Additive manufacturing lens surface optimization method based on real-time feedback control

Abstract

The application relates to the technical field of additive manufacturing, and discloses an additive manufacturing lens surface optimization method based on real-time feedback control. First, the eye biometric data and personalized visual demand parameters of a wearer are acquired, and a personalized optical model of an eyeball is established and a retinal defocus distribution map is derived according to the parameters. The spatial arrangement and geometric shape parameters of a distributed defocus microstructure layer on a lens are determined according to the map. In the additive manufacturing process, an optical material is deposited and solidified layer by layer, and three-dimensional topographic information of the surface of each layer after deposition is acquired in real time. The topographic information is compared and analyzed with a theoretical model to generate topographic deviation results. Subsequently, a correction model is used to calculate correction instructions of printing parameters, and the printing parameters of the additive manufacturing equipment are adjusted in real time to realize real-time compensation and optimization of the topographic deviation, and finally a lens product with high precision and meeting the requirements of personalized optical design is manufactured.

Description

Technical Field

[0001] This invention relates to the field of additive manufacturing, specifically to a method for optimizing the surface of additive manufacturing lenses based on real-time feedback control. Background Technology

[0002] Traditional optical lens manufacturing has long relied on mold making and grinding / polishing processes. These methods are typically designed based on standardized human eye models, producing lenses that only offer basic refractive correction and cannot meet the diverse needs of individual wearers in different visual scenarios. As people's demands for visual quality increase, this mass production model is gradually revealing its limitations, especially in controlling peripheral defocus, reducing aberrations, and improving visual comfort.

[0003] In terms of personalized lens design, existing technologies have begun to explore optical optimization by incorporating the user's ocular biometrics. For example, by measuring data such as corneal curvature and pupillary distance, more accurate optometry and lens fitting can be achieved. However, such optimizations are mostly concentrated in the optical center area of ​​the lens, and effective control over astigmatism, field curvature, and defocus distribution at large field of view is still lacking. The fundamental reason is that traditional processing techniques struggle to efficiently and economically manufacture lens surfaces with complex freeform surfaces or micro / nanostructures, especially lenses containing numerous asymmetric, aspherical, or localized microstructures.

[0004] Additive manufacturing, or 3D printing, has brought new possibilities to the molding of complex optical components. This technology constructs objects by layering materials, theoretically enabling the manufacture of any shape. While research has attempted to use this technology to create optical lenses, several challenges remain in practical applications. Firstly, factors such as material curing shrinkage, ambient temperature fluctuations, and nozzle positioning errors during printing can introduce morphological deviations, leading to discrepancies between the final optical surface and the theoretical design, thus affecting image quality. Secondly, current additive manufacturing processes lack effective online monitoring and closed-loop control mechanisms. Typically, quality assessment is performed offline after the entire lens printing is complete, using methods such as interferometers and profilometers. If defective products are found, the entire unit must be scrapped, resulting in material waste and reduced production efficiency.

[0005] Therefore, although additive manufacturing offers process possibilities, ensuring that the printed optical surface accurately conforms to the preset optical model remains a major technical bottleneck hindering its widespread application in the field of high-precision optical manufacturing. Developing a manufacturing method capable of online monitoring, real-time feedback, and dynamic adjustment of printing parameters is essential for improving the performance and yield rate of additively manufactured optical lenses. Summary of the Invention

[0006] The purpose of this invention is to provide a method for optimizing the surface of additive manufacturing lenses based on real-time feedback control, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides a method for optimizing the surface of additive manufacturing lenses based on real-time feedback control, the method comprising:

[0008] Acquire the wearer's ocular biometric data and personalized visual needs parameters;

[0009] A personalized optical model of the eyeball was established using ocular biometric data and personalized visual requirement parameters, and the defocus distribution map of the retina under different visual angles was derived.

[0010] The spatial arrangement parameters and geometric morphological parameters of the distributed defocus microstructure layer are determined based on the defocus amount distribution map.

[0011] Optical materials are loaded into additive manufacturing equipment and cured by layer-by-layer deposition; during the deposition and curing of each layer, the three-dimensional morphology information of the surface of that layer is acquired in real time.

[0012] The real-time acquired 3D topography information is compared and analyzed with the pre-stored theoretical topography model to generate topography deviation results;

[0013] Based on the morphological deviation results, the correction instructions for printing parameters are calculated using the calibration model; the printing parameters of the additive manufacturing equipment are adjusted in real time according to the correction instructions.

[0014] Preferably, acquiring the wearer's ocular biometric data includes using a non-contact biometric instrument to collect corneal curvature, axial length, pupil diameter, and accommodative hysteresis. Personalized visual needs parameters are obtained through a standardized questionnaire, including the distribution of daily eye use scenarios, glare sensitivity, and preference for lens color change speed. The non-contact biometric instrument scans the anterior segment of the eyeball by emitting a low-coherence beam and reconstructs the corneal curvature and axial length based on the interference signal. The pupil diameter is measured as the equivalent circle diameter under standard lighting conditions using an infrared camera system. The accommodative hysteresis is recorded by using dynamic retinoscopy to record the hysteresis difference between the accommodative response curve and the ideal curve.

[0015] Preferably, a personalized optical model of the eyeball is established using a ray tracing algorithm, which simplifies the eyeball into a multi-layer optical system composed of the cornea, aqueous humor, lens, and vitreous body. The refractive index and thickness of each layer are assigned by ocular biometric data. The model input is the direction of parallel incident light rays at different field of view angles, and the output is the focal position of the light rays on the retina. The defocus amount is calculated by comparing the focal position with the ideal retinal plane. The defocus amount distribution map is stored in the form of a two-dimensional matrix, with the matrix row and column indices corresponding to the horizontal and vertical field of view angles, and the element value being the defocus amount for the corresponding field of view angle.

[0016] Preferably, determining the spatial arrangement parameters of the distributed defocus microstructure layer includes dividing the front surface of the lens into a hexagonal grid region, with the center point of each grid defined by latitude and longitude coordinates. The geometric parameters include the radius of curvature, sag, and edge transition slope of each microstructure unit. The radius of curvature is calculated using the thin lens formula based on the defocus distribution map, the sag is determined by the radius of curvature and the diameter of the microstructure bottom surface, and the edge transition slope is limited to a preset threshold range.

[0017] Preferably, the real-time acquisition of the three-dimensional morphology information of the surface of the layer is achieved by a laser confocal microscopy imaging module integrated inside the additive manufacturing equipment. The laser confocal microscopy imaging module emits a laser beam of a preset wavelength, performs grid scanning through a galvanometer system, receives reflected light signals, and reconstructs three-dimensional point cloud data. The scan covers the entire surface of the current layer, and the point cloud data is stored in the form of an array, with each point containing three-dimensional coordinate information.

[0018] Preferably, the real-time acquired 3D topographic information is compared and analyzed with the pre-stored theoretical topographic model, using a combination of morphological contour decomposition and dynamic waveform matching, including:

[0019] The real-time 3D topography data is decomposed into a series of continuous cross-sectional profile waveforms along the printing path. Each waveform consists of a timing signal composed of the height values ​​of each point on the cross section.

[0020] Simultaneously, the theoretical morphology model is decomposed into cross-sectional profiles in the same direction to generate a reference waveform sequence;

[0021] The real-time waveform and the reference waveform are dynamically time-normalized and aligned, and the root mean square value of the height difference of the corresponding points after alignment is calculated as the shape deviation result.

[0022] Preferably, the dynamic waveform matching process includes the following processing steps:

[0023] Wavelet transform is used to denoise the real-time acquired cross-sectional contour waveform, and key feature points in the waveform are extracted.

[0024] The denoised real-time waveform and the reference waveform are dynamically time-warped to find the optimal nonlinear alignment path.

[0025] Calculate the height difference between corresponding points of the two waveforms along the alignment path and statistically analyze the distribution characteristics of the height difference. When the root mean square of the height difference exceeds the threshold, it is determined that there is a shape deviation in the cross-sectional area, and the amplitude and spatial distribution of the deviation are recorded.

[0026] Preferably, the correction instructions for printing parameters are calculated using a correction model, and a decision architecture that integrates multiple algorithms is adopted.

[0027] The decision architecture executes two processing flows in parallel: genetic algorithm optimization and fuzzy logic reasoning. The genetic algorithm uses the spatial distribution characteristics of the shape deviation results as the initial population input and iteratively searches for the optimal ink droplet volume adjustment coefficient through selection, crossover, and mutation operations. At the same time, the fuzzy logic reasoning system derives the fuzzy decision result of the landing point coordinate offset based on the magnitude and distribution density of the shape deviation through a preset membership function and rule base.

[0028] Finally, the genetic algorithm output and fuzzy inference results are integrated through a weighted fusion module to generate the final correction instruction parameters.

[0029] Preferred implementations of the multi-algorithm fusion decision architecture include:

[0030] The genetic algorithm part uses real number encoding to represent the ink droplet volume adjustment coefficient, and the fitness function is defined as the similarity between the corrected predicted morphology and the theoretical morphology;

[0031] The fuzzy logic reasoning system establishes a three-dimensional input variable space, including three dimensions: the magnitude of the shape deviation, the area of ​​the deviation region, and the deviation gradient value.

[0032] The weighted fusion module adopts a dynamic weight allocation strategy, which adjusts the weight ratio of the genetic algorithm results and the fuzzy inference results alternately according to the parity of the current printed layer number. Odd-numbered layers take priority to use the genetic algorithm output, while even-numbered layers take priority to use the fuzzy inference results.

[0033] Preferably, the printing parameters of the additive manufacturing equipment are adjusted in real time according to the correction instructions, including adjusting the piezoelectric crystal driving voltage to change the ink droplet ejection volume and fine-tuning the position of the printhead in the horizontal and vertical directions to correct the landing point deviation. The adjustment operation is executed immediately in the next ejection cycle to ensure that the current layer topography deviation is compensated in real time.

[0034] Compared with the prior art, the beneficial effects of the present invention are:

[0035] The method proposed in this invention integrates the entire process from personalized optical design to real-time control of additive manufacturing, offering advantages in several aspects. First, the core of the method lies in fully considering the individual differences of the human eye as a complex optical system, moving away from reliance on a single standard eye model. By incorporating refined biological parameters, including accommodative hysteresis and pupillary dynamic response, and combining them with the user's subjective visual preferences, the established eye optical model can more realistically simulate the wearer's actual visual state. This results in a final defocus distribution map and its corresponding microstructure arrangement possessing strong personal attributes, effectively adapting to the visual needs of different users in different scenarios, and fundamentally improving lens fit and visual quality.

[0036] Introducing a real-time feedback control mechanism into the manufacturing process, by immediately scanning and comparing the 3D morphology after each layer of material deposition, allows for the timely detection of minute deviations caused by equipment fluctuations or changes in material properties. This online inspection avoids the accumulation of defects, nipping problems in the bud and significantly reducing the risk of subsequent repair work or overall scrap. Compared to the traditional method of quality inspection after printing, this significantly improves the first-pass yield and material utilization rate, while also reducing reliance on offline high-end inspection equipment. The morphology deviation analysis algorithm used in this invention possesses high intelligence and robustness. By decomposing the 3D morphology into cross-sectional contour waveforms and performing dynamic time warping, this method effectively overcomes interference caused by slight misalignment of the scanning starting point or local noise, accurately identifying true, optically significant morphology errors rather than measurement noise. This lays a reliable data foundation for subsequent precise correction.

[0037] The multi-algorithm fusion decision architecture designed in this invention exhibits powerful capabilities in handling nonlinear problems. Genetic algorithms excel at global optimization, while fuzzy logic reasoning is adept at handling empirical and uncertain problems. Running both in parallel and fusion with weights allows the correction model to both search for the optimal parameter combination from a large number of possible solutions and incorporate expert experience to make fast and robust decisions on specific types of deviations. This hybrid strategy enhances adaptability to various anomalies in complex printing processes, making correction instructions more scientific and effective. In summary, this method promotes the development of personalized optical lens manufacturing technology from multiple dimensions, including design accuracy, manufacturing efficiency, process reliability, and cost control. Attached Figure Description

[0038] Figure 1 This is a schematic diagram illustrating the working principle of the additive manufacturing lens surface optimization method based on real-time feedback control described in this invention.

[0039] Figure 2 A flowchart for obtaining ocular biometric data and personalized visual requirement parameters;

[0040] Figure 3 A flowchart for real-time acquisition of 3D topographic information;

[0041] Figure 4 A diagram analyzing the defocus distribution and compensation effect of the human eye;

[0042] Figure 5 This is a statistical analysis chart of surface morphology deviations. Detailed Implementation

[0043] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] Please see Figure 1 This invention provides a method for optimizing the surface of additively manufactured lenses based on real-time feedback control. Through a closed-loop control mechanism, it dynamically corrects morphological deviations during the manufacturing process, thereby achieving precise customization of the lens's optical performance. This method acquires the wearer's ocular biometric data and personalized visual requirement parameters. Subsequently, this model is used to derive a defocus distribution map of the retina at different field of view angles. This map directly determines the design parameters of the distributed defocus microstructure layer on the lens. During the manufacturing stage, optical materials are loaded into the additive manufacturing equipment and deposited and cured layer by layer. Crucially, during the deposition and curing of each layer, a detection system integrated into the equipment captures the three-dimensional morphological information of the surface of that layer in real time. The acquired real-time morphological data is compared and analyzed with a pre-stored theoretical morphological model with high precision, thereby generating quantified morphological deviation results. This deviation result is input into an intelligent correction model, which generates correction instructions for the current printing parameters through algorithm calculations. The additive manufacturing equipment immediately adjusts its printing parameters, such as ink droplet ejection characteristics or nozzle pose, based on these instructions, thereby compensating for the morphology of the current or subsequent layers being manufactured, ensuring that the surface morphology of the final product highly matches the optical design intent.

[0045] Example 1

[0046] Acquiring wearer ocular biometric data is accomplished using a high-performance, non-contact biometric instrument that integrates optical interferometry and infrared imaging technology. During operation, the instrument emits a safe, low-coherence near-infrared laser beam. This beam scans the anterior anatomical structures of the eyeball at a specific angle, including the anterior corneal surface and the anterior lens surface. When the beam contacts these optical interfaces, it is reflected. The reflected light waves interfere with a reference light wave inside the instrument. By calculating the phase difference and optical path difference of these interference signals, the system can accurately reconstruct the radius of curvature of the anterior corneal surface and the total axial length of the eyeball from the anterior surface to the retinal pigment epithelium. To obtain reliable pupil diameter data, measurements are performed in a dark room under standard lighting conditions. A high-resolution infrared imaging system continuously captures dynamic images of the eyeball. After the images are transmitted to the processing software, the algorithm automatically identifies the pupil edge and fits it to an equivalent circle, then calculates its diameter value. This value is the average of multiple measurements to eliminate the influence of instantaneous fluctuations. The measurement of accommodative lag relies on the principle of dynamic retinoscopy. In a controllable target traction system, the wearer fixates on a target that moves at a constant speed along the optical axis. At the same time, an infrared retinoscope is used to observe the movement of the reflected light and shadow on the retina. The curve of the actual accommodative response generated by the eyeball changes over time is recorded. This curve is compared with the ideal accommodative response curve that the moving stimulus should theoretically elicit. The systematic lag difference between the two in terms of time phase and amplitude is the required accommodative lag data.

[0047] The collection of personalized visual needs parameters aims to capture the wearer's subjective visual habits and preferences. These parameters are obtained through a standardized visual needs questionnaire containing structured questions across multiple dimensions. A survey of daily eye usage scenarios covers the time spent and visual task types in different scenarios, such as reading at work distance, using a computer at medium distance, driving long distances, and engaging in outdoor activities, thereby inferring the wearer's primary eye usage patterns and corresponding field of view. Glare sensitivity assessment uses a series of images simulating visual scenes with glare sources of varying intensities and angles, allowing wearers to report their discomfort levels and quantify their tolerance threshold to glare. Lens photochromic speed preference is directly related to the selection of photochromic lenses, asking wearers about their desired speed of lens darkening from a transparent state and returning to transparency from a dark state—whether they prefer rapid adaptation to environmental changes or a more gradual and uniform tone transition. All these collected biometric data and questionnaire responses are aggregated and encoded into a standardized data format, then transmitted securely to the subsequent modeling and computation unit through a data interface, forming the basic input parameters for the personalized optical model.

[0048] Creating a personalized optical model of the eye is a digital process based on the principles of geometric optics. Its core lies in simulating the real optical path using ray tracing algorithms. This model simplifies and abstracts the human eye into a system composed of multiple rotationally symmetric coaxial optical surfaces. These surfaces sequentially represent the anterior and posterior surfaces of the cornea and the lens, and divide the intraocular media into the aqueous humor (with uniform refractive index), the lens nucleus, and the vitreous humor. The refractive index and geometric parameters of each optical element, such as central corneal thickness, lens thickness, anterior chamber depth, and the radius of curvature of each surface, are directly assigned from the corresponding biometric data obtained in the preceding steps, thus constructing a unique computational model representing the optical characteristics of the wearer's eye. The ray tracing simulation of the model assumes parallel light incidence, simulating light rays emitted from an infinitely distant object. The direction of the incident light ray is defined by the field of view angle, covering the maximum effective peripheral field of view perceptible from directly in front of the eye (zero field of view) to the retina. The algorithm traces the path of each ray as it passes through the cornea, aqueous humor, lens, and vitreous humor, calculating its refraction at each optical interface until it finally intersects the retinal plane. By comparing the focal point of the ray calculated from the tracing with the axial position of the theoretically ideal retinal image plane, it can accurately calculate whether the retinal imaging plane is in a state of relative anterior or posterior defocus at each specific field of view angle, and obtain the specific amount of defocus, which is usually expressed in diopters. Finally, the system systematically organizes and stores the defocus calculation results of all sampling points in the entire field of view into a two-dimensional matrix. The row and column indices of this matrix correspond to the horizontal and vertical field of view angles on the standardized discrete grid, respectively. The value of each element in the matrix is ​​the defocus amount calculated under that specific combination of viewing angles. This defocus distribution map completely quantifies the optical defects of the wearer's eyeball in the entire field of view.

[0049] Example 2

[0050] Based on the defocus distribution map, the specific parameters of the distributed defocus microstructure layer on the lens are determined. This microstructure layer is designed within a ring-shaped area on the front surface of the lens to correspond to the defocus field of view of the human eye. The spatial arrangement design of the microstructure layer first maps the target processing area onto a virtual spherical coordinate system, and then systematically divides the area into countless closely adjacent hexagonal grid units according to latitude and longitude lines. This hexagonal tiling structure is considered to provide the most uniform distribution and the highest filling efficiency for a given area. The spatial position of the center point of each independent hexagonal grid is uniquely determined by its longitude and latitude values ​​on the sphere. This coordinate system provides a precise location address for each microstructure unit. After completing the spatial layout planning, specific geometric parameters need to be assigned to the microstructures within each grid unit. These parameters directly determine the optical performance of the microstructure, with the radius of curvature of its optical working surface being the most critical. This value is obtained through inversion calculations using the target defocus value at the corresponding latitude and longitude coordinates in the defocus distribution map, combined with the refractive index of the lens substrate and the preset microstructure substrate diameter, applying basic principles of optical design. This ensures that the microstructure can generate the required additional optical power to accurately compensate for the defocus at that field of view. The purpose of the inversion calculation is to accurately transform the macroscopic optical performance target into specific microscopic geometric parameters. The design system receives the defocus distribution map data transmitted from the personalized optical model, which clearly indicates the defocus value that needs to be compensated at a specific latitude and longitude coordinate position on the lens surface. For example, the system processes a hexagonal grid cell located in the temporal peripheral region of the lens, coded with latitude and longitude coordinates (Theta_i, Phi_j). The target defocus amount corresponding to this cell is labeled as -2.00 diopters. This negative value indicates that in this field of view, the imaging focus is located behind the retina, and the microstructure needs to generate an equivalent positive lens effect to move the focus forward onto the retina.

[0051] The design software accesses its internally stored material database, retrieves the refractive index parameters of the currently used optical lens substrate (assuming a value of 1.60), and loads the pre-defined substrate diameter design value for this type of microstructure, such as 50 micrometers. These parameters constitute the known conditions for the inversion calculation. The core objective of the calculation is to determine the spherical radius of curvature that produces an optical power of exactly +2.00. The calculation process is based on the fundamental principles of optics, namely that the optical power of a microlens is related to the refractive index, radius of curvature of its constituent materials, and the surrounding medium. The system solves this equation using a built-in iterative algorithm, calculating a specific value on the order of several millimeters for the theoretical radius of curvature required under these specific conditions.

[0052] After obtaining the theoretical value of the radius of curvature, the system immediately initiates a geometric constraint verification process. This process aims to ensure that the calculated geometry is manufacturable and smoothly integrates with the surrounding structure. The system automatically calculates the sag of the microlens, i.e., the maximum height of its center point, using the radius of curvature and a preset substrate diameter. The calculated sag value is immediately compared with multiple process thresholds set within the system. These thresholds include the maximum allowable sag (determined by layer printing accuracy and material flowability) and the minimum radius of curvature (limited by the curing capability of the light spot). If the calculated value exceeds the allowable range, the system triggers a parameter adjustment loop. A common adjustment strategy is to keep the optical power target unchanged while fine-tuning the substrate diameter. The system will attempt to slightly increase or decrease the substrate diameter value within the allowable range and re-execute the above calculation process. Through several rapid iterations, a set of radii of curvature and substrate diameter combinations that both meet the optical performance requirements (generating +2.00D optical power) and fully comply with all geometric manufacturing constraints (sag and curvature within the allowable range) is finally obtained. Next, the system addresses the transition slope of the microstructure edges. Slope control is crucial for ensuring visual comfort and preventing glare. Based on the final determined radius of curvature and substrate diameter, the design software calculates the theoretical slope value at the connection point between the microstructure edge and the substrate. This calculated value is also compared with pre-stored empirical slope thresholds in the database. If the edge slope is too steep, exceeding the upper limit of the threshold, the system initiates a smoothing process. This process may make very subtle aspherical adjustments to the surface of the microstructure edge region without significantly affecting the central optical power, allowing it to blend into the substrate more smoothly. This process also involves repeated local calculations and verifications until all parameters meet both optical and geometric requirements.

[0053] Finally, for this specific mesh cell with coordinates (Theta_i, Phi_j), the inversion calculation process outputs a set of defined, executable geometric parameters, including a validated and optimized radius of curvature value, a precisely matched base diameter value, a derived sag value, and an edge transition slope identifier confirming compliance with smoothness requirements. The sag of the microstructure defines the height of the highest point at the center of the microstructure relative to its edge reference plane. This value is not an independent variable but is directly calculated from the determined radius of curvature and the diameter of the microstructure's base surface through geometric relationships. A higher sag usually means a steeper surface and stronger optical power. Another parameter that needs to be strictly controlled is the transition slope of the area where the microstructure edge connects to the lens substrate. An excessively steep slope can produce sharp edges or even breakage during processing, and may cause strong stray light interference with vision during use. Therefore, this slope is limited to a threshold range preset based on material properties and optical experience to ensure a smooth and gradual edge transition. All calculated spatial arrangement parameters and geometric parameters are integrated into a digital microstructure layer design document, which defines in detail the target three-dimensional coordinates of each point on the lens surface.

[0054] Selected optical-grade liquid resin or photosensitive polymer material is loaded into the sealed tank of the additive manufacturing equipment. The equipment control system begins the layer-by-layer deposition and selective curing process based on the layer data generated by the slicing software. After each layer of material is deposited and cured by inkjet printing or photopolymerization, the system pauses the manufacturing cycle for the next layer and initiates an integrated online morphology inspection program. Real-time acquisition of the three-dimensional morphology information of the cured layer surface is accomplished by a laser confocal microscopy module deeply integrated into the equipment's manufacturing cavity. This module shares the same motion platform with the printhead or has a precise positioning linkage. Its core is a low-power laser that emits a specific wavelength (e.g., a 650 nm red laser). The laser beam is focused onto the surface of the material under test through a miniaturized objective lens to form a tiny spot.

[0055] To scan the entire surface of the current layer, the focused laser spot is controlled by a galvanometer system consisting of two high-speed galvanometers. The two galvanometers are responsible for deflecting the laser spot along the X and Y axes, respectively, thus driving the spot to perform a high-speed grid scan with a preset step size and path, covering the entire region of interest in the current layer. As the laser spot scans the material surface, its reflected light returns along its original path and is guided by a beam splitter to a high-sensitivity photodetector. The confocal optical system effectively blocks stray light scattered from off-focal areas by placing a confocal pinhole at the focal plane of the objective lens, ensuring that only reflected light from the focal point can pass through efficiently and be received by the detector. By recording the Z-axis height value corresponding to the peak light intensity at each XY point as the objective lens moves precisely along the Z-axis, the system can reconstruct a high-precision three-dimensional surface topography of the entire scanned area. The massive amount of 3D point cloud data collected is processed and organized into a multi-dimensional array structure for storage within the system. Each element in the array corresponds to a scanning point, and its contents include the X and Y coordinates of the point in the device coordinate system and the calculated Z-axis height value. This complete dataset accurately represents the actual 3D morphology of the surface of the newly cured layer.

[0056] See Figure 4This study systematically presents the visualization analysis results of the defocus distribution characteristics of the human eye and compensation techniques. The defocus distribution at different field of view angles is displayed as a two-dimensional heatmap, with the horizontal axis representing longitude and the vertical axis representing latitude (defocus). Values ​​are indicated by grayscale scales, ranging from +1.50D (white) to -0.25D (black). The distribution pattern shows that the defocus in the central field of view approaches zero, while the defocus in the peripheral field of view, especially in the horizontal direction, increases significantly, reaching a maximum of +1.50D. This reveals the inherent hyperopic defocus phenomenon of the human eye. This distribution map is consistent with the defocus matrix calculated by the personalized optical model using a ray tracing algorithm. The defocus compensation effect diagrams for different fields of view are compared using a bilinear model to show the change in defocus amount before and after compensation with the field of view angle (0-60 degrees). The curve before compensation (black solid line) rises sharply from the origin, with a defocus amount of approximately 36D at 60 degrees. The curve after compensation (gray solid line) stabilizes within the range of 0-2D, verifying the corrective efficiency of the distributed defocus microstructure layer. At the technical implementation level, the defocus distribution map, as the output of the personalized optical model, is used to derive the spatial arrangement parameters and geometric morphology parameters of the microstructure layer. The microstructure design is calculated through thin lens formula inversion, combined with the refractive index of the lens substrate and the substrate diameter, ensuring that each micro-unit generates the target optical power to compensate for specific field-of-view aberrations. During additive manufacturing, real-time 3D morphology acquisition relies on a laser confocal microscopy imaging module, generating point cloud data through grid scanning. Morphology deviation analysis employs a dynamic waveform matching algorithm to calculate the root mean square value of the height difference. The correction model integrates genetic algorithms and fuzzy logic reasoning to dynamically adjust the droplet volume and landing point coordinates, achieving real-time parameter compensation. Key parameter configurations include a defocus range of -0.25 to +1.50D, a field of view accuracy of ±1 degree, and a microstructure edge slope threshold preset based on material properties to ensure a balance between optical performance and manufacturing feasibility.

[0057] Example 3

[0058] After each layer of material is deposited and cured, the laser confocal microscopy module integrated into the manufacturing equipment immediately starts and performs a fully automated 3D scan of the current layer surface. This module controls the laser focus to move at high speed in the XY plane via a galvanometer system, while a precision Z-axis driver moves the objective lens vertically. By detecting the peak signal intensity of the confocal spot, it acquires the precise 3D coordinates of tens of thousands of measurement points, thereby reconstructing the actual surface morphology of the entire current layer. This massive amount of point cloud data is converted into a regular grid height map and sent to the real-time processing system. Upon receiving the actual height map data of the current layer, the real-time processing system immediately initiates a comparative analysis process with a pre-stored theoretical morphology model. This process first performs data preprocessing and coordinate registration. The pre-stored theoretical morphology model is a high-resolution digital surface model generated during the previous optical design phase, representing the ideal target geometry of the current layer. The system identifies specific reference markers or feature structures in the current layer scan data and precisely aligns the measured data with the theoretical model in translation and rotation, ensuring that subsequent comparisons are performed within the same coordinate framework.

[0059] After the alignment operation is completed, the analysis algorithm will virtually section the measured 3D topography data along the main scanning path direction of the print head during the manufacturing of this layer, generating a series of continuous cross-sectional contour lines. Each contour line is a one-dimensional discrete-time signal composed of the height values ​​of hundreds of points arranged in the path order on the cross section. The same virtual sectioning process is synchronously applied to the pre-stored theoretical model, thereby generating an ideal reference contour signal with the same length and number of sampling points. Next, the system calls the dynamic time warping algorithm to process the measured contour signal and the reference contour signal. The core purpose of the dynamic time warping algorithm is to find the optimal nonlinear mapping path between the two signals to eliminate the stretching and deformation of the signal on the time axis caused by the small speed fluctuations that may exist in the actual manufacturing process of the print head. The algorithm constructs a cumulative cost matrix and uses the principle of dynamic programming to search for a path from the starting point to the ending point that minimizes the overall alignment cost. This path defines which sampling point in the measured signal should be compared with which sampling point in the reference signal. After completing the 3D topographic scan of the current deposition layer and obtaining full-area point cloud data, the analysis algorithm immediately initiates the processing flow. The system first retrieves the printing task file for this layer from the manufacturing execution system. This file precisely records the motion trajectory and scanning path sequence of the print head during the construction of this layer. For example, the path may consist of a series of parallel, fixed-spaced straight-line scanning vectors, each vector bearing a timestamp and a corresponding XY coordinate point sequence. This motion trajectory defines the spatial order and directionality of material deposition. The analysis algorithm loads the printing path data into memory and uses it as a reference to spatially reconstruct the measured 3D point cloud. The algorithm does not simply perform uniform meshing on the entire layer surface, but strictly follows the actual movement direction of the print head to define the analysis coordinate system. It defines the tangent direction of the printing path as the primary direction of the analysis for this layer, and the normal direction of the path as the secondary direction, thereby establishing a dynamic local coordinate system strongly correlated with the printing process. Along this main direction, virtual cross-section contour extraction is performed. It first selects a starting point on the printing path, and then sets an analysis cross-section perpendicular to the tangent of the path at certain intervals (e.g., every 10 micrometers). These cross-sections are not physically existing, but are virtual cutting planes defined by the algorithm. Each such cross-section intersects with high-density measured point cloud data. The algorithm extracts all measurement points located within a certain tolerance range on both sides of the cross-section and sorts them according to their coordinate values ​​along the path direction, thereby generating a contour line that reflects the actual height change of the cross-section position. This contour line consists of hundreds of height value points arranged in the printing order.The exact same cross-section settings and extraction operations are synchronously applied to the pre-stored theoretical morphology model, which is itself a high-precision digital surface model. The algorithm uses the exact same virtual cross-section to cut this ideal model and extracts the sequence of height values ​​that the theoretical model should have at the exact same spatial location, thereby generating an ideal reference profile with the same length and the same number of sampling points as the measured profile.

[0060] After determining the optimal alignment path, the system calculates the height difference between each pair of matching points along the path. These height differences intuitively reflect the geometric deviation between the actual manufactured surface and the ideal design surface at each local position. The system quickly calculates the root mean square value of all these height differences and uses it as a quantitative indicator of the overall shape deviation of the profile. At the same time, it records the position of all points whose height difference exceeds the preset threshold and their deviation amount.

[0061] After all parallel section profile analyses of the current layer are completed, the system aggregates the deviation analysis results of all sections and generates a two-dimensional deviation distribution map that comprehensively reflects the morphological quality of the current layer. This distribution map clearly indicates which areas have insufficient material deposition (negative deviation), which areas have excessive material deposition (positive deviation), and the specific values ​​and spatial ranges of the deviations. This deviation distribution map, together with its related quantitative indicators (such as overall RMS deviation, maximum positive and negative deviation values ​​and their locations), constitutes the morphological deviation results of the current layer. This result is immediately packaged into a data packet and sent to the next stage of the calibration model, providing accurate input basis for real-time printing parameter adjustment.

[0062] See Figure 5 In the surface optimization of additive manufacturing lenses based on real-time feedback control, the root mean square deviation distribution curve along the X-axis shows the fluctuation of the root mean square deviation as the X-coordinate changes, reflecting the local topographic deviation characteristics of the lens surface along the X-axis direction. The statistical distribution of topographic deviation is presented in the form of a histogram. The topographic deviation results are obtained by real-time acquisition and comparative analysis of the three-dimensional topography of each layer of the additive manufacturing lens surface. The deviation is distributed around the average value (0.014μm), and the range of ±1σ (0.265μm) reflects the degree of dispersion of the deviation. This intuitively shows the overall statistical characteristics of the topographic deviation, providing key statistical basis for subsequent calculation of printing parameter correction instructions and adjustment of printing parameters of additive manufacturing equipment based on the deviation results.

[0063] Example 4

[0064] During the implementation of real-time feedback-based printing parameter correction, the multi-algorithm fusion decision architecture begins processing the deviation data packets transmitted from the morphology analysis module. These packets contain a height difference distribution map between the current printed layer surface morphology and the theoretical model, the overall root mean square error, and the spatial location information of out-of-range points. The decision architecture launches two independent computation threads in parallel. The first thread runs a genetic algorithm optimization routine. This algorithm converts the received morphology deviation data into an initial population. Each individual in the population directly represents a set of possible droplet volume adjustment coefficients using a real-number encoding method. For example, an individual might be encoded as [1.05, 0.98, 1.12], representing the suggested droplet volume adjustment multiplier for the next three printing cycles. The algorithm's fitness function is defined as the similarity index between the expected new surface morphology and the theoretical morphology after applying the adjustment coefficients represented by the current population individuals to the prediction model for the next printing cycle. The algorithm iteratively evolves the population through selection, crossover, and mutation operations, seeking the optimal combination of adjustment coefficients that minimizes the predicted morphology deviation. Meanwhile, a second thread starts the fuzzy logic inference engine. This engine receives the same topographic deviation data packets, but its processing method is completely different from that of the genetic algorithm. The fuzzy system first performs fuzzification processing on the input data. Its three-dimensional input variables include: the average value of the overall topographic deviation amplitude of the current layer (in micrometers), the proportion of the area of ​​the deviation region to the total area of ​​the current layer (percentage), and the maximum value of the spatial gradient of the deviation distribution (reflecting the severity of the deviation change). Each input variable is mapped to a preset fuzzy set. For example, the deviation amplitude may be divided into three fuzzy sets: "small", "medium", and "significant". The system's built-in membership function (such as the triangular or trapezoidal function) is responsible for converting the precise input value into the membership degree of the corresponding fuzzy set. Subsequently, the inference engine calls a rule base containing multiple IF-THEN rules to perform fuzzy inference; see Table 1.

[0065] Table 1: Fuzzy Logic Reasoning Rules Table

[0066]

[0067] The fuzzy inference process follows the rule base example shown in the table above. The system evaluates all activated rules and calculates the strength of each rule's output conclusion based on the input membership degree. Finally, it aggregates the output conclusions of all rules into a precise, directly executable suggested value for the landing point coordinate offset using a defuzzification method (such as the centroid method), for example, a suggested offset of -0.5 micrometers in the X direction. After their respective calculations, the genetic algorithm thread and the fuzzy logic inference thread output their results to the weighted fusion module. This module does not simply average the two results but employs a dynamic weight allocation strategy. This strategy alternately adjusts the trust weights of the two algorithm results based on the parity of the current printing layer number. Assuming the current printing layer is layer 7 (an odd layer), the fusion module will assign a higher weight (e.g., 0.7) to the droplet volume adjustment coefficient output by the genetic algorithm and a relatively lower weight (e.g., 0.3) to the landing point coordinate offset output by the fuzzy logic inference. Then, a weighted calculation is performed to generate the final, fused set of correction instruction parameters. Conversely, if the current printing layer is the 8th layer (an even-numbered layer), the weighting scheme is reversed, prioritizing the reasoning result of fuzzy logic. This alternating priority mechanism aims to comprehensively utilize the advantages of the two algorithms at different levels, avoiding the bias or local optima that might be introduced by over-reliance on a single algorithm. The final generated set of correction instruction parameters is encapsulated and immediately sent to the printing controller of the additive manufacturing equipment. The instruction set explicitly includes the specific values ​​of the droplet volume adjustment coefficient for the next one or more printing cycles, as well as the fine-tuning of the nozzle's landing point coordinates in the XY plane. After receiving the instructions, the controller converts them into adjustment signals for the piezoelectric crystal driving voltage and displacement commands for the precision servo motor, thereby compensating for the discovered morphological deviations in real time during the subsequent physical manufacturing process, forming an efficient closed-loop control.

[0068] Example 5

[0069] Upon receiving the comprehensive correction instruction parameter set generated by the upper-level correction model, the underlying motion and jetting control system of the additive manufacturing equipment immediately initiates the parameter parsing and execution process. This system maintains a high-precision internal state machine, which constantly tracks key parameters such as the current printing cycle number, the absolute coordinates of the printhead, and the current operating voltage of the piezoelectric actuator. The correction instruction set usually arrives in the form of a structured data, which typically includes, but is not limited to: an array of droplet volume adjustment coefficients for the Nth to N+Kth subsequent jetting cycles, such as [cycle N: 1.05, cycle N+1: 0.98, cycle N+2: 1.02], and a coordinate offset that needs to be superimposed for the printhead in the next XY plane movement, such as (Δx: -0.7μm, Δy: +0.3μm).

[0070] The system first processes the droplet volume adjustment command. Precise control of the droplet volume is achieved by altering the high-voltage pulse waveform parameters driving the piezoelectric crystal. This crystal is precisely encapsulated outside the liquid path cavity of the printhead, and its deformation is directly proportional to the applied driving voltage amplitude. This deformation directly determines the volume of fluid extruded from the nozzle orifice. Based on the adjustment coefficient specified in the command (e.g., 1.05, representing a 5% increase in volume), the control system calculates a new target driving voltage value by consulting a pre-stored, rigorously calibrated "driving voltage-droplet volume" correspondence table. For example, the voltage is adjusted from the default 20.0V to 21.0V. This calibration table was obtained before the equipment was put into use by measuring the actual mass of thousands of droplets using a high-precision balance and calculating their volumes, ensuring the accuracy of the control relationship. The voltage adjustment command generates a corresponding analog signal via a digital-to-analog converter, which is then amplified by a high-voltage amplifier and applied to the piezoelectric crystal. This entire process is completed within milliseconds, ensuring that the new drive voltage is already in effect at the start of the next specified ejection cycle (e.g., cycle N), thus ejecting ink droplets with a volume precisely increased by 5%. Next, the system processes fine-tuning commands for the printhead's landing point coordinates. The printhead is mounted on an XY two-dimensional motion platform composed of a high-precision linear servo motor and a grating ruler feedback system, with a theoretical positioning accuracy reaching sub-micron levels. The offset (Δx, Δy) given in the command is added in real-time to the original interpolation trajectory planning data in the motion controller. In the next tiny interpolation cycle, the motion controller adds this tiny offset component to the planned movement vector. For example, if the original plan was to move linearly from point (X_i, Y_j) to point (X_{i+1}, Y_{j+1}), after incorporating the offset command, the target point is corrected to (X_{i+1} + Δx, Y_{j+1} + Δy. After receiving the new target position, the servo driver will drive the motor to complete the positioning with offset with extremely high precision and rigidity. The grating ruler provides real-time feedback on the actual position and compares it with the target position to form a closed-loop control, ensuring that the nozzle finally stops precisely at the expected correction point.

[0071] All these adjustments, whether voltage switching or position fine-tuning, are calculated and prepared within a very short time before the next physical action (jetting or moving) begins, within one manufacturing cycle of the current printed layer. This delay is far less than the length of a manufacturing cycle, ensuring immediate adjustment for the next manufacturing action. This proactive control allows for immediate and targeted compensation in subsequent deposition processes for topographic deviations identified in the previous inspection cycle (e.g., a slight material shortage in a localized area). This compensation directly affects the topographic correction of unprinted areas in the current layer and may also indirectly optimize the deposition basis of subsequent layers by influencing the stacking effect. This gradually converges within a closed-loop control framework, ensuring that the final manufactured lens's overall surface topography faithfully reproduces the optical design intent, meeting personalized visual correction needs.

[0072] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for optimizing the surface of additive manufacturing lenses based on real-time feedback control, characterized in that, The method includes: Acquire the wearer's ocular biometric data and personalized visual needs parameters; A personalized optical model of the eyeball was established using ocular biometric data and personalized visual requirement parameters, and the defocus distribution map of the retina under different visual angles was derived. The spatial arrangement parameters and geometric morphological parameters of the distributed defocus microstructure layer are determined based on the defocus amount distribution map. Optical materials are loaded into additive manufacturing equipment and cured by layer-by-layer deposition; during the deposition and curing of each layer, the three-dimensional morphology information of the surface of that layer is acquired in real time. The real-time acquired 3D topography information is compared and analyzed with the pre-stored theoretical topography model to generate topography deviation results; Based on the morphological deviation results, the correction instructions for printing parameters are calculated using the calibration model; the printing parameters of the additive manufacturing equipment are adjusted in real time according to the correction instructions. A personalized optical model of the eyeball is established using a ray tracing algorithm, which simplifies the eyeball into a multi-layer optical system composed of the cornea, aqueous humor, lens, and vitreous body. The refractive index and thickness of each layer are assigned by ocular biometric data. The model input is the direction of parallel incident light rays at different field of view angles, and the output is the focal position of the light rays on the retina. The defocus amount is calculated by comparing the focal position with the ideal retinal plane. The defocus amount distribution map is stored in the form of a two-dimensional matrix, with the matrix row and column indices corresponding to the horizontal and vertical field of view angles, and the element value being the defocus amount for the corresponding field of view angle. Determining the spatial arrangement parameters of the distributed defocus microstructure layer includes dividing the front surface of the lens into hexagonal grid regions, with the center point of each grid defined by latitude and longitude coordinates. The geometric parameters include the radius of curvature, sag, and edge transition slope of each microstructure unit. The radius of curvature is calculated using the thin lens formula based on the defocus distribution map, the sag is determined by the radius of curvature and the diameter of the microstructure bottom surface, and the edge transition slope is limited to a preset threshold range.

2. The additive manufacturing lens surface optimization method based on real-time feedback control according to claim 1, characterized in that, Acquiring wearer ocular biometric data includes collecting corneal curvature, axial length, pupil diameter, and accommodative hysteresis using a non-contact biometric instrument. Personalized visual needs parameters are obtained through a standardized questionnaire, including distribution of daily eye use scenarios, glare sensitivity, and preference for lens photochromic speed. The non-contact biometric instrument scans the anterior segment of the eyeball by emitting a low-coherence beam and reconstructs corneal curvature and axial length based on interference signals. The pupil diameter is measured as the equivalent circle diameter under standard lighting conditions using an infrared imaging system. The accommodative hysteresis is recorded by dynamic retinoscopy to show the hysteresis difference between the accommodative response curve and the ideal curve.

3. The additive manufacturing lens surface optimization method based on real-time feedback control according to claim 1, characterized in that, Real-time acquisition of the three-dimensional morphology information of the surface layer is achieved by a laser confocal microscopy imaging module integrated inside the additive manufacturing equipment. The laser confocal microscopy imaging module emits a laser beam of a preset wavelength, performs grid scanning through a galvanometer system, receives reflected light signals, and reconstructs three-dimensional point cloud data. The scan covers the entire surface of the current layer, and the point cloud data is stored in the form of an array, with each point containing three-dimensional coordinate information.

4. The additive manufacturing lens surface optimization method based on real-time feedback control according to claim 1, characterized in that, The real-time acquired 3D topographic information is compared and analyzed with the pre-stored theoretical topographic model using a combination of morphological contour decomposition and dynamic waveform matching, including: The real-time 3D topography data is decomposed into a series of continuous cross-sectional profile waveforms along the printing path. Each waveform consists of a timing signal composed of the height values ​​of each point on the cross section. Simultaneously, the theoretical morphology model is decomposed into cross-sectional profiles in the same direction to generate a reference waveform sequence; The real-time waveform and the reference waveform are dynamically time-normalized and aligned, and the root mean square value of the height difference of the corresponding points after alignment is calculated as the shape deviation result.

5. The additive manufacturing lens surface optimization method based on real-time feedback control according to claim 4, characterized in that, The dynamic waveform matching process includes the following steps: Wavelet transform is used to denoise the real-time acquired cross-sectional contour waveform, and key feature points in the waveform are extracted. The denoised real-time waveform and the reference waveform are dynamically time-warped to find the optimal nonlinear alignment path. Calculate the height difference between corresponding points of the two waveforms along the alignment path, and statistically analyze the distribution characteristics of the height difference. When the root mean square of the height difference exceeds the threshold, it is determined that there is a morphological deviation in the cross-sectional area, and the magnitude and spatial distribution of the deviation are recorded.

6. The additive manufacturing lens surface optimization method based on real-time feedback control according to claim 1, characterized in that, The correction instructions for printing parameters are calculated using a calibration model, and a decision architecture that integrates multiple algorithms is employed. The decision architecture executes two processing flows in parallel: genetic algorithm optimization and fuzzy logic reasoning. The genetic algorithm uses the spatial distribution characteristics of the shape deviation results as the initial population input and iteratively searches for the optimal ink droplet volume adjustment coefficient through selection, crossover, and mutation operations. At the same time, the fuzzy logic reasoning system derives the fuzzy decision result of the landing point coordinate offset based on the magnitude and distribution density of the shape deviation through a preset membership function and rule base. Finally, the genetic algorithm output and fuzzy inference results are integrated through a weighted fusion module to generate the final correction instruction parameters.

7. The additive manufacturing lens surface optimization method based on real-time feedback control according to claim 6, characterized in that, The specific implementation of the multi-algorithm fusion decision architecture includes: The genetic algorithm part uses real number encoding to represent the ink droplet volume adjustment coefficient, and the fitness function is defined as the similarity between the corrected predicted morphology and the theoretical morphology; The fuzzy logic reasoning system establishes a three-dimensional input variable space, including three dimensions: the magnitude of the shape deviation, the area of ​​the deviation region, and the deviation gradient value. The weighted fusion module adopts a dynamic weight allocation strategy, which adjusts the weight ratio of the genetic algorithm results and the fuzzy inference results alternately according to the parity of the current printed layer number. Odd-numbered layers take priority to use the genetic algorithm output, while even-numbered layers take priority to use the fuzzy inference results.

8. The method for optimizing the surface of additive manufacturing lenses based on real-time feedback control according to claim 1, characterized in that, The printing parameters of the additive manufacturing equipment are adjusted in real time according to the correction instructions, including adjusting the piezoelectric crystal driving voltage to change the ink droplet ejection volume and fine-tuning the position of the printhead in the horizontal and vertical directions to correct the landing point deviation. The adjustment operation is executed immediately in the next ejection cycle to ensure that the current layer topography deviation is compensated in real time.

Images