A method for predicting the distribution of contamination layer of optical components based on neural network

Abstract

The present invention relates to the field of optical processing technology, and in particular to a method for predicting the distribution of contamination layers of optical components based on a neural network. n A set of optical component processing experiments under different processing parameters were conducted to collect process parameter data and establish n A data set is formed; the contamination layer distribution data corresponding to the process parameter data is collected; n The data set is divided into a training set and a test set in proportion, and the multi-layer perceptron neural network is trained using the training set to obtain a neural network prediction model; the new process parameters in the optical element contamination layer distribution prediction task are obtained, the new process parameter vector is input and copied three times, and then the three process parameter vectors are respectively input into the three sub-networks of the neural network prediction model, and the neural network prediction coordinate sub-matrices,,, are respectively output, and then input into the reconstruction module for combination to obtain the predicted contamination distribution point cloud data. The advantages are: no need to re-sample and analyze, low cost, and rapid response.

Description

Technical Field

[0001] The present invention relates to the field of optical processing technology, and in particular to a method for predicting the distribution of contamination layers of optical elements based on a neural network. Background Art

[0002] In a high-power laser system, optical components need to withstand extremely high energy density laser beams, which places extremely high demands on the surface accuracy, surface quality and residual contamination of optical components. In particular, residual contamination of optical components can induce laser damage under high-energy density laser irradiation, reduce the life of optical components, and even completely destroy optical components in severe cases, adding a considerable burden to the stable operation of high-power laser systems. Therefore, in the post-processing process of optical components, the contamination layer in the processing of optical components must be removed to increase the laser loss threshold of the optical components. Before using post-processing methods to remove the contamination layer, it is necessary to know the distribution of the contamination layer to achieve accurate removal.

[0003] However, existing methods for determining the distribution of the contamination layer, such as the TOF-SIMS (static time-of-flight secondary ion mass spectrometer) depth analysis method and the XPS (X-ray photoelectron spectrometer) depth analysis method, can only analyze the distribution of the contamination layer under one processing technology (corresponding to a set of processing parameters); when the processing technology is changed, re-sample preparation and analysis are required to determine the distribution of the contamination layer. This process greatly increases the manufacturing cycle of optical components. Summary of the invention

[0004] In order to solve the above problems, the present invention provides a method for predicting the distribution of contamination layer of optical elements based on neural network.

[0005] The present invention aims to provide a method for predicting the distribution of contamination layers of optical components based on a neural network, which specifically comprises the following steps:

[0006] S1. Conduct n Conduct optical component processing experiments under different processing parameters and collect process parameter data ,Establish n Group datasets; Represents the row vector formed by each set of process parameters;

[0007] S2. Collect the pollution layer distribution data corresponding to the process parameter data; the pollution layer distribution data is expressed as follows:

[0008] ;

[0009] S3. nThe data set is divided into a training set and a test set in proportion, and the multilayer perceptron neural network is trained using the training set to obtain a neural network prediction model; the multilayer perceptron neural network includes three subnetworks, and the inputs of the three subnetworks are Row vector, the outputs of the three sub-networks are x-coordinate matrix, y-coordinate matrix and z-coordinate matrix respectively;

[0010] S4. Obtain new process parameters in the optical element contamination layer distribution prediction task, input a new process parameter vector and copy it three times, then input the three process parameter vectors into the three sub-networks of the neural network prediction model respectively, and output the neural network prediction coordinate sub-matrices respectively. , , ;

[0011] S5. The neural network predicts the coordinate sub-matrix , , Input into the reconstruction module for combination to obtain the predicted pollution distribution point cloud data.

[0012] Preferably, the loss functions of the three sub-networks are:

[0013] ;

[0014] Among them, m represents a matrix with m rows and m columns, , , is the coordinate submatrix predicted by the neural network, and , , is the actual coordinate sub-matrix.

[0015] Preferably, in step S3, when the loss function is less than 0.0001, training is stopped to obtain the neural network prediction model.

[0016] Preferably, the optimization algorithm used in the multilayer perceptron neural network training is the Adam algorithm, with a step size of 0.001 and an iteration number of 100,000 times.

[0017] Preferably, in step S4, a new process parameter vector is input into an input module and copied three times, and the input module is represented as follows:

[0018] ;

[0019] in, represents the feature vector input to the neural network, Represents the new process parameter vector input, w Describe the number of process parameters, p Indicates pressure,r Indicates the speed, ph Indicates polishing fluid.

[0020] Preferably, the processing parameters include pressure, rotation speed, abrasive particle size and polishing liquid pH.

[0021] Preferably, in step S2, a static time-of-flight secondary ion mass spectrometer method is used to collect the contamination layer distribution data.

[0022] Preferably, the ratio of training set to test set is 8:2.

[0023] Compared with the prior art, the present invention can achieve the following beneficial effects:

[0024] The present invention proposes a method for predicting the distribution of the contamination layer of an optical element based on a neural network. The input parameters are various parameters in the polishing process of the optical element (such as pressure, rotation speed, abrasive particle size, polishing liquid pH, etc.), and the output is the distribution of the contamination layer of the optical element (three-dimensional point cloud data). The input and output data are trained by a neural network to realize the prediction of the distribution of the contamination layer of the optical element under any combination of parameters. After changing the processing technology, there is no need to re-prepare the sample for analysis. By establishing a neural network prediction structure, the relationship between the processing technology parameters and the contamination layer distribution data is established, thereby realizing accurate prediction of the distribution of the contamination layer of the optical element under any processing technology. The present invention can greatly shorten the processing cycle of optical elements and promote the rapid manufacturing of high-power laser optical elements.

[0025] Although the initial data set collection of this method is relatively large, once the neural network calculation model is built and trained, the prediction method of the optical component contamination layer distribution based on the neural network will be very fast. Compared with traditional methods, it has the characteristics of low cost and rapid response. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 The present invention is a flowchart of a method for predicting the distribution of contamination layers of optical components based on a neural network according to an embodiment of the present invention.

[0027] Figure 2 1 is a schematic diagram of the distribution of some elements provided according to an embodiment of the present invention; in the figure, A represents the distribution of copper ions, and B represents the distribution of iron ions.

[0028] Figure 3 Schematic diagram of the overall prediction of pollution distribution by a neural network according to an embodiment of the present invention. DETAILED DESCRIPTION

[0029] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. In the following description, the same modules are represented by the same reference numerals. In the case of the same reference numerals, their names and functions are also the same. Therefore, the detailed description thereof will not be repeated.

[0030] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and do not constitute a limitation of the present invention.

[0031] The present invention provides a method for predicting the distribution of contamination layer of optical components based on neural network (see flowchart for details) Figure 1 ), specifically including the following steps:

[0032] S1. Conduct n Conduct optical component processing experiments under different processing parameters and collect process parameter data , establish a data set;

[0033] Represents the row vector composed of the process parameters of each group, where each element is a processing parameter such as pressure, rotation speed, abrasive particle size, polishing liquid pH, etc.; Suppose the number of process parameters is w , the process parameter data set actually collected is a The pollution layer distribution data is actually expressed as Figure 2 The point cloud shown (taking copper ions and iron ions as examples); assuming that the length and width of the distribution unit area are m , in data form, is a A 4D tensor of size .

[0034] S2. Use the static time-of-flight secondary ion mass spectrometer (TOF-SIMS) method to collect the contamination layer distribution data corresponding to the process parameter data (as output data); expressed as follows: .

[0035] S3. The dataset (a total of n The group) is divided into a training set and a test set with a ratio of 8:2. Then the training set is used to train the multi-layer perceptron (MLP) neural network to obtain the neural network prediction model;

[0036] like Figure 3 The figure shows the overall schematic diagram of the neural network prediction method for pollution distribution. During training, the three MLP sub-networks are trained separately for the xyz coordinate matrix, and the three inputs are Row vector, the three outputs are x-coordinate matrix, y-coordinate matrix and z-coordinate matrix;

[0037] The optimization algorithm used for network training is the adam algorithm, with a step size of 0.001 and 100,000 iterations. The L2 norm of the difference between the actual coordinate submatrix and the predicted coordinate submatrix is ​​used as the loss function, that is, the loss functions of the three MLP subnetworks are:

[0038] ;

[0039] Where m represents a matrix with m rows and m columns (m is the subscript of the matrix, and the square root of the sum of the differences between all elements of the predicted matrix and the actual matrix is ​​the LOSS expression). , , is the coordinate submatrix predicted by the neural network, and , , is the actual coordinate submatrix; when the loss function is less than 0.0001, the training is stopped and the neural network prediction model is obtained.

[0040] S4. When a new optical component contamination layer distribution prediction task comes, input a new process parameter vector into the input module and copy it three times, input it into the three MLP sub-networks of the neural network prediction model respectively, and output the coordinate sub-matrices obtained by the neural network prediction respectively. , , ;

[0041] The input module is represented as follows:

[0042] ;

[0043] in represents the feature vector input into the entire neural network, Represents the new process parameter vector input, w Describe the number of process parameters, p Indicates pressure, r Indicates the speed, ph Indicates polishing fluid.

[0044] S5. Sub-matrix the coordinates , , Input to the reconstruction module for combination to obtain the predicted pollution distribution point cloud data;

[0045] The reconstruction module concatenates the three coordinate sub-matrices. Since the XYZ sub-matrices are all in the shape of M*M during data processing, the reconstruction module can use the stack function in the numpy function library:

[0046] Stack( , , )=[ , , ];

[0047] [ , , ] is the pollution distribution point cloud matrix.

[0048] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the disclosure of the present invention can be performed in parallel, sequentially or in different orders, as long as the desired results of the technical solution disclosed in the present invention can be achieved, and this document does not limit this.

[0049] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting the distribution of contamination layer of optical components based on neural network, characterized in that: The specific steps include: S1. Conduct n Conduct optical component processing experiments under different processing parameters and collect process parameter data ,Establish n Group datasets; Represents the row vector formed by each set of process parameters; S2. Collect the pollution layer distribution data corresponding to the process parameter data; the pollution layer distribution data is expressed as follows: ； S3. n The data set is divided into a training set and a test set in proportion, and the multilayer perceptron neural network is trained using the training set to obtain a neural network prediction model; the multilayer perceptron neural network includes three subnetworks, and the inputs of the three subnetworks are Row vector, the outputs of the three sub-networks are x-coordinate matrix, y-coordinate matrix and z-coordinate matrix respectively; S4. Obtain new process parameters in the optical element contamination layer distribution prediction task, input the new process parameter vector into the input module and copy it three times, then input the three process parameter vectors into the three sub-networks of the neural network prediction model respectively, and output the neural network prediction coordinate sub-matrices respectively. , , ; The input module is represented as follows: ； in, represents the feature vector input to the neural network, Represents the new process parameter vector input, w Describe the number of process parameters, p Indicates pressure, r Indicates the speed, ph Indicates polishing fluid; S5. The neural network predicts the coordinate sub-matrix , , Input into the reconstruction module for combination to obtain the predicted pollution distribution point cloud data.

2. The method for predicting the distribution of contamination layer of an optical component based on a neural network according to claim 1, characterized in that: The loss functions of the three sub-networks are: ； Among them, m represents a matrix with m rows and m columns. , , is the coordinate submatrix predicted by the neural network, and , , is the actual coordinate sub-matrix.

3. The method for predicting the distribution of contamination layer of an optical component based on a neural network according to claim 2, characterized in that: In step S3, when the loss function is less than 0.0001, the training is stopped to obtain the neural network prediction model.

4. The method for predicting the distribution of contamination layer of an optical component based on a neural network according to claim 3, characterized in that: The optimization algorithm used in the multi-layer perceptron neural network training is the Adam algorithm, with a step size of 0.001 and an iteration number of 100,000 times.

5. The method for predicting the distribution of contamination layer of an optical component based on a neural network according to claim 4, characterized in that: The processing parameters include pressure, rotation speed, abrasive particle size and polishing liquid ph .

6. The method for predicting the distribution of contamination layer of an optical component based on a neural network according to claim 5, characterized in that: In the step S2, a static time-of-flight secondary ion mass spectrometer method is used to collect the contamination layer distribution data.

7. The method for predicting the distribution of contamination layer of an optical component based on a neural network according to claim 6, characterized in that: The ratio of the training set to the test set is 8:2.

Images