Ritchey-mangon detection method and system based on global power wave aberration prediction

Abstract

The application belongs to the technical field of optical detection, and particularly relates to a Ritchey-Common detection method and system based on global power wave aberration prediction. The method first establishes a three-dimensional optical model through a standard spherical mirror, a to-be-detected plane mirror and interferometer parameters, combines a preset Ritchey angle and a defocus / astigmatism coefficient, generates an enhanced training data space by using Monte Carlo simulation to inject vibration and temperature noise; secondly, a phase error adjustment model is constructed based on a generative adversarial network, a multi-scale attention mechanism and a radial basis function network are used to decouple and adjust the mapping relationship between the error and the wave aberration distortion, so that the adaptive correction of the sub-aperture phase data is realized; finally, a graph neural network is used to splice the sub-aperture phase and fuse the multi-angle full-aperture measurement data, and the to-be-detected plane mirror surface is accurately reconstructed; through the physical constraint adversarial training and dynamic error compensation, the detection efficiency and reliability of the large-aperture optical element are significantly improved.

Description

Technical Field

[0001] The present invention belongs to the field of optical detection technology, and in particular relates to a Richcomon detection method and system based on global power wave aberration prediction. Background Art

[0002] In the field of optical manufacturing, high-precision surface shape detection of large-aperture plane mirrors is a core link to ensure the performance of high-end optical systems such as aerospace and astronomy. Among traditional detection methods, the Richey-Common method has attracted much attention because it does not require an ultra-large aperture interferometer. It uses a small-aperture standard spherical mirror to perform multi-angle sub-aperture scanning on the plane mirror to be measured, and then realizes full-aperture surface shape reconstruction through a phase stitching algorithm. However, the existing Richey-Common sub-aperture stitching algorithm has fundamental limitations: it assumes that the coordinate transformation between sub-apertures is only caused by ideal mechanical motion, ignoring the nonlinear coupling effect of adjustment error and mirror surface shape error in actual detection; for example, during the two Richey angle adjustments, the slight displacement or angle of the spherical mirror The deviation will introduce additional wave aberrations and superimpose them with the local surface error of the measured mirror, resulting in mixed distortion in the sub-aperture phase data. Traditional algorithms eliminate rigid displacement through least squares fitting or iterative optimization, but cannot analyze the global transmission characteristics of such coupling errors, resulting in low-frequency error accumulation and high-frequency detail distortion of the full-aperture surface after stitching. In addition, the existing methods rely on multiple mechanical adjustments and repeated measurements, which are inefficient and difficult to ensure the stability of the adjustment error. Therefore, there is an urgent need for a new stitching algorithm that can decouple the adjustment error and surface error and is based on the global wave aberration propagation model to achieve high-precision and high-efficiency Richey-Common detection and break through the precision bottleneck of large-aperture plane mirror manufacturing. Summary of the Invention

[0003] In response to the shortcomings of the existing technology, the present invention proposes a Richey-Common detection method and system based on global power wave aberration prediction. The method first establishes a three-dimensional optical model through a standard spherical mirror, a plane mirror to be measured and interferometer parameters, and combines the preset Richey angle and defocus / astigmatism coefficient to generate an enhanced training data space through Monte Carlo simulation of injected vibration and temperature noise; secondly, a phase error adjustment model is constructed based on an adversarial network, and the mapping relationship between the adjustment error and the wave aberration distortion is decoupled through a multi-scale attention mechanism and a radial basis function network to achieve adaptive correction of sub-aperture phase data; finally, a graph neural network is used to splice the sub-aperture phase and fuse multi-angle full-aperture measurement data to accurately reconstruct the surface shape of the plane mirror to be measured; the present invention significantly improves the detection efficiency and reliability of large-aperture optical components through physically constrained adversarial training and dynamic error compensation.

[0004] To achieve the above object, the present invention provides the following technical solutions:

[0005] A Rich Common detection method based on global power wave aberration prediction, comprising:

[0006] A three-dimensional optical model is established based on the preset standard spherical mirror, the plane mirror to be measured, and the interferometer lens parameter information combined with the preset three-dimensional coordinate system;

[0007] Based on the three-dimensional optical model combined with the preset first Richter angle and the defocus and astigmatism coefficients corresponding to the plane mirror, an enhanced training data space is obtained through the Monte Carlo simulation sampling algorithm combined with preset random noise;

[0008] Obtaining first sub-aperture phase adjustment data based on the measured enhanced training data space in combination with a preset phase error countermeasure adjustment model and a preset adjustment error-waveform aberration distortion mapping;

[0009] The phase error adversarial adjustment model is obtained by combining K groups of first training set sequences and second training set sequences with an adversarial network algorithm for training;

[0010] The adjustment error-wave aberration distortion mapping is obtained by fitting the convolution radial function in the coding layer of the adversarial network algorithm with the wave aberration data of the sub-aperture overlapping area in the enhanced training data space and the interferometer adjustment error;

[0011] Obtaining first full-aperture measurement data through a phase map stitching algorithm based on the first sub-aperture phase adjustment data;

[0012] Adjust the first Richey angle, repeat the first full-aperture measurement data acquisition process, and obtain the second full-aperture measurement data;

[0013] Based on the first full-aperture measurement data and the second full-aperture measurement data, a surface shape corresponding to the plane mirror to be measured is obtained.

[0014] Specifically, the process of establishing a three-dimensional optical model includes:

[0015] The interferometer focus parameter is used as the origin of the preset three-dimensional coordinate system, and the plane parallel to the plane mirror to be measured is used as the XOY plane of the three-dimensional coordinate system, and the optical path direction is used as the Z axis to construct the preset three-dimensional coordinate system;

[0016] Configure the preset three-dimensional coordinate system into the optical simulation software, and configure the standard spherical mirror aperture parameters according to the plane mirror aperture to be measured;

[0017] The configured standard spherical mirror aperture parameter set properties, the properties of the plane mirror to be measured, and the interferometer lens parameter information are input into the optical simulation, and an initial three-dimensional optical model is established based on the preset three-dimensional coordinate system.

[0018] Specifically, the process of establishing the three-dimensional optical model also includes:

[0019] Obtain historical adjustment parameter information corresponding to the standard spherical mirror, the plane mirror to be measured, and the interferometer lens parameter information ,in represents the defocus corresponding to the plane mirror, represents the 90° astigmatism coefficient corresponding to the plane mirror, It represents the interferometer adjustment error, that is, the deviation of the distance from the interferometer focus to the plane mirror to be measured. represents the power spectrum density of the vibration noise in the detection environment, Indicates the temperature drift value corresponding to the test environment;

[0020] Based on the historical adjustment parameter information, the parameter distribution function corresponding to each adjustment parameter is obtained through statistical analysis software;

[0021] Based on the parameter distribution function corresponding to each adjustment parameter and the corresponding attribute information and detection accuracy requirements of the standard spherical mirror, the plane mirror to be measured and the interferometer, the physical constraint space corresponding to the standard spherical mirror, the plane mirror to be measured and the interferometer is set;

[0022] Configuring the acquired physical constraint space into the attribute information corresponding to the virtual standard spherical mirror, the plane mirror to be measured, and the interferometer in the initial three-dimensional optical model;

[0023] At the same time, under the parameter distribution function corresponding to each adjustment parameter, the finite element analysis algorithm is used to obtain right The correlation coefficient of each parameter in and Corresponding superimposed deformation parameters right The correlation influence coefficient of each parameter in;

[0024] Based on the correlation influence coefficient, the noise linear function space is obtained through linear algorithm;

[0025] The noise linear function space is given by and The noise linear function of each parameter in and The noise linear function of each parameter in is constructed;

[0026] The noise linear function space is built into the corresponding attribute information of the virtual standard spherical mirror, the plane mirror to be measured and the interferometer in the initial three-dimensional optical model to obtain a three-dimensional optical model with noise.

[0027] Specifically, the process of acquiring the enhanced training data space includes:

[0028] exist In this case, set the initial defocus , initial astigmatism coefficient The number of scans is preset based on the standard spherical mirror aperture parameters;

[0029] Based on the three-dimensional optical model with noise, the initial defocus, the initial astigmatism coefficient, the standard spherical mirror aperture parameters, the preset number of scans and the number of sub-apertures, the position of the plane mirror to be measured is moved in the XOY plane to scan each preset sub-aperture;

[0030] When scanning each sub-aperture, a preset interferometer adjustment error sequence is combined to obtain a first wave of phase difference training data set formed by wave aberration data of different sub-apertures;

[0031] Adjust the initial defocus and initial astigmatism coefficient and repeat the above steps. Combined with the preset interferometer adjustment error sequence during each scan of each sub-aperture, a second wave of phase difference training data set formed by wave aberration data of different sub-apertures is obtained.

[0032] Specifically, the process of acquiring the enhanced training data space also includes:

[0033] exist In this case, based on the first-wave phase difference training data set and the second-wave phase difference training data set, the first-wave phase difference training data set and the second-wave phase difference training data set are randomly divided into layers by a random division algorithm to obtain a first-wave phase difference segmented training data set and a second-wave phase difference segmented training data set;

[0034] Based on the first-wave phase difference segmentation training data set and the second-wave phase difference segmentation training data set, the Latin hypercube sampling algorithm is combined with the preset sampling number K-2 to perform random sampling in each first-wave phase difference segmentation training data or the second-wave phase difference segmentation training data to obtain K-2 groups of phase difference random sampling sequences;

[0035] Feeding back K-2 groups of random sampling sequences of wave phase differences to the three-dimensional optical model with noise, performing random noise injection simulation on each group of random sampling sequences of wave phase differences through the noise linear function space and the physical constraint space, and obtaining K-2 groups of random sampling sequences of wave phase differences after noise injection;

[0036] Based on the K-2 groups of random sampling sequences of wave phase differences after noise injection, the enhanced training data space is obtained by combining the first wave phase difference training data set and the second wave phase difference training data set.

[0037] Specifically, the process of acquiring the first full-aperture measurement data includes:

[0038] Inputting the sub-aperture wavelet aberration, adjustment error, and environmental noise parameters in the enhanced training data space into the coding layer of the phase error adversarial adjustment model to obtain adjusted sub-aperture wavelet phase data, predicted adjustment error, and adjustment error-wavelet aberration distortion mapping;

[0039] The adjusted sub-aperture wave difference phase data, the predicted adjustment error and the actual sub-aperture wave aberration obtained by measurement are input into the decoding layer in the phase error adversarial adjustment model, and the error between the wave phase difference of the adjusted sub-aperture and the actual sub-aperture wave aberration and the loss function constructed by the adjustment error-wave aberration distortion mapping are combined for training, and the adjusted sub-aperture wave difference phase data and the coverage relationship corresponding to all adjusted sub-aperture wave difference phase data are output.

[0040] Specifically, the process of acquiring the first full-aperture measurement data further includes:

[0041] The adjusted wave difference phase data of each sub-aperture is modeled as a sub-aperture graph node, and a splicing edge is established based on the coverage relationship corresponding to all the adjusted wave difference phase data of the sub-aperture. At the same time, the difference weight of the splicing edge is constructed by the phase difference of the overlapping area of ​​two adjacent sub-apertures.

[0042] The modeled sub-aperture graph nodes and the corresponding stitching edges and stitching edge difference weights are input into the graph algorithm to obtain the sub-aperture stitching graph and the stitching phase difference loss corresponding to the overlapping area of ​​two adjacent sub-apertures. The sub-aperture stitching graph and the stitching phase difference loss are fed back into the decoding layer in the phase error adversarial adjustment model for training to obtain the trained phase error adversarial adjustment model, and the first full-aperture measurement data after stitching is output.

[0043] Specifically, the process of acquiring the adjusted sub-aperture wave difference phase data includes:

[0044] The sub-aperture wavelet aberration and adjustment error are converted into a single-channel phase map and input into the first branch sub-layer in the coding layer for local feature extraction. Meanwhile, the sub-aperture wavelet aberration, adjustment error, and environmental noise parameters are input into the second branch sub-layer to obtain the global features of the sub-aperture wavelet aberration and adjustment error under the influence of environmental noise;

[0045] The results of local feature extraction and global features are input into the deconvolution network for feature fusion. At the same time, the fused features are input into the back-propagation RBF network to obtain the predicted adjusted wave phase difference, the predicted adjustment error and the adjustment error-wave aberration distortion mapping of the RBF function fitting;

[0046] The predicted adjusted wave aberration is obtained by subtracting the predicted adjustment error from the actual sub-aperture wave aberration.

[0047] A Rich Common detection system based on global power wave aberration prediction, comprising: a modeling module and a data enhancement module;

[0048] The modeling module establishes a three-dimensional optical model based on the preset standard spherical mirror, the plane mirror to be measured, and the interferometer lens parameter information in combination with the preset three-dimensional coordinate system;

[0049] The data enhancement module obtains an enhanced training data space based on a three-dimensional optical model combined with a preset first Richter angle and the defocus and astigmatism coefficients corresponding to the plane mirror, through a Monte Carlo simulation sampling algorithm combined with preset random noise.

[0050] Specifically, the Rich Common detection system also includes an error adjustment module and a plane mirror calculation module;

[0051] The error adjustment module includes a first adjustment unit, a first splicing unit and a second adjustment and splicing unit;

[0052] The first adjustment unit obtains first sub-aperture phase adjustment data based on the measured enhanced training data space in combination with a preset phase error countermeasure adjustment model and a preset adjustment error-waveform aberration distortion mapping;

[0053] The phase error adversarial adjustment model is obtained by combining K groups of first training set sequences and second training set sequences with an adversarial network algorithm for training;

[0054] The adjustment error-wave aberration distortion mapping is obtained by fitting the convolution radial function in the coding layer of the adversarial network algorithm with the wave aberration data of the sub-aperture overlapping area in the enhanced training data space and the interferometer adjustment error;

[0055] The first stitching unit obtains first full-aperture measurement data through a phase map stitching algorithm based on the first sub-aperture phase adjustment data;

[0056] The second adjustment and splicing unit adjusts the first Richey angle and repeats the first full-aperture measurement data acquisition process to obtain second full-aperture measurement data;

[0057] The plane mirror calculation module obtains the surface shape corresponding to the plane mirror to be measured based on the first full-aperture measurement data and the second full-aperture measurement data.

[0058] Compared with the prior art, the present invention has the following beneficial effects:

[0059] In response to the shortcomings of the existing technology, the present invention achieves global decoupling of the error coupling relationship and high-precision surface reconstruction in Richey-Common detection by integrating three-dimensional optical modeling, Monte Carlo noise simulation and adversarial network optimization algorithm. Specifically, based on the physical constraint space and noise linear function space constructed by the parameter distribution function and the finite element analysis to quantify the correlation effect of the adjustment parameters, the enhanced training data generated by the Monte Carlo simulation can more realistically reflect the complex noise coupling in actual detection; through the dual-branch feature extraction structure in the phase error adversarial adjustment model, the local phase characteristics of the coding layer are deconvolved with the global characteristics of the environmental noise, and the RBF network is used to fit the distortion mapping relationship between the adjustment error and the wave aberration, which effectively separates the coupling effect of the mechanical adjustment error and the real surface error; at the same time, the graph algorithm is used to construct the sub-aperture stitching topology, and the stitching weight is dynamically calculated by the phase difference in the overlapping area. Combined with the iterative optimization of the adversarial network loss function, the low-frequency accumulation of the adjustment error in the stitching process is suppressed, and the high-frequency surface details are retained through the nonlinear correction of the sub-aperture phase map; the present invention breaks through the limitations of the traditional method that relies on iterative adjustment through the collaborative optimization of the physical model and the data-driven method, and simultaneously realizes error compensation and phase stitching in a single measurement, which significantly improves the detection efficiency and repeatability. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 This is a flow chart of a Richcomon detection method based on global power wave aberration prediction according to the present invention;

[0061] Figure 2 This is a module diagram of a Richcomon detection system based on global power wave aberration prediction according to the present invention. DETAILED DESCRIPTION

[0062] Example 1

[0063] See also Figure 1 The present invention provides an embodiment of a Richmond-Commons detection method based on global power wave aberration prediction, comprising the following steps:

[0064] S1, establishing a three-dimensional optical model based on a preset standard spherical mirror, a plane mirror to be measured, and interferometer lens parameter information in combination with a preset three-dimensional coordinate system;

[0065] Furthermore, the process of establishing the three-dimensional optical model in this embodiment includes:

[0066] The interferometer focus parameter is used as the origin of the preset three-dimensional coordinate system, and the plane parallel to the plane mirror to be measured is used as the XOY plane of the three-dimensional coordinate system, and the optical path direction is used as the Z axis to construct the preset three-dimensional coordinate system;

[0067] Configure the preset three-dimensional coordinate system into the optical simulation software, and configure the standard spherical mirror aperture parameters according to the plane mirror aperture to be measured;

[0068] Input the configured standard spherical mirror aperture parameter set properties, the plane mirror properties to be measured, and the interferometer lens parameter information into the optical simulation, and establish an initial three-dimensional optical model based on the preset three-dimensional coordinate system;

[0069] Obtain historical adjustment parameter information corresponding to the standard spherical mirror, the plane mirror to be measured, and the interferometer lens parameter information ,in represents the defocus corresponding to the plane mirror, represents the 90° astigmatism coefficient corresponding to the plane mirror, It represents the interferometer adjustment error, that is, the deviation of the distance from the interferometer focus to the plane mirror to be measured. represents the power spectrum density of the vibration noise in the detection environment, Indicates the temperature drift value corresponding to the test environment;

[0070] Based on the historical adjustment parameter information, the parameter distribution function corresponding to each adjustment parameter is obtained through statistical analysis software;

[0071] Based on the parameter distribution function corresponding to each adjustment parameter and the corresponding attribute information and detection accuracy requirements of the standard spherical mirror, the plane mirror to be measured and the interferometer, the physical constraint space corresponding to the standard spherical mirror, the plane mirror to be measured and the interferometer is set;

[0072] Furthermore, the physical constraint space corresponding to the standard spherical mirror, the plane mirror to be measured, and the interferometer in this embodiment is Specifically:

[0073] ;

[0074] in, represents the coefficient of thermal expansion, f represents the focal length of the interferometer; NA represents the numerical aperture, Indicates wavelength of light; defocus represents the wavefront error caused by the axial position deviation of the plane mirror to be measured, and D represents the aperture of the plane mirror to be measured. It should be further explained that the parameters in the above formula in this embodiment are all dimensionally standardized.

[0075] Constraint principle: Based on the Rayleigh criterion, the wavefront error caused by defocus is limited to within a quarter of the wavelength to ensure that the imaging is close to the diffraction limit; the larger the numerical aperture (NA), the more sensitive the system is to defocus, so the allowable defocus range needs to be reduced in inverse proportion to the square of the NA.

[0076] Astigmatism Characterizes the focus separation error in the orthogonal direction caused by the measured mirror or installation tilt.

[0077] Constraint principle: The astigmatism error must be smaller than the measured mirror shape accuracy. In this embodiment, it is set to λ / 10 to retain a safety margin.

[0078] Adjustment error Indicates the distance deviation from the interferometer focus to the mirror.

[0079] Constraint principle: The maximum allowable deviation is set to a1% of the interferometer focal length, which not only ensures the visibility of the interference fringes but also matches the mechanical travel limit of the translation stage.

[0080] Reason for setting: To balance optical performance and hardware feasibility, and to prevent mechanical over-limit from causing adjustment failure.

[0081] Deformation Overlay Value parameter Represents the total deformation of the mirror caused by the combined effects of gravity and thermal expansion.

[0082] Constraint principle: The upper limit of deformation is set to one two thousandth of the diameter of the mirror being measured. This standard is derived from the surface tolerance requirements of high-precision optical components.

[0083] Reason for setting: Verify the support structure and thermodynamic model through finite element analysis to ensure that the deformation distribution meets the optical performance indicators.

[0084] Temperature drift value Represents thermally induced deformation excitation caused by ambient temperature fluctuations.

[0085] Constraint principle: The temperature variation range is limited to ±b1°C, matching the laboratory's active temperature control capabilities and suppressing dimensional drift caused by thermal expansion.

[0086] Reason for setting: To avoid unpredictable mirror distortion caused by nonlinear temperature changes and to reduce the risk of thermal stress fatigue.

[0087] The obtained physical constraint space is configured into the attribute information corresponding to the virtual standard spherical mirror, the plane mirror to be measured, and the interferometer in the initial three-dimensional optical model;

[0088] Furthermore, the dynamic constraints corresponding to the above variables are used for:

[0089] Temperature change rate: Limit the temperature rise rate to ≤c1°C / minute to ensure quasi-static heat conduction and prevent transient stress from damaging the mirror body.

[0090] Vibration noise: High-frequency vibration energy must be lower than the interferometer's anti-vibration threshold to suppress random jitter interference on phase measurement.

[0091] Adjustment speed: The interferometer translation stage movement speed should be ≤ h1 mm / s to avoid mechanical shock causing instability in the adjustment. The values ​​of a1, b1, c1, and h1 should be determined by those skilled in the art based on actual conditions.

[0092] Furthermore, the coupling effect constraints corresponding to the above variables are used to:

[0093] Defocus-astigmatism synthesis: The total error of the vector synthesis of the two must be less than one-eighth of the wavelength to prevent the composite error from exceeding the system tolerance.

[0094] Temperature-deformation coupling: The superposition of thermal expansion and gravitational deformation must satisfy the total deformation constraint to avoid the uncontrolled combined effects of multiple physical fields.

[0095] Vibration-alignment error: The root mean square sum of vibration noise and alignment error must be less than the tolerance to suppress error accumulation in dynamic environments.

[0096] At the same time, under the parameter distribution function corresponding to each adjustment parameter, the finite element analysis algorithm is used to obtain 、 Respectively The correlation coefficient of each parameter in and Corresponding superimposed deformation parameters right The correlation influence coefficient of each parameter in;

[0097] based on 、 Respectively The correlation coefficient of each parameter in and Corresponding superimposed deformation parameters right The correlation influence coefficient of each parameter in is used to obtain the noise linear function space through linear algorithm;

[0098] The noise linear function space is given by 、 Respectively The noise linear function of each parameter in and The noise linear function of each parameter in is constructed;

[0099] Furthermore, in this embodiment, the functions and meanings of each linear function in the linear function space are:

[0100] The linear relationship between deformation superposition and defocus is used to quantify the direct impact of mirror deformation due to gravity or thermal forces on defocus. This means that when the mirror bends due to support structure deformation or temperature fluctuations, its position along the optical axis shifts, causing defocus. This linear relationship indicates that with each increase in deformation, defocus increases proportionally, helping to predict the need for adjustment of mirror position deviations. Its application is that during assembly and adjustment, by monitoring deformation values, defocus can be estimated and compensated in real time to ensure clear interference fringes.

[0101] The linear relationship between temperature drift and defocus: This function reveals how temperature fluctuations alter the mirror's position or curvature through thermal expansion, thereby causing defocus. This means that as temperature rises, the mirror material expands, causing the mirror's curvature to change and the focus position to shift backward; the opposite occurs when temperature decreases. The linear coefficient reflects the sensitivity of the material's thermal expansion coefficient to defocus. Its application is to dynamically correct defocus errors using temperature sensor data in unstable temperature control environments, improving measurement stability.

[0102] The linear relationship between deformation superposition and astigmatism is used to assess the contribution of asymmetric mirror deformation to astigmatism. This means that when a mirror is distorted by gravity or uneven clamping, differences in curvature in different directions can lead to astigmatism. This linear relationship indicates that the greater the deformation, the more pronounced the astigmatism, necessitating support optimization to mitigate uneven deformation distribution. This can be applied to support structure design by predicting deformation distribution through finite element analysis and optimizing it to reduce astigmatism.

[0103] The linear relationship between temperature drift and astigmatism is used to analyze the impact of anisotropic mirror expansion caused by temperature gradients on astigmatism. This means that uneven temperature distribution causes different regions of the mirror to expand to varying degrees, resulting in curvature differences in orthogonal directions and, consequently, astigmatism. The linear coefficient reflects the strength of the correlation between temperature gradient and astigmatism. Applications include preheating the system before testing to reduce temperature gradients or incorporating temperature compensation algorithms into data processing.

[0104] The linear relationship between deformation superposition and adjustment error describes how mirror deformation indirectly causes deviations in the interferometer's focal position. This means that mirror deformation can change the relative position between the mirror surface and the interferometer, causing focus misalignment. This linear relationship quantifies the degree to which deformation interferes with alignment accuracy. Its application is to improve alignment efficiency by using deformation monitoring data to reversely correct interferometer position.

[0105] The linear relationship between temperature drift and adjustment error measures the focus position deviation caused by thermal deformation of the mechanical structure due to temperature changes. This means that thermal expansion and contraction of the interferometer bracket or mirror body will change the actual distance between the focus and the mirror surface. The linear coefficient reflects the impact of temperature changes on mechanical stability. Applications include using low-expansion materials for key components or suppressing thermal drift through closed-loop temperature control systems.

[0106] The noise linear function space is built into the corresponding attribute information of the virtual standard spherical mirror, the plane mirror to be measured and the interferometer in the initial three-dimensional optical model to obtain a three-dimensional optical model with noise.

[0107] S2, based on the 3D optical model combined with the preset first Richter angle and the defocus and astigmatism coefficients corresponding to the plane mirror, obtains an enhanced training data space through a Monte Carlo simulation sampling algorithm combined with preset random noise;

[0108] Furthermore, the process of acquiring the enhanced training data space in this embodiment includes:

[0109] exist In this case, set the initial defocus , initial astigmatism coefficient and the number of scans preset based on the standard spherical mirror aperture parameters; further, in this embodiment, when Under ideal conditions, there is no influence of temperature, deformation and temperature-deformation coupling;

[0110] Scan times M: M = (1.5N) × 1.5NM = (1.5N) × 1.5N, where N is the standard spherical mirror aperture;

[0111] Based on the three-dimensional optical model with noise, the initial defocus, the initial astigmatism coefficient, the standard spherical mirror aperture parameters, the preset number of scans and the number of sub-apertures, the position of the plane mirror to be measured is moved in the XOY plane to scan each preset sub-aperture;

[0112] When scanning each sub-aperture, a preset interferometer adjustment error sequence is combined to obtain a first wave of phase difference training data set formed by wave aberration data of different sub-apertures;

[0113] Adjust the initial defocus and initial astigmatism coefficient and repeat the above steps. Combined with the preset interferometer adjustment error sequence during each scan of each sub-aperture, a second wave of phase difference training data set formed by wave aberration data of different sub-apertures is obtained.

[0114] exist In this case, based on the first-wave phase difference training data set and the second-wave phase difference training data set, the first-wave phase difference training data set and the second-wave phase difference training data set are randomly divided into layers by a random division algorithm to obtain a first-wave phase difference segmented training data set and a second-wave phase difference segmented training data set;

[0115] Furthermore, in this embodiment, when In actual conditions, there are the effects of temperature, deformation and temperature-deformation coupling;

[0116] Based on the first-wave phase difference segmented training data set and the second-wave phase difference segmented training data set, random sampling is performed in each first-wave phase difference segmented training data set or the second-wave phase difference segmented training data set using a Latin hypercube sampling algorithm combined with a preset sampling number K-2 to obtain K-2 sets of phase difference random sampling sequences; further, in this embodiment, K is a positive integer greater than 2;

[0117] Feeding back K-2 groups of random sampling sequences of wave phase differences to the three-dimensional optical model with noise, performing random noise injection simulation on each group of random sampling sequences of wave phase differences through the noise linear function space and the physical constraint space, and obtaining K-2 groups of random sampling sequences of wave phase differences after noise injection;

[0118] Based on the K-2 groups of random sampling sequences of wave phase differences after noise injection, the enhanced training data space is obtained by combining the first wave phase difference training data set and the second wave phase difference training data set.

[0119] S3, obtaining first sub-aperture phase adjustment data based on the measured enhanced training data space in combination with a preset phase error countermeasure adjustment model and a preset adjustment error-waveform aberration distortion mapping;

[0120] The phase error adversarial adjustment model is obtained by combining K groups of first training set sequences and second training set sequences with an adversarial network algorithm for training;

[0121] The adjustment error-wave aberration distortion mapping is obtained by fitting the convolution radial function in the coding layer of the adversarial network algorithm with the wave aberration data of the sub-aperture overlapping area in the enhanced training data space and the interferometer adjustment error;

[0122] S4, obtaining first full-aperture measurement data through a phase map stitching algorithm based on the first sub-aperture phase adjustment data;

[0123] Furthermore, the process of acquiring the first full-aperture measurement data in this embodiment includes:

[0124] Inputting the sub-aperture wavelet aberration, adjustment error, and environmental noise parameters in the enhanced training data space into the coding layer of the phase error adversarial adjustment model to obtain adjusted sub-aperture wavelet phase data, predicted adjustment error, and adjustment error-wavelet aberration distortion mapping;

[0125] The adjusted sub-aperture wave difference phase data, the predicted adjustment error and the actual sub-aperture wave aberration obtained by measurement are input into the decoding layer in the phase error adversarial adjustment model, and the error between the wave phase difference of the adjusted sub-aperture and the actual sub-aperture wave aberration and the loss function constructed by the adjustment error-wave aberration distortion mapping are combined for training, and the adjusted sub-aperture wave difference phase data and the coverage relationship corresponding to all adjusted sub-aperture wave difference phase data are output.

[0126] The adjusted wave difference phase data of each sub-aperture is modeled as a sub-aperture graph node, and a splicing edge is established based on the coverage relationship corresponding to all the adjusted wave difference phase data of the sub-aperture. At the same time, the difference weight of the splicing edge is constructed by the phase difference of the overlapping area of ​​two adjacent sub-apertures.

[0127] The modeled sub-aperture graph nodes and the corresponding stitching edges and stitching edge difference weights are input into the graph algorithm to obtain the sub-aperture stitching graph and the stitching phase difference loss corresponding to the overlapping area of ​​two adjacent sub-apertures. The sub-aperture stitching graph and the stitching phase difference loss are fed back into the decoding layer in the phase error adversarial adjustment model for training to obtain the trained phase error adversarial adjustment model, and the first full-aperture measurement data after stitching is output.

[0128] Furthermore, the detailed process of implementing sub-aperture image modeling and stitching in this embodiment includes:

[0129] Obtain the node features corresponding to each sub-aperture i to construct a mosaic graph node, such as the Zernike coefficient of the adjusted wave aberration; the sub-aperture center coordinates (x, y); and the coverage weight;

[0130] The edge connection weight is constructed using the overlapping area of ​​sub-aperture i and sub-aperture j. If the overlapping area ratio of sub-aperture i and sub-aperture j is greater than the preset ratio threshold, an undirected edge is established.

[0131] Furthermore, the edge weight in this embodiment is calculated by the RMS value of the phase difference in the overlapping area;

[0132] Graph algorithms are used to combine subapertures with edge weights. After three layers of message passing, the global context subaperture node features are fused. The output layer of the graph algorithm then maps the node features into a full-aperture phase map. Bicubic interpolation is used to fuse adjacent subaperture data for phase smoothing in the overlapping region.

[0133] Furthermore, the process of acquiring the adjusted sub-aperture wave difference phase data in this embodiment includes:

[0134] The sub-aperture wavelet aberration and adjustment error are converted into a single-channel phase map and input into the first branch sub-layer of the coding layer for local feature extraction. At the same time, the sub-aperture wavelet aberration, adjustment error, and environmental noise parameters are input into the second branch sub-layer to obtain the global features of the sub-aperture wavelet aberration and adjustment error under the influence of environmental noise.

[0135] The results of local feature extraction and global features are input into the deconvolution network for feature fusion. At the same time, the fused features are input into the back-propagation RBF network to obtain the predicted adjusted wave phase difference, the predicted adjustment error and the adjustment error-wave aberration distortion mapping of the RBF function fitting;

[0136] Furthermore, the more detailed process of acquiring the adjusted sub-aperture wave difference phase data in this embodiment includes:

[0137] Through normalization, the 256×256 phase map, the adjustment error scalar, and the 6-dimensional environmental noise parameter are input into the local feature branch and the global feature branch, respectively, and a 64×64×128 local feature map and a 512-dimensional global feature vector are output. Furthermore, in this embodiment, the local feature branch preferably uses 3D convolution to extract high-frequency details, and the global feature branch preferably uses Transformer to model long-range dependencies.

[0138] After weighted concatenation of the local feature map and the global vector, a linear projection layer is used to enforce physical constraints and a truncation function is used to limit the output range, ensuring that the feature map conforms to the system linear relationship.

[0139] The fusion features are modeled based on the node RBF network initialized by K-means. The wavelet aberration adjustment value, error prediction and regional sensitivity distortion mapping are generated through nonlinear sensitivity learning to guide the optimization of local compensation strategy.

[0140] The predicted adjusted wavefront aberration is obtained by subtracting the predicted adjustment error from the true subaperture wavefront aberration.

[0141] S5, adjust the first Richey angle, repeat the first full-aperture measurement data acquisition process, obtain second full-aperture measurement data, and simultaneously obtain the surface shape corresponding to the plane mirror to be measured based on the first full-aperture measurement data and the second full-aperture measurement data.

[0142] This process achieves error analysis and global optimization by constructing a three-dimensional optical simulation model and an enhanced data generation strategy, combined with a physically constrained neural network architecture. The specific implementation path includes: first, establishing a multi-physics field model of thermal-mechanical-optical coupling to convert mirror deformation and temperature changes into optical parameter corrections; second, using a parameter space sampling method to generate simulation data covering mechanical adjustment errors and environmental noise, and simulating the wavefront distortion characteristics under industrial field conditions through a dynamic perturbation model; further designing an adversarial network structure with physical mechanism constraints, and using nonlinear function mapping to achieve feature decoupling of adjustment errors and optical distortion; finally, introducing a graph neural network framework, and constructing an iterative optimization mechanism for sub-aperture splicing through node features and edge weight definitions; this technical system drives data generation through physical simulation, achieves error decoupling through deep learning, and ensures global consistency through graph structure optimization, forming a systematic solution for optical detection for complex working conditions.

[0143] Example 2

[0144] See also Figure 2 , another embodiment provided by the present invention: a Rich Common detection system based on global power wave aberration prediction, comprising: a modeling module, a data enhancement module, an error adjustment module and a plane mirror calculation module;

[0145] The modeling module establishes a three-dimensional optical model based on the preset standard spherical mirror, the plane mirror to be measured, and the interferometer lens parameter information in combination with the preset three-dimensional coordinate system;

[0146] The data enhancement module obtains an enhanced training data space based on a three-dimensional optical model combined with a preset first Richter angle and the defocus and astigmatism coefficients corresponding to the plane mirror, using a Monte Carlo simulation sampling algorithm combined with preset random noise;

[0147] The error adjustment module includes a first adjustment unit, a first splicing unit and a second adjustment and splicing unit;

[0148] The first adjustment unit obtains first sub-aperture phase adjustment data based on the measured enhanced training data space in combination with a preset phase error countermeasure adjustment model and a preset adjustment error-waveform aberration distortion mapping;

[0149] The phase error adversarial adjustment model is obtained by combining K groups of first training set sequences and second training set sequences with an adversarial network algorithm for training;

[0150] The adjustment error-wave aberration distortion mapping is obtained by fitting the convolution radial function in the coding layer of the adversarial network algorithm with the wave aberration data of the sub-aperture overlapping area in the enhanced training data space and the interferometer adjustment error;

[0151] The error adjustment module is used for adjusting the phase difference of the sub-aperture waves and splicing the phases;

[0152] The first stitching unit obtains first full-aperture measurement data through a phase map stitching algorithm based on the first sub-aperture phase adjustment data;

[0153] The second adjustment and splicing unit adjusts the first Richey angle and repeats the first full-aperture measurement data acquisition process to obtain second full-aperture measurement data;

[0154] The plane mirror calculation module obtains the surface shape corresponding to the plane mirror to be measured based on the first full-aperture measurement data and the second full-aperture measurement data.

[0155] Example 3

[0156] An electronic device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, a Richmond-Commons detection method based on global power wave aberration prediction is implemented.

[0157] A computer-readable storage medium stores computer instructions, which, when executed, execute a Richmond-Commons detection method based on global power wave aberration prediction.

[0158] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in this field can also change, modify, replace and modify the above-mentioned embodiments without departing from the scope of protection of the purpose of the present invention and the claims, and all of these are protected by the present invention.

Claims

1. A Richcomon detection method based on global power wave aberration prediction, characterized in that: include: A three-dimensional optical model is established based on the preset standard spherical mirror, the plane mirror to be measured, and the interferometer lens parameter information combined with the preset three-dimensional coordinate system; Based on the three-dimensional optical model combined with the preset first Richter angle and the defocus and astigmatism coefficients corresponding to the plane mirror, an enhanced training data space is obtained through the Monte Carlo simulation sampling algorithm combined with preset random noise; Obtaining first sub-aperture phase adjustment data based on the measured enhanced training data space in combination with a preset phase error countermeasure adjustment model and a preset adjustment error-waveform aberration distortion mapping; The phase error adversarial adjustment model is obtained by training K groups of first training set sequences and second training set sequences in combination with an adversarial network algorithm; The adjustment error-wave aberration distortion mapping is obtained by fitting the convolution radial function in the coding layer of the adversarial network algorithm with the wave aberration data of the sub-aperture overlapping area in the enhanced training data space and the interferometer adjustment error; Obtaining first full-aperture measurement data through a phase map stitching algorithm based on the first sub-aperture phase adjustment data; Adjust the first Richey angle, repeat the first full-aperture measurement data acquisition process, and obtain the second full-aperture measurement data; Based on the first full-aperture measurement data and the second full-aperture measurement data, a surface shape corresponding to the plane mirror to be measured is obtained.

2. The Richcomon detection method based on global power wave aberration prediction according to claim 1, wherein: The process of establishing the three-dimensional optical model includes: The interferometer focus parameter is used as the origin of the preset three-dimensional coordinate system, and the plane parallel to the plane mirror to be measured is used as the XOY plane of the three-dimensional coordinate system, and the optical path direction is used as the Z axis to construct the preset three-dimensional coordinate system; Configure the preset three-dimensional coordinate system into the optical simulation software, and configure the standard spherical mirror aperture parameters according to the plane mirror aperture to be measured; The configured standard spherical mirror aperture parameter set properties, the properties of the plane mirror to be measured, and the interferometer lens parameter information are input into the optical simulation, and an initial three-dimensional optical model is established based on the preset three-dimensional coordinate system.

3. A Rich Common detection method based on global power wave aberration prediction as claimed in claim 2, characterized in that: The process of establishing the three-dimensional optical model further includes: Obtain historical adjustment parameter information corresponding to the standard spherical mirror, the plane mirror to be measured, and the interferometer lens parameter information ,in represents the defocus corresponding to the plane mirror, represents the 90° astigmatism coefficient corresponding to the plane mirror, It represents the interferometer adjustment error, that is, the deviation of the distance from the interferometer focus to the plane mirror to be measured. represents the power spectrum density of the vibration noise in the detection environment, Indicates the temperature drift value corresponding to the test environment; Based on the historical adjustment parameter information, the parameter distribution function corresponding to each adjustment parameter is obtained through statistical analysis software; Based on the parameter distribution function corresponding to each adjustment parameter and the corresponding attribute information and detection accuracy requirements of the standard spherical mirror, the plane mirror to be measured and the interferometer, the physical constraint space corresponding to the standard spherical mirror, the plane mirror to be measured and the interferometer is set; Configuring the acquired physical constraint space into the attribute information corresponding to the virtual standard spherical mirror, the plane mirror to be measured, and the interferometer in the initial three-dimensional optical model; At the same time, under the parameter distribution function corresponding to each adjustment parameter, the finite element analysis algorithm is used to obtain right The correlation coefficient of each parameter in and Corresponding superimposed deformation parameters right The correlation influence coefficient of each parameter in; Based on the correlation influence coefficient, the noise linear function space is obtained through linear algorithm; The noise linear function space is given by and The noise linear function of each parameter in and The noise linear function of each parameter in is constructed; The noise linear function space is built into the corresponding attribute information of the virtual standard spherical mirror, the plane mirror to be measured and the interferometer in the initial three-dimensional optical model to obtain a three-dimensional optical model with noise.

4. A Rich Common detection method based on global power wave aberration prediction as claimed in claim 3, characterized in that: The process of acquiring the enhanced training data space includes: exist In this case, set the initial defocus , initial astigmatism coefficient The number of scans is preset based on the standard spherical mirror aperture parameters; Based on the three-dimensional optical model with noise, the initial defocus, the initial astigmatism coefficient, the standard spherical mirror aperture parameters, the preset number of scans and the number of sub-apertures, the position of the plane mirror to be measured is moved in the XOY plane to scan each preset sub-aperture; When scanning each sub-aperture, a preset interferometer adjustment error sequence is combined to obtain a first wave of phase difference training data set formed by wave aberration data of different sub-apertures; Adjust the initial defocus and initial astigmatism coefficient and repeat the above steps. Combined with the preset interferometer adjustment error sequence during each scan of each sub-aperture, a second wave of phase difference training data set formed by wave aberration data of different sub-apertures is obtained.

5. The Richcomon detection method based on global power wave aberration prediction according to claim 4, wherein: The process of acquiring the enhanced training data space further includes: exist In this case, based on the first-wave phase difference training data set and the second-wave phase difference training data set, the first-wave phase difference training data set and the second-wave phase difference training data set are randomly divided into layers by a random division algorithm to obtain a first-wave phase difference segmented training data set and a second-wave phase difference segmented training data set; Based on the first-wave phase difference segmentation training data set and the second-wave phase difference segmentation training data set, the Latin hypercube sampling algorithm is combined with the preset sampling number K-2 to perform random sampling in each first-wave phase difference segmentation training data or the second-wave phase difference segmentation training data to obtain K-2 groups of phase difference random sampling sequences; Feeding back K-2 groups of random sampling sequences of wave phase differences to the three-dimensional optical model with noise, performing random noise injection simulation on each group of random sampling sequences of wave phase differences through the noise linear function space and the physical constraint space, and obtaining K-2 groups of random sampling sequences of wave phase differences after noise injection; Based on the K-2 groups of random sampling sequences of wave phase differences after noise injection, the enhanced training data space is obtained by combining the first wave phase difference training data set and the second wave phase difference training data set.

6. The Richcomon detection method based on global power wave aberration prediction according to claim 5, characterized in that: The process of acquiring the first full-aperture measurement data includes: Inputting the sub-aperture wavelet aberration, adjustment error, and environmental noise parameters in the enhanced training data space into the coding layer of the phase error adversarial adjustment model to obtain adjusted sub-aperture wavelet phase data, predicted adjustment error, and adjustment error-wavelet aberration distortion mapping; The adjusted sub-aperture wave difference phase data, the predicted adjustment error and the actual sub-aperture wave aberration obtained by measurement are input into the decoding layer in the phase error adversarial adjustment model, and the error between the wave phase difference of the adjusted sub-aperture and the actual sub-aperture wave aberration and the loss function constructed by the adjustment error-wave aberration distortion mapping are combined for training, and the adjusted sub-aperture wave difference phase data and the coverage relationship corresponding to all adjusted sub-aperture wave difference phase data are output.

7. The Richcomon detection method based on global power wave aberration prediction according to claim 6, wherein: The process of acquiring the first full-aperture measurement data further includes: The adjusted wave difference phase data of each sub-aperture is modeled as a sub-aperture graph node, and a splicing edge is established based on the coverage relationship corresponding to all the adjusted wave difference phase data of the sub-aperture. At the same time, the difference weight of the splicing edge is constructed by the phase difference of the overlapping area of ​​two adjacent sub-apertures. The modeled sub-aperture graph nodes and the corresponding stitching edges and stitching edge difference weights are input into the graph algorithm to obtain the sub-aperture stitching graph and the stitching phase difference loss corresponding to the overlapping area of ​​two adjacent sub-apertures. The sub-aperture stitching graph and the stitching phase difference loss are fed back into the decoding layer in the phase error adversarial adjustment model for training to obtain the trained phase error adversarial adjustment model, and the first full-aperture measurement data after stitching is output.

8. The Richcomon detection method based on global power wave aberration prediction according to claim 7, wherein: The process of acquiring the adjusted sub-aperture wave difference phase data includes: The sub-aperture wavelet aberration and adjustment error are converted into a single-channel phase map and input into the first branch sub-layer in the coding layer for local feature extraction. Meanwhile, the sub-aperture wavelet aberration, adjustment error, and environmental noise parameters are input into the second branch sub-layer to obtain the global features of the sub-aperture wavelet aberration and adjustment error under the influence of environmental noise; The results of local feature extraction and global features are input into the deconvolution network for feature fusion. At the same time, the fused features are input into the back-propagation RBF network to obtain the predicted adjusted wave phase difference, the predicted adjustment error and the adjustment error-wave aberration distortion mapping of the RBF function fitting; The predicted adjusted wave aberration is obtained by subtracting the predicted adjustment error from the actual sub-aperture wave aberration.

9. A Richcomon detection system based on global power wave aberration prediction, which is used to implement a Richcomon detection method based on global power wave aberration prediction according to any one of claims 1 to 8, characterized in that: include: Modeling module and data enhancement module; The modeling module establishes a three-dimensional optical model based on the preset standard spherical mirror, the plane mirror to be measured, and the interferometer lens parameter information in combination with the preset three-dimensional coordinate system; The data enhancement module obtains an enhanced training data space based on a three-dimensional optical model combined with a preset first Richter angle and the defocus and astigmatism coefficients corresponding to the plane mirror, through a Monte Carlo simulation sampling algorithm combined with preset random noise.

10. The Richcomon detection system based on global power wave aberration prediction according to claim 9, characterized in that: The Richcomang detection system also includes an error adjustment module and a plane mirror calculation module; The error adjustment module includes a first adjustment unit, a first splicing unit and a second adjustment and splicing unit; The first adjustment unit obtains first sub-aperture phase adjustment data based on the measured enhanced training data space in combination with a preset phase error countermeasure adjustment model and a preset adjustment error-waveform aberration distortion mapping; The phase error adversarial adjustment model is obtained by training K groups of first training set sequences and second training set sequences in combination with an adversarial network algorithm; The adjustment error-wave aberration distortion mapping is obtained by fitting the convolution radial function in the coding layer of the adversarial network algorithm with the wave aberration data of the sub-aperture overlapping area in the enhanced training data space and the interferometer adjustment error; The first stitching unit obtains first full-aperture measurement data through a phase map stitching algorithm based on the first sub-aperture phase adjustment data; The second adjustment and splicing unit adjusts the first Richey angle and repeats the first full-aperture measurement data acquisition process to obtain second full-aperture measurement data; The plane mirror calculation module obtains the surface shape corresponding to the plane mirror to be measured based on the first full-aperture measurement data and the second full-aperture measurement data.

Images