Automatic correction method and device for aspheric mirror interference measurement installation and adjustment error

Abstract

The invention relates to an automatic correction method and device for an aspheric mirror interference measurement installation and adjustment error, which can realize automatic, accurate and rapid automatic correction of the installation and adjustment error. The method comprises the following steps: (1) building a virtual interferometer; (2) making a data set; (3) neural network construction and training: constructing a classification neural network, inputting an interference pattern and outputting an installation and adjustment error category, and performing network training by using the data set obtained in the step (2), so that the neural network has the installation and adjustment error detection capability; (4) assembly and adjustment error prediction: obtaining an interferogram under the current assembly and adjustment in an actual interference light path, inputting the interferogram into the trained neural network, and completing the prediction of an assembly and adjustment error category; (5) correcting the installation and adjustment error: according to an installation and adjustment error prediction result obtained in the step (4), controlling a corresponding mechanical part to correct the installation and adjustment error; (6) if the neural network detects that the installation and adjustment error does not meet the requirement, executing the step (4); otherwise, executing the step (7); and (7) ending.

Description

Technical Field

[0001] The present invention relates to the technical field of photoelectric detection, and in particular to a method for automatically correcting the interferometric measurement error of an aspheric mirror, aiming to achieve automatic and accurate correction of the interferometric measurement error of an aspheric mirror, and also to an automatic correction device for the interferometric measurement error of an aspheric mirror. Background Art

[0002] With the development of my country's industrial technology, the industrial sector has an increasing demand for the application of various optical systems. The optical system manufacturing process, in addition to completing the processing of each optical component, also requires the precise assembly of each optical component into an optical system, which is the optical system assembly process. The accuracy of the optical system assembly is an important factor affecting the imaging quality and detection accuracy of the system. The development and progress of assembly technology is particularly important. Especially in the field of optical interferometry of aspheric mirrors, the alignment accuracy of the interferometer and the measured mirror will have a great impact on the accuracy of aspheric mirror surface shape detection. Therefore, the industrial detection field places high demands on the alignment of the interferometer measured mirror. How to quickly, accurately and automatically complete the correction of the aspheric mirror interferometry measurement error has become a major research issue.

[0003] Traditional optical alignment refers to the use of high-precision mechanical devices to perform centering processing on the system, using alignment experience to achieve optical system alignment and ensure that the relative positions of optical components meet design requirements, mainly relying on alignment experience and simple alignment tools. However, traditional optical alignment is highly blind and the alignment cycle is also very long. The existing mechanical alignment accuracy can no longer meet the system alignment requirements and is not suitable for high-precision alignment of large and complex optical systems. For aspheric reflectors in large-aperture optical systems, it is impossible to ensure that the optical axis of the aspheric component is coaxial with the mechanical axis through centering processing methods. Traditional alignment tools and methods are no longer suitable for the alignment process of reflective optical systems.

[0004] Another type of alignment method is known as computer-aided alignment technology. Its basic principle is to determine the misalignment of an optical system based on the correspondence between the optical system's misalignment state and its actual wavefront aberration characteristics. First, the wavefront aberration characteristics generated during the alignment process are studied. Then, optical design software is used to establish a computer-aided alignment model, representing the relationship between component misalignment and aberration in the optical system. The imaging quality of different fields of view is then measured in real time during the alignment process. Based on the computer-aided alignment model, the real-time wavefront aberration data is calculated to determine the misalignment of each optical component. Finally, adjustment mechanisms and positioning tools are used to return the optical components to their ideal design positions. Existing computer-aided alignment technologies rely on auxiliary techniques such as sensitivity matrices and evaluation function calculations, resulting in complex algorithms and a tendency for alignment error correction to fall into local optimal solutions. In summary, there is a need for an automatic, accurate, and rapid method for correcting alignment errors in aspheric mirror interferometry. Summary of the Invention

[0005] In order to overcome the defects of the prior art, the technical problem to be solved by the present invention is to provide an automatic correction method for the interferometric measurement of the adjustment error of an aspheric mirror, which can realize automatic, accurate and rapid automatic correction of the adjustment error.

[0006] The technical solution of the present invention is: the automatic correction method of the aspheric mirror interferometric measurement adjustment error comprises the following steps:

[0007] (1) Construction of virtual interferometer: Based on the actual interference optical path, a virtual interferometer is constructed in the optical simulation software;

[0008] (2) Dataset production: simulate interference patterns under different alignment errors in a virtual interferometer and batch produce data sets for neural network training;

[0009] (3) Neural network construction and training: Build a classification neural network with the interference pattern as input and the adjustment error category as output. Use the data set obtained in step (2) to train the network so that the neural network has the ability to detect adjustment errors.

[0010] (4) Adjustment error prediction: Obtain the interference under the current adjustment in the actual interference optical path

[0011] Figure, the interference pattern is input into the trained neural network to complete the prediction of the adjustment error category;

[0012] (5) Correction of installation error: According to the prediction result of the installation error obtained in step (4), the corresponding mechanical parts are controlled to correct the installation error;

[0013] (6) If the neural network detects that the adjustment error does not meet the requirements, execute step (4); otherwise, execute step (7);

[0014] (7)End.

[0015] This paper establishes a method for automatically correcting the alignment error of aspheric mirror interferometry by sequentially building a virtual interferometer, creating a data set, building and training a neural network, predicting alignment errors, and correcting alignment errors. The method utilizes a virtual interferometer to help the neural network learn the relationship between the interference pattern and the alignment error. During the actual alignment process, the neural network predicts the alignment error contained in the interference pattern and controls a mechanical translation stage to automatically correct the alignment error. The entire process is convenient and fast, with a low detection threshold, achieving automatic and accurate correction of aspheric interferometry alignment errors, making it more convenient for industrial testing.

[0016] Also provided is an automatic correction device for aspheric mirror interferometric measurement adjustment error, which includes:

[0017] A virtual interferometer building module, which is configured to build a virtual interferometer in optical simulation software based on the actual interference light path;

[0018] A dataset creation module, configured to simulate interference patterns under different alignment errors in a virtual interferometer and batch create datasets for neural network training;

[0019] The neural network building and training module is configured to build a classification neural network with the interference pattern as input and the alignment error category as output. The network is trained using the dataset obtained by the dataset creation module so that the neural network has the ability to detect alignment errors.

[0020] The setup error prediction module is configured to obtain the interference pattern under the current setup in the actual interference optical path, input the interference pattern into the trained neural network, and complete the prediction of the setup error category;

[0021] an assembly error correction module configured to control corresponding mechanical parts to correct assembly errors according to the assembly error prediction result obtained by the assembly error prediction module;

[0022] The assembly error prediction module and the assembly error correction module are iterated repeatedly until the neural network detects that the assembly error has been corrected. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a flow chart of an automatic correction method for aspheric mirror interferometric measurement adjustment error of the present invention.

[0024] Figure 2 The present invention is a structural diagram of an automatic correction system for aspheric mirror interferometric measurement adjustment errors.

[0025] Figure 3 It is the interference pattern of an ideal aspheric mirror with no adjustment error.

[0026] Figure 4 It is the interference pattern of an aspheric mirror with adjustment error.

[0027] Among them: 1-aspheric mirror interferometry measurement module, 2-mechanical translation stage, 3-aspheric mirror to be measured, 4-compensation mirror, 5-standard plane mirror, 6-interferometer, 7-computer. DETAILED DESCRIPTION

[0028] like Figure 1 As shown, the automatic correction method of the aspheric mirror interferometric measurement adjustment error includes the following steps:

[0029] (1) Construction of virtual interferometer: Based on the actual interference optical path, a virtual interferometer is constructed in the optical simulation software;

[0030] (2) Dataset production: simulate interference patterns under different alignment errors in a virtual interferometer and batch produce data sets for neural network training;

[0031] (3) Neural network construction and training: Build a classification neural network with the interference pattern as input and the adjustment error category as output. Use the data set obtained in step (2) to train the network so that the neural network has the ability to detect adjustment errors.

[0032] (4) Adjustment error prediction: Obtain the interference under the current adjustment in the actual interference optical path

[0033] Figure, the interference pattern is input into the trained neural network to complete the prediction of the adjustment error category;

[0034] (5) Correction of installation error: According to the prediction result of the installation error obtained in step (4), the corresponding mechanical parts are controlled to correct the installation error;

[0035] (6) If the neural network detects that the adjustment error does not meet the requirements, execute step (4); otherwise, execute step (7);

[0036] (7)End.

[0037] This paper establishes a method for automatically correcting the alignment error of aspheric mirror interferometry by sequentially building a virtual interferometer, creating a data set, building and training a neural network, predicting alignment errors, and correcting alignment errors. The method utilizes a virtual interferometer to help the neural network learn the relationship between the interference pattern and the alignment error. During the actual alignment process, the neural network predicts the alignment error contained in the interference pattern and controls a mechanical translation stage to automatically correct the alignment error. The entire process is convenient and fast, with a low detection threshold, achieving automatic and accurate correction of aspheric interferometry alignment errors, making it more convenient for industrial testing.

[0038] Preferably, in step (1), Zemax optical simulation software is used to perform interference light path simulation.

[0039] Preferably, in step (2), the wavefront distribution of the optical simulation is saved, disturbances such as noise, phase shift, and vignetting are added, and then the interference pattern is calculated to simulate the actual interference pattern; the adjustment errors of each degree of freedom should be comprehensively simulated in the simulation software according to the actual adjustment degrees of freedom, so that the adjustment errors in the data set cover all situations in the actual experiment.

[0040] Preferably, in step (3), a classification neural network of Restnet50 is selected to fully train the neural network and avoid overfitting.

[0041] Preferably, in step (4), a camera is used to obtain the interference pattern of the actual interferometer, and the image size and value range of the interference pattern are guaranteed to be consistent with the neural network through cropping and scaling.

[0042] Preferably, in step (5), the mechanical translation stage is controlled to perform adjustment error correction according to the adjustment error identification result of the network, and the step size is gradually reduced from large to small so as to eventually converge.

[0043] Those skilled in the art will appreciate that all or part of the steps in the above-described embodiment method can be implemented by instructing the relevant hardware through a program. The program can be stored in a computer-readable storage medium. When executed, the program includes the steps of the above-described embodiment method. The storage medium can be ROM / RAM, a magnetic disk, an optical disk, a memory card, etc. Therefore, corresponding to the method of the present invention, the present invention also includes an automatic correction device for aspheric mirror interferometric measurement alignment errors. The device is generally represented in the form of functional modules corresponding to the steps of the method. The device includes:

[0044] A virtual interferometer building module, which is configured to build a virtual interferometer in optical simulation software based on the actual interference light path;

[0045] A dataset creation module, configured to simulate interference patterns under different alignment errors in a virtual interferometer and batch create datasets for neural network training;

[0046] The neural network building and training module is configured to build a classification neural network with the interference pattern as input and the alignment error category as output. The network is trained using the dataset obtained by the dataset creation module so that the neural network has the ability to detect alignment errors.

[0047] The setup error prediction module is configured to obtain the interference pattern under the current setup in the actual interference optical path, input the interference pattern into the trained neural network, and complete the prediction of the setup error category;

[0048] an assembly error correction module configured to control corresponding mechanical parts to correct assembly errors according to the assembly error prediction result obtained by the assembly error prediction module;

[0049] The assembly error prediction module and the assembly error correction module are iterated repeatedly until the neural network detects that the assembly error has been corrected.

[0050] Preferably, in the virtual interferometer building module, Zemax optical simulation software is used to perform interference light path simulation;

[0051] In the dataset creation module, the wavefront distribution of the optical simulation is saved, and disturbances such as noise, phase shift, and vignetting are added before the interferogram calculation is performed to simulate the actual interferogram. The adjustment errors of each degree of freedom should be comprehensively simulated in the simulation software based on the actual adjustment degrees of freedom so that the adjustment errors in the dataset cover all situations in the actual experiment.

[0052] Preferably, in the neural network building and training module, the classification neural network of Restnet50 is selected to fully train the neural network and avoid overfitting.

[0053] Preferably, in the adjustment error prediction module, a camera is used to obtain the interference pattern of the actual interferometer, and the image size and value range of the interference pattern are ensured to be consistent with those of the neural network through cropping and scaling;

[0054] In the adjustment error correction module, the mechanical translation stage is controlled to perform adjustment error correction according to the adjustment error identification result of the network, and the step size gradually decreases from large to small so as to eventually converge.

[0055] The following is a detailed description of a specific embodiment of the present invention. A method for automatically correcting the alignment error of an aspheric mirror by interferometric measurement is implemented in the following manner:

[0056] The process of establishing an automatic correction method for aspheric mirror interferometry error is as follows: Figure 1 As shown, the specific implementation steps are:

[0057] Step 1: Virtual interferometer construction

[0058] In the example, according to Figure 2 The aspheric mirror interferometry optical path shown in the figure uses a virtual interferometer built in Zemax.

[0059] Step 2: Dataset creation

[0060] In this example, a Zemax virtual interferometer was used to analyze the alignment errors of the test piece for X-axis eccentricity, Y-axis eccentricity, and Z-axis defocus. Randomly adding three types of alignment errors (X, Y, and Z) within a ±5mm error range was performed. Data enhancement was then performed by adding noise, phase shifting, and vignetting to each resulting wavefront. This resulted in a dataset of 500,000 data points. The input of the dataset was an image, and the output was the alignment error categories.

[0061] Step 3: Neural Network Construction and Training

[0062] In this example, a Resnet50 classification network with a 512×512 input dimension and a 7-category output is built. The network is trained using the dataset. The cross-entropy loss function is used. For the first 400 epochs, the learning rate is set to 1e-4 to accelerate optimization. During the 400-500 epoch period, the learning rate is reduced to 5e-5 for fine-tuning the network.

[0063] Step 4: Prediction of installation error

[0064] In this example, a camera is used to capture the interference pattern of the current alignment. After cropping and scaling, the image is input into Resnet50 with a dimension of 512×512 for alignment error prediction.

[0065] Step 5: Correction of installation errors

[0066] In this example, manual correction of the adjustment error is performed based on the prediction results of the neural network, with each correction being 0.05 mm. Finally, the adjustment error correction is completed after 7 iterations.

[0067] The beneficial effects of the present invention are as follows:

[0068] 1. The automatic correction method for the interferometric measurement of the aspheric mirror adjustment error disclosed in the present invention can realize the automatic correction of the adjustment error of the aspheric mirror without manual intervention or complex post-processing algorithms.

[0069] 2. The automatic correction device for interferometric measurement of adjustment errors of aspheric mirrors disclosed in the present invention has a simple structure, does not require the addition of any other optical elements, and does not require phase decomposition. It can complete the automatic correction of adjustment errors only by relying on the interference pattern and the corresponding neural network algorithm.

[0070] The above description is merely a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention are still within the scope of protection of the technical solution of the present invention.

Claims

1. An automatic correction method for aspheric mirror interferometry error, characterized by: It includes the following steps: (1) Construction of virtual interferometer: Based on the actual interference optical path, a virtual interferometer is constructed in the optical simulation software; (2) Dataset production: simulate interference patterns under different alignment errors in a virtual interferometer and batch produce data sets for neural network training; (3) Neural network construction and training: Build a classification neural network with the interference pattern as input and the adjustment error category as output. Use the data set obtained in step (2) to train the network so that the neural network has the ability to detect adjustment errors. (4) Prediction of setup error: Obtain the interference pattern under the current setup in the actual interference optical path, input the interference pattern into the trained neural network, and complete the prediction of the setup error category; (5) Correction of installation error: According to the prediction result of the installation error obtained in step (4), the corresponding mechanical parts are controlled to correct the installation error; (6) If the neural network detects that the adjustment error does not meet the requirements, execute step (4); otherwise, execute step (7); (7)End.

2. The automatic correction method for aspheric mirror interferometry error according to claim 1, characterized in that: In the step (1), Zemax optical simulation software is used to perform interference light path simulation.

3. The automatic correction method for aspheric mirror interferometry error according to claim 2, characterized in that: In the step (2), the wavefront distribution of the optical simulation is saved, and disturbances such as noise, phase shift, and vignetting are added, and then the interference pattern is calculated to simulate the actual interference pattern; the adjustment errors of each degree of freedom should be comprehensively simulated in the simulation software according to the actual adjustment degree of freedom, so that the adjustment errors in the data set cover all situations in the actual experiment.

4. The automatic correction method for aspheric mirror interferometry error according to claim 3, characterized in that: In step (3), the classification neural network of Restnet50 is selected; Make the neural network fully trained and avoid overfitting.

5. The automatic correction method for aspheric mirror interferometry error according to claim 4, characterized in that: In step (4), a camera is used to obtain the interference pattern of the actual interferometer, and the image size and value range of the interference pattern are guaranteed to be consistent with those of the neural network through cropping and scaling.

6. The automatic correction method for aspheric mirror interferometry error according to claim 5, characterized in that: In the step (5), the mechanical translation stage is controlled to perform adjustment error correction according to the adjustment error identification result of the network, and the step size is gradually reduced from large to small so as to finally converge.

7. An automatic correction device for aspheric mirror interferometric measurement error, characterized by: It includes: A virtual interferometer building module, which is configured to build a virtual interferometer in optical simulation software based on the actual interference light path; A dataset creation module, configured to simulate interference patterns under different alignment errors in a virtual interferometer and batch create datasets for neural network training; A neural network building and training module, which is configured to build a classification neural network with an interference pattern as input and an alignment error category as output; Use the dataset generated by the dataset creation module to train the network so that the neural network has the ability to detect assembly errors; The setup error prediction module is configured to obtain the interference pattern under the current setup in the actual interference optical path, input the interference pattern into the trained neural network, and complete the prediction of the setup error category; an assembly error correction module configured to control corresponding mechanical parts to correct assembly errors according to the assembly error prediction result obtained by the assembly error prediction module; The assembly error prediction module and the assembly error correction module are iterated repeatedly until the neural network detects that the assembly error has been corrected.

8. The automatic correction device for aspheric mirror interferometry error according to claim 7, characterized in that: In the virtual interferometer construction module, Zemax optical simulation software is used to simulate the interference light path; In the dataset preparation module, the wavefront distribution of the optical simulation is saved, and disturbances such as noise, phase shift, and vignetting are added before the interferogram calculation is performed to simulate the actual interferogram. The adjustment errors of each degree of freedom should be comprehensively simulated in the simulation software based on the actual adjustment degrees of freedom so that the adjustment errors in the dataset cover all situations in the actual experiment.

9. The automatic correction method for aspheric mirror interferometry error according to claim 8, characterized in that: In the neural network building and training module, the Restnet50 classification neural network is selected; Make the neural network fully trained and avoid overfitting.

10. The automatic correction method for aspheric mirror interferometry error according to claim 9, characterized in that: In the adjustment error prediction module, a camera is used to obtain the interference pattern of the actual interferometer, and the image size and value range of the interference pattern are guaranteed to be consistent with those of the neural network through cropping and scaling; In the adjustment error correction module, the mechanical translation stage is controlled to perform adjustment error correction according to the adjustment error identification result of the network, and the step size gradually decreases from large to small so as to eventually converge.

Images