Intelligent processing control system for special-shaped curved surface of optical element

Abstract

The invention discloses an intelligent processing control system for a special-shaped curved surface of an optical element. The intelligent processing control system comprises a dynamic visual angle vision measurement module, a dual-process processing module, an intelligent control module and a data storage and interaction module. Collecting image data of a processing area in real time through a dynamic visual angle vision measurement module; a deep learning analysis unit, a dual-process collaborative decision-making unit, a cross-process error compensation unit and a real-time control unit in the intelligent control module are used for processing the data and generating a control instruction; the dual-process machining module executes abrasive jet rough machining or laser micro-cladding finish machining / repairing. All the modules form a real-time closed loop, the dynamic visual angle vision measurement module continuously collects machining results and feeds the machining results back to the intelligent control module, and self-adaptive adjustment based on real-time detection is achieved. According to the method, efficient, high-precision and self-adaptive intelligent machining of the special-shaped curved surface is achieved.

Description

Technical Field

[0001] This invention relates to the fields of precision manufacturing and automatic control technology, and in particular to an intelligent processing and control system for irregular curved surfaces of optical elements. Background Technology

[0002] As optical systems evolve towards higher performance, miniaturization, and more complex structures, optical components (such as lenses, prisms, and optical windows) are increasingly adopting irregularly shaped surface structures such as aspherical and freeform surfaces to improve image quality, reduce system aberrations, and decrease the number of optical components. These irregularly shaped surfaces typically feature large curvature variations, complex local characteristics, and high precision requirements. Their fabrication quality directly affects the overall performance of the optical system, thus placing higher demands on the precision control, stability, and consistency of the fabrication process.

[0003] In the manufacturing process of irregularly shaped curved optical components, it is often necessary to balance material removal efficiency and final surface quality. On the one hand, high material removal capacity is required in the early stages of processing to quickly form the target curved surface; on the other hand, precise control over the surface morphology and surface quality is required in the later stages of forming and the repair of local defects. This requirement for both "high-efficiency processing" and "high-precision control" makes the processing process exhibit multi-stage and multi-mode collaborative characteristics, posing a challenge to the comprehensive capabilities of the processing control system.

[0004] However, in existing processing methods, the control of processing irregularly shaped curved optical elements typically exhibits the following shortcomings: First, the process coordination capability is limited. Since the surface morphology may change dynamically during the processing, the requirements for process type and parameters vary significantly at different processing stages. However, processing control often relies on preset process flow or fixed parameter strategies, which makes it difficult to respond in a timely manner to real-time changes in surface features and states. This can easily lead to problems such as uneven distribution of processing allowance, local over-processing, or under-processing.

[0005] Second, the ability to perceive the processing status is insufficient. Due to limitations in measurement methods and observation angles, some areas of irregular curved surfaces are difficult to perceive comprehensively and continuously during the processing, especially in areas with high curvature, edge areas, or areas of dynamic processing changes. Information such as surface contours, surface defects, and processing tool status is difficult to obtain in a timely and complete manner, resulting in a lack of sufficient data support for processing control decisions.

[0006] Third, the comprehensive compensation capability for machining errors needs to be improved. During the machining process, changes in tool condition, motion errors, and connection deviations between different machining methods can all lead to cumulative errors on the curved surface. If compensation is only applied to a single machining process or a single error source, it is difficult to effectively suppress machining deviations caused by the coupling of multiple factors, thereby affecting the continuity of the curved surface and the overall machining accuracy.

[0007] Fourth, the level of intelligence in processing decisions is limited. The processing quality of irregular curved surfaces is not only related to their static geometric characteristics, but also affected by real-time state changes during processing. However, existing control methods lack the ability to comprehensively analyze multi-dimensional information and make dynamic decisions, making it difficult to achieve an adaptive balance between processing efficiency and processing accuracy.

[0008] Therefore, how to achieve real-time perception of the state of the processing area, intelligent decision-making and collaborative control of the processing technology, and dynamic compensation for multi-source errors during the processing of irregular curved surface optical components, and form a stable and efficient closed-loop processing control mechanism, has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0009] To address the shortcomings of existing technologies, the present invention aims to provide an intelligent processing control system for irregular curved surfaces of optical components. This system integrates dynamic visual measurement, deep learning intelligent analysis, dual-process collaborative decision-making, and cross-process error compensation into a full-link closed-loop control, enabling efficient and high-precision intelligent processing of irregular curved surfaces through real-time state perception, adaptive process switching, and dynamic error correction.

[0010] To achieve the above objectives, the present invention provides the following technical solution: an intelligent processing and control system for irregular curved surfaces of optical elements, comprising: The dynamic perspective vision measurement module is used to acquire and process images of the processing area in real time, and generate image data, including real-time contour data, surface defect data and tool status data. The dual-processing module includes an abrasive jetting unit, a laser micro-cladding unit, and a motion platform unit; An intelligent control module, connecting the dynamic perspective vision measurement module and the dual-process processing module, includes: The deep learning analysis unit is used to receive the real-time contour data and the surface defect data, process them, and output analysis results containing surface features, deformation prediction, and defect information. A dual-process collaborative decision-making unit, connected to the deep learning analysis unit, is used to generate process types and corresponding optimized process parameter sets based on the analysis results; A cross-process error compensation unit, connected to the dual-process collaborative decision-making unit, is used to calculate the compensation amount for tool pose and process parameters based on the tool status data and the process type. The real-time control unit is connected to the dual-process collaborative decision-making unit and the cross-process error compensation unit, respectively, and is used to generate control signals based on the optimized process parameter set and the compensation amount to drive the dual-process processing module. The data storage and interaction module connects the dynamic perspective visual measurement module, the dual-process processing module, and the intelligent control module, and is used to store data and provide an interaction interface; During the processing, the dynamic perspective vision measurement module acquires image data of the processing area and inputs it into the intelligent control module; the intelligent control module generates control signals from the image data to drive the dual-process processing module to operate; the dynamic perspective vision measurement module acquires image data of the dual-process processing module after it operates again and feeds it back to the intelligent control module.

[0011] Furthermore, the deep learning analysis unit is specifically used for: Receives real-time contour data and surface defect data from the dynamic viewpoint vision measurement module, as well as data from externally provided external sensors; The real-time contour data is input into a pre-trained surface feature recognition model, which outputs surface feature information including the actual curvature distribution of the current processing area and the deviation from the design contour. The external sensor data and the real-time contour data are input together into a pre-trained dynamic deformation prediction model, which outputs the micro-deformation prediction information of the current processing area. The surface defect data is input into a pre-trained defect classification and localization model, which outputs defect information including defect type, severity, and coordinate location. The surface feature information, the micro-deformation prediction information, and the defect information together constitute the analysis result and are transmitted to the dual-process collaborative decision-making unit.

[0012] Furthermore, the dual-process collaborative decision-making unit is specifically used for: The analysis results received from the deep learning analysis unit include at least the deviation from the design profile, micro-deformation prediction information, and defect information. The judgment is based on the deviation from the design profile: If the machining allowance in the current processing area is greater than the first preset allowance threshold, and the defect information indicates that there is no specific serious defect, then the process type is determined to be abrasive jet roughing, and the optimized process parameter set including jet pressure and moving speed is generated. If the current processing allowance in the processing area is not greater than the first preset allowance threshold, or if the defect information indicates the existence of a defect but its severity is lower than the second preset defect threshold, then the process type is switched to laser micro-cladding finishing or repair, and the optimized process parameter set containing laser power, spot diameter and scanning path is generated. If the defect information indicates the existence of a defect and its severity reaches or exceeds the second preset defect threshold, then the decision is made to prioritize laser micro-cladding repair, and after the repair is completed, switch to laser micro-cladding finishing to generate the corresponding set of optimized process parameters. The determined process type and the corresponding optimized process parameter set are output to the real-time control unit and the cross-process error compensation unit.

[0013] Furthermore, the cross-process error compensation unit is specifically used for: The tool status data is periodically received from the dynamic perspective vision measurement module, and the tool status data includes at least the wear status data of the abrasive jetting unit and the spot status data of the laser micro-cladding unit. Receive the process type of the current decision from the dual-process collaborative decision-making unit; Based on the aforementioned process type, select the appropriate error compensation model: When the process type is abrasive jet roughing, the position compensation amount of the abrasive jetting unit in three-dimensional space is calculated based on the wear state data. When the process type is laser micro-cladding finishing or repair, the correction coefficient of laser power and the offset compensation amount of the spot focusing position are calculated based on the spot state data. The calculated position compensation amount, the correction coefficient, and the offset compensation amount are used as the compensation amount and output to the real-time control unit.

[0014] Furthermore, the real-time control unit is specifically used for: Receive the process type and the optimized process parameter set from the dual-process collaborative decision-making unit, and the compensation amount from the cross-process error compensation unit; According to the process type, the corresponding optimized process parameter set and the compensation amount are fused to generate a unified drive instruction set that includes spatial position instructions and process parameter instructions; At a frequency not greater than a preset control cycle, the drive instruction set is converted into a low-level control signal and synchronously sent to the motion platform unit, the abrasive jetting unit, or the laser micro-cladding unit in the dual-process processing module. The real-time control unit is also used to complete the entire process from receiving data to issuing the underlying control signal within a single control cycle, and its response time is lower than a preset response time threshold.

[0015] Furthermore, the dynamic viewpoint visual measurement module includes: The image acquisition unit is used to drive the gimbal equipped with an industrial camera to move along a preset path according to the control instructions from the intelligent control module during the processing, so as to acquire the image data of the processing area without blind spots, wherein the preset path is planned based on the curvature distribution in the design model of the optical element. The image processing unit, connected to the image acquisition unit, is used to perform noise reduction and contour extraction on the acquired image data, generate standardized real-time contour data, surface defect data, and tool status data, and output them to the intelligent control module.

[0016] Furthermore, the abrasive jetting unit includes: The abrasive supply unit is used to provide nanoscale abrasives within a set particle size range; The pressure regulation unit, connected to the abrasive supply unit, is used to receive control signals and adjust the injection pressure to a first preset pressure range; The nozzle actuator, connected to the pressure control unit, is used to perform jetting at a moving speed within a second preset speed range; The set particle size range, the first preset pressure range, and the second preset speed range are configured to work together to achieve the set material removal efficiency when roughing irregular curved surfaces, and to control the surface roughness to be lower than the first preset roughness threshold.

[0017] Furthermore, the laser micro-cladding unit includes: The laser generator is used to generate a laser beam with adjustable wavelength and power, wherein the power can be continuously adjusted within a third preset power range; The optical path control unit is connected to the laser generator unit and is used to adjust the spot diameter of the laser beam to a set focusing range and control the laser beam to move at a scanning speed within a fourth preset speed range. The powder feeding section is used to transport cladding material powder that matches the optical element substrate to the processing area; The working parameters of the optical path control unit and the powder feeding unit are set based on the process type and the optimized process parameter set output by the dual process collaborative decision unit, so as to achieve the repair of surface defects or the finishing of the processed surface, and make the surface roughness after finishing lower than the second preset roughness threshold.

[0018] Furthermore, the motion platform unit includes: A multi-axis drive unit is used to drive the abrasive jetting unit or the laser micro-cladding unit to move along a planned path in three-dimensional space according to the received control signal. The position feedback unit, connected to the multi-axis drive unit, is used to detect the spatial position of the abrasive jetting unit or the laser micro-cladding unit in real time and generate a position feedback signal. The multi-axis drive unit and the position feedback unit form a closed-loop control, which ensures that the spatial positioning accuracy of the motion platform unit is not lower than a preset positioning accuracy threshold and the repeatability positioning accuracy is not lower than a preset repeatability positioning accuracy threshold, so as to ensure that the abrasive spraying unit or the laser micro-cladding unit moves according to the planned path.

[0019] Furthermore, the data storage and interaction module includes: The data recording unit is used to continuously store the image data, control signals, optimized process parameter set, and compensation amount in the entire processing flow in a time sequence, forming a traceable processing process database; The external interface unit is used to receive surface design model data from external input and output the final contour data and process report after processing to the external system. The human-computer interaction unit is used to display images of the processing area, key information of the analysis results, the process type and processing progress in real time, and to receive parameter input or intervention instructions from the operator.

[0020] The beneficial effects of this invention are: 1. Achieving intelligent collaborative closed-loop control for dual-process machining: This invention organically combines dynamic perspective visual measurement, intelligent analysis, and dual-process machining to construct a closed-loop control mechanism covering machining status perception, process decision-making, execution control, and error compensation. It can dynamically adjust the process type and parameters according to the real-time status changes during the machining of irregular curved surfaces, avoiding the problems of rigid process switching and delayed response in traditional segmented machining, and improving the overall coordination and controllability of the machining process.

[0021] 2. Effectively improve the machining accuracy and efficiency of irregular curved surfaces: By acquiring information on the surface contour, surface defects and tool status in real time, and continuously controlling the machining process based on intelligent analysis results, the occurrence of uneven machining allowance distribution, overmachining or undermachining can be reduced. While ensuring the forming accuracy and surface quality of the curved surface, the probability of invalid machining and rework is reduced, thereby achieving a simultaneous improvement in machining accuracy and machining efficiency.

[0022] 3. Enhanced adaptability to complex curved surfaces and multiple working conditions: This invention can flexibly select and coordinate the corresponding processing methods and parameter configurations for different curvature characteristics, different processing stages and different processing states. It is applicable to a variety of irregular curved surface optical elements and different material conditions, and has strong process adaptability and application versatility.

[0023] 4. Improve processing stability and consistency: By uniformly modeling and compensating for changes in tool condition, deformation during processing, and errors between different processes, error accumulation and process connection defects can be effectively suppressed, thereby improving the stability of the processing process and ensuring dimensional consistency and repeatability between different processing batches.

[0024] 5. Facilitates system integration and process optimization: This invention adopts a modular control architecture and stores and manages processing data in a unified manner, which facilitates integration and upgrading with existing processing equipment, and is also beneficial for subsequent process parameter optimization and quality traceability, thereby reducing system application and maintenance costs. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the intelligent processing and control system for irregular curved surfaces of optical elements in this invention.

[0026] Reference numerals: 1. Dynamic viewpoint visual measurement module; 11. Image acquisition unit; 12. Image processing unit; 2. Dual-process processing module; 21. Abrasive spraying unit; 22. Laser micro-cladding unit; 23. Motion platform unit; 3. Intelligent control module; 31. Deep learning analysis unit; 32. Dual-process collaborative decision-making unit; 33. Cross-process error compensation unit; 34. Real-time control unit; 4. Data storage and interaction module; 41. Data recording unit; 42. External interface unit; 43. Human-computer interaction unit. Detailed Implementation

[0027] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Identical components are denoted by the same reference numerals. It should be noted that the terms "front," "rear," "left," "right," "upper," and "lower" used in the following description refer to directions in the accompanying drawings, and the terms "bottom surface," "top surface," "inner," and "outer" refer to directions toward or away from the geometric center of a specific component, respectively.

[0028] like Figure 1 As shown in Embodiment 1, an intelligent processing and control system for irregular curved surfaces of optical elements is described.

[0029] I. Description of Implementation Examples; The intelligent processing control system and processing method for irregular curved surfaces of optical elements according to the present invention will be described in detail below with reference to specific embodiments. These embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention. Equivalent modifications or substitutions made to the embodiments by those skilled in the art without departing from the technical concept of the present invention should all fall within the scope of protection of the present invention.

[0030] This embodiment takes the processing of an aspherical optical lens as an example. The optical lens is made of quartz glass, has a diameter of φ50mm, and the surface roughness required by the design is Ra≤0.01μm, with a contour deviation requirement of ±0.01mm.

[0031] II. System Composition and Hardware Selection; The intelligent processing and control system for irregular curved surfaces of optical elements used in this embodiment includes the following modules: (a) Dynamic Viewpoint Visual Measurement Module 1; It is used to acquire and process images of the processing area in real time during the processing, and generate real-time contour data, surface defect data and tool status data.

[0032] In this embodiment, the dynamic viewing angle visual measurement module 1 specifically includes: Industrial camera: Basler acA2500-14gm CCD camera, 5 megapixel resolution, maximum frame rate 14fps; Dynamic viewing mechanism: PT-Gimbal motorized gimbal with an angle positioning accuracy of 0.01°, supporting continuous rotation within the range of ±90° horizontally and ±60° vertically; Light source system: Ring LED fill light with a wavelength of 600nm, used to enhance the contrast of lens surface contours and defects; Image acquisition cycle: During the processing, the camera acquires an image of the processing area every 10ms.

[0033] With the above configuration, the dynamic viewing angle visual measurement module 1 can continuously observe irregular curved surfaces from different viewing angles, avoiding measurement blind spots caused by fixed viewing angles.

[0034] (ii) Dual-process machining module 2; The dual-processing module 2 is connected to the dynamic perspective vision measurement module 1, and includes an abrasive spraying unit 21, a laser micro-cladding unit 22, and a motion platform unit 23.

[0035] In this embodiment: The abrasive jetting unit 21 uses nano-diamond abrasive with a particle size of 200nm, and the jetting pressure is adjustable in the range of 0.1~1.0MPa; The laser micro-cladding unit 22 uses an IPG YLR-500 fiber laser with an output power range of 50 to 500W and supports spot diameter adjustment of 0.2 to 0.5 mm. The motion platform unit 23 is an XYZ three-axis linear motor platform with a positioning accuracy of ±0.005mm, used to support optical lenses and realize multi-axis linkage machining.

[0036] (iii) Intelligent control module 3; The intelligent control module 3 is connected to the dynamic perspective vision measurement module 1 and the dual-process processing module 2, and includes the following functional units: The deep learning analysis unit 31 is used to receive real-time contour data and surface defect data, perform intelligent analysis, and output analysis results including surface features, deformation prediction, and defect information.

[0037] In this embodiment, the deep learning analysis unit 31 runs on an industrial computer (CPU is Intel i7, memory is 16GB) and is built using the TensorFlow 2.8 deep learning framework.

[0038] The dual-process collaborative decision-making unit 32 is used to generate process types and corresponding optimized process parameter sets based on the analysis results of the deep learning analysis unit 31.

[0039] The cross-process error compensation unit 33 is used to calculate the compensation amount for tool pose and process parameters based on tool status data and the current process type.

[0040] The real-time control unit 34 is used to generate control signals based on the optimized process parameter set and compensation amount to drive the abrasive spraying unit 21, the laser micro-cladding unit 22 and the motion platform unit 23 to work together.

[0041] (iv) Data storage and interaction module 4; It is used to store image data, process parameters, error compensation data and processing result data during the processing, and provides a human-machine interface to facilitate parameter setting, status monitoring and quality traceability.

[0042] III. Working principle of Example 1; The working principle of this embodiment is based on a closed-loop control mechanism of "dynamic visual perception - intelligent analysis and decision-making - dual-process collaborative processing - cross-process error compensation", as detailed below.

[0043] (a) Pre-processing stage; First, the STL format CAD model of the aspherical optical lens is imported into the intelligent control module 3 to extract the design contour data and curvature distribution information (curvature range of 100-500mm). -1 And processing accuracy requirements (contour deviation ±0.01mm, surface roughness Ra≤0.01μm).

[0044] Subsequently, the optical lens is fixed on the tooling fixture of the motion platform unit 23. The dynamic viewing angle vision measurement module 1 collects the initial contour data at the initial viewing angle (0° horizontal, 0° vertical) and compares it with the design contour data to generate an initial machining allowance distribution map, in which the maximum machining allowance is 0.2mm and the minimum machining allowance is 0.08mm.

[0045] (ii) Dynamic perspective visual measurement and data acquisition; During the processing, the electric gimbal plans the rotation path according to the curvature distribution, enabling the industrial camera to continuously observe the processing area from multiple perspectives.

[0046] The dynamic perspective visual measurement module 1 preprocesses the acquired images, extracts real-time contour point cloud data, surface defect data, and tool status data, and transmits them to the intelligent control module 3 in real time.

[0047] By using a dynamic perspective measurement method, the true processing status of high curvature areas and edge areas can be continuously acquired during the processing, thereby improving the completeness of state perception.

[0048] (III) Deep learning analysis and process decision-making; The deep learning analysis unit 31 performs comprehensive analysis on real-time contour data, temperature data, and vibration data to identify the surface features, machining allowance, and potential defects of the current processing area.

[0049] In this embodiment, the analysis results show that: The curvature of the lens edge area is approximately 500mm. -1 The machining allowance was 0.15mm, and no defects were detected. The curvature of the central region of the lens is approximately 100mm. -1 The machining allowance is 0.04 mm, and there is a slight material residue with a size of about 0.02 mm.

[0050] The dual-process collaborative decision-making unit 32 makes process judgments based on the following threshold rules: When the machining allowance is >0.05mm and there are no serious defects, the abrasive blasting process should be selected; When the machining allowance is ≤0.05mm or there are slight / moderate defects, laser micro-cladding process is selected; When a severe defect with a crack depth greater than 0.02 mm is detected, laser repair should be prioritized.

[0051] (iv) Dual-process collaborative processing and real-time control; The real-time control unit 34 converts the optimized process parameters into control signals, driving the dual-process machining module 2 to execute machining actions. For example: Abrasive blasting was performed on the edge area, with the blasting pressure set at 0.6 MPa and the moving speed at 15 mm / s. The central area was switched to laser micro-cladding processing, with the laser power set to 180W, the spot diameter to 0.3mm, and the scanning path planned along the defect contour.

[0052] During the processing, the dynamic perspective vision measurement module 1 continuously updates the image data with a period of 10ms, forming a high-frequency closed-loop control.

[0053] (v) Cross-process error compensation; During the processing, the dynamic perspective vision measurement module 1 collects tool status data every 5 seconds, including the wear of the abrasive nozzle and the drift of the laser spot.

[0054] The cross-process error compensation unit 33 calculates the compensation amount based on a preset error compensation model, for example: When the nozzle wear is 0.002mm, the output X-direction pose compensation is +0.002mm; When the laser spot drift is 0.001mm, the output laser power is corrected by +3%.

[0055] By uniformly compensating for the errors of the two processes, defects such as steps and texture breaks can be effectively avoided in the process switching area.

[0056] (vi) Inspection upon completion of processing; After processing, the dynamic perspective visual measurement module 1 performs a full-area scan of the lens's irregular curved surface to collect the final contour data and surface roughness data.

[0057] The intelligent control module 3 judges the detection results: when the contour deviation is ±0.01mm and the surface roughness Ra≤0.01μm, the processing is deemed qualified, and the complete processing data is stored in the data storage and interaction module 4.

[0058] In this embodiment, the present invention does not simply connect dynamic perspective visual measurement, deep learning analysis, dual-process processing and error compensation as independent functional modules, but constructs a driven, integrated intelligent closed-loop collaborative control mechanism through the real-time multi-dimensional data output by the dynamic perspective visual measurement module 1.

[0059] Specifically, the dynamic perspective visual measurement module 1 continuously acquires multi-source information during processing, including real-time surface contour data, surface defect data, and tool status data. This data is not only used for judging the processing results or for post-processing correction, but is also used as a unified state input quantity, which is input to the intelligent control module 3 in real time. The intelligent control module 3, through a deep learning analysis unit 31, performs feature fusion and correlation modeling on the multi-dimensional data, thereby simultaneously outputting surface feature representations of the current processing state, dynamic deformation prediction results, and information on defect types and spatial distribution.

[0060] Based on this, the dual-process collaborative decision-making unit 32 does not switch processes according to fixed rules or a single threshold. Instead, it uses deep learning analysis results as the core decision-making variable to dynamically select between abrasive jetting and laser micro-cladding processes, and generates corresponding optimized process parameter sets in real time for different processing areas. This decision-making process is synchronized with the processing execution process, enabling the process type and parameters to be continuously and adaptively adjusted during processing, rather than being discretely switched between processing stages.

[0061] Furthermore, the cross-process error compensation unit 33 and the dual-process collaborative decision-making unit 32 operate in parallel. Their compensation calculations are not only based on changes in the tool state of a single process, but also integrate abrasive jet tool wear, laser spot drift, and path deviations introduced by process switching into the same error model for comprehensive calculation. The obtained compensation amount and process parameter optimization results are input to the real-time control unit 34, enabling the control signal to simultaneously complete process adaptive adjustment and cross-process error correction when driving the dual-process machining module 2.

[0062] Through the above mechanism, this embodiment realizes a closed-loop control method "actively driven" by the dynamic perspective vision measurement module 1. That is, changes in the processing state can simultaneously affect process selection, parameter adjustment and error compensation strategy within the same control cycle, thereby forming a highly coupled integrated closed-loop processing control process of "state perception - intelligent analysis - collaborative decision-making - error correction - execution control".

[0063] Compared with the traditional control method that separates multi-process processing, visual measurement and error compensation, this embodiment can maintain the synergistic consistency between the two processes during the processing, avoiding processing defects caused by process switching delays or error accumulation, thereby significantly improving processing accuracy, stability and overall processing efficiency in the processing of irregular curved surface optical components.

[0064] IV. Technical Effects; As can be seen from the above embodiments, the present invention achieves real-time closed-loop control of the dual-process machining through the synergistic effect of dynamic perspective visual measurement and intelligent control module 3. It can dynamically adjust the process type and parameters according to the changes in the processing state and uniformly compensate for cross-process errors, thereby significantly improving the processing accuracy, processing efficiency and processing stability of irregular curved surface optical elements.

[0065] Example 2 is the second embodiment of the present invention.

[0066] I. Description of Implementation Examples; Based on the intelligent processing control system for irregular curved surfaces of optical elements described in Embodiment 1, this embodiment further elaborates on the deep learning analysis mechanism, dual-process collaborative decision-making logic, and collaborative method of cross-process error compensation and real-time control within the intelligent control module 3. It highlights the technical implementation method of this invention in multi-model fusion analysis and integrated closed-loop control, as well as the resulting technical effects.

[0067] II. Working principle of deep learning analysis unit 31; In this embodiment, the deep learning analysis unit 31 is used to perform fusion analysis on the multi-source data acquired by the dynamic viewpoint visual measurement module 1, and its specific working method is as follows.

[0068] The dynamic perspective vision measurement module 1 outputs real-time contour data and surface defect data during the processing. At the same time, external sensors continuously provide external sensor data to the intelligent control module 3, including at least temperature data and vibration data of the processing area.

[0069] After receiving the above data, the deep learning analysis unit 31 inputs it into multiple pre-trained deep learning models for parallel processing: Analysis process of surface feature recognition model: Input real-time contour data into pre-trained surface feature recognition model, extract features from point cloud data of current processing area, and output actual curvature distribution information of current processing area and deviation from design contour.

[0070] This model can identify key processing areas such as high curvature regions and edge regions, and quantify the current processing allowance.

[0071] Detailed description of the surface feature recognition model: 1. Model structure; The surface feature recognition model is used to extract geometric feature information of the current processing area from the real-time contour data output by the dynamic viewpoint visual measurement module 1, including the actual curvature distribution and the deviation from the design contour.

[0072] The model employs a network structure that combines multi-scale convolutional feature extraction with geometrically constrained regression. Its structure includes: Input layer: The input is standardized real-time contour data, which is in the form of local surface height matrix and normal vector matrix obtained from image or point cloud reconstruction, used to describe the geometry of the current processing area.

[0073] Multi-scale convolutional feature extraction layer: A multi-layer two-dimensional convolutional network structure is adopted, with different convolutional layers corresponding to different receptive field scales, which are used to extract local micro curvature features and macro surface morphology features simultaneously; each convolutional layer is followed by a nonlinear activation function and a normalization layer to enhance the model's ability to distinguish different curvature regions (high curvature region, flat region, edge region).

[0074] Geometric feature fusion layer: The feature maps output by the multi-scale convolutional layers are fused at the channel level, and geometric prior information based on the design contour is introduced, so that the network considers the design model constraints at the feature expression stage.

[0075] Curvature and Deviation Regression Output Layer: The fully connected regression layer outputs the actual curvature distribution of the current processing area and the deviation data between it and the design profile, which serves as an important input for subsequent dual-process collaborative decision-making.

[0076] This model structure enables the system to acquire the geometric state of irregular curved surfaces in real time and stably during the processing, providing a reliable basis for process selection.

[0077] 2. Construction steps; Step S201: Spatial resampling and scale unification are performed on the real-time contour point cloud to construct the processing region domain Ω; Step S202: Calculate the local principal curvature κ at each point in the domain, and extract the spatial gradient components along the principal curvature direction and the secondary curvature direction; Step S203: Based on the dynamic viewing angle change frequency, assign different orders of Bessel functions to different viewing angles and perform spectral domain expansion on the curvature change; Step S204: Scale and bias enhancement of the curvature spectrum energy using the Gamma function and the Riemann Zeta function; Step S205: Apply L2 norm mapping and normalize the integral result, and output the surface feature evaluation quantity. .

[0078] 3. Formulas and explanations; ; in, The curvature spectrum energy index output by the surface feature recognition model. Define the point cloud domain for the current processing area. Let Gamma be the function, where These are the point cloud scale modulation parameters. For local principal curvature, For a Bessel function of the first kind, where For the first Each visual frequency order and These are the spatial gradient components along the principal curvature and secondary curvature directions, respectively. For Riemann Zeta function, where To design the order of curvature deviation, It is the L2 norm in the sense of square integrability. This is an adaptive normalization operator for noise.

[0079] The Gamma function is used to describe the nonlinear relationship between point cloud density and curvature scale, the Bessel function is used to characterize the oscillation characteristics of curvature at different viewpoint frequencies, and the Riemann Zeta function is used to enhance higher-order statistical features related to design curvature deviations.

[0080] When the output surface feature evaluation quantity When the corresponding machining allowance, after mapping, is greater than the first preset allowance threshold of 0.05mm, it indicates that the current area is still in the high allowance stage.

[0081] 4. Model technical effects; Through this model, the present invention transforms the complex geometric problem of irregular curved surfaces into a stable and quantifiable spectral energy problem, solving the technical problems of difficult accurate identification of high curvature regions and unstable evaluation of machining allowance in the prior art, and providing a reliable basis for subsequent dual-process collaborative decision-making.

[0082] Analysis process of dynamic deformation prediction model: After synchronizing the external sensor data with the real-time contour data, they are input into the pre-trained dynamic deformation prediction model to predict the trend of surface micro-deformation caused by factors such as thermal effect and vibration during the processing, and output micro-deformation prediction information.

[0083] In this embodiment, the micro-deformation prediction information is used to determine whether there is a controllable micro-deformation of no more than 0.005 mm in the current processing area, or whether there is a deformation trend that needs to be compensated in advance.

[0084] Detailed description of the dynamic deformation prediction model: 1. Model structure; The dynamic deformation prediction model is used to predict the trend of surface micro-deformation caused by factors such as thermal effects and vibration during the processing.

[0085] This model employs a network structure that combines temporal modeling with multimodal data fusion. Its structure includes: Multi-source input layer: The input includes real-time contour data, temperature data collected by infrared sensors, and vibration data collected by piezoelectric sensors. All types of data are aligned on the time axis.

[0086] Temporal feature encoding layer: For time-series data such as temperature and vibration, a recurrent neural network structure is used for encoding to extract time-related features during the processing; at the same time, the change sequence of contour data is encoded in parallel.

[0087] Cross-modal feature fusion layer: It fuses geometric change features with thermo-mechanical temporal features to construct a joint feature representation that reflects the relationship between "processing state and deformation response".

[0088] Micro-deformation prediction output layer: The regression layer outputs the predicted micro-deformation of the current processing area within the future control cycle.

[0089] This model structure can detect potential deformation risks in advance, enabling the intelligent control module 3 to complete parameter adjustments before significant deformation occurs.

[0090] 2. Construction steps; Step S301: Time-align the temperature data and vibration data collected by the external sensor; Step S302: Map the temperature gradient and vibration phase to the elliptic integral parameter space to describe the anisotropic response of the material; Step S303: Weight the historical inputs using an exponentially decaying kernel function to form a temporal convolution; Step S304: Use the exponential integral function and the error function to jointly model the short-term shock and the long-term cumulative effect; Step S305: Output the predicted value of micro-deformation at the current moment.

[0091] 3. Formulas and explanations; ; in, For a moment Predicting microvariables under the following conditions For time integration variables, The thermal-vibration coupling attenuation coefficient, For the third kind of elliptic integral, where For temperature gradient mapping parameters, For vibration phase, The anisotropy factor of the material. Let be an exponential integral function, where Energy dissipation rate, Let be the error function where This is the time sensitivity coefficient.

[0092] Elliptic integrals are used to characterize the non-uniform deformation capacity of materials in different directions, exponential integral functions are used to describe the energy dissipation behavior over time, and error functions are used to suppress the influence of measurement noise on short-term predictions.

[0093] When the predicted micro-deformation approaches or reaches 0.005 mm, the system determines that error compensation needs to be performed in advance.

[0094] 4. Model technical effects; This model solves the problem in existing technologies that rely solely on instantaneous measurements and cannot detect the trend of thermal-vibration coupling deformation in advance, enabling the system to achieve feedforward error control during the processing.

[0095] Analysis process of defect classification and localization model: Input surface defect data into pre-trained defect classification and localization model, classify and locate detected defects, and output defect type, severity and coordinate position.

[0096] Among them, the defect types include at least microcracks, material residues and steps; the severity of defects is quantified according to the defect size or depth. When the defect size or depth is less than 0.02 mm, it is defined as not reaching the second preset defect threshold. When the defect size or depth reaches or exceeds 0.02 mm, it is defined as reaching the second preset defect threshold.

[0097] Detailed description of the defect classification and localization model: 1. Model structure; The defect classification and location model is used to identify and label the type, severity, and spatial location of defects on the machined surface.

[0098] This model employs a deep neural network structure that integrates detection and classification, and its structure includes: Image feature input layer: The input is the defect candidate region image data output by the image processing unit 12, which includes surface texture, brightness variation and edge anomaly information.

[0099] Feature extraction backbone network: Deep convolutional structure is used to extract texture and structural features of defect areas, enabling the model to distinguish different defect morphologies such as microcracks, material residues, and steps.

[0100] Defect region localization branch: A spatial localization branch is constructed based on the feature map to output the precise coordinate position of the defect in the processing area.

[0101] Defect Classification and Severity Assessment Branch: A classification output layer is set up in parallel to output the defect type and its corresponding severity level, and is associated with the second preset defect threshold.

[0102] This model structure enables the system to identify defects and quantify their risk levels in real time during the manufacturing process, providing a basis for dual-process switching and repair strategies.

[0103] 2. Construction steps; Step S401: Extract candidate defect regions based on dynamic viewpoint visual measurement module 1; Step S402: Calculate the equivalent radius, texture complexity, and grayscale intensity contrast of the defect region; Step S403: Describe the oscillation characteristics of the defect edge using the Bessel function, and describe the higher-order statistical distribution of texture complexity using the Riemann Zeta function; Step S404: Integrate the area over the entire defect region to form a defect accumulation effect; Step S405: Output the probability of defect severity in an exponentially complementary form.

[0104] 3. Formulas and explanations; ; in, This is the probability that the defect reaches the severity threshold. The coordinate domain of the surface where the defect is located. For Riemann Zeta function, where The order of the defect texture complexity; Here, the Bessel function is... For edge oscillation frequency, The equivalent radius of the defect. For the grayscale intensity contrast of defects. Let be the complementary error function, where This represents the imaging noise parameter.

[0105] The model outputs the value Naturally falling within the [0,1] interval, when When the value is ≥0.02, the defect is determined to have reached or exceeded the second preset defect threshold, corresponding to the decision condition of "prioritizing laser micro-fusion repair".

[0106] 4. Model technical effects; By using probabilistic modeling, this invention avoids the problem of traditional hard threshold classification being sensitive to noise and achieves direct coupling between defect identification results and process priority decisions.

[0107] The aforementioned surface feature information, micro-deformation prediction information, and defect information together constitute the analysis results of the deep learning analysis unit 31, and are uniformly transmitted to the dual-process collaborative decision-making unit 32.

[0108] III. Working principle of dual-process collaborative decision-making unit 32; The dual-process collaborative decision-making unit 32 receives the analysis results from the deep learning analysis unit 31 and completes the selection of process type and the generation of optimized process parameter set based on the analysis results.

[0109] In this embodiment, the dual-process collaborative decision-making unit 32 uses machining allowance, defect severity, and dynamic deformation prediction information as the main decision-making basis, and its decision-making logic is as follows: When the analysis results show that the machining allowance in the current machining area is greater than the first preset allowance threshold, and the defect information indicates that there are no defects that reach the second preset defect threshold, the current process type is determined to be abrasive jet roughing, and an optimized process parameter set including jet pressure and moving speed is generated. When the analysis results show that the processing allowance of the current processing area is not greater than the first preset allowance threshold, or there are defects but the severity of the defects is lower than the second preset defect threshold, the decision will switch the process type to laser micro-cladding finishing or repair, and generate an optimized process parameter set including laser power, spot diameter and scanning path. When the analysis results indicate the presence of defects and the severity of the defects reaches or exceeds the second preset defect threshold, the decision is made to prioritize laser micro-cladding repair. After the repair is completed, the process is switched to laser micro-cladding finishing, and optimized process parameter sets for the corresponding stages are generated respectively.

[0110] In this embodiment, the first preset margin threshold is set to 0.05mm, and the second preset defect threshold is set to a defect size or depth of 0.02mm.

[0111] The dual-process collaborative decision-making unit 32 simultaneously outputs the final decision-made process type and the corresponding optimized process parameter set to the real-time control unit 34 and the cross-process error compensation unit 33.

[0112] IV. Working principle of cross-process error compensation unit 33; In this embodiment, the cross-process error compensation unit 33 is used to uniformly compensate for tool errors generated during dual-process machining. Its specific working process is as follows.

[0113] The cross-process error compensation unit 33 periodically receives tool status data from the dynamic perspective vision measurement module 1. The tool status data includes at least the wear status data of the abrasive jetting unit 21 and the spot status data of the laser micro-cladding unit 22. Simultaneously, this unit receives the current process type from the dual-process collaborative decision-making unit 32.

[0114] The cross-process error compensation unit 33 selects the appropriate error compensation model based on the process type: When the current process type is abrasive jet roughing, the cross-process error compensation unit 33 calculates the position compensation amount of the abrasive jet unit 21 in three-dimensional space based on the wear state data, so as to correct the machining path deviation caused by the wear of the jet nozzle.

[0115] When the current process type is laser micro-cladding finishing or repair, the cross-process error compensation unit 33 calculates the correction coefficient of laser power and the offset compensation amount of the laser spot focusing position based on the spot state data to ensure the consistency of laser energy distribution and processing accuracy requirements.

[0116] The calculated position compensation, correction coefficient, and offset compensation are output as compensation values ​​to the real-time control unit 34.

[0117] Detailed description of the error compensation model: 1. Model structure; The cross-process error compensation model is used to establish a unified error compensation mechanism between abrasive jetting and laser micro-cladding processes.

[0118] The model adopts a process state awareness + parameter mapping model structure, which includes: Tool status input layer: The input is the tool status data output by the dynamic view vision measurement module 1, including the wear status data of the abrasive jetting unit 21 and the spot status data of the laser micro-cladding unit 22.

[0119] Process type perception layer: Introduces the process type information currently output by the dual-process collaborative decision-making unit 32, enabling the model to distinguish different process states during error calculation.

[0120] Error Correlation Modeling Layer: Based on historical processing data, this layer establishes the correlation between tool state changes and processing errors, used to characterize the impact of wear and spot drift on the actual processing trajectory and energy input.

[0121] Compensation output layer: outputs three-dimensional spatial position compensation amount under abrasive spraying process, outputs laser power correction coefficient and spot focusing position offset compensation amount under laser micro cladding process, and outputs them uniformly to real-time control unit 34.

[0122] This model structure enables unified modeling and compensation for cross-process errors, avoiding the connection defects caused by traditional single-process independent correction.

[0123] 2. Construction steps; Step S501: Collect wear status data of abrasive jetting unit 21 and spot status data of laser micro-cladding unit 22; Step S502: Select the corresponding error mapping parameters according to the current process type; Step S503: Construct a statistical mapping relationship for the error using the Gamma function and the Riemann Zeta function; Step S504: Establish the energy-displacement coupling relationship using the modified Bessel function and the error function; Step S505: Solve the implicit equations using numerical iteration to obtain the compensation amount.

[0124] 3. Formulas and explanations; ; in, For compensation amount, For tool state sampling interval, Let Gamma be the function, where For the wear evolution order, For Riemann Zeta function, where For the statistical order of light spot drift, For the second type of modified Bessel function, where For the process mapping order, The energy decay coefficient, Let be the error function where To compensate for sensitivity.

[0125] Compensation amount The value range is [-0.003mm, 0.003mm]. Its sign and size represent the feedforward compensation or pullback compensation of the machining path, respectively. The real-time control unit completes the solution and issues control commands within a single control cycle.

[0126] 4. Model technical effects; This model solves the problems of inconsistent modeling of different process errors and the easy generation of step defects at process switching points in existing technologies, and realizes integrated and coordinated correction of dual process errors.

[0127] V. Working principle of real-time control unit 34; The real-time control unit 34 is used to integrate the process decision results with the error compensation results and generate the underlying control signals that drive the dual-process processing module 2 to perform processing actions.

[0128] Specifically, the real-time control unit 34 receives the process type and optimized process parameter set from the dual-process collaborative decision-making unit 32, and simultaneously receives the compensation amount from the cross-process error compensation unit 33. Based on the current process type, the real-time control unit 34 merges the corresponding optimized process parameter set and compensation amount to generate a unified drive instruction set, which includes at least spatial position instructions and process parameter instructions.

[0129] The real-time control unit 34 generates and issues drive instruction sets at a frequency no greater than a preset control cycle. Within a single control cycle, it completes the entire process from data reception and parameter fusion to control signal output, with an overall response time lower than a preset response time threshold. In this embodiment, the response time threshold is 10ms.

[0130] VI. Overall Technical Effects; As can be seen from the above embodiment 2, the present invention introduces a multi-model deep learning analysis mechanism inside the intelligent control module 3, and uses its analysis results as a unified input for dual-process collaborative decision-making and cross-process error compensation, thereby realizing real-time adaptive collaborative control of abrasive jetting and laser micro-cladding processes during the processing.

[0131] Compared with control methods based solely on a single process parameter or static rules, this embodiment can dynamically respond to changes in surface features, processing status, and tool status during the machining process, effectively reducing the accumulation of errors caused by process switching and improving the accuracy, stability, and consistency of machining irregular curved surfaces.

[0132] Example 3 is the third embodiment of the present invention.

[0133] I. Description of Implementation Examples; This embodiment further describes the intelligent processing control system for irregular curved surfaces of optical elements from the perspective of system composition and its collaborative operation. It focuses on the working principle and collaborative operation of the dynamic viewing angle vision measurement module 1, dual process processing module 2, motion platform unit 23, and data storage and interaction module 4 in the actual processing process, so as to demonstrate the comprehensive technical effect of the present invention in terms of processing perception, execution accuracy, and process traceability.

[0134] II. Structure and working principle of dynamic perspective visual measurement module 1; In this embodiment, the dynamic viewing angle vision measurement module 1 is used to perform blind-spot-free and continuous state perception of irregular curved surfaces during the processing, and it includes an image acquisition unit 11 and an image processing unit 12.

[0135] (a) Image acquisition unit 11; During the processing, the image acquisition unit 11 drives the motorized pan-tilt head equipped with an industrial camera to move along a preset path according to the control instructions from the intelligent control module 3. The preset path is planned based on the curvature distribution in the optical element design model, so that the pan-tilt head can be dynamically adjusted within the range of ±120° horizontally and ±90° vertically, thereby covering high curvature areas, edge areas and areas with dynamic changes in processing, and avoiding measurement blind spots caused by fixed viewing angles.

[0136] During the processing, the image acquisition unit 11 performs millisecond-level image acquisition on the processing area at a frame rate of not less than 100fps, providing continuous status input to the intelligent control module 3.

[0137] (ii) Image processing unit 12; The image processing unit 12 performs real-time processing on the acquired image data, including noise reduction using a Gaussian filtering algorithm and contour extraction using a Canny edge detection algorithm, to generate standardized real-time contour data.

[0138] Meanwhile, the image processing unit 12 identifies and labels surface anomalies in the image, generates surface defect data, and generates tool status data through image analysis of nozzle shape and laser spot morphology.

[0139] The aforementioned real-time contour data, surface defect data, and tool status data are uniformly output to the intelligent control module 3, providing basic data support for subsequent deep learning analysis, process decision-making, and error compensation.

[0140] III. Structure and working principle of dual-process machining module 2; (a) Working principle of abrasive jetting unit 21; The abrasive injection unit 21 includes an abrasive supply unit, a pressure control unit, and a nozzle actuator.

[0141] The abrasive supply unit is used to provide nanoscale abrasive with a particle size in the range of 50–500 nm; the pressure control unit is used to adjust the abrasive injection pressure to a first preset pressure range of 0.1–1.0 MPa; and the nozzle actuator moves along the planned path within a second preset speed range of 5–50 mm / s.

[0142] By coordinating the configuration of abrasive particle size, injection pressure and nozzle movement speed, the abrasive injection unit 21 can achieve the set material removal efficiency when roughing irregular curved surfaces, and make the surface roughness after processing consistently lower than the first preset roughness threshold Ra 0.8μm, thereby reserving a uniform processing allowance for subsequent finishing or repair stages.

[0143] (II) Working principle of laser micro-cladding unit 22; The laser micro-cladding unit 22 includes a laser generator, an optical path control unit, and a powder feeding unit.

[0144] The laser generating unit generates a laser beam with a wavelength of 1064nm and a power that is continuously adjustable within the third preset power range of 50–500W; the optical path control unit is used to adjust the spot diameter of the laser beam to a focusing range of 0.1–1.0mm and control the laser beam to move at a scanning speed within the fourth preset speed range of 10–100mm / s; the powder feeding unit transports cladding material powder that matches the optical element substrate to the processing area.

[0145] The various operating parameters of the laser micro-cladding unit 22 are set according to the process type and optimized process parameter set output by the dual process collaborative decision-making unit 32. These parameters are used to repair surface defects or to perform finishing on the processed surface, so that the surface roughness after finishing is lower than the second preset roughness threshold Ra 0.1μm.

[0146] IV. Structure and working principle of motion platform unit 23; The motion platform unit 23 includes a multi-axis drive unit and a position feedback unit.

[0147] The multi-axis drive unit is used to drive the abrasive jetting unit 21 or the laser micro-cladding unit 22 to move along a planned path in three-dimensional space under the control signal of the intelligent control module 3; the position feedback unit is used to detect the current spatial position of the tool in real time and generate a position feedback signal.

[0148] The multi-axis drive unit and the position feedback unit form a closed-loop control structure, ensuring that the spatial positioning accuracy of the motion platform unit 23 is not lower than the preset positioning accuracy threshold ±0.005mm, and the repeatability is not lower than the preset repeatability threshold ±0.002mm. This ensures that the dual-process machining unit strictly follows the planned path to perform machining actions and avoids machining errors caused by path deviation.

[0149] V. Structure and working principle of data storage and interaction module 4; The data storage and interaction module 4 includes a data recording unit 41, an external interface unit 42, and a human-computer interaction unit 43.

[0150] The data recording unit 41 continuously stores image data, control signals, optimized process parameter sets, and compensation amounts throughout the entire processing process in a time sequence, forming a complete and traceable processing process database.

[0151] The external interface unit 42 is used to receive externally input surface design model data and output the final contour data and process report after processing to the external system.

[0152] The human-machine interaction unit 43 is used to display images of the processing area, key information of the analysis results, the current process type and processing progress in real time, and to receive parameter input or intervention instructions from the operator.

[0153] VI. Overall working principle and technical effects of Example 3; In this embodiment, the dynamic perspective visual measurement module 1 achieves continuous perception of the processing status of irregular curved surfaces through blind-spot-free image acquisition and processing; the dual-process processing module 2, under the coordination of the intelligent control module 3, performs abrasive spraying or laser micro-cladding processing according to different processing stages and processing statuses; the motion platform unit 23 ensures the execution accuracy of the processing path through high-precision closed-loop control; and the data storage and interaction module 4 records and displays the entire processing process.

[0154] Through the above-mentioned system-level collaborative working method, this embodiment can simultaneously improve machining accuracy, machining stability and process controllability during the machining of irregular curved surfaces, and provide a reliable data foundation for process optimization and quality traceability, thereby effectively solving the problems of blind spots in measurement, insufficient execution accuracy and lack of traceability in the existing technology.

[0155] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principle of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. An intelligent processing and control system for irregular curved surfaces of optical elements, characterized in that, include: The dynamic perspective vision measurement module is used to acquire and process images of the processing area in real time, and generate image data, including real-time contour data, surface defect data and tool status data. The dual-processing module includes an abrasive jetting unit, a laser micro-cladding unit, and a motion platform unit; An intelligent control module, connecting the dynamic perspective vision measurement module and the dual-process processing module, includes: The deep learning analysis unit is used to receive the real-time contour data and the surface defect data, process them, and output analysis results containing surface features, deformation prediction, and defect information. A dual-process collaborative decision-making unit, connected to the deep learning analysis unit, is used to generate process types and corresponding optimized process parameter sets based on the analysis results; A cross-process error compensation unit, connected to the dual-process collaborative decision-making unit, is used to calculate the compensation amount for tool pose and process parameters based on the tool status data and the process type. The real-time control unit is connected to the dual-process collaborative decision-making unit and the cross-process error compensation unit, respectively, and is used to generate control signals based on the optimized process parameter set and the compensation amount to drive the dual-process processing module. The data storage and interaction module connects the dynamic perspective visual measurement module, the dual-process processing module, and the intelligent control module, and is used to store data and provide an interaction interface; During the processing, the dynamic perspective vision measurement module acquires image data of the processing area and inputs it into the intelligent control module; the intelligent control module generates control signals from the image data to drive the dual-process processing module to operate; the dynamic perspective vision measurement module acquires image data of the dual-process processing module after it operates again and feeds it back to the intelligent control module.

2. The intelligent processing control system for irregular curved surfaces of optical elements according to claim 1, characterized in that, The deep learning analysis unit is specifically used for: Receives real-time contour data and surface defect data from the dynamic viewpoint vision measurement module, as well as data from externally provided external sensors; The real-time contour data is input into a pre-trained surface feature recognition model, which outputs surface feature information including the actual curvature distribution of the current processing area and the deviation from the design contour. The external sensor data and the real-time contour data are input together into a pre-trained dynamic deformation prediction model, which outputs the micro-deformation prediction information of the current processing area. The surface defect data is input into a pre-trained defect classification and localization model, which outputs defect information including defect type, severity, and coordinate location. The surface feature information, the micro-deformation prediction information, and the defect information together constitute the analysis result and are transmitted to the dual-process collaborative decision-making unit.

3. The intelligent processing control system for irregular curved surfaces of optical elements according to claim 1, characterized in that, The dual-process collaborative decision-making unit is specifically used for: The analysis results received from the deep learning analysis unit include at least the deviation from the design profile, micro-deformation prediction information, and defect information. The judgment is based on the deviation from the design profile: If the machining allowance in the current processing area is greater than the first preset allowance threshold, and the defect information indicates that there is no specific serious defect, then the process type is determined to be abrasive jet roughing, and the optimized process parameter set including jet pressure and moving speed is generated. If the current processing allowance in the processing area is not greater than the first preset allowance threshold, or if the defect information indicates the existence of a defect but its severity is lower than the second preset defect threshold, then the process type is switched to laser micro-cladding finishing or repair, and the optimized process parameter set containing laser power, spot diameter and scanning path is generated. If the defect information indicates the existence of a defect and its severity reaches or exceeds the second preset defect threshold, then the decision is made to prioritize laser micro-cladding repair, and after the repair is completed, switch to laser micro-cladding finishing to generate the corresponding set of optimized process parameters. The determined process type and the corresponding optimized process parameter set are output to the real-time control unit and the cross-process error compensation unit.

4. The intelligent processing control system for irregular curved surfaces of optical elements according to claim 1, characterized in that, The cross-process error compensation unit is specifically used for: The tool status data is periodically received from the dynamic perspective vision measurement module, and the tool status data includes at least the wear status data of the abrasive jetting unit and the spot status data of the laser micro-cladding unit. Receive the process type of the current decision from the dual-process collaborative decision-making unit; Based on the aforementioned process type, select the appropriate error compensation model: When the process type is abrasive jet roughing, the position compensation amount of the abrasive jetting unit in three-dimensional space is calculated based on the wear state data. When the process type is laser micro-cladding finishing or repair, the correction coefficient of laser power and the offset compensation amount of the spot focusing position are calculated based on the spot state data. The calculated position compensation amount, the correction coefficient, and the offset compensation amount are used as the compensation amount and output to the real-time control unit.

5. The intelligent processing control system for irregular curved surfaces of optical elements according to claim 1, characterized in that, The real-time control unit is specifically used for: Receive the process type and the optimized process parameter set from the dual-process collaborative decision-making unit, and the compensation amount from the cross-process error compensation unit; According to the process type, the corresponding optimized process parameter set and the compensation amount are fused to generate a unified drive instruction set that includes spatial position instructions and process parameter instructions; At a frequency not greater than a preset control cycle, the drive instruction set is converted into a low-level control signal and synchronously sent to the motion platform unit, the abrasive jetting unit, or the laser micro-cladding unit in the dual-process processing module. The real-time control unit is also used to complete the entire process from receiving data to issuing the underlying control signal within a single control cycle, and its response time is lower than a preset response time threshold.

6. The intelligent processing control system for irregular curved surfaces of optical elements according to claim 1, characterized in that, The dynamic perspective visual measurement module includes: The image acquisition unit is used to drive the gimbal equipped with an industrial camera to move along a preset path according to the control instructions from the intelligent control module during the processing, so as to acquire the image data of the processing area without blind spots, wherein the preset path is planned based on the curvature distribution in the design model of the optical element. The image processing unit, connected to the image acquisition unit, is used to perform noise reduction and contour extraction on the acquired image data, generate standardized real-time contour data, surface defect data, and tool status data, and output them to the intelligent control module.

7. The intelligent processing control system for irregular curved surfaces of optical elements according to claim 1, characterized in that, The abrasive jetting unit includes: The abrasive supply unit is used to provide nanoscale abrasives within a set particle size range; The pressure regulation unit, connected to the abrasive supply unit, is used to receive control signals and adjust the injection pressure to a first preset pressure range; The nozzle actuator, connected to the pressure control unit, is used to perform jetting at a moving speed within a second preset speed range; The set particle size range, the first preset pressure range, and the second preset speed range are configured to work together to achieve the set material removal efficiency when roughing irregular curved surfaces, and to control the surface roughness to be lower than the first preset roughness threshold.

8. The intelligent processing control system for irregular curved surfaces of optical elements according to claim 1, characterized in that, The laser micro-cladding unit includes: The laser generator is used to generate a laser beam with adjustable wavelength and power, wherein the power can be continuously adjusted within a third preset power range; The optical path control unit is connected to the laser generator unit and is used to adjust the spot diameter of the laser beam to a set focusing range and control the laser beam to move at a scanning speed within a fourth preset speed range. The powder feeding section is used to transport cladding material powder that matches the optical element substrate to the processing area; The working parameters of the optical path control unit and the powder feeding unit are set based on the process type and the optimized process parameter set output by the dual process collaborative decision unit, so as to achieve the repair of surface defects or the finishing of the processed surface, and make the surface roughness after finishing lower than the second preset roughness threshold.

9. The intelligent processing control system for irregular curved surfaces of optical elements according to claim 1, characterized in that, The motion platform unit includes: A multi-axis drive unit is used to drive the abrasive jetting unit or the laser micro-cladding unit to move along a planned path in three-dimensional space according to the received control signal. The position feedback unit, connected to the multi-axis drive unit, is used to detect the spatial position of the abrasive jetting unit or the laser micro-cladding unit in real time and generate a position feedback signal. The multi-axis drive unit and the position feedback unit form a closed-loop control, which ensures that the spatial positioning accuracy of the motion platform unit is not lower than a preset positioning accuracy threshold and the repeatability positioning accuracy is not lower than a preset repeatability positioning accuracy threshold, so as to ensure that the abrasive spraying unit or the laser micro-cladding unit moves according to the planned path.

10. The intelligent processing control system for irregular curved surfaces of optical elements according to claim 1, characterized in that, The data storage and interaction module includes: The data recording unit is used to continuously store the image data, control signals, optimized process parameter set, and compensation amount in the entire processing flow in a time sequence, forming a traceable processing process database; The external interface unit is used to receive surface design model data from external input and output the final contour data and process report after processing to the external system. The human-computer interaction unit is used to display images of the processing area, key information of the analysis results, the process type and processing progress in real time, and to receive parameter input or intervention instructions from the operator.

Images