An intelligent calibration system for optical lens production

Abstract

The present invention belongs to the technical field of lens calibration, and specifically discloses an intelligent calibration system for optical lens production. The system includes: an environmental coupling module, an optical detection module, an adaptive calibration decision-making module, and an automatic focusing adjustment module. The present invention collects various environmental data in real time through the environmental coupling module to form a spatio-temporal correlation parameter set, providing a basis for suppressing environmental interference; the optical detection module generates a dynamic target sequence and synchronously obtains various types of lens parameters, which helps to dynamically optimize the calibration path; the adaptive calibration decision-making module can not only establish the mapping relationship between the environment and optical distortion, but also calculate the focusing adjustment parameters based on multiple parameters when the error exceeds the threshold; the automatic focusing adjustment module controls the focusing device in real time according to the adjustment parameters and verifies, realizing the sorting and re-calibration of qualified and unqualified lenses, comprehensively improving the calibration accuracy, and enhancing the scene adaptability of the AF algorithm.

Description

Technical Field

[0001] The present invention belongs to the technical field of lens calibration, and more particularly, relates to an intelligent calibration system for optical lens production. Background Art

[0002] With the continuous development of technology, optical products are increasingly widely used in various fields. Consumer electronic products such as smartphones, digital cameras, security monitoring, and automotive cameras have a continuous growing demand for optical lenses, and there are also more stringent requirements for the accuracy and performance of optical lenses. In order to meet the large market demand for high-quality optical lenses, the calibration link in the production process needs to be more precise.

[0003] Existing technologies such as the lens calibration method and system disclosed in the Chinese invention patent application with the application number 201710373514.5, in which the lens calibration method is applied to the lens calibration system, involving the lens calibration method and system, emphasizing automated control to improve efficiency and accuracy.

[0004] Existing technologies such as the optical lens, camera module, and their production management method disclosed in the Chinese invention patent application with the application number 201910226653.4, regarding the production management of optical lenses, trace component information through identification patterns for quality traceability.

[0005] In view of the above technical solutions, obviously, there are still the following deficiencies in the current optical lens production calibration: 1. There is a lack of a real-time compensation mechanism for interferences such as temperature fluctuations and mechanical vibrations during the calibration process, and the ability to suppress environmental interferences is weak, resulting in the calibration results being affected by environmental interferences and the accuracy deteriorating.

[0006] 2. The existing calibration path planning depends on fixed logic such as preset lifting time sequences and cannot dynamically optimize the motion trajectory based on real-time detection data such as specular reflectivity and distortion coefficients, resulting in a certain lack in the improvement rate of calibration efficiency.

[0007] 3. Currently, the calibration focuses on the mechanical accuracy of the lens such as optical axis alignment and distortion correction, but does not incorporate the dynamic response characteristics of the AF module such as focusing speed and stepping motor accuracy into the calibration system, resulting in the AF performance parameters such as focusing time and accuracy not being quantitatively controlled.

[0008] 4. Existing calibrations mostly use static targets and lack dynamic target simulations, unable to verify the scene adaptability of AF algorithms, and thus the improvement of the optical lens calibration efficiency is not obvious enough. Summary of the Invention

[0009] In view of this, to solve the problems raised in the above background art, an intelligent calibration system for optical lens production is proposed.

[0010] The object of the present invention can be achieved by the following technical solutions: The present invention provides an intelligent calibration system for optical lens production, which includes: an environmental coupling module that collects temperature, particle concentration, and vibration data through a multi-source environmental sensor array to generate a spatio-temporally correlated set of environmental parameters.

[0011] An optical detection module that generates a dynamic multi-spectral target sequence and simultaneously obtains geometric parameter data and optical distortion parameter data of the lens.

[0012] An adaptive calibration decision module, including an error traceability unit and a focusing interference modeling unit.

[0013] The error traceability unit, based on the set of environmental parameters and the optical distortion parameter data, establishes a dynamic mapping relationship between the environmental parameters and the optical distortion through a dual-branch neural network technology, and outputs the correlation degree of optical calibration error and the corresponding causal environmental parameter items.

[0014] The focusing interference modeling unit, when the correlation degree of calibration error exceeds its set threshold, calculates the adjustment parameters required for focusing based on the set of environmental parameters and the geometric parameter data.

[0015] An automatic focusing adjustment module that controls the focusing device in real time according to the adjustment parameters, and simultaneously performs focusing verification, sorts qualified lenses, and recalibrates unqualified lenses.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) The present invention collects various environmental data in real time through the environmental coupling module to form a spatio-temporally correlated parameter set, providing a basis for suppressing environmental interference. The optical detection module generates a dynamic target sequence and simultaneously obtains various types of parameters of the lens, which helps to dynamically optimize the calibration path. The adaptive calibration decision module can not only establish the mapping relationship between the environment and the optical distortion, but also calculate the focusing adjustment parameters based on multiple parameters when the error exceeds the threshold. The automatic focusing adjustment module controls the focusing device in real time according to the adjustment parameters and verifies, realizing the sorting and recalibration of qualified and unqualified lenses, comprehensively improving the calibration accuracy, and enhancing the scene adaptability of the AF algorithm.

[0017] (2) The present invention collects temperature, particle concentration, and vibration data through a multi-source environmental sensor array to generate a spatio-temporally correlated set of environmental parameters. It can perceive the environmental conditions in real time and comprehensively, providing a rich and accurate data basis for subsequent analysis of the impact of environmental factors on the optical system, effectively making up for the defects of untimely and incomplete detection of environmental interference in the prior art, and improving the ability to cope with environmental interference.

[0018] (3) The present invention breaks the limitation of the traditional static target by generating a dynamic multi-spectral target sequence, and can simulate complex dynamic scenarios in actual applications and verify the scene adaptability of the AF algorithm.

[0019] (4) By obtaining the geometric parameter data and optical distortion parameter data of the lens, the present invention provides comprehensive and real-time lens status information for the calibration process, which helps to dynamically adjust the calibration path according to the actual situation of the lens, solves the problem that the existing calibration path planning depends on fixed logic, and significantly improves the calibration efficiency.

[0020] (5) By using the dual-branch neural network technology, the present invention establishes a dynamic mapping relationship based on the environmental parameter set and the optical distortion parameter data, outputs the correlation degree of optical calibration error and the corresponding environmental parameter items of the causative factors, can accurately locate the influence of environmental factors on the optical calibration error, and then can take targeted measures to suppress environmental interference, solving the problems of weak environmental interference suppression ability and poor calibration accuracy in the existing technology.

[0021] (6) By calculating the adjustment parameters required for focusing, the present invention takes into account the dynamic response characteristics of the AF module, fills the gap that the existing calibration system does not consider the dynamic characteristics of the AF module, and realizes the quantitative control of AF performance parameters such as focusing time and accuracy.

[0022] (7) By controlling the focusing device in real time and performing focusing verification, sorting qualified lenses and recalibrating unqualified lenses, the present invention realizes the automatic closed-loop control of the calibration process, improves the accuracy and efficiency of calibration, and changes the situation of untimely and inaccurate treatment of unqualified lenses in the existing calibration process. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings without creative efforts based on these drawings.

[0024] Figure 1 It is a schematic diagram of the system module structure process of the present invention.

[0025] Figure 2 It is a schematic diagram of the method implementation step process of the present invention.

[0026] Figure 3 It is a schematic diagram of the composition structure of the adaptive calibration decision module of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0027] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0028] Please refer to Figure 1 and Figure 2 As shown, the present invention provides an intelligent calibration system for optical lens production, which includes: an environmental coupling module, an optical detection module, an adaptive calibration decision module, and an autofocus adjustment module.

[0029] Among the above, the adaptive calibration decision module is respectively connected to the environmental coupling module, the optical detection module, and the autofocus adjustment module.

[0030] In the present invention, the calibration of optical lens production involves initial detection to determine the current state of the lens. Then parameter adjustment is carried out, such as adjusting the focal length, optical axis alignment, etc. Next, autofocus is performed to ensure clear imaging. Finally, quality verification is carried out to ensure compliance with the standards. The environmental coupling module and the optical detection module are applied to the calibration detection process of the initial detection. By collecting various environmental data in real time to form a spatio-temporal correlation parameter set, it provides a basis for suppressing environmental interference. The optical detection module generates a dynamic target sequence and synchronously obtains various types of lens parameters, which helps to dynamically optimize the calibration path. The adaptive calibration decision module is applied to the calibration detection process of parameter adjustment. By calculating the adjustment parameters required for focusing and outputting the correlation degree of optical calibration error and the corresponding environmental parameter items of the cause, it can take targeted measures to suppress environmental interference and fill the gap that the existing calibration system does not consider the dynamic characteristics of the AF module. The autofocus adjustment module is applied to the calibration detection process of autofocus. By controlling the focusing device in real time and performing focusing verification, it sorts qualified lenses and recalibrates unqualified lenses. It realizes the automatic closed-loop control of the calibration process.

[0031] The embodiment of the present invention collects various environmental data in real time through the environmental coupling module to form a spatio-temporal correlation parameter set, providing a basis for suppressing environmental interference. The optical detection module generates a dynamic target sequence and synchronously obtains various types of lens parameters, which helps to dynamically optimize the calibration path. The adaptive calibration decision module can not only establish the mapping relationship between the environment and optical distortion, but also calculate the focusing adjustment parameters based on multiple parameters when the error exceeds the threshold. The autofocus adjustment module controls the focusing device in real time according to the adjustment parameters and verifies, realizes the sorting and recalibration of qualified and unqualified lenses, comprehensively improves the calibration accuracy, and enhances the scene adaptability of the AF algorithm.

[0032] The environmental coupling module collects temperature, particulate concentration, and vibration data through a multi-source environmental sensor array, and generates a spatio-temporally correlated set of environmental parameters.

[0033] Specifically, the multi-source environmental sensor array consists of a distributed temperature sensor, a laser scattering particulate detector, and a three-axis vibration accelerometer. The three-axis vibration accelerometer, temperature sensor, and laser scattering particulate detector are deployed at key workstations on the assembly line to collect vibration, temperature, and particulate concentration data in real time at a sampling rate of ≥1 kHz.

[0034] More specifically, generating the spatio-temporally correlated set of environmental parameters includes: performing time synchronization and alignment on the temperature, particulate concentration, and vibration data, and eliminating noise through sliding window mean filtering, and outputting the processed environmental parameters.

[0035] Extract the spatio-temporal features of the processed environmental parameters through a convolutional LSTM network, and output a set of environmental parameters including the temperature gradient range, the PM2.5-PM10 concentration ratio, and the main frequency vibration energy.

[0036] It can be understood that performing time synchronization and alignment on the temperature, particulate concentration, and vibration data can use the IEEE1588 Precision Clock Protocol to perform millisecond-level time synchronization on the temperature, particulate concentration, and vibration data. The sliding window length can specifically be 5-10 production cycle periods. The convolutional LSTM network has three layers. The first layer has a convolutional kernel size of 5×5 and a stride of 2. The second layer has a convolutional kernel of 3×3 and a dilation factor of 2. The third layer has a convolutional kernel of 1×1, and outputs a feature matrix including the temperature gradient range, the PM2.5-PM10 concentration ratio, and the main frequency vibration energy as the set of environmental parameters.

[0037] In the embodiment of the present invention, the multi-source environmental sensor array collects temperature, particulate concentration, and vibration data, and generates a spatio-temporally correlated set of environmental parameters. It can perceive the environmental conditions in real time and comprehensively, providing a rich and accurate data basis for subsequent analysis of the impact of environmental factors on the optical system, effectively making up for the deficiencies of untimely and incomplete detection of environmental interference in the prior art, and improving the ability to respond to environmental interference.

[0038] The optical detection module generates a dynamic multi-spectral target sequence, and synchronously obtains the geometric parameter data and optical distortion parameter data of the lens.

[0039] It can be understood that the dynamic multi-spectral target sequence is generated by a dynamic target generator, where the dynamic target generator is configured with a tunable LED array and can generate a sine stripe pattern containing 6 groups of characteristic frequencies in the range of 380-950 nm.

[0040] It can also be understood that obtaining the geometric parameter data of the lens reflects the change in the physical shape of the lens, mainly including axial offset, radial distortion, thickness deviation, curvature error, etc. It can be detected by a laser interferometer. The optical distortion parameters describe the aberrations and distortions introduced by the lens during the imaging process, which directly affect the imaging quality, such as spherical aberration, coma, astigmatism, field curvature, distortion, etc. It can be detected by using a wavefront sensor, such as a Hartmann-Shack wavefront sensor.

[0041] In the embodiment of the present invention, by obtaining the geometric parameter data and optical distortion parameter data of the lens, comprehensive and real-time lens state information is provided for the calibration process, which helps to dynamically adjust the calibration path according to the actual situation of the lens, solves the problem that the existing calibration path planning depends on fixed logic, and significantly improves the calibration efficiency.

[0042] The embodiment of the present invention also breaks the limitation of the traditional static target by generating a dynamic multi-spectral target sequence, and can simulate complex dynamic scenes in actual applications and verify the scene adaptability of the AF algorithm.

[0043] Please refer to Figure 3 As shown, the adaptive calibration decision module includes an error traceability unit and a focus interference modeling unit.

[0044] The error traceability unit, according to the environmental parameter set and the optical distortion parameter data, establishes a dynamic mapping relationship between the environmental parameters and the optical distortion through the dual-branch neural network technology, and outputs the optical calibration error correlation degree and the corresponding causal environmental parameter items.

[0045] In the embodiment of the present invention, by using the dual-branch neural network technology to establish a dynamic mapping relationship according to the environmental parameter set and the optical distortion parameter data, and outputting the optical calibration error correlation degree and the corresponding causal environmental parameter items, it can accurately locate the influence of environmental factors on the optical calibration error, and then can take targeted measures to suppress environmental interference, solving the problems of weak environmental interference suppression ability and poor calibration accuracy in the prior art.

[0046] Specifically, the dual-branch neural network technology mainly consists of an environmental branch network, an optical branch network, a multi-head attention mechanism, an error coefficient regressor, and a causal parameter projection layer. Among them, the environmental branch network inputs an environmental parameter set and extracts temporal features, outputting an environmental feature matrix. The optical branch network is composed of a convolutional network, inputs optical distortion parameter data and extracts spatial features, outputting an optical feature matrix. The multi-head attention mechanism dynamically calculates the correlation weight matrix between the environmental feature matrix and the optical feature matrix to generate a fused feature matrix. Exemplarily, its embedding dimension is 96, which can comprehensively process the encoded environmental features and optical parameter features. The number of heads is 4, and the attention weights are calculated in parallel by multiple heads to enhance the model's ability to capture different feature information, enabling the model to better focus on the mutual relationships between different features. The error coefficient regressor is based on the fused feature matrix and regresses and outputs the optical calibration error correlation degree, which is a sequential model composed of a linear layer and a Sigmoid activation function. The input dimension of the linear layer is 96, which matches the output dimension of the cross-modal attention mechanism, and the output dimension is 1, used to output the optical calibration error correlation degree. The Sigmoid activation function maps the output of the linear layer to the interval from 0 to 1 to obtain a normalized optical calibration error coefficient as the optical calibration error correlation degree. The causal parameter projection layer screens out the environmental parameters with a correlation degree higher than a preset threshold as causal parameters through weight matrix mapping. Among them, each row of the correlation weight matrix corresponds to an environmental parameter, and each column corresponds to an optical distortion parameter.

[0047] It should be added that the loss function of the dual-branch neural network is a weighted combination of the Huber loss and the KL divergence. Among them, the Huber loss is used for optical calibration error coefficient regression, and the KL divergence is used to constrain the consistency between the causal weight distribution and prior knowledge.

[0048] Among them, the specific execution process of the error traceability unit includes: U1. Use a bidirectional LSTM network to extract the temporal features of the environmental parameter set to generate an environmental parameter sequence. At the same time, the input layer of the bidirectional LSTM network receives the environmental parameter sequence, and after two layers of Bi-LSTM, an environmental feature matrix is generated and used as the input of the environmental branch network.

[0049] U2. Use a convolutional layer to extract the spatial features of the optical distortion parameters, and combine the lens curvature radius and the assembly torque value to output an optical feature matrix as the input of the optical branch network.

[0050] U3. Use the multi-head attention mechanism to perform cross-modal fusion of the environmental feature matrix and the optical feature matrix, calculate the correlation weight matrix between the two, and then output an environmental-optical fusion feature matrix.

[0051] U4. Based on the fused feature matrix, regression outputs the correlation values between each environmental parameter and the optical distortion parameter, and filters out the maximum value as the correlation degree of the optical calibration error.

[0052] U5. Set an environmental correlation threshold, compare the correlation value of each environmental parameter with the environmental correlation threshold, and filter out the environmental parameters with a correlation higher than the preset threshold as the leading environmental parameter items.

[0053] Further, the cross-modal fusion of the environmental feature matrix and the optical feature matrix in step U3 includes setting a fusion ratio, and the specific settings are as follows: set the environmental weight matrix and the optical weight matrix according to the historical detection environmental parameter set during the production calibration of the optical lens and the optical lens calibration detection report data respectively, and perform an initialization process on the environmental weight matrix and the optical weight matrix.

[0054] Multiply the environmental feature matrix and the optical feature matrix by the corresponding weight matrices respectively, and record the output results as the environmental feature weighted score and the optical feature weighted score respectively.

[0055] Record the sum of the environmental feature weighted score, the optical feature weighted score, and the preset bias as the comprehensive score of the environment and the optical feature, and use it as the input of the Sigmoid function, and then output the fusion ratio.

[0056] Understandably, by linearly combining the environmental features and the optical features, adding a bias term, and then activating through the Sigmoid function, a value between 0 and 1 is obtained as the fusion ratio of the two features. This method belongs to the gate control mechanism. It can dynamically adjust the contribution degrees of the two features in the final fusion, rather than a fixed ratio, so as to more flexibly adapt to the changes in the feature importance in different scenarios.

[0057] It should be explained that the gate control mechanism can map the environmental features and the optical features to the same latent space through a linear transformation, compress them to the interval [0, 1] through a weighted sum and then through the Sigmoid function, and generate a dynamic fusion ratio. This method dynamically adjusts the fusion weights of the cross-modal features through learnable parameters, not only retains the temporal dependence of the environmental parameters and the spatial characteristics of the optical parameters, but also enhances the expression ability of the model through non-linear activation. At the same time, experiments show that compared with the fixed ratio fusion strategy, the prediction error is significantly reduced, and at the same time, the robustness of the leading parameter screening is significantly improved, providing a highly generalized solution for the real-time calibration of the optical system.

[0058] It is also understandable that in the production and calibration process of optical lenses, various types of data, features, or sensor information are often used. For example, when detecting the imaging quality of a lens, both optical imaging features need to be considered, and it may also be combined with environmental data such as temperature and humidity to analyze their impact on lens parameters. Different types of data have different importance and contribution degrees to calibration, so it is necessary to set a fusion ratio to make a reasonable trade-off. By reasonably setting the fusion ratio, the advantages of different data or features can be comprehensively utilized to make up for the limitations of single data or features. For example, when calibrating a lens in a complex environment, by combining environmental data and optical imaging data and appropriately setting the fusion ratio, the calibration process can more comprehensively consider various factors, enabling the lens to reach a better performance state and meet production standards and actual usage requirements. In terms of calibration accuracy, a reasonable fusion ratio can significantly improve the accuracy of calibration and reduce errors. For example, in the production of some precision optical lenses, by fusing multiple sensor data and optimizing the fusion ratio, the calibration accuracy can be improved to the micron level or even higher. In terms of stability, setting an appropriate fusion ratio can enhance the stability of the lens under different environments and usage conditions, reducing performance fluctuations caused by changes in single factors.

[0059] Further, the environmental correlation threshold in step U5 is a dynamic threshold, and its specific setting process is as follows: For the th input sample, a normalized attention weight matrix with a dimension of is generated through the multi-head attention mechanism, represents the total number of time steps of the environmental parameter sequence, represents the dimension of the optical feature matrix output by the optical branch network.

[0060] Perform a global average calculation on all elements in the attention weight matrix , and record the calculation result as .

[0061] Calculate the degree of dispersion of the elements in the attention weight matrix relative to the through the standard deviation formula, and mark it as .

[0062] Generate a dynamic threshold according to and , , represents the setting of a dynamic adjustment factor, is a pre-set upper limit value for environmental correlation setting.

[0063] In a specific embodiment, the preset upper limit value of the environmental correlation degree varies according to different production scenarios, lens types, and calibration requirements. For ordinary consumer-grade optical lenses, the upper limit value of the environmental correlation degree may be between 0.6 and 0.8. For high-precision professional optical lenses, since they are more sensitive to environmental factors, the upper limit value may be relatively higher, between 0.8 and 0.9. Based on past practical experience in optical lens production calibration and similar projects, combined with the correlation between common environmental parameters and lens performance, a suitable upper limit value can be directly given. For example, after multiple tests, it is found that the correlation degree between certain environmental parameters and lens imaging quality rarely exceeds 0.8, so the upper limit value can be set to 0.8.

[0064] In another specific embodiment, is a constant, and its specific value can be dynamically adjusted according to factors such as the complexity of the current task, such as the severity of environmental changes in the lens calibration task, and the change trend of historical data, such as the fluctuation of recent calibration errors. In this way, the threshold for screening causal parameters can better adapt to different situations, improve the accuracy and adaptability of screening. Exemplarily, it can take a value of 0 in a stable environment, 0.5 in a severely fluctuating environment, and 0.2 in a periodically changing environment.

[0065] It can be understood that in actual production, environmental parameters such as temperature change at all times, and lenses have different sensitivities to parameters in different environments. The dynamic threshold can be adjusted in real time according to environmental changes to accurately screen out the causal parameters that truly affect lens performance. Therefore, in complex environments such as sudden temperature changes and mechanical vibration interference, the dynamic threshold can flexibly distinguish noise from true correlations. Exemplarily, dynamic verification examples are given for input samples in three scenarios: stable environment, severely fluctuating environment, and periodically changing environment, as shown in Table 1 specifically.

[0066] Table 1 Dynamic verification example

[0067]

[0068] According to Table 1, when in a stable environment, the threshold is low, and time segments with medium correlation intensity are retained, which is suitable for scenarios with small environmental fluctuations to ensure the effectiveness of causal parameter screening. When in a severely fluctuating environment, the threshold increases significantly, so that high-noise correlations introduced by sudden environmental changes can be effectively filtered, and only strongly significant time segments are retained. When in a periodically changing environment, the threshold can adapt to the periodic changes, avoiding misjudging periodic fluctuations as abnormal causal factors and improving stability.

[0069] Preferably, establishing the dynamic mapping relationship between the environmental parameters and the optical distortion in the present invention further includes setting a compensation controller, and the specific setting process is as follows: H1. Generate a temperature field distribution matrix based on the temperature data, calculate the temperature gradient, and update the change amount of the lens spacing in a discrete integral form.

[0070] Exemplarily, arrange N temperature sensors such as PT100 and thermocouples around the optical system to obtain the position coordinates and the real-time temperature , where represents the temperature sensor number, , and represents the time sequence number. .

[0071] Construct a temperature field matrix through Kriging interpolation , which satisfies , where represents the weight coefficient of the th temperature sensor, , is the spatial correlation scale, representing the spatial smoothness of the temperature field, usually taken as 1 / 2 of the sensor spacing, is the natural constant, is the reference input position coordinate.

[0072] Calculate the temperature gradient through the finite element heat conduction model , , where represents the transpose symbol, and represents the execution of partial derivative calculation.

[0073] Update the change amount of the lens spacing in a discrete integral form , and its specific calculation formula is: , where represents the coefficient of thermal expansion of the material, such as for fused silica / °C, represents the characteristic dimension of the lens, such as the diameter or side length, represents the sampling time interval, which needs to satisfy the Nyquist frequency, for example, a sampling rate of 1 kHz corresponds to Δt = 1 ms, represents the total number of cumulative sampling time points.

[0074] H2. Extract the main vibration frequency and the amplitude from the vibration data, fit the exponential term based on the vibration energy decay curve, and simultaneously generate the vibration compensation amount using the defocus amount prediction equation.

[0075] Among them, the fitting formula for fitting the exponential term based on the vibration energy decay curve is: , where represents the th The amplitude at a time step, Indicates that the damping ratio is dimensionless and is determined by fitting experimental data using the least squares method, Is the natural constant, Indicates the initial amplitude, which is determined by the vibration shock intensity, Is the phase angle, which determines the starting position of the waveform.

[0076] Furthermore, the defocus equation is: , Indicates the defocus at the th time step, Indicates the system vibration sensitivity coefficient, which is determined through a calibration experiment.

[0077] Furthermore, the compensation amount is generated by the reverse displacement of the piezoelectric ceramic actuator, and its specific calculation formula is: , Is the piezoelectric ceramic displacement sensitivity, such as 1.2 μm / V, Indicates the compensation amount at the th time step.

[0078] H3. Based on the change in lens spacing and the vibration compensation amount, set up the state space equation and predict the wavefront error in the next M time steps based on the state space equation.

[0079] Among them, the specific expression formula of the state space equation is: , where, Indicates the actual displacement of the lens at the th time step, Indicates the actual displacement of the lens at the th time step, Indicates the control input at the th time step. The control input is an externally applied drive or adjustment signal, such as force, voltage, etc., Indicates the wavefront error of the optical system at the th time step, Indicates the calibration coefficient, which is obtained through the linear regression of the Zernike coefficient and the RMS.

[0080] H4. Set the actuator displacement limit constraint condition and the response delay constraint condition, and solve the constrained quadratic programming problem for each time step to generate the optimal control sequence.

[0081] Understandably, the actuator displacement limit constraint condition is: , response delay ≤ 2 ms.

[0082] H5. Set the target compensation function, apply the first control quantity in the generated optimal control sequence, and re-optimize based on the target compensation function in the next time step.

[0083] Specifically, the target compensation function is specifically represented by the formula: , represents the tracking error term, which is used to minimize the deviation between and is the control increment term, which is used to minimize the mutation of the control input ; is the optimization variable, representing the control inputs at the first three moments that need to be optimized, such as t = 0, 1, 2, represents the control input at the th time step, is the weight coefficient of the tracking error term, which is used to adjust the sensitivity of the controller to the wavefront error. Exemplarily, the value is 0.7 and needs to be adjusted according to the system dynamic characteristics. is the weight coefficient of the control increment term, which is used to suppress the frequent and drastic actions of the actuator. Exemplarily, the value is 0.3.

[0084] When the calibration error correlation exceeds its set threshold, the focus interference modeling unit calculates the adjustment parameters required for focusing based on the environmental parameter set and geometric parameter data.

[0085] By calculating the adjustment parameters required for focusing, the embodiment of the present invention takes into account the dynamic response characteristics of the AF module, fills the gap in the existing calibration system that does not consider the dynamic characteristics of the AF module, and realizes the quantitative control of AF performance parameters such as focusing time and accuracy.

[0086] Specifically, calculating the adjustment parameters required for focusing includes: B1. Matching and comparing the environmental parameter items of the cause with the pre-set relationship table between environmental parameters and focusing adjustment items, and outputting the matching focusing adjustment items. The focusing adjustment items include axial compensation, radial compensation, and thermal expansion compensation.

[0087] B2. Setting the geometric deformation compensation amount based on the geometric parameters of the lens.

[0088] B3. Based on the focusing adjustment items, calculating the deviation values of the focusing adjustment items corresponding to the associated environmental parameter items of the cause, and inputting them into the Sigmoid function to output the environmental deformation compensation coefficient. Taking the product of the environmental deformation compensation coefficient and the pre-set environmental deformation setting compensation amount as the environmental deformation compensation amount.

[0089] B4. Weightedly superposing the environmental deformation compensation amount and the geometric deformation compensation amount to generate the final adjustment compensation amount, which is used as the adjustment parameter required for the corresponding focusing adjustment item.

[0090] Understandably, the relationship table between environmental parameters and focus adjustment items can be specifically shown in Table 2 as follows.

[0091] Table 2 Mapping Relationship Based on Environmental Parameters - Calibration Parameters - Verification Results

[0092]

[0093] It should be added that axial compensation is used to offset the deviation in the optical axis direction, radial compensation is used to correct the jitter perpendicular to the optical axis direction, thermal expansion / contraction compensation: dynamically adjusts the focus position according to temperature changes, surface cleaning compensation can trigger the cleaning device to remove contaminants on the lens surface, and anti-fog compensation activates the anti-fog device to prevent fogging on the lens surface.

[0094] It should be added that according to the causative environmental parameter items, such as vibration, temperature, and particle concentration, the corresponding focus adjustment items are found from the relationship table. If multiple factors are triggered simultaneously, such as high-frequency vibration and high temperature, a comprehensive adjustment parameter set is generated according to the priority.

[0095] Furthermore, in step B2, the geometric deformation compensation amount is set, including: B21. Extract the measured values of the axial offset, radial distortion, and curvature deviation from the geometric parameter data of the lens, compare them with the preset reference value intervals respectively, generate the deviation values of each parameter, and obtain the deviation values of each geometric parameter.

[0096] B22. Count the number of parameters whose deviation values exceed the preset threshold, perform a ratio operation with the total number of geometric parameters, and output the geometric parameter deviation coverage ratio.

[0097] B23. Standardize the absolute values of each deviation value and perform a weighted average operation according to the preset weight distribution rule to generate the geometric parameter deviation degree.

[0098] B24. Set the weights of the geometric parameter deviation coverage ratio and the geometric parameter deviation degree, obtain the geometric deformation compensation coefficient through weighted summation, and use the product of the geometric deformation compensation coefficient and the pre-set geometric deformation setting compensation amount as the geometric deformation compensation amount.

[0099] Exemplarily, the geometric parameter deviation coverage ratio is analyzed from the overall dimension, and the geometric parameter deviation degree is analyzed from the data dimension. Specifically, they can take values of 0.55 and 0.45 in sequence.

[0100] The autofocus adjustment module controls the focusing device in real time according to the adjustment parameters, while performing focus verification, sorting qualified lenses and recalibrating unqualified lenses.

[0101] Specifically, the focus verification includes: restarting the optical detection module again, and then outputting the geometric parameter data and optical distortion parameter data of each lens after focus adjustment.

[0102] Compare the adjusted geometric parameter data and optical distortion parameter data with their corresponding preset tolerance ranges respectively.

[0103] If the adjusted geometric parameters exceed the corresponding preset tolerance range, mark that the geometric parameter verification fails. If the adjusted optical distortion parameters exceed the corresponding preset tolerance range, mark that the optical distortion parameter verification fails.

[0104] If both the adjusted geometric parameters and optical distortion parameters are within their corresponding preset tolerance ranges, calculate the current environmental parameter data through the standard deviation, output the current environmental fluctuation degree, and when the current environmental fluctuation degree is lower than or equal to the set permitted fluctuation degree, mark it as calibration verification passed.

[0105] It should be added that when the current environmental fluctuation degree is higher than the set permitted fluctuation degree, the temperature / vibration can be continuously monitored, and the compensation amount can be updated in real time. Exemplarily, for example, when the temperature rises by 1°C → an additional compensation of +0.4μm.

[0106] It can be understood that the preset tolerance standards are specifically referred to as follows: 1) Geometric tolerance: Axial offset ≤ 0.1μm. Radial distortion ≤ 0.05°. Curvature deviation ≤ 0.2%.

[0107] 2) Optical tolerance: Wavefront aberration (RMS) ≤ 0.05λ. Coma (Z7 / Z8) ≤ 0.03λ. Astigmatism (Z5 / Z6) ≤ 0.04λ.

[0108] 3) Environmental threshold: Temperature fluctuation ≤ ±1°C. Vibration amplitude ≤ ±0.5μm.

[0109] Exemplarily, if the axial offset = 0.12μm (exceeding the limit of 0.1μm) → mark it as "geometric parameter verification fails", if the wavefront aberration RMS = 0.06λ (exceeding the limit of 0.05λ) → mark it as "optical distortion parameter verification fails".

[0110] Another specifically, the recalibration of the unqualified lens includes: projecting a single - spectrum target and collecting geometric parameters within a pre - set detection area.

[0111] If the geometric parameters meet the standards, mark them as qualified and enter the sorting process. If the geometric parameters do not meet the standards, trigger the pre - set primary calibration rule and perform calibration.

[0112] Project a multi - spectrum target to cover the entire area of the lens, monitor the environmental parameters in real time, and calculate the environmental parameter fluctuation degree through the standard deviation calculation formula.

[0113] If the environmental parameter fluctuation degree is within the set permitted range, mark the lens as qualified. Otherwise, trigger the pre - set secondary calibration rule and perform calibration.

[0114] The projection full-band spectral target detects the optical distortion parameters and the overall aberration of the lens for the corresponding central area, edge area, off-axis area, and focal area of the lens. If the optical distortion parameters of a certain area do not meet the standard or the overall aberration of the lens does not meet the standard, the lens is marked as abnormal, and the unqualified data is recorded.

[0115] Exemplarily, in a specific embodiment, the primary calibration rule is that the adjustment amplitude is increased by 20%. For example, the temperature compensation coefficient is increased from 0.4 μm / °C to 0.48 μm / °C, and when high-frequency vibration is detected, the vibration compensation priority is increased. The secondary calibration is that the adjustment amplitude is increased by 50%.

[0116] The embodiment of the present invention realizes the automatic closed-loop control of the calibration process by controlling the focusing device in real time, performing focusing verification, sorting qualified lenses, and recalibrating unqualified lenses, improving the accuracy and efficiency of calibration, and changing the situation of untimely and inaccurate processing of unqualified lenses in the existing calibration process.

[0117] The above content is only an example and illustration of the concept of the present invention. Those skilled in the art of the present technology can make various modifications or supplements to the described specific embodiments or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, they should fall within the protection scope of the present invention.

Claims

1. An intelligent calibration system for optical lens production, characterized in that The system includes the following modules: An environmental coupling module that collects temperature, particulate concentration, and vibration data through a multi-source environmental sensor array to generate a spatio-temporally correlated set of environmental parameters; An optical detection module that generates a dynamic multi-spectral target sequence and simultaneously obtains geometric parameter data and optical distortion parameter data of the lens; An adaptive calibration decision module that includes an error traceability unit and a focusing interference modeling unit; The error traceability unit establishes a dynamic mapping relationship between environmental parameters and optical distortion through a dual-branch neural network technology based on the set of environmental parameters and optical distortion parameter data, and outputs the correlation degree of optical calibration error and the corresponding causal environmental parameter items; The focusing interference modeling unit calculates the adjustment parameters required for focusing based on the set of environmental parameters and geometric parameter data when the correlation degree of calibration error exceeds its set threshold; An automatic focusing adjustment module that controls the focusing device in real time according to the adjustment parameters, and simultaneously performs focusing verification, sorts qualified lenses, and recalibrates unqualified lenses; The specific execution process of the error traceability unit includes: Using a bidirectional LSTM network to extract the temporal features of the set of environmental parameters to generate a sequence of environmental parameters. At the same time, the input layer of the bidirectional LSTM network receives the sequence of environmental parameters, and after two layers of Bi-LSTM, an environmental feature matrix is generated and used as the input of the environmental branch network; Using a convolutional layer to extract the spatial features of the optical distortion parameters, and combining the lens curvature radius and the assembly torque value to output an optical feature matrix, which is used as the input of the optical branch network; Using a multi-head attention mechanism to perform cross-modal fusion of the environmental feature matrix and the optical feature matrix, calculate the correlation weight matrix between the two, and then output an environmental-optical fusion feature matrix; Based on the fusion feature matrix, regression outputs the correlation degree values between each environmental parameter and the optical distortion parameter, and selects the maximum value as the correlation degree of optical calibration error; Set an environmental correlation degree threshold, compare the correlation degree value of each environmental parameter with the environmental correlation degree threshold, and select the environmental parameters with a correlation degree higher than the preset threshold as the causal environmental parameter items.

2. An intelligent calibration system for optical lens production according to claim 1, characterized in that: The generation of the spatio-temporally correlated set of environmental parameters includes: Performing time synchronization alignment on the temperature, particulate concentration, and vibration data, and eliminating noise through sliding window mean filtering to output the processed environmental parameters; Extracting the spatio-temporal features of the processed environmental parameters through a convolutional LSTM network to output a set of environmental parameters including the temperature gradient range, the PM2.5-PM10 concentration ratio, and the main frequency vibration energy.

3. An intelligent calibration system for optical lens production according to claim 1, characterized in that: The cross-modal fusion of the environmental feature matrix and the optical feature matrix includes setting a fusion ratio, and the specific settings are as follows: Setting an environmental weight matrix and an optical weight matrix respectively according to the detected environmental parameter set during the calibration of historical optical lens production and the optical lens calibration detection report data, and initializing the environmental weight matrix and the optical weight matrix; Multiplying the environmental feature matrix and the optical feature matrix by the corresponding weight matrices respectively, and recording the output results as the weighted score of the environmental features and the weighted score of the optical features; The sum of the weighted environmental feature score, the weighted optical feature score, and a preset bias is denoted as the comprehensive score of the environment and optical features, which serves as the input to the Sigmoid function, and then the fusion ratio is output.

4. An intelligent calibration system for optical lens production according to claim 1, characterized in that: The environmental correlation threshold is a dynamic threshold, and its specific setting process is as follows: For the th input sample, a normalized attention weight matrix of dimension is generated through the multi-head attention mechanism , denotes the total number of time steps of the environmental parameter sequence, denotes the dimension of the optical feature matrix output by the optical branch network; Perform a global average calculation on all elements in the attention weight matrix and denote the calculation result as ; Calculate the attention weight matrix through the standard deviation formula The elements in are relative to the said Degree of dispersion, and marked as ; According to and Generate a dynamic threshold , , Indicates setting a dynamic adjustment factor Set an upper limit value for the pre-set environmental relevance 5. An intelligent calibration system for optical lens production according to claim 1, characterized in that: The establishment of the dynamic mapping relationship between environmental parameters and optical distortion also includes setting a compensation controller, and its specific setting process is as follows: Generate a temperature field distribution matrix based on temperature data, calculate the temperature gradient, and update the change in lens spacing in the form of discrete integration; Extract the main vibration frequency and amplitude from vibration data, fit the exponential term based on the vibration energy decay curve, and simultaneously generate a vibration compensation amount using the defocus amount prediction equation; Based on the change in lens spacing and the vibration compensation amount, set up a state space equation, and predict the wavefront error in the next M time steps based on the state space equation; Set the actuator displacement limit constraint condition and the response delay constraint condition, solve the constrained quadratic programming problem for each time step, and generate an optimal control sequence; Set up an objective compensation function, apply the first control quantity in the generated optimal control sequence, and re-optimize based on the objective compensation function in the next time step.

6. The intelligent calibration system for optical lens production according to claim 1, characterized in that: The calculation of the adjustment parameters required for focusing includes: Match and compare the causal environmental parameter terms with a preset relationship table of environmental parameters and focusing adjustment terms, and output the matching focusing adjustment terms. The focusing adjustment terms include axial compensation, radial compensation, and thermal expansion compensation; Based on the geometric parameters of the lens, set the geometric deformation compensation amount; Based on the focusing adjustment terms, calculate the deviation values of the causal environmental parameters corresponding to the focusing adjustment terms, and input them into the Sigmoid function to output the environmental deformation compensation coefficient. Multiply the environmental deformation compensation coefficient by the preset environmental deformation setting compensation amount to obtain the environmental deformation compensation amount; Weightedly superimpose the environmental deformation compensation amount and the geometric deformation compensation amount to generate the final adjustment compensation amount, which serves as the adjustment parameter required for the corresponding focusing adjustment term.

7. An intelligent calibration system for optical lens production according to claim 6, characterized in that: The setting of the geometric deformation compensation amount includes: Extract the measured values of the axial offset, radial distortion, and curvature deviation from the geometric parameter data of the lens, compare them with the preset reference value intervals respectively, generate the deviation values of each parameter, and obtain the deviation values of each geometric parameter; Count the number of parameters whose deviation values exceed the preset threshold, perform a ratio operation with the total number of geometric parameters, and output the geometric parameter deviation coverage ratio; Perform standardization processing on the absolute values of each deviation value, and perform a weighted average operation according to the preset weight distribution rule to generate the geometric parameter deviation degree; Set the weights of the geometric parameter deviation coverage ratio and the geometric parameter deviation degree, obtain the geometric deformation compensation coefficient through weighted summation, and multiply it by the preset geometric deformation setting compensation amount to obtain the geometric deformation compensation amount.

8. An intelligent calibration system for optical lens production according to claim 1, characterized in that: The performance of focusing verification includes: Restart the optical detection module again, and then output the geometric parameter data and optical distortion parameter data of each lens after focusing adjustment; Compare the adjusted geometric parameter data and optical distortion parameter data with the corresponding preset tolerance ranges respectively; If the adjusted geometric parameters exceed the corresponding preset tolerance range, mark that the verification of geometric parameters fails. If the adjusted optical distortion parameters exceed the corresponding preset tolerance range, mark that the verification of optical distortion parameters fails. If both the adjusted geometric parameters and optical distortion parameters are within the corresponding preset tolerance ranges, calculate the current environmental parameter data through the standard deviation, and output the current environmental fluctuation. When the current environmental fluctuation is lower than or equal to the set permitted fluctuation, mark that the calibration verification passes.

9. An intelligent calibration system for optical lens production according to claim 1, characterized in that: The re-calibration of the unqualified lens includes: Project a single-spectrum target and collect the geometric parameters within the preset detection area. If the geometric parameters meet the standards, mark them as qualified and enter the sorting process. If the geometric parameters do not meet the standards, trigger the preset primary calibration rule and perform calibration. Project a multi-spectrum target to cover the entire area of the lens, monitor the environmental parameters in real time, and calculate the environmental parameter fluctuation through the standard deviation calculation formula. If the environmental parameter fluctuation is within the set permitted range, mark the lens as qualified. Otherwise, trigger the preset secondary calibration rule and perform calibration. Project a full-band spectrum target to detect the optical distortion parameters and the overall aberration of the lens in the corresponding central area, edge area, off-axis area, and focal area of the lens. If the optical distortion parameters in a certain area do not meet the standards or the overall aberration of the lens does not meet the standards, mark the lens as abnormal and record the non-compliant data.

Images