A method and system for measuring and adjusting the optical axis position of a coaxial optical component

Abstract

The present invention discloses a method for measuring and adjusting the optical axis posture of a coaxial optical component. The method adopts a measurement method based on interferometry combined with a computational hologram of a subwavelength structure to separate the spatial posture of the optical axis of each optical component, and reflects the spatial posture information of each optical axis into an interference diagram; combined with a method for solving the misalignment parameters based on deep learning, the misalignment of the optical component can be quickly iterated and fed back, thereby completing the rapid adjustment of the coaxial optical component. The advantages of the present invention are that the method adopts a measurement method based on interferometry combined with a computational hologram of a subwavelength structure, which can separate the spatial posture information of the optical axis of each optical component while ensuring the measurement accuracy; the mapping relationship between the spatial posture of the optical axis and the misalignment is analyzed through a deep learning neural network model, thereby improving the efficiency of spatial posture solution and iteration.

Description

Technical Field

[0001] The invention belongs to the technical field of interference measurement and image processing, and in particular relates to a method and system for measuring and assembling the optical axis posture of a coaxial optical component. Background Art

[0002] High-precision optical components are key core components of high-end optoelectronic equipment such as aerospace detection and high-end photolithography machines. Their assembly and adjustment accuracy directly determines imaging resolution, clarity and other indicators. The number of lenses in a general optical component is positively correlated with the imaging index, and the difficulty of assembly and adjustment is also positively correlated with the number of lenses. Therefore, high-end optoelectronic equipment has a large number of lenses and is difficult to assemble and adjust.

[0003] At present, the assembly of high-precision optical components is mainly done by manual and computer-assisted adjustment to gradually measure and adjust the position, tilt and offset of the lens, so as to ensure the adjustment requirements of the optical components. However, this method has poor accuracy, low efficiency and unstable product performance, and can no longer meet the future requirements of ultra-high precision, short cycle and batch processing of optical components in various fields.

[0004] The existing coaxial optical component assembly and adjustment devices are mainly divided into two categories, namely, the assembly and adjustment based on the self-centering instrument (Zhou Yunyi, Lin Lina, Peng Zhangxian. Design of precision reflective centering instrument equipment [J]. Optics and Optoelectronics, 2020, 18(05): 68-74.) and the assembly and adjustment based on the interferometer (Chen Taixi, Wang Jinxin, Huang He, et al. Optical assembly and adjustment method of near-infrared catadioptric system with non-ideal imaging of the intermediate image plane [J]. Infrared and Laser Engineering, 2024, 53(05): 216-225.).

[0005] However, the spot size of the self-centering instrument is large, and it can only achieve micron-level posture measurement. In addition, only a single laser spot can feedback the coupled posture information of each optical lens, and it is impossible to separate the posture misalignment of each optical lens.

[0006] Interferometer-based adjustment can achieve hundreds of nanometers of measurement using interference measurement, but it is also impossible to obtain the position information of each coaxial lens at the same time. It requires manual or computer-assisted repeated measurement-adjustment and iterative optimization, and the iterative adjustment time depends on the operator's expertise and experience. When the number of optical component lenses is large, iterative optimization is difficult and time-consuming. In addition, for the complex nonlinear relationship between system aberrations and misalignment, the traditional computer-assisted adjustment algorithm performs linear approximation on it, and constructs a system sensitivity matrix to iteratively calculate the misalignment in an analytical form. Summary of the invention

[0007] In view of the above-mentioned deficiencies in the prior art, the present invention provides a method and system for measuring and adjusting the optical axis posture of a coaxial optical component.

[0008] The present invention first provides a method for measuring and adjusting the optical axis posture of a coaxial optical component, wherein the coaxial optical component comprises N lenses, all of which are optical rotationally symmetrical structures, where N is a positive integer ≥ 2, and each of the N lenses is arranged on a five-dimensional adjustment frame. The method comprises the following steps:

[0009] (1) A computer-generated hologram is arranged on one side of the coaxial optical component, a standard interferometer is arranged on the outer side of the computer-generated hologram, the standard interferometer has a standard transmission plane mirror, and a standard reflection plane mirror is arranged on the other side of the coaxial optical component;

[0010] (2) The light emitted from the standard interferometer is divided into two beams by the standard transmission plane mirror, one of which is reflected back into the standard interferometer as reference light; the other light passes through the standard transmission plane mirror and irradiates the computer generated hologram, then diffracts and propagates to the coaxial optical component. The light passing through the coaxial optical component is reflected by the standard reflection plane mirror and then propagates in sequence until it returns to the standard interferometer, and interferes with the reference light as the test light, thereby obtaining an interference pattern carrying the spatial position information of the optical axis of the coaxial optical component;

[0011] When the positions of the lenses in the coaxial optical assembly are correct, the interference pattern is an interference pattern without aberration; when the positions of the lenses are misaligned, the interference pattern is an interference pattern containing aberration;

[0012] (3) Inputting the interference pattern containing aberrations into a neural network model based on deep learning, outputting the spatial posture of each lens, and converting it into the five-dimensional posture misalignment of each lens, and adjusting the corresponding five-dimensional adjustment frame according to the five-dimensional posture misalignment of the lens;

[0013] (4) Repeat steps (2) and (3) until the positions of the lenses of the coaxial optical assembly are correct.

[0014] Preferably, before placing the coaxial optical component and the computer-generated hologram, the position of the standard reflecting plane mirror is first determined by a standard interferometer and a standard transmitting plane mirror, and then the position of the computer-generated hologram is determined by a standard interferometer with a standard transmitting plane mirror and a standard plane reflecting mirror, and finally the coaxial optical component is measured and adjusted.

[0015] Preferably, the five adjustment dimensions of the five-dimensional adjustment frame include three mutually perpendicular directions of x, y, and z, and θ x ,θ y Two angles of adjustment, where the z direction is the axial direction of the coaxial optical component, θ x ,θ y The two angles are adjusted around the x direction and around the y direction respectively.

[0016] Preferably, the area divided by the computer-generated hologram includes N annular areas arranged in concentric circles, and the N annular areas respectively correspond to the spatial posture measurement of N lenses of the coaxial optical component;

[0017] There are also 2N+2 square areas on the periphery of the N annular areas. The 2N+2 square areas are divided into N+1 groups, each group has two areas, one of which is used for the position alignment of the computer-generated hologram itself, and the other N groups are used for the axial position alignment of the N lenses of the coaxial optical component.

[0018] Preferably, the neural network model construction method based on deep learning comprises the following steps:

[0019] S1, constructing an interference system model in the simulation software, confirming the initial error range of the N lenses of the coaxial optical component in the actual adjustment, quantitatively sampling within the initial error range, inputting the five-dimensional spatial posture, randomly combining the position errors of the N lenses, and calculating and outputting the corresponding interference pattern data set,

[0020] According to the Zernike aberration theory, each interference pattern in the interference pattern data set is converted into 36 Zernike polynomial coefficients, and a data set in which the spatial posture corresponds to the 36 Zernike polynomial coefficients is established;

[0021] S2, constructing an initial neural network model, using the data set obtained in step S1 as training data based on a deep learning neural network model, dividing the training data into a training set and a test set, the input of the neural network is an interference pattern, and according to the Zernike aberration theory, the interference pattern is converted into corresponding 36 Zernike polynomial coefficients, and the output is the actual spatial posture, and the neural network model based on deep learning is obtained through training.

[0022] More preferably, the neural network model based on deep learning adopts a convolutional network model, and the loss function adopts MSEloss.

[0023] More preferably, the convergence condition of the neural network model is that the residual root mean square of the aberration is less than 0.05 wavelength, and the model parameters are saved when the prediction accuracy of the neural network model meets the requirements.

[0024] The present invention further provides a system for measuring and adjusting the optical axis posture of a coaxial optical component, and a method for measuring and adjusting the optical axis posture of the coaxial optical component. The coaxial optical component includes N lenses, all of which are optical rotationally symmetrical structures, where N is a positive integer ≥ 2, and each of the N lenses is arranged on a five-dimensional adjustment frame. The system for measuring and adjusting the optical axis posture of the coaxial optical component includes a standard interferometer, a computer-generated hologram, a standard reflecting plane mirror, and a control device. The standard interferometer has a standard transmitting plane mirror.

[0025] When in use, the computer-generated hologram is arranged on one side of the coaxial optical component, the standard interferometer is arranged on the outer side of the computer-generated hologram, and a standard reflecting plane mirror is arranged on the other side of the coaxial optical component;

[0026] The control device includes a data processing unit and a control unit. The data processing unit is used to input an interference pattern containing aberrations into a neural network model based on deep learning, output the spatial posture of each lens, and convert it into a five-dimensional posture misalignment amount of each lens; the control unit is used to adjust the corresponding five-dimensional adjustment frame according to the five-dimensional posture misalignment amount of each lens output by the data processing unit.

[0027] Preferably, the area divided by the computer-generated hologram includes N annular areas arranged in concentric circles, and the N annular areas respectively correspond to the spatial posture measurement of N lenses of the coaxial optical component;

[0028] There are also 2N+2 square areas on the periphery of the N annular areas. The 2N+2 square areas are divided into N+1 groups, each group has two areas, one of which is used for the position alignment of the computer-generated hologram itself, and the other N groups are used for the axial position alignment of the N lenses of the coaxial optical component.

[0029] Compared with the prior art, the present invention has the following significant advantages:

[0030] (1) A measurement method based on interferometry combined with sub-wavelength structured computer-generated holograms is adopted. While ensuring the accuracy of interferometry measurement, the spatial position information of the optical axis of each coaxial optical component can be separated by taking advantage of the strong phase modulation capability of the sub-wavelength structured computer-generated hologram.

[0031] (2) The deep learning-based method has obvious advantages in dealing with the nonlinear relationship between system aberrations and misalignment. The mapping relationship between the spatial pose of the optical axis and the misalignment is analyzed through the deep learning neural network model, thereby improving the accuracy of spatial pose solution and iteration efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a schematic diagram of the method for measuring and adjusting the optical axis posture of a coaxial optical component of the present invention.

[0033] Figure 2 Schematic diagram of the structure of a computer-generated hologram with a sub-wavelength structure and two mirrors.

[0034] Figure 3 Flowchart of the neural network training process for two-mirror alignment. DETAILED DESCRIPTION

[0035] A system for measuring and adjusting the optical axis posture of a coaxial optical component, wherein the coaxial optical component to be measured includes N lenses of optical rotationally symmetrical structure, N is a positive integer ≥ 2, and each of the N lenses is arranged on a five-dimensional adjustment frame. The system for measuring and adjusting the optical axis posture of the coaxial optical component includes a standard interferometer, a computer-generated hologram, a standard reflecting plane mirror and a control device, wherein the standard interferometer has a standard transmitting plane mirror. The computer-generated hologram is a computer-generated hologram of a sub-wavelength structure.

[0036] The five adjustment dimensions of the five-dimensional adjustment frame include three mutually perpendicular directions of x, y, and z, and θ x ,θ y Two angles of adjustment, where the z direction is the axial direction of the coaxial optical component, θ x ,θ y The two angles are adjusted around the x direction and around the y direction respectively.

[0037] The area divided by the computational hologram includes N annular areas arranged in concentric circles, and the N annular areas correspond to the spatial posture measurement of the N lenses of the coaxial optical component respectively; there are also 2N+2 square areas on the periphery of the N annular areas, and the 2N+2 square areas are divided into N+1 groups, each group has two areas, one of which is used for the position alignment of the computational hologram itself, and the other N groups are used for the axial position alignment of the N lenses of the coaxial optical component.

[0038] When in use, a computer-generated hologram is set on one side of the coaxial optical component to be measured, a standard interferometer is set on the outside of the computer-generated hologram, and a standard reflecting plane mirror is set on the other side of the coaxial optical component. Before placing the coaxial optical component and the computer-generated hologram, the standard interferometer and the standard transmitting plane mirror are used to determine the position and posture of the standard reflecting plane mirror, and then the standard interferometer with the standard transmitting plane mirror and the standard plane reflecting mirror are used to determine the position and posture of the computer-generated hologram, and finally the coaxial optical component is measured and adjusted.

[0039] The control device includes a data processing unit and a control unit. The data processing unit is used to input the interference pattern containing aberrations into a neural network model based on deep learning, output the spatial posture of each lens, and convert it into a five-dimensional posture misalignment amount of each lens; the control unit is used to adjust the corresponding five-dimensional adjustment frame according to the five-dimensional posture misalignment amount of each lens output by the data processing unit.

[0040] A method for measuring and adjusting the optical axis position of a coaxial optical component, wherein the coaxial optical component to be measured includes N lenses, all of which are optical rotationally symmetrical structures, where N is a positive integer ≥ 2, and each of the N lenses is arranged on a five-dimensional adjustment frame. The method includes the following steps:

[0041] (1) A CGM is placed on one side of the coaxial optical component to be measured. A standard interferometer is placed on the outside of the CGM. The standard interferometer has a standard transmission plane mirror. A standard reflection plane mirror is placed on the other side of the coaxial optical component. Before placing the coaxial optical component and the CGM, the standard interferometer and the standard transmission plane mirror are used to determine the position of the standard reflection plane mirror. Then, the standard interferometer with the standard transmission plane mirror and the standard plane reflection mirror are used to determine the position of the CGM. Finally, the position of the coaxial optical component is measured and adjusted.

[0042] (2) The light emitted from the standard interferometer is divided into two beams by the standard transmission plane mirror, one of which is reflected back into the standard interferometer as reference light; the other light passes through the standard transmission plane mirror and irradiates the computer generated hologram, then diffracts and propagates to the coaxial optical component. The light passing through the coaxial optical component is reflected by the standard reflection plane mirror and propagates in sequence until it returns to the standard interferometer, and interferes with the reference light as the test light, thereby obtaining an interference pattern carrying the spatial position information of the optical axis of the coaxial optical component;

[0043] When the positions of the lenses in the coaxial optical assembly are correct, the interference pattern is an interference pattern without aberration; when the positions of the lenses are misaligned, the interference pattern is an interference pattern containing aberration.

[0044] (3) The interference pattern containing aberrations is input into a neural network model based on deep learning, and the spatial posture of each lens is output and converted into the five-dimensional posture misalignment of each lens. The corresponding five-dimensional adjustment frame is adjusted according to the five-dimensional posture misalignment of the lens.

[0045] The method for building a neural network model based on deep learning includes the following steps:

[0046] S1, constructing an interference system model in the simulation software, confirming the initial error range of the N lenses of the coaxial optical component in the actual adjustment, quantitatively sampling within the initial error range, inputting the five-dimensional spatial posture, randomly combining the position errors of the N lenses, and calculating and outputting the corresponding interference pattern data set,

[0047] According to the Zernike aberration theory, each interference pattern in the interference pattern data set is converted into 36 Zernike polynomial coefficients, and a data set in which the spatial posture corresponds to the 36 Zernike polynomial coefficients is established;

[0048] S2, construct an initial neural network model, use the data set obtained in step S1 as training data based on the deep learning neural network model, divide the training data into a training set and a test set, the input of the neural network is an interference pattern, and the interference pattern is converted into the corresponding 36 Zernike polynomial coefficients according to the Zernike aberration theory, and the output is the actual spatial posture, and the neural network model based on deep learning is obtained through training. When constructing the initial neural network model, python or other programming languages ​​can be used. When the training data is divided into a training set and a test set, it can be divided in a ratio of 8:2.

[0049] The neural network model based on deep learning adopts a convolutional network model, including convolutional layers and fully connected layers. It selects appropriate parameters such as the number of convolutional layers, convolution kernel format, convolution kernel size, and the number of fully connected layers. It uses the interference pattern as input and the spatial posture as output. The loss function uses MSEloss. The convergence condition of the neural network model is that the residual aberration root mean square is less than 0.05 wavelengths. When the prediction accuracy of the neural network model meets the requirements, the model parameters are saved. After the training of the neural network, a set of interference patterns are input to directly obtain the actual spatial posture and spatial position misalignment.

[0050] (4) Repeat steps (2) and (3) until the positions of the lenses of the coaxial optical assembly are correct.

[0051] like Figure 1 It is a schematic diagram of the method for measuring and adjusting the optical axis posture of the coaxial optical component of the present invention when the coaxial optical component to be measured includes two lenses.

[0052] Figure 2 Schematic diagram of the sub-wavelength structure of the computer-generated hologram for the two-mirror assembly. The sub-wavelength structure of the computer-generated hologram is designed in different areas according to the number of lenses of the coaxial optical component to be measured. The divided areas include: the two central annular areas correspond to the spatial position information measurement of the two lenses respectively, and the six square areas on the edge are used for the alignment of the position of the computer-generated hologram itself, the two middle square areas are used for the axial position alignment of one of the lenses, and the two lower square areas are used for the axial position alignment of the other lens.

[0053] Figure 3 Flowchart of the neural network training process for two-mirror alignment. The steps include:

[0054] 1. Construct an interference system model to confirm the initial error range of lenses 1 and 2 to be adjusted in actual adjustment;

[0055] 2. Quantitatively sample the initial error range of the lens, randomly combine the position errors, calculate and output the corresponding interference pattern data, and construct a data set;

[0056] 3. Using the data set, train a deep learning-based neural network model to predict the relationship between the pose error and the Zernike coefficient corresponding to the interference pattern;

[0057] 4. In the actual adjustment process, the collected interference pattern is input into the neural network model, and the lens adjustment error is output. It is adjusted through the five-dimensional adjustment frame until it meets the accuracy requirements.

[0058] Example 1

[0059] Take the objective lens of a large-aperture theodolite with a diameter of 600 mm as an example. The objective lens contains two optical lenses with a diameter of 600 mm placed close together. Usually, the position of one lens is adjusted based on the position of the other lens to ensure the consistency of the spatial position of the optical system.

[0060] Table 1 Posture parameters of the objective lens system of the large-aperture theodolite before and after adjustment

[0061]

[0062] In the method of this embodiment, the optical design convergence value is reached after only one round of adjustment. The adjustment effect data shown in Table 1 are obtained in the experiment. The adjustment accuracy reaches 0.35μm displacement deviation and 0.11 second angle deviation. The entire adjustment process is controlled within 1 minute.

Claims

1. A method for measuring and adjusting the optical axis position of a coaxial optical component, wherein the coaxial optical component comprises N lenses of optical rotationally symmetric structure, N is a positive integer ≥ 2, and each of the N lenses is arranged on a five-dimensional adjustment frame, characterized in that: The method comprises the following steps: (1) A computer-generated hologram is arranged on one side of the coaxial optical component, a standard interferometer is arranged on the outer side of the computer-generated hologram, the standard interferometer has a standard transmission plane mirror, and a standard reflection plane mirror is arranged on the other side of the coaxial optical component; (2) The light emitted from the standard interferometer is divided into two beams by the standard transmission plane mirror, one of which is reflected back into the standard interferometer as reference light; the other light passes through the standard transmission plane mirror and irradiates the computer generated hologram, then diffracts and propagates to the coaxial optical component. The light passing through the coaxial optical component is reflected by the standard reflection plane mirror and then propagates in sequence until it returns to the standard interferometer, and interferes with the reference light as the test light, thereby obtaining an interference pattern carrying the spatial position information of the optical axis of the coaxial optical component; When the positions of the lenses in the coaxial optical assembly are correct, the interference pattern is an interference pattern without aberration; when the positions of the lenses are misaligned, the interference pattern is an interference pattern containing aberration; (3) Inputting the interference pattern containing aberrations into a neural network model based on deep learning, outputting the spatial posture of each lens, and converting it into the five-dimensional posture misalignment of each lens, and adjusting the corresponding five-dimensional adjustment frame according to the five-dimensional posture misalignment of the lens; (4) Repeat steps (2) and (3) until the positions of the lenses of the coaxial optical assembly are correct; The method for constructing a neural network model based on deep learning comprises the following steps: S1, constructing an interference system model in the simulation software, confirming the initial error range of the N lenses of the coaxial optical component in the actual adjustment, quantitatively sampling within the initial error range, inputting the five-dimensional spatial posture, randomly combining the position errors of the N lenses, and calculating and outputting the corresponding interference pattern data set, According to the Zernike aberration theory, each interference pattern in the interference pattern data set is converted into 36 Zernike polynomial coefficients, and a data set in which the spatial posture corresponds to the 36 Zernike polynomial coefficients is established; S2, constructing an initial neural network model, using the data set obtained in step S1 as training data based on a deep learning neural network model, dividing the training data into a training set and a test set, the input of the neural network is an interference pattern, and according to the Zernike aberration theory, the interference pattern is converted into corresponding 36 Zernike polynomial coefficients, and the output is the actual spatial posture, and the neural network model based on deep learning is obtained through training.

2. The method for measuring and adjusting the optical axis position of a coaxial optical component according to claim 1, characterized in that: Before placing the coaxial optical component and the computer-generated hologram, the position and posture of the standard reflecting plane mirror are first determined by a standard interferometer and a standard transmitting plane mirror, and then the position and posture of the computer-generated hologram are determined by a standard interferometer with a standard transmitting plane mirror and a standard plane reflecting mirror. Finally, the position and posture of the coaxial optical component are measured and adjusted.

3. The method for measuring and adjusting the optical axis position of a coaxial optical component according to claim 1, characterized in that: The five adjustment dimensions of the five-dimensional adjustment frame include three mutually perpendicular directions of x, y, and z, and θ x ,θ y Two angles of adjustment, where the z direction is the axial direction of the coaxial optical component, θ x ,θ y The two angles are adjusted around the x direction and around the y direction respectively.

4. The method for measuring and adjusting the optical axis position of a coaxial optical component according to claim 1, characterized in that: The area divided by the computerized hologram includes N annular areas arranged in concentric circles, and the N annular areas respectively correspond to the spatial posture measurement of the N lenses of the coaxial optical component; There are also 2N+2 square areas on the periphery of the N annular areas. The 2N+2 square areas are divided into N+1 groups, each group has two areas, one of which is used for the position alignment of the computer-generated hologram itself, and the other N groups are used for the axial position alignment of the N lenses of the coaxial optical component.

5. The method for measuring and adjusting the optical axis position of a coaxial optical component according to claim 1, characterized in that: The deep learning-based neural network model adopts a convolutional network model, and the loss function adopts MSEloss.

6. The method for measuring and adjusting the optical axis position of a coaxial optical component according to claim 5, characterized in that: The convergence condition of the neural network model is that the residual root mean square of the aberration is less than 0.05 wavelength, and the model parameters are saved when the prediction accuracy of the neural network model meets the requirements.

7. A system for measuring and adjusting the optical axis posture of a coaxial optical component, used in the method for measuring and adjusting the optical axis posture of a coaxial optical component according to any one of claims 1 to 6, wherein the coaxial optical component comprises N lenses, all of which are optical rotationally symmetric structures, N is a positive integer ≥ 2, and each of the N lenses is arranged on a five-dimensional adjustment frame, characterized in that: The system for measuring and adjusting the optical axis position of the coaxial optical component comprises a standard interferometer, a computer generated hologram, a standard reflecting plane mirror and a control device. The standard interferometer has a standard transmitting plane mirror. When in use, the computer-generated hologram is arranged on one side of the coaxial optical component, the standard interferometer is arranged on the outer side of the computer-generated hologram, and a standard reflecting plane mirror is arranged on the other side of the coaxial optical component; The control device includes a data processing unit and a control unit. The data processing unit is used to input an interference pattern containing aberrations into a neural network model based on deep learning, output the spatial posture of each lens, and convert it into a five-dimensional posture misalignment amount of each lens; the control unit is used to adjust the corresponding five-dimensional adjustment frame according to the five-dimensional posture misalignment amount of each lens output by the data processing unit.

8. The system for measuring and adjusting the optical axis position of a coaxial optical component according to claim 7, characterized in that: The area divided by the computerized hologram includes N annular areas arranged in concentric circles, and the N annular areas respectively correspond to the spatial posture measurement of the N lenses of the coaxial optical component; There are also 2N+2 square areas on the periphery of the N annular areas. The 2N+2 square areas are divided into N+1 groups, each group has two areas, one of which is used for the position alignment of the computer-generated hologram itself, and the other N groups are used for the axial position alignment of the N lenses of the coaxial optical component.

Images