[0017] Step S1.6: Repeat S1.1-S1.5 until sufficient damage growth characteristic data are collected;

[0018] Step S2: Construction and training of a prediction model for damage growth characteristics of fused quartz components. The model is constructed using a decision tree algorithm. The measured damage point growth characteristics include but are not limited to the equivalent diameter of the initial damage point, laser energy, pulse width, and local laser irradiation flux as characteristic quantities. The equivalent diameter of the damage point after actual growth obtained by measurement is used as a hyperparameter in the true value training decision tree model.

[0019] Step S3: inputting actual data such as the equivalent diameter of the initial damage point of the fused silica component to be tested, laser energy, pulse width, and local laser irradiation flux into the trained fused silica component damage growth characteristic prediction model to obtain the equivalent diameter prediction value after the damage point grows, that is, the fused silica component damage growth characteristic prediction result.

[0020] The step S2 of constructing and training the damage growth characteristic prediction model of the fused silica component specifically includes:

[0021] Step S2.1: constructing a prediction model for damage growth characteristics of fused silica components based on a deep learning decision tree algorithm, wherein the nodes of the decision tree are divided into branches and leaves according to the information entropy gain of the damage growth characteristics;

[0022] Step S2.2: m damage points (m<n) related data among n damage point sample data are used to train the decision tree algorithm model, wherein the initial damage point equivalent diameter, laser energy, laser pulse width, damage point local flux, etc. are used as prediction input features, and the equivalent diameter after the damage point grows is used as the prediction output feature;

[0023] Step S2.3: Establish the loss function:

[0024]

[0025] Where m is the total number of damage points used to train the model; Y i is the true value of the equivalent diameter after the i-th damage point grows; y i For the i Prediction value of equivalent diameter of damage point;

[0026]

[0027] Step S2.4: According to the loss function MAE and R 2 The score evaluation and prediction accuracy of the evaluation model are used to obtain the candidate model, where the lower the MAE value, the higher the R 2 The closer it is to 1, the higher the prediction accuracy; It is the average value of the predicted equivalent diameters of multiple damage points;

[0028] Step S2.5: Use the relevant data of nm damage points to verify the model to prevent overfitting;

[0029] Step S2.6: Continuously adjust the decision tree algorithm model hyperparameter - maximum depth Max_depth. When Max_depth = K, the damage function is minimized. When Max_depth is greater than K, the accuracy of the training data is improved but the accuracy of the verification data is reduced. Therefore, Max_depth = K is determined to be the optimal model and the model is saved.

[0030] Compared with the prior art, the method of the present invention has the following advantages:

[0031] 1. The present invention utilizes a deep learning decision tree algorithm to predict the growth characteristics of initial damage points in batches through characteristics such as the initial scale of the damage point, the local flux of the damage point, the laser energy, and the laser pulse width, with high prediction capability and accuracy.

[0032] 2. The model of the present invention requires less sample data, and the method is simple and easy to expand, which is suitable for small sample data. It can accurately predict the damage growth characteristics of batch fused quartz components under different energies, different initial damage sizes and other conditions; BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 The present invention is a flowchart of the method for predicting damage growth characteristics of fused quartz components.

[0034] Figure 2 Preparation diagram of damage point array, where a is the prepared damage point array and b is the growth result of the damage point array after subsequent laser irradiation.

[0035] Figure 3 Decision tree prediction model hyperparameter Max_depth=3 structure diagram.

[0036] Figure 4 The decision tree damage function varies with the hyperparameter Max_depth. DETAILED DESCRIPTION

[0037] The present invention will be further described below in conjunction with embodiments and drawings, but the protection scope of the present invention should not be limited by these embodiments.

[0038] A method for predicting damage growth characteristics of a fused quartz component comprises the following steps: Step S1. A method for obtaining sample data of damage growth characteristics of a fused quartz component;

[0039] Step S1.1: Converge the laser beam to 1 mm 2 Then, the fused quartz element was irradiated, and a damage point array was prepared by adjusting the laser energy: the damage points were spaced 1.6 mm horizontally and 2 mm vertically, and a 4×5 damage point array was prepared. Each array contained 20 damage points, of which 2 damage points were used as marking points, and 18 damage points were numbered as non-marking points.

[0040] Step S1.2: Use a microscopic imaging system to record the initial equivalent area S of the 18 damage points in the array i , and calculate the initial equivalent diameter D of the damage point i , the formula is as follows:

[0041]

[0042] Where i = 1, 2, 3, ..., 18, i represents the i-th damage point;

[0043] Step S1.3: Adjust the position of the fused quartz element, cover the prepared damage point array with a flat-top laser spot with a diameter of 8 mm, and record the laser irradiation parameters, including laser energy, pulse width, near-field light intensity distribution of the laser spot, etc. According to the relative spatial position relationship between the damage point and the light intensity distribution, combined with the laser energy, calculate the local laser irradiation flux of each damage point;

[0044] Step S1.4 Use the microscopic imaging system to record the equivalent area S of each damage point in the array after irradiation. i ', calculate the equivalent diameter D after the damage point grows i ';

[0045]

[0046] Step S1.5: taking the initial equivalent diameter, laser energy, pulse width, local laser irradiation flux, etc. of each damage point in the array as prediction input features, and taking the equivalent diameter of each damage point after growth as prediction output feature;

[0047] Step S1.6: Repeat S1.1-S1.5 until the number of damage growth characteristic data sets collected is 400;

[0048] Step S2: Construction and training of a prediction model for damage growth characteristics of fused quartz components. The model is constructed using a decision tree algorithm. The measured damage point growth characteristics include but are not limited to the equivalent diameter of the initial damage point, laser energy, pulse width, and local laser irradiation flux as characteristic quantities. The equivalent diameter of the damage point after actual growth obtained by measurement is used as a hyperparameter in the true value training decision tree model.

[0049] The construction and training of the damage growth characteristics prediction model of fused silica components include:

[0050] Step S2.1: constructing a prediction model for damage growth characteristics of fused silica components based on a deep learning decision tree algorithm, wherein the nodes of the decision tree are divided into branches and leaves according to the information entropy gain of the damage growth characteristics;

[0051] Step S2.2: The relevant data of 320 damage points (m<n) among the 400 damage point sample data are used to train the decision tree algorithm model, wherein the initial damage point equivalent diameter, laser energy, laser pulse width, damage point local flux, etc. are used as prediction input features, and the equivalent diameter after the damage point grows is used as the prediction output feature;

[0052] Step S2.3: Establish the loss function:

[0053]

[0054] Where m is the total number of damage points used to train the model; Y i is the true value of the equivalent diameter after the i-th damage point grows; y i is the predicted value of the equivalent diameter of the i-th damage point;

[0055]

[0056] Step S2.4: According to the loss function MAE and R 2 The score evaluation and prediction accuracy of the evaluation model are used to obtain the candidate model, where the lower the MAE value, the higher the R 2 The closer it is to 1, the higher the prediction accuracy; It is the average value of the predicted equivalent diameter of 320 damage points;

[0057] Step S2.5: The relevant data of 80 damage points are used to verify the model to prevent overfitting;

[0058] Step S2.6: Continuously adjust the decision tree algorithm model hyperparameters - maximum depth Max_depth, such as Figure 4 As shown in Figure 2, when the maximum depth of the decision tree increases, R 2 _train keeps increasing, but R 2 _test increases first and then decreases, and its inflection point is at Max_depth = 6; MAE_train keeps decreasing, MAE_test decreases first and then increases, and its inflection point is at Max_depth = 6; therefore, Max_depth = 6 is determined to be the optimal solution, and the model is saved.

[0059] Step S3: inputting actual data such as the equivalent diameter of the initial damage point of the fused silica component to be tested, laser energy, pulse width, and local laser irradiation flux into the trained fused silica component damage growth characteristic prediction model to obtain the equivalent diameter prediction value after the damage point grows, that is, the fused silica component damage growth characteristic prediction result.

[0060] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for predicting damage growth characteristics of fused quartz components, characterized in that: The steps include: Step S1. Obtain sample data of damage growth characteristics of fused silica components: S1.1 A damage point array of a fused silica element is irradiated by laser, each damage point array comprising a plurality of damage points; S1.2 using a microscopic imaging system to record the area of ​​each damage point in the damage point array, and calculating the initial equivalent diameter of each damage point; S1.3 Adjust the position of the fused quartz element so that the laser spot covers the entire damage point array, record the laser irradiation parameters, including laser energy, pulse width, and near-field light intensity distribution of the laser spot, and calculate the local laser irradiation flux of each damage point based on the relative spatial position relationship between the damage point and the light intensity distribution; S1.4 again using the microscopic imaging system to record the equivalent area of ​​each damage point in the damage point array after irradiation, and calculate the equivalent diameter of each damage point after growth; Step S2. Input the actual initial damage point equivalent diameter, laser energy, pulse width and local laser irradiation flux of the fused quartz component to be tested into the pre-trained fused quartz component damage growth characteristic prediction model, and the output is the equivalent diameter prediction value after the damage point grows, that is, the prediction result of the damage growth characteristic of the fused quartz component.

2. The method for predicting damage growth characteristics of fused silica components according to claim 1, characterized in that: The damage growth characteristic prediction model of the fused silica component in step S2 is constructed and trained using a deep learning decision tree algorithm, specifically including: Step S2.1: A deep learning decision tree algorithm is used to construct a prediction model for damage growth characteristics of fused silica components, wherein the nodes of the decision tree are divided according to the information entropy gain of the damage growth characteristics; Step S2.2: m damage point data (m<n) among n damage point sample data are used to train the damage growth characteristic prediction model of fused silica components, wherein the initial damage point equivalent diameter, laser energy, laser pulse width, damage point local flux, etc. are used as prediction input features, and the equivalent diameter after the damage point grows is used as the prediction output feature; Step S2.3: Establish a loss function to evaluate the difference between the model prediction value and the actual value, and use it as the goal of model optimization; Step S2.4: By continuously iterating and adjusting the model parameters, including the depth of the decision tree and the size of the leaf nodes, the prediction accuracy of the model is evaluated using the loss function, and the model with the highest prediction accuracy is selected as the candidate model; Step S2.5: The data of the remaining nm damage points are used to verify the generalization ability of the damage growth characteristic prediction model of the fused quartz component to prevent overfitting; Step S2.6: Further adjust the hyperparameters of the decision tree, including the maximum depth of the decision tree, until the optimal model that performs well on both the training data and the validation data is found, and save the model.

Images