High-precision VR lens cone processing device

Abstract

The invention relates to the technical field of VR lens cone processing, in particular to a high-precision VR lens cone processing device which comprises a processing table. A placing table is fixedly connected to the top end of the outer wall of the processing table; a square through groove is formed in the top end of the outer wall of the placement table; the top end of the outer wall of the machining table is fixedly connected with driving equipment. A double-section threaded shaft is arranged at the output end of the driving equipment, and one end of the outer wall of the double-section threaded shaft extends into the square through groove; the driving device drives the double-section threaded shaft to rotate, so that the double-section threaded shaft drives the pair of clamping plates to clamp and fix the outer side wall of the lens barrel, and meanwhile, the double-section threaded shaft rotates to drive the double-section threaded rod to rotate through the pair of bevel gears, so that the double-section threaded rod drives the pair of inner supporting plates to abut against the inner side wall of the lens barrel; the mechanical stability of the lens cone in the machining state is improved, the clamping force and the supporting force are perpendicular to each other in the circumferential direction and are evenly dispersed, and micro-deformation or vibration of the lens cone in the machining process is avoided.

Description

Technical Field

[0001] This invention relates to the field of VR lens barrel processing technology, specifically a high-precision VR lens barrel processing device. Background Technology

[0002] The high-precision VR lens barrel processing device does not refer to a single piece of equipment. It includes a complete set of technologies and equipment from mold forming and ultra-precision component processing to automated assembly and testing. When assembling and fitting high-precision VR lens barrels, clamping equipment is often required to fix the lens barrels so that the lens barrels can be easily assembled and fitted into subsequent assembly processes.

[0003] In existing technologies, high-precision VR lens barrel processing devices generally use a single external clamping method to fix the lens barrel when clamping it, lacking internal support coordination. This has obvious mechanical defects, making the simple external clamping prone to radial shrinkage deformation of the lens barrel. Especially when the clamping force is large or the wall thickness is uneven, the clamping stress will cause a decrease in the micro-geometric accuracy of the lens barrel, directly affecting the fitting clearance and optical center alignment accuracy of subsequent lens assembly. When using external clamping and internal support, traditional external clamping and internal support are mostly completed step by step by two independent mechanisms, which not only increases the complexity of the equipment, but also makes it difficult to ensure the synchronization and matching degree of internal and external forces. It is easy to have situations where the clamping is locked but the support is not in place, or the support is pushed out but the clamping is loose, resulting in micro-displacement or secondary deformation of the lens barrel during processing, leading to defects in the processed lens barrel. Summary of the Invention

[0004] The purpose of this invention is to solve the problems mentioned in the background art and to propose a high-precision VR lens barrel processing device.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] This invention provides a high-precision VR lens barrel processing device, including a processing table; a placement platform is fixedly connected to the top of the outer wall of the processing table; a square through groove is formed on the top of the outer wall of the placement platform; a driving device is fixedly connected to the top of the outer wall of the processing table; the output end of the driving device is provided with a double-segment threaded shaft, and one end of the outer wall of the double-segment threaded shaft extends into the square through groove; a pair of clamping plates are threadedly connected to the outer wall of the double-segment threaded shaft, and the outer walls of the pair of clamping plates are slidably connected in the square through groove; a pair of arc-shaped locking blocks are fixedly connected to the top of the outer wall of the placement platform; a double-segment threaded rod is rotatably connected to the top of the outer wall of the placement platform through a connecting block; a pair of inner support plates are threadedly connected to the outer wall of the double-segment threaded rod, and the outer walls of the pair of inner support plates are in contact with one side of the outer wall of the pair of arc-shaped locking blocks; helical gears are fixedly connected to the outer walls of both the double-segment threaded rod and the double-segment threaded shaft, and the pair of helical gears mesh.

[0007] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0008] 1. The drive device rotates the double-segment threaded shaft, causing it to move a pair of clamping plates towards the center simultaneously, clamping and fixing the outer wall of the lens barrel. At the same time, the double-segment threaded shaft drives the double-segment threaded rod to rotate via a pair of helical gears, thereby causing a pair of inner support plates to move outward simultaneously, pressing against the inner wall of the lens barrel. This improves the mechanical stability of the lens barrel during processing. Furthermore, the inner support plates apply outward compressive force from the inner wall of the lens barrel, forming opposing supports with the outer clamping plates. This ensures that the clamping force and support force are perpendicular and evenly distributed in the circumferential direction, effectively preventing micro-deformation or vibration of the lens barrel during processing. The single power source design ensures the synchronization of internal and external movements, eliminating the problems of unstable clamping or inadequate support caused by timing deviations in traditional multi-power source solutions, making the lens barrel more stable during processing.

[0009] 2. The double-segment threaded shaft drives the first bevel gear to rotate, which in turn drives the second bevel gear to rotate, thereby driving the reciprocating rod to rotate. The rotation of the reciprocating rod causes the reciprocating plate to move up and down. During the clamping process and the inner support plate to squeeze, the reciprocating plate drives the circular plate, cleaning rod, and cleaning brush to move down first and then up to reset, cleaning the inside of the lens barrel. This effectively removes residual processing debris and dust from the inner wall of the lens barrel, preventing impurities from interfering with subsequent processing. The simultaneous design of fixing and cleaning integrates the clamping and fixing with the pre-treatment cleaning process, eliminating the separate cleaning step in the traditional processing flow, shortening the processing preparation time, and ensuring that the lens barrel is in a clean and stable state before entering the formal processing stage, reducing the risk of processing errors caused by foreign objects. Attached Figure Description

[0010] Figure 1 This is a structural diagram of the main body of the present invention;

[0011] Figure 2 This is an exploded view of the clamping plate and placement platform of the present invention;

[0012] Figure 3 This is an exploded view of the connecting block and the double-segment threaded rod of the present invention;

[0013] Figure 4 This is a structural diagram of the drive device, reciprocating rod, reciprocating plate, and cleaning brush of the present invention.

[0014] Figure 5 This is a structural diagram of the rotating rod, the fourth sprocket, the fifth sprocket, and the chain of the present invention;

[0015] Figure 6 This is an exploded view of the rotating rod and the fifth sprocket of the present invention;

[0016] Figure 7 This is an exploded view of the reciprocating plate and the circular plate of the present invention;

[0017] Figure 8 The flowchart shows the adaptive clamping force intelligent control method based on multi-physics field coupling sensing.

[0018] Figure 9 This is a block diagram of an adaptive clamping force intelligent control system based on multi-physics field coupling sensing. Detailed Implementation

[0019] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. 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.

[0020] Example 1: Please refer to Figures 1-7 As shown, a high-precision VR lens barrel processing device includes a processing table 1; a placement table 2 is fixedly connected to the top of the outer wall of the processing table 1; a square through groove 3 is formed on the top of the outer wall of the placement table 2; a driving device 4 is fixedly connected to the top of the outer wall of the processing table 1; a double-segment threaded shaft 5 is provided at the output end of the driving device 4, and one end of the outer wall of the double-segment threaded shaft 5 extends into the square through groove 3; a pair of clamping plates 6 are threadedly connected to the outer wall of the double-segment threaded shaft 5, and the outer walls of the pair of clamping plates 6 are slidably connected in the square through groove 3; a pair of arc-shaped locking blocks 7 are fixedly connected to the top of the outer wall of the placement table 2; a double-segment threaded rod 9 is rotatably connected to the top of the outer wall of the placement table 2 through a connecting block 8; a pair of inner support plates 10 are threadedly connected to the outer wall of the double-segment threaded rod 9, and the outer walls of the pair of inner support plates 10 are in contact with one side of the outer wall of the pair of arc-shaped locking blocks 7; the outer walls of the double-segment threaded rod 9 and the double-segment threaded shaft 5 are both fixedly connected to the outer wall of the placement table 2. A pair of helical gears 11 are connected and meshed. When the lens barrel to be assembled or polished is placed on the placement table 2, a pair of arc-shaped locking blocks 7 are inserted into the bottom of the inner wall of the lens barrel. At this time, the drive device 4 drives the double-segment threaded shaft 5 to rotate, which drives a pair of clamping plates 6 to move towards the center at the same time, clamping and fixing the outer wall of the lens barrel. At the same time, the rotation of the double-segment threaded shaft 5 drives the double-segment threaded rod 9 to rotate through the pair of helical gears 11, which drives a pair of inner support plates 10 to move outward at the same time, abutting against the inner wall of the lens barrel. While the pair of clamping plates 6 clamp and fix the outer wall of the lens barrel, the pair of inner support plates 10 abut against and support the inner wall of the lens barrel, improving the mechanical stability of the lens barrel in the processing state, and applying outward extrusion force from the inner wall of the lens barrel through the inner support plates 10, forming an inner and outer opposing support with the outer clamping plates 6.

[0021] The connecting block 8 is inverted U-shaped; the connecting block 8, together with a pair of arc-shaped locking blocks 7 and a square through slot 3, forms a shielding space, and a pair of helical gears 11 are located within the shielding space; the top of the outer wall of the processing table 1 is rotatably connected to a reciprocating rod 13 via a connecting plate 12; a reciprocating plate 14 is provided on the outer wall of the reciprocating rod 13; a rotating rod 15 is provided on the top of the outer wall of the connecting plate 12, and the top of the outer wall of the rotating rod 15 passes through the reciprocating plate 14; a first bevel gear 16 is fixedly connected to the outer wall of the double-segment threaded shaft 5; a second bevel gear 17 is fixedly connected to the bottom of the outer wall of the reciprocating rod 13, and the first bevel gear 16 and the second bevel gear 17 mesh; a circular plate 18 is provided on the inner wall of the reciprocating plate 14; a set of cleaning rods 19 is provided on the bottom of the outer wall of the circular plate 18; the cleaning rods A set of cleaning brushes 20 is provided on the outer wall of lens 19. The double-segment threaded shaft 5 drives the first bevel gear 16 to rotate, which in turn drives the second bevel gear 17 to rotate, thereby driving the reciprocating rod 13 to rotate. The rotation of the reciprocating rod 13 causes the reciprocating plate 14 to move up and down. During the clamping process of the clamping plate 6 and the squeezing process of the inner support plate 10, the reciprocating plate 14 drives the circular plate 18, cleaning rod 19 and cleaning brushes 20 to move down first and then up to reset, cleaning the inside of the lens barrel. When the clamping plate 6 is clamped and the inner support plate 10 is squeezed, the cleaning brushes 20 have finished cleaning and the reciprocating plate 14 has reset. This completes the cleaning of the lens barrel during the lens barrel fixing process, effectively removing the machining debris and dust remaining on the inner wall of the lens barrel. Because the connecting block 8, together with a pair of arc-shaped locking blocks 7 and a square through slot 3, forms a shielding space, and a pair of helical gears 11 are located in the shielding space, the pair of helical gears 11 are shielded by the connecting block 8, making it less likely for the pair of helical gears 11 to be disturbed during operation.

[0022] The outer wall of the circular plate 18 is rotatably connected to the inner wall of the reciprocating plate 14; the outer walls of both the rotating rod 15 and the reciprocating rod 13 are fixedly connected to a third gear 21, and a pair of third gears 21 mesh; the top of the outer wall of the circular plate 18 is fixedly connected to a fourth sprocket 22; a groove 23 is provided on the outer wall of the reciprocating rod 13; the top of the outer wall of the reciprocating plate 14 is rotatably connected to a fifth sprocket 24, and the inner wall of the fifth sprocket 24 is slidably connected to the inner wall of the groove 23; the fourth sprocket 22 and the fifth sprocket 24 are connected by a chain 25; the bottom of the outer wall of the rotating rod 15 is rotatably connected to the top of the outer wall of the connecting plate 12, and the reciprocating rod 13 is driven by the pair of third gears 21 because the outer walls of both the rotating rod 15 and the reciprocating rod 13 are fixedly connected to a third gear 21, and a pair of third gears 21 mesh. Rotating rod 15 rotates, and the rotation of rotating rod 15 drives the fifth sprocket 24 to rotate through slide groove 23. When reciprocating plate 14 moves up and down, it drives the fifth sprocket 24 to move within slide groove 23. The fifth sprocket 24 rotates and moves at the same time, always keeping it on the same plane as the fourth sprocket 22. Thus, the fifth sprocket 24 drives the fourth sprocket 22 to rotate through chain 25. The fourth sprocket 22 drives the circular rod to rotate, and the circular rod drives the cleaning rod 19 and cleaning brush 20 to rotate. This increases the cleaning range of a single cleaning brush 20 and improves the cleaning effect. When the cleaning brush 20 contacts the inner wall of the lens barrel, the centrifugal force generated by the rotation makes the bristles closely adhere to the curved surface. During the up and down movement, it performs axial sweeping of the inner wall of the lens barrel, while the rotation achieves circumferential coverage, which can effectively remove residual debris on the inner wall.

[0023] A ring rack 26 is fixedly connected to the bottom of the outer wall of the reciprocating plate 14; the top of the outer wall of a set of cleaning rods 19 is rotatably connected to the bottom of the outer wall of the circular plate 18; a sixth gear 27 is fixedly connected to the outer wall of a set of cleaning rods 19; the sixth gear 27 meshes with the ring rack 26. When the circular rod drives the cleaning rod 19 to rotate, the cleaning rod 19 drives the sixth gear 27 to rotate, so that the sixth gear 27 rotates through the ring rack 26, thereby driving the cleaning rod 19 to rotate, so that the cleaning rod 19 drives a set of cleaning brushes 20 to rotate, so that a single cleaning brush 20 maintains its own rotation while revolving around the revolution, increasing the number of brush bristles friction per unit area, and improving the cleanliness of the inner wall compared to the traditional single cleaning method. In addition, the rotatable connection design between the cleaning rod 19 and the circular plate 18 allows the cleaning brush 20 to generate an adaptive swing angle when contacting the inner wall with different curvatures, ensuring that the bristles always fit the curved surface at the optimal angle, avoiding the problem of incomplete cleaning in some areas due to rigid contact, thereby achieving a full-area cleaning without dead corners from the end of the mirror barrel to the depth of the inner wall.

[0024] In use, the lens barrel that needs to be assembled, polished, or otherwise processed is placed on the placement table 2, so that a pair of arc-shaped locking blocks 7 are inserted into the bottom of the inner wall of the lens barrel. At this time, the drive device 4 drives the double-segment threaded shaft 5 to rotate, so that the double-segment threaded shaft 5 drives a pair of clamping plates 6 to move towards the center at the same time, clamping and fixing the outer wall of the lens barrel. At the same time, the rotation of the double-segment threaded shaft 5 drives the double-segment threaded rod 9 to rotate through a pair of helical gears 11, so that the double-segment threaded rod 9 drives a pair of inner support plates 10 to move outward at the same time, abutting against the inner wall of the lens barrel. While the pair of clamping plates 6 clamps and fixes the outer wall of the lens barrel, the pair of inner support plates 10 abut against and support the inner wall of the lens barrel. The rotation of the double-segment threaded shaft 5 synchronously drives the first bevel gear 16 to rotate, which in turn drives the second bevel gear 17 to rotate, thereby driving the reciprocating rod 13 to rotate. The rotation of the reciprocating rod 13 causes the reciprocating plate 14 to move up and down reciprocally. During the process of clamping the clamping plate 6 and squeezing the inner support plate 10, the reciprocating plate 14 drives the circular plate 18, cleaning rod 19 and cleaning brush 20 to move down first and then up to reset, cleaning the inside of the lens barrel. When the clamping plate 6 is clamped and the inner support plate 10 is squeezed, the cleaning brush 20 completes the cleaning, and the reciprocating plate 14 resets, thus completing the cleaning of the lens barrel during the lens barrel fixing process. While the reciprocating rod 13 rotates, it drives the rotating rod 15 to rotate via a pair of third gears 21. The rotation of the rotating rod 15 drives the fifth sprocket 24 to rotate via the slide groove 23. When the reciprocating plate 14 moves up and down, it drives the fifth sprocket 24 to move within the slide groove 23. The fifth sprocket 24 rotates and moves at the same time, always keeping it on the same plane as the fourth sprocket 22. This causes the fifth sprocket 24 to drive the fourth sprocket 22 to rotate via the chain 25. The fourth sprocket 22 drives the circular rod to rotate, which in turn drives the cleaning rod 19 and the cleaning brush 20 to rotate, increasing the cleaning range of a single cleaning brush 20. When the cleaning brush 20 contacts the inner wall of the lens barrel, the centrifugal force generated by the rotation makes the bristles fit tightly against the curved surface. During the up and down movement, it performs axial sweeping of the inner wall of the lens barrel, while the rotation achieves circumferential coverage. When the circular rod drives the cleaning rod 19 to rotate, the cleaning rod 19 drives the sixth gear 27 to rotate. When the sixth gear 27 rotates, it rotates through the ring rack 26, thereby driving the cleaning rod 19 to rotate. This causes the cleaning rod 19 to drive a set of cleaning brushes 20 to rotate, so that each cleaning brush 20 maintains its own rotation while revolving around the central axis. The rotational connection between the cleaning rod 19 and the circular plate 18 allows the cleaning brushes 20 to generate an adaptive swing angle when contacting inner walls with different curvatures. This ensures that the bristles always fit the curved surface at the optimal angle, avoiding the problem of incomplete cleaning in certain areas due to rigid contact. This achieves a thorough cleaning of the entire area from the end of the lens barrel to the depths of the inner wall without any dead angles.

[0025] Example 2: Please refer to Figures 8-9As shown, this embodiment, based on embodiment 1, provides an adaptive clamping force intelligent control scheme based on multi-physics field coupling sensing. This scheme is used to monitor the mechanical state of the clamping system in real time during the lens barrel processing and dynamically adjust the balance between clamping force and support force to avoid micro-deformation of the lens barrel caused by processing vibration, friction heat generation and other factors. This embodiment specifically involves the deployment of three types of sensors, signal acquisition methods and intelligent control process based on multi-physics field coupling.

[0026] In this embodiment, PZT piezoelectric ceramic patches are symmetrically attached to the inner contact surfaces of a pair of clamping plates 6. These PZT piezoelectric ceramic patches are used simultaneously as a high-frequency sweep excitation source and a mechanical impedance response sensor via an impedance analyzer circuit. Fiber Bragg grating strain sensor arrays are embedded along the generatrix direction on the arc surfaces of a pair of inner support plates 10. Each fiber Bragg grating strain sensor array contains at least two gratings with different center wavelengths to achieve synchronous decoupled measurement of strain and temperature. A pair of differential eddy current micro-displacement sensors are symmetrically installed at the top of the outer wall of the placement stage 2. The sensor probes are pointed to the outer wall of the lens barrel and arranged symmetrically about the central axis of the lens barrel. An embedded real-time signal processing and control unit is fixedly connected to the top of the outer wall of the processing table 1. The embedded real-time signal processing and control unit is electrically connected to the driving device 4, the PZT piezoelectric ceramic patch, the fiber Bragg grating strain sensor array and the differential eddy current micro-displacement sensor. The embedded real-time signal processing and control unit executes an adaptive clamping force intelligent control process based on multi-physics field coupling sensing. This control process includes steps S100, S200, S300, S400 and S500.

[0027] Two PZT piezoelectric ceramic patches are symmetrically attached to the inner contact surface of each clamping plate 6, at positions one-third and two-thirds of the height of the contact surface. The PZT piezoelectric ceramic patches are connected to the impedance analyzer circuit module via shielded coaxial wires. The impedance analyzer circuit module is integrated into the embedded real-time signal processing and control unit. When an alternating voltage signal is applied, the patches generate mechanical vibrations that propagate to the contact interface between the clamping plate 6 and the lens barrel wall. Changes in the mechanical state of the contact interface alter the mechanical admittance of the structure, thereby affecting the impedance characteristics of the patches.

[0028] The fiber Bragg grating strain sensor array is arranged as follows: an optical fiber containing at least two different center wavelength gratings is embedded at equal intervals along the generatrix direction on the arc surface of each inner support plate 10. For example, three fiber Bragg gratings are arranged in series on each optical fiber. The three gratings are distributed at equal intervals along the generatrix direction of the arc surface of the inner support plate 10. The spacing between adjacent gratings is 1 / 4 of the arc surface length of the inner support plate 10. The optical fiber is embedded into the surface of the inner support plate 10 through a pre-made groove. The surface of the optical fiber is flush with the arc surface of the inner support plate 10. The lead-out end of the optical fiber is connected to the optical fiber demodulator through a miniature optical fiber connector. The optical fiber demodulator is integrated into the embedded real-time signal processing and control unit.

[0029] The differential eddy current micro-displacement sensor is installed as follows: a pair of differential eddy current micro-displacement sensors are installed on the top of the outer wall of the placement stage 2, with their probe axis pointing in the radial direction of the outer wall of the lens barrel. The two differential eddy current micro-displacement sensors are arranged symmetrically about the central axis of the lens barrel. The initial gap between the probe of the differential eddy current micro-displacement sensor and the outer wall of the lens barrel is set to the middle value of the range.

[0030] The hardware architecture of the embedded real-time signal processing and control unit is as follows: It uses an ARM Cortex A72 processor as the main control core, with ample memory and storage space. Internally, this unit integrates an impedance analyzer circuit module, a fiber optic demodulator module, an eddy current signal conditioning module, an analog-to-digital converter module, and a digital-to-analog converter module. The impedance analyzer circuit module includes a DDS signal generator and a high-speed analog-to-digital converter. The embedded real-time signal processing and control unit communicates with the driver device 4 via an RS485 bus. This allows for the integration of sensor data acquisition, signal processing, algorithm execution, and control command output onto a single embedded platform, meeting the time constraints of real-time control during the lens barrel manufacturing process.

[0031] Step S100: Mechanical impedance spectrum excitation acquisition and contact interface mechanical fingerprint extraction: Through the self-excitation and self-sensing characteristics of the PZT piezoelectric ceramic patch, the mechanical state information of the contact interface between the clamping plate 6 and the lens barrel wall is obtained, and the information is condensed into a quantifiable contact interface mechanical fingerprint vector. Step S100 specifically includes steps S101, S102 and S103.

[0032] Step S101: The embedded real-time signal processing and control unit drives each PZT piezoelectric ceramic patch to apply a linear frequency-modulated sweep voltage signal within a preset frequency range, synchronously acquires the voltage and current response signals of each PZT piezoelectric ceramic patch, and calculates the mechanical impedance spectrum Z(ω) of each patch based on Fourier transform.

[0033] Specifically, the DDS signal generator produces a linear frequency-modulated sweep voltage signal, with the sweep signal starting at a frequency of... Termination frequency and sweep time Based on the structural characteristics of the clamping plate 6 and the mirror barrel, the method for determining the initial frequency is as follows: Free modal analysis is performed on the contact structure composed of the clamping plate 6 and the mirror barrel wall using finite element modal analysis software to extract its first-order natural frequency. The initial frequency is set to 0.5 times the first natural frequency. The termination frequency is set to 10 times the first-order natural frequency. This ensures that the frequency sweep range covers the first 10 modes.

[0034] When a linear frequency-modulated sweep voltage signal is applied to the PZT piezoelectric ceramic patch, the patch generates minute mechanical vibrations due to the inverse piezoelectric effect. This vibration is transmitted through the adhesive layer to the contact interface between the clamping plate 6 and the lens barrel wall. The mechanical properties of the contact interface, including contact stiffness and contact damping, affect the mechanical admittance of the structure. Changes in mechanical admittance are reflected in the impedance of the PZT piezoelectric ceramic patch through the direct piezoelectric effect. The embedded real-time signal processing and control unit synchronously acquires the voltage signal V(t) applied to the PZT piezoelectric ceramic patch and the current signal I(t) flowing through the patch. After acquisition, discrete Fourier transforms are performed on the voltage signal V(t) and current signal I(t) to obtain the frequency domain voltage V(f) and frequency domain current I(f). Then, the computer calculates the impedance spectrum Z(ω) = V(f) / I(f), where Z(ω) is a complex function, with its real part representing the resistance component and its imaginary part representing the reactance component. This process is repeated for each PZT piezoelectric ceramic patch to obtain the impedance spectrum corresponding to each patch position.

[0035] Step S102: Perform multi-order resonance feature extraction on the mechanical impedance spectrum, extracting the offset of the first m order resonance frequencies relative to the free-state reference spectrum. The change in damping ratio corresponding to each modal order and the change in Q factor at the anti-resonance frequency .

[0036] Specifically, the method for obtaining the free-state reference spectrum is as follows: In the free state where the clamping plate 6 is not in contact with the lens barrel, the frequency sweep acquisition process of step S101 is performed on each PZT piezoelectric ceramic patch to obtain the electromechanical impedance spectrum in the free state. The free-state reference spectrum is stored in the memory of the embedded real-time signal processing and control unit as a reference. The free-state reference spectrum needs to be recalibrated each time the lens barrel specifications are changed or the PZT piezoelectric ceramic patch is replaced.

[0037] The specific process for extracting multi-order resonance features is as follows: Peak detection is performed on the real part curve of the electromechanical impedance spectrum Z(ω) and the conductivity spectrum. The peak values ​​in the conductivity spectrum correspond to the resonant frequencies of the structure. Peak detection uses the first derivative zero-crossing method: the difference between adjacent frequency points in the conductivity spectrum is calculated; when the difference changes from positive to negative, the corresponding frequency point is the resonant frequency. The first m resonant frequencies are then extracted. n = 1, 2, ..., m, where m is determined as follows: The number of peak values ​​with a signal-to-noise ratio greater than 20 dB in the conductivity spectrum within the swept frequency range is counted, and this number is used as the value of m. For example, when the swept frequency range is 7.5 kHz to 150 kHz and the clamping plate 6 is made of aluminum alloy, the first 6 to 8 resonant frequencies are extracted, i.e., m is taken as 6 to 8. The resonant frequency of each order is calculated relative to the corresponding order resonant frequency in the free-state reference spectrum. offset .

[0038] For the extraction of modal damping ratio, the half-power bandwidth method is used: for the nth resonance peak in the conductivity spectrum, find 0.707 times the peak value, which is the half-power point, and the two corresponding frequencies. and Modal damping ratio Calculate the modal damping ratio relative to the free-state reference value. Change .

[0039] For extracting the Q factor at the anti-resonance frequency, which corresponds to the valley point in the conductivity spectrum, the Q factor is defined as the ratio of the anti-resonance frequency to the corresponding half-power bandwidth. The Q factor is calculated relative to the free-state reference value. Change .

[0040] Resonance frequency offset This reflects changes in the equivalent stiffness of the contact interface. When the contact stiffness increases, the overall structural stiffness increases, and the resonant frequency shifts towards higher frequencies. Positive value; change in modal damping ratio The damping ratio increases when there is micro-slippage or loosening at the contact interface, reflecting changes in the energy dissipation characteristics of the contact interface. Positive value; Q-factor change at the anti-resonance frequency It reflects the dynamic response characteristics of the structure at the anti-resonance frequency and is sensitive to local defects at the contact interface.

[0041] Step S103: Based on the coupling relationship between the one-dimensional piezoelectric constitutive equation and the structural admittance, establish an electromechanical coupling analysis model including the boundary conditions of the contact interface between the clamping plate 6 and the mirror tube wall, using the offset in step S102. and change As the observation data for the inverse problem, the equivalent contact stiffness between the clamping plate 6 and the mirror tube wall is solved using the inverse eigenvalue inversion algorithm. and equivalent contact damping Equivalent contact stiffness and equivalent contact damping Combined with the characteristic parameters of each order in step S102, they form the mechanical fingerprint vector of the contact interface. Output to step S200.

[0042] The method for establishing the electromechanical coupling analysis model is as follows: Based on the one-dimensional piezoelectric constitutive equation, the theoretical expression for the electromechanical impedance of the PZT piezoelectric ceramic patch is: ,in The free capacitance of the PZT piezoelectric ceramic patch. The electromechanical coupling coefficient is... The mechanical impedance of the PZT piezoelectric ceramic patch. The structural mechanical impedance sensed by the PZT piezoelectric ceramic patch. Depending on the geometry, material properties, and boundary conditions of the contact interface between the clamping plate 6 and the lens barrel wall, the contact interface is equivalent to a spring element with a stiffness of... and damping elements, with a damping coefficient of If the parallel lumped parameter model is used, then the impedance of the contact interface is... ,Will Substitute structural mechanical resistance In the expression, make Become about Parameterized functions.

[0043] The specific process of the inverse eigenvalue inversion algorithm is as follows: Define the objective function. ,in and These represent the theoretical resonant frequency shift and the change in modal damping ratio calculated based on the parametric electromechanical coupling analysis model, respectively. and These are the measured values ​​extracted in step S102. and The weighting coefficients for frequency offset and damping ratio change are determined as follows: , ,in and These represent the standard deviations of the frequency offset and the change in damping ratio over 10 repeated measurements. These standard deviations were obtained by repeating steps S101 and S102 10 times under the condition that the lens barrel clamping state remained unchanged. The Levenberg-Marquardt optimization algorithm was used to solve the objective function. Minimum value, initial value of iteration Obtained through a rough estimate using Hertzian contact theory in mechanics of materials. Set as The convergence criterion for iteration is that the relative change of the objective function in two consecutive iterations is less than 1 / 3. Or the number of iterations reaches 1000.

[0044] The equivalent contact stiffness is obtained by solving. and equivalent contact damping Then, it is combined with the feature parameters of each order extracted in step S102 to form the contact interface mechanical fingerprint vector. Output to step S200.

[0045] Step S200: Multiphysics coupling parameter identification and strain / temperature decoupling: Mechanical strain information of the inner support plate 10 is acquired through a fiber Bragg grating strain sensor array, while eliminating temperature interference caused by frictional heat generation during processing. The strain field of the inner support plate 10 is inverted into the normal pressure distribution of the contact surface, and compared with the mechanical fingerprint vector of the contact interface output in step S100. Fusion to construct multi-physics contact state feature vectors This includes steps S201, S202, and S203.

[0046] Step S201: The embedded real-time signal processing and control unit synchronously receives the reflected wavelength signals of each grating output by the fiber Bragg grating strain sensor array, and performs strain component analysis at the same measurement point based on the dual-wavelength sensing matrix method. With temperature component Linear decoupling was performed to obtain the pure mechanical micro-strain distribution field of each inner support plate 10 along the generatrix direction. and temperature gradient distribution .

[0047] The reflected wavelength of a fiber Bragg grating With strain and temperature changes The relationship between them is ,in The change in the reflected wavelength, The center wavelength of the grating For the effective elastic coefficient, This is the coefficient of thermal expansion of the optical fiber. is the thermo-optic coefficient of the optical fiber. For two gratings with different center wavelengths at the same measurement point, the center wavelengths are respectively... and The wavelength changes are respectively and Establish the sensing matrix equation:

[0048]

[0049] in, and These are the effective photoelastic coefficients of the two gratings at their respective center wavelengths. and These are the thermo-optical coefficients of the two gratings at their respective center wavelengths. Since the center wavelengths of the two gratings are different, their... and The values ​​differ, making the aforementioned 2×2 coefficient matrix a non-singular matrix. This can be solved simultaneously using matrix inversion. and .

[0050] The calibration method for each coefficient in the sensing matrix is ​​as follows: The inner support plate 10 embedded with the fiber Bragg grating is placed in a constant temperature and constant strain calibration device. Five different temperature points are set under zero strain conditions, and the wavelength changes of the two gratings at each temperature point are recorded. The results are then obtained through linear regression fitting.

[0051] and Then, under isothermal conditions, five different known strain values ​​were applied, and the wavelength changes of the two gratings at each strain value were recorded. The results were then obtained through linear regression. and During the calibration process, each operating condition is measured three times and the average value is taken to reduce random measurement errors.

[0052] The embedded real-time signal processing and control unit receives the reflected wavelength signals of each grating from the fiber optic demodulator in real time, and performs the aforementioned inversion operation of the sensing matrix on the data at each sampling time to obtain the strain components at each measuring point. and temperature components ,in Let represent the arc length coordinate of the i-th measuring point along the generatrix of the inner support plate 10. The strain components of all measuring points are combined to form a purely mechanical micro-strain distribution field. The temperature components of all measuring points are combined to form a temperature gradient distribution. .

[0053] Step S202: Using the purely mechanical micro-strain distribution field from step S201 Using the bending beam elasticity theory as input, an integral equation relating the arc deformation of the inner support plate 10 to the normal pressure at the contact surface is established. The integral equation is then stably solved using the Tikhonov regularization method, and the distribution of normal pressure at the contact surface between the inner support plate 10 and the inner wall of the mirror tube is obtained by inversion. .

[0054] The inner support plate 10 has an arc-shaped structure, which exerts normal pressure on the inner wall of the lens barrel. Under the action of [something], bending deformation occurs, and the strain on the outer surface of the inner support plate 10 [is related to this]. With normal pressure The following integral relationship equation applies between them: ;in Let Green's function be denoted by in arc-length coordinates. When a unit normal force is applied at a point, the arc length coordinate is... The strain response generated at the point. Green's function. The construction method is as follows: A finite element model of a bending beam is established for the inner support plate 10 using the finite element method. A unit normal force is applied sequentially at each discrete node, and the values ​​are recorded. The strain response at the location can be used to obtain the discretized Green's function matrix. The material parameters of the inner support plate 10 in the finite element model, such as the elastic modulus and Poisson's ratio, are determined according to the inspection report provided by the material supplier. The geometric parameters, such as the radius of curvature, cross-sectional width and thickness, are determined according to the design drawings.

[0055] Discretizing the integral relation equations yields a system of linear equations. ,in Let G be the column vector of the purely mechanical micro-strain distribution field output in step S201, G be the discretized Green's function matrix, and p be the column vector of the normal pressure distribution to be solved. Since the Green's function matrix G is usually ill-conditioned and has a large condition number, directly solving this system of linear equations will lead to severe oscillations and instability of the solution. Therefore, the Tikhonov regularization method is used for stable solution, i.e., solving the minimization problem:

[0056] ,in For regularization parameters, Denotes the Euclidean norm and the regularization parameter. Determined using the L-curve method: Plot the residual norm in a double logarithmic coordinate system. with solution norm The relationship curve is shaped like the letter L. The value corresponding to the corner of the L-shaped curve is taken. The value is used as the optimal regularization parameter. For example, when the arc surface of the inner support plate 10 is discretized into 20 nodes, the condition number of the Green's function matrix G is approximately 10. 4 At that time, the optimal regularization parameter determined by the L-curve method Approximately 10 -3 Up to 10 -2 Order of magnitude. The solution yields the normal pressure distribution. , The physical meaning is that the inner support plate 10 and the inner wall of the lens tube are in arc length coordinates. The normal pressure value at the contact surface, in Pascals (Pa).

[0057] Step S203: Concatenate and splice the contact interface mechanical fingerprint vector output in step S103 with the values ​​of the contact surface normal pressure distribution at each discrete measurement point in step S202 to construct a multi-physics contact state feature vector. Simultaneously, the temperature gradient distribution in step S201 will be... The thermally induced additional strain compensation was calculated by converting the coefficient of linear expansion. Output to step S300.

[0058] Will and Cascaded splicing is performed to construct multi-physics contact state feature vectors. For example, when m=6 and n=6, The dimensions are 2+6+6+6+6=26.

[0059] Thermally induced additional strain compensation The calculation method is as follows:

[0060] ,in Let be the coefficient of linear expansion of the lens barrel wall material. When the lens barrel material is the engineering plastic polycarbonate... The coefficient of linear expansion is determined based on the material certificate of the lens barrel material.

[0061] Step S300: Cross-domain heterogeneous information fusion state estimation: The multi-physics contact state feature vector output from step S200, the joint information in the frequency and spatial domains, and the time-domain radial runout signal acquired by the differential eddy current micro-displacement sensor are fused together to perform cross-domain heterogeneous information fusion. The comprehensive state vector of the clamping system is then output using an improved capacitive Kalman filter algorithm. This provides accurate and real-time state estimates for subsequent control decisions, including steps S301, S302, and S303.

[0062] Step S301: The embedded real-time signal processing and control unit acquires the timing signal of the radial runout of the outer wall of the lens barrel output by the differential eddy current micro-displacement sensor in real time at a preset sampling frequency. And subtract the thermally induced additional strain compensation amount in step S203. By mapping the thermally induced displacement component to the radial direction using the thin-shell geometry, the net dynamic radial runout signal after eliminating thermal effects is obtained. .

[0063] The differential eddy current micro-displacement sensor outputs the radial displacement signal of the outer wall of the lens barrel at two symmetrical measurement points in real time at a sampling frequency of not less than 5 kHz. and The differential processing procedure is as follows: radial runout signal Differential operation eliminates the common-mode displacement component caused by the vibration of the placement stage 2 or the deformation of the base of the processing stage 1, and retains only the radial deformation component of the mirror barrel itself.

[0064] The calculation method for the thermally induced displacement component is as follows: based on the geometric relationship of the thin shell, the temperature change of the lens tube wall... Radial expansion resulting Where R is the average radius of the lens tube. The temperature gradient distribution output in step S201 The average value across all measuring points will from Subtracting from the middle yields the net dynamic radial runout signal. For example, when the average radius of the lens barrel R = 20 mm and the material is aluminum alloy, At Celsius, Micrometer.

[0065] Step S302: Establish a dynamic model of the thin-walled shell of the mirror tube based on lumped parameters as the state transition equation. Inject the equivalent contact stiffness and equivalent contact damping in step S103 as time-varying system parameters into the stiffness matrix and damping matrix of the state transition equation. Take the normal pressure distribution of the contact surface in step S202 as the external excitation input of the system.

[0066] The microscope tube is equivalent to a mass in the radial direction. Stiffness and damping A single-degree-of-freedom vibration system, equivalent mass It is calculated using the actual mass of the lens barrel and the modal effective mass coefficient. For example, = Lens barrel mass × First-order modal effective mass coefficient, which is obtained through finite element modal analysis, with a typical value of 0.3 to 0.5. Equivalent stiffness Due to the structural rigidity of the lens barrel itself Equivalent stiffness of clamping contact interface Series composition:

[0067] ,in The equivalent damping was obtained through finite element static analysis. Damping by the structure of the lens barrel itself Equivalent damping of contact interface Parallel configuration: ,in The results were determined by the free decay method.

[0068] The state transition equation, expressed in state-space form, is as follows:

[0069] Where the state vector , Let be the radial deformation at time k. The radial deformation rate at time k; system matrix It is a 2×2 matrix whose elements contain , and ,reflect and The influence of time-varying parameters on the dynamic characteristics of the system; the input matrix B is a 2×1 matrix; external excitation. The contact surface normal pressure distribution output from step S202 The equivalent radial resultant force obtained after spatial integration constitutes the force. Let be the process noise vector, which has a mean of zero and a covariance matrix of... Gaussian distribution, This was determined through a systematic identification experiment.

[0070] Step S303: Using the net dynamic radial runout signal from step S301 As an observation, the multiphysics contact state feature vector in step S203 The mechanical parameters in the system are used as auxiliary observation constraints. An improved capacitive Kalman filter algorithm is employed. Through capacitive point transformation, the state transition equation in step S302 is nonlinearly propagated and multi-source information is fused to output the comprehensive state vector of the clamping system. Proceed to step S400.

[0071] The specific implementation process of the capacitive Kalman filter algorithm is as follows: Initialization phase: Set the initial state estimate Assuming the mirror tube is initially undeformed, the initial error covariance matrix is... Multiply the 2×2 identity matrix by the initial uncertainty coefficient , The noise level is determined by measuring a differential eddy current micro-displacement sensor, as exemplified. =1 micrometer.

[0072] Prediction Phase: At each sampling time k, 2n1 volume points are generated, where n1 is the dimension of the state vector. Here, n1=2, so the number of volume points is 4. The volume points are decomposed into P(k-1) by Cholesky and distributed in the state space according to the volume rule. Each volume point is substituted into the state transition equation of step S302 for propagation to obtain the propagated volume points. The prior state estimate is calculated by weighting the propagated volume points. and prior error covariance matrix .

[0073] Update phase: The observation equation is ,in The net dynamic radial runout signal output in step S301 H=[1,0] is the observation matrix, v(k) is the observation noise, which follows a normal distribution with zero mean and variance. Gaussian distribution, The variance of the differential eddy current micro-displacement sensor is obtained by acquiring the output signal without a mirror tube and calculating it. The Kalman gain matrix K(k) is calculated, and the prior estimate and observations are fused using K(k) to obtain the posterior state estimate. and posterior error covariance matrix .

[0074] The auxiliary observation constraint is introduced by: using the multi-physics contact state feature vector Equivalent contact stiffness in As in the state transition equation The online update value will be equivalent to the contact damping. As The online update value updates the system matrix during the prediction phase of each filtering cycle. At the same time, Normal pressure distribution at the contact surface Integral form is used as an external incentive. The online update value.

[0075] Output clamping system integrated state vector ,in The circumferential deformation distribution of the mirror barrel after fusion estimation is obtained by circumferential interpolation of the radial deformation output of the filter combined with the spatial positions of two differential eddy current micro-displacement sensors. Deformation rate distribution; To estimate the distribution of normal pressure at the contact surface, take the output of step S202. The filtered value; and This is the latest online identification value output in step S103.

[0076] Step S400: Reduced-order digital twin-driven feedforward and feedback composite control: based on the comprehensive state vector output from step S300. The model uses an intrinsic orthogonal decomposition reduced-order digital twin model to predict the overall deformation trend of the lens barrel, and combines feedforward compensation and sliding mode feedback control to generate the optimal correction torque command. The output of the drive device 4 is adjusted in real time so that the clamping force and the support force dynamically tend to the optimal balance throughout the processing, including steps S401, S402 and S403.

[0077] Step S401: The embedded real-time signal processing and control unit pre-stores a reduced-order model based on intrinsic orthogonal decomposition. The offline construction process of the reduced-order model is as follows: In the offline stage, finite element simulation software is used to perform parametric simulation of the deformation field of the lens barrel under different combinations of clamping force and internal support force. The working condition design method for parametric simulation is as follows: the clamping force... Within its scope of work Take evenly inside This value represents the internal support force. Within its scope of work Take evenly inside Each value forms A combination of operating conditions. For example, when... The operating range is 50N to 200N. When =10, The values ​​are 50N, 66.7N, 83.3N, 100N, 116.7N, 133.3N, 150N, 166.7N, 183.3N, and 200N; when The operating range is 30N to 150N. There are a total of 80 combined working conditions. Finite element static analysis is performed on each combined working condition, and the radial displacement of all nodes on the outer surface of the mirror tube is extracted as a deformation field snapshot. Each snapshot is an M-dimensional column vector, where M is the total number of nodes. The 80 snapshots are arranged in columns to form an M-row, 80-column deformation field snapshot matrix S.

[0078] Perform eigenorthogonal decomposition on the snapshot matrix S: calculate the autocorrelation matrix of S. Eigenvalue decomposition of C ,in It is a diagonal eigenvalue matrix. Let be the eigenvector matrix. Calculate the eigenorthogonal decomposition basis vector matrix. ,in for The first r columns, for The truncation order r of the first r × r submatrix is ​​determined by calculating the ratio of the sum of the first r eigenvalues ​​to the sum of all eigenvalues. The value of r corresponding to the first time this ratio exceeds 99.9% is the truncation order. For example, for a typical VR lens barrel structure, when the number of snapshots is 80, the truncation order r is usually between 3 and 5, meaning that the first 3 to 5 basis vectors can capture more than 99.9% of the deformation field energy.

[0079] During online execution, the comprehensive state vector output in step S303 is... In As a boundary condition for real-time loading, and As the contact interface parameter, the generalized coordinate vector q(t) is solved in the reduced-order subspace. The governing equations in the reduced-order subspace are:

[0080] ,in These are the mass, damping, and stiffness matrices after order reduction, each with dimension r×r. For the reason The resulting external force vector. The solution speed of this r-dimensional equation system is much faster than that of the original M-dimensional finite element equations.

[0081] After obtaining q(t), through Reconstruct the full-field deformation prediction distribution of the lens barrel, and output the results from step S303. and The residual at the corresponding measuring point of the differential eddy current micro-displacement sensor The projection coefficients of the reduced-order basis matrix are updated online using recursive least squares. The forgetting factor for recursive least squares is set to 0.98, which was determined through cross-validation experiments on 10 different lens barrel sizes. The forgetting factor value that minimizes the root mean square value of the prediction residual is selected by iterating through the range of 0.95 to 0.99 with a step size of 0.01. The corrected full-field deformation prediction is then output. and deformation trend gradient field .

[0082] Step S402: Based on the corrected full-field deformation prediction from step S401 and deformation trend gradient field As the input to the feedforward channel, the theoretical clamping force increment required to bring the deformation field to zero is calculated through the inverse mapping of the reduced-order model. ;Based on the circumferential deformation distribution of the lens barrel in step S303 The deviation from the target zero-deformation reference is used as the input to the feedback channel. A sliding mode control strategy based on the exponential decay approach law is adopted. The sliding surface is constructed using the deformation deviation and its rate of change, and the increment of the feedback correction force is calculated. .

[0083] The specific calculation process of the feedforward channel is as follows: The inverse mapping of the reduced-order model refers to setting the target deformation to zero in the governing equations of the reduced-order subspace, that is... According to the current deformation state and deformation trend The reverse calculation ensures that the deformation is within the preset feedforward time domain. The increase in external force required for the internal force to approach zero The specific calculation method is as follows:

[0084] This means the increment of external force required to offset the elastic restoring force and damping force corresponding to the current deformation state. (Feedforward time domain) It is set to 3 times the electromechanical time constant of drive device 4 to ensure that feedforward compensation can take effect within the drive system response time.

[0085] The specific calculation process for the feedback channel is as follows: Define deformation deviation ,in The circumferential deformation distribution of the lens barrel output in step S303 is the average value at the corresponding angle positions of the clamping plate 6 and the inner support plate 10 at the key angle positions. Define the sliding surface as the reference for zero deformation. ,in The sliding surface coefficient, The value is determined by the system closed-loop bandwidth requirements. , For the desired closed-loop control bandwidth, an example is... ,but Using the exponential decay approach law ,in To approximate the velocity parameters, The exponential decay coefficient is... For symbolic functions, The value of must be greater than the upper bound of the system uncertainty, which is obtained by measuring the maximum value of the system uncertainty under 10 different processing conditions and multiplying it by 1.2 times the safety margin. The value of is determined through closed-loop pole placement to ensure that the convergence speed of the sliding surface meets the dynamic response requirements of the machining process. The increment of the feedback correction force is solved according to the reaching law. .

[0086] Step S403: The steps in step S402... and The superposition of the clamping mechanism transmission ratio between the clamping plate 6 and the inner support plate 10 is converted into an optimal corrected torque command. The optimal corrected torque command will be used. The output is sent to the drive device 4 to achieve real-time fine-tuning of the rotational speed of the double-segment threaded shaft 5, so that the clamping force of the clamping plate 6 and the supporting force of the inner support plate 10 dynamically tend to the optimal balance throughout the machining process, while simultaneously sending the optimal correction torque command. The time series data is recorded and output to step S500. Total correction force increment. The transmission ratio of the clamping mechanism Defined as the ratio between the torque of the double-segment threaded shaft 5 and the clamping force output by the clamping plate 6. The following calculations were performed using the thread lead of the double-segment threaded shaft 5, the thread friction coefficient, and the sliding friction coefficient between the clamping plate 6 and the square through groove 3: ,in For thread lead, The coefficient of friction of the thread. The thread pitch diameter. Optimal corrected torque command. . The signal is converted into an analog voltage signal by a digital-to-analog converter module and transmitted to the driver of drive device 4 via an RS485 bus. The driver then... Adjusting the motor output torque causes a slight change in the rotational speed of the double-segment threaded shaft 5. This, in turn, drives a corresponding change in the displacement of the clamping plate 6 and the inner support plate 10 through threaded transmission, ultimately achieving dynamic adjustment of the clamping force and the supporting force.

[0087] Step S500: Bayesian self-tuning optimization of control parameters based on a probabilistic surrogate model: Using the control effect data output in step S400, a probabilistic surrogate model between control parameters and processing quality is established through sparse Gaussian process regression. Based on the Bayesian optimization framework, the optimal control parameters are automatically searched, enabling the present application to automatically adapt to and converge to the optimal clamping control strategy when processing different batches and different wall thicknesses of lens barrels. This includes steps S501, S502, and S503.

[0088] Step S501: Use the optimal corrected torque command from step S403 The time series and the integrated state vector in step S303 The historical sequence data within the sliding time window and the projection coefficient update residuals generated during the online correction of the reduced-order model in step S401 are used as the feature input set to define the comprehensive quality evaluation index Q1 of the processing process. The feature input set contains three types of data: the first type is... The first category is the statistical characteristics of the time series within the most recent processing cycle, including five scalars: mean, standard deviation, maximum, minimum, and root mean square value; the second category is the comprehensive state vector. Within the sliding time window, with the window length set to 1 / 10 of the processing cycle, statistical characteristics are analyzed. The mean and standard deviation of each component are calculated separately; the third category is the root mean square value of the projection coefficient update residual generated during the recursive least squares update process in step S401, which reflects the degree of matching between the reduced-order model and the actual system.

[0089] Comprehensive evaluation index of processing quality ,in Based on The calculated standard deviation of the deformation distribution reflects the degree of non-uniformity of the deformation; The maximum absolute value of the corrected full-field deformation prediction output in step S401 reflects the maximum degree of deformation. The output of step S403 The variance of the time series within a processing cycle reflects the degree of fluctuation in the control torque. The weighting coefficients α, β, and γ are determined as follows: Using the analytic hierarchy process (AHP), with processing accuracy requirements, structural safety, and control stability as evaluation dimensions, three engineering technicians with experience in precision machining are invited to conduct pairwise comparisons and scores on the relative importance of the three dimensions. A judgment matrix is ​​constructed, and a weight vector is calculated. The average of the weight vectors of the three experts is taken as the final values ​​of α, β, and γ. For example, when processing accuracy is the highest priority factor, α = 0.5, β = 0.35, and γ = 0.15. The smaller the Q1 value, the better the processing quality.

[0090] Step S502: Using the sparse Gaussian process regression algorithm, with the control parameter space formed by the sliding mode control reaching law parameters and feedforward gain coefficients in step S402 as input and the comprehensive quality evaluation index Q1 of the processing process in step S501 as output, a probabilistic proxy model between the control parameter space and the processing quality is established to obtain the posterior mean prediction and uncertainty estimate of the comprehensive quality evaluation index Q1 with respect to each control parameter.

[0091] The control parameter space consists of the four adjustable parameters in step S402: sliding surface coefficient Approaching velocity parameters Exponential decay coefficient and feedforward gain coefficient , The feedforward force increment in step S402 The scaling factor is initially set to 1.0, and the range of values ​​for each parameter is determined through system stability analysis. The range is [100, 1000]. The range is [0.1, 10]. The range is [1, 100]. The range is [0.5, 2], and the input dimension of the control parameter space is 4-dimensional.

[0092] The network structure and training method for sparse Gaussian process regression are as follows: Sparse Gaussian process regression is an approximation of standard Gaussian process regression, using a set of data of size... To reduce computational complexity, we can use the guiding points. The value of is determined by balancing prediction accuracy and computation speed, for example... Set it to 10 times the dimension of the control parameter space, that is =40, the initial position of the induction point was selected from the existing training data using the K-means clustering algorithm. One cluster center was obtained.

[0093] The covariance function is determined by automatically correlated quadratic exponent kernel function:

[0094] ,in This is the signal variance hyperparameter. The reason for choosing the automatically correlated quadratic exponential kernel function as the feature length scale hyperparameter corresponding to the d-th control parameter dimension is that the four control parameters may have different degrees of influence on the processing quality Q1, and the automatically correlated quadratic exponential kernel function assigns an independent feature length scale to each dimension. It can automatically identify the importance of each parameter. A smaller value indicates that Q1 is sensitive to changes in the d-th parameter. A larger value indicates that Q1 is not sensitive to this parameter.

[0095] The hyperparameter training method is as follows: Optimize the marginal likelihood function by maximizing the marginal likelihood function and using logarithmic evidence.

[0096]

[0097] ,in Given the existing Q1 observation vector, X is the corresponding control parameter matrix. For the kernel matrix, To observe the standard deviation of noise, Let I be the set of hyperparameters, n be the identity matrix, and n be the number of observations. The LBFGSB optimization algorithm is used to maximize the marginal likelihood function. The upper limit for the number of iterations is set to 100, and the convergence criterion is that the gradient norm is less than 10. -5 .

[0098] The initial method for obtaining training data is as follows: before the start of the first processing batch, Latin hypercube sampling is used to generate data in the control parameter space. One initial sampling point, Set it to 5 times the dimension of the control parameter space, that is =20. For each sampling point, set the corresponding control parameters into the controller in step S402, execute a complete processing cycle, and record the Q1 value of that processing cycle. Use 20 sets of control parameters, Q1 values, and data pairs as the initial training set to train a sparse Gaussian process regression model.

[0099] Step S503: Based on the posterior distribution of the probabilistic surrogate model in step S502, the expected improvement acquisition function is used to search for the candidate parameter combination that maximizes the expected improvement in the control parameter space after each processing cycle. The optimal candidate parameters obtained by the search are fed back to the sliding mode control strategy and feedforward gain in step S402 to realize the cross-batch Bayesian self-tuning optimization closed loop.

[0100] The definition of the desired improvement in the acquisition function is: ,in This is the smallest Q1 value that has been observed so far. To predict the Q1 value at the control parameter x. Based on the posterior distribution of Gaussian process regression. It can be parsed and calculated as follows:

[0101]

[0102] ,in and Let x be the posterior mean and posterior standard deviation of the Gaussian process regression at x, respectively. The cumulative distribution function of the standard normal distribution. It is the probability density function of the standard normal distribution.

[0103] The desired maximization search process for improving the acquisition function is as follows: 1000 candidate points are generated in the control parameter space using Latin hypercube sampling. The EI value of each candidate point is calculated. The top 10 candidate points with the largest EI values ​​are selected as the initial points for local optimization. The LBFGSB algorithm is used for local optimization at each initial point. The solution with the largest EI value among the 10 local optima is selected as the recommended control parameter combination for the next processing cycle. .

[0104] Will In and Update the sliding mode controller and feedforward controller in step S402 respectively. Use the updated parameters to execute control in the next machining cycle. Calculate the new Q1 value after the machining cycle ends. Add it to the training set, retrain the sparse Gaussian process regression model, update the hyperparameters, and enter the next round of optimization loop.

[0105] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to specific implementations. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A high-precision VR lens barrel processing device, comprising a processing table (1); a placement table (2) is fixedly connected to the top of the outer wall of the processing table (1); a square through groove (3) is opened on the top of the outer wall of the placement table (2); a driving device (4) is fixedly connected to the top of the outer wall of the processing table (1); a double-segment threaded shaft (5) is provided at the output end of the driving device (4), and one end of the outer wall of the double-segment threaded shaft (5) extends into the square through groove (3); a pair of clamping plates (6) are threadedly connected to the outer wall of the double-segment threaded shaft (5), and the outer walls of the pair of clamping plates (6) are slidably connected in the square through groove (3); a pair of arc-shaped locking blocks (7) are fixedly connected to the top of the outer wall of the placement table (2); characterized in that, The top of the outer wall of the placement platform (2) is rotatably connected to a double-segment threaded rod (9) via a connecting block (8); the outer wall of the double-segment threaded rod (9) is threadedly connected to a pair of inner support plates (10), and the outer walls of the pair of inner support plates (10) are in contact with one side of the outer wall of a pair of arc-shaped locking blocks (7); the outer walls of the double-segment threaded rod (9) and the double-segment threaded shaft (5) are both fixedly connected to helical gears (11), and the pair of helical gears (11) are meshed.

2. The high-precision VR lens barrel processing device according to claim 1, characterized in that, The connecting block (8) is inverted U-shaped; the connecting block (8) cooperates with a pair of arc-shaped locking blocks (7) and a square through slot (3) to form a shielding space, and a pair of helical gears (11) are located in the shielding space; the top of the outer wall of the processing table (1) is rotatably connected to a reciprocating rod (13) through a connecting plate (12); a reciprocating plate (14) is provided on the outer wall of the reciprocating rod (13); a rotating rod (15) is provided on the top of the outer wall of the connecting plate (12), and the top of the outer wall of the rotating rod (15) is... The end of the reciprocating plate (14) is connected to the reciprocating rod (5); a first bevel gear (16) is fixed to the outer wall of the double-segment threaded shaft (5); a second bevel gear (17) is fixed to the bottom of the outer wall of the reciprocating rod (13), and the first bevel gear (16) and the second bevel gear (17) mesh; a circular plate (18) is provided on the inner wall of the reciprocating plate (14); a set of cleaning rods (19) is provided on the bottom of the outer wall of the circular plate (18); a set of cleaning brushes (20) is provided on the outer wall of the cleaning rods (19).

3. The high-precision VR lens barrel processing device according to claim 2, characterized in that, The outer wall of the circular plate (18) is rotatably connected to the inner wall of the reciprocating plate (14); the outer walls of the rotating rod (15) and the reciprocating rod (13) are both fixedly connected to a third gear (21), and a pair of third gears (21) mesh; the top of the outer wall of the circular plate (18) is fixedly connected to a fourth sprocket (22); the outer wall of the reciprocating rod (13) is provided with a sliding groove (23); the top of the outer wall of the reciprocating plate (14) is rotatably connected to a fifth sprocket (24), and the inner wall of the fifth sprocket (24) is slidably connected to the inner wall of the sliding groove (23); the fourth sprocket (22) and the fifth sprocket (24) are connected by a chain (25); the bottom of the outer wall of the rotating rod (15) is rotatably connected to the top of the outer wall of the connecting plate (12).

4. The high-precision VR lens barrel processing device according to claim 3, characterized in that, The bottom of the outer wall of the reciprocating plate (14) is fixedly connected to an annular rack (26); the top of the outer wall of a set of cleaning rods (19) is rotatably connected to the bottom of the outer wall of the circular plate (18); the outer wall of a set of cleaning rods (19) is fixedly connected to a sixth gear (27); the sixth gear (27) meshes with the annular rack (26).

5. The high-precision VR lens barrel processing device according to claim 4, characterized in that, PZT piezoelectric ceramic patches are symmetrically pasted on the inner contact surfaces of the pair of clamping plates (6); the PZT piezoelectric ceramic patches are used as both a high-frequency sweep excitation source and a mechanical impedance response sensor through an impedance analyzer circuit; fiber Bragg grating strain sensor arrays are embedded in the arc surfaces of the pair of inner support plates (10) along the generatrix direction; the fiber Bragg grating strain sensor array contains at least two gratings with different center wavelengths to achieve synchronous decoupled measurement of strain and temperature; a pair of differential eddy current micro-displacement sensors are symmetrically installed on the top of the outer wall of the placement stage (2); the probes of the pair of differential eddy current micro-displacement sensors point to the outer wall of the lens barrel and are symmetrically arranged about the central axis of the lens barrel; an embedded real-time signal processing and control unit is fixedly connected to the top of the outer wall of the processing stage (1); the embedded real-time signal processing and control unit is electrically connected to the driving device (4), the PZT piezoelectric ceramic patches, the fiber Bragg grating strain sensor array, and the differential eddy current micro-displacement sensors respectively; The embedded real-time signal processing and control unit performs an adaptive clamping force intelligent control process based on multi-physics field coupling sensing, the control process including: Step S100: Mechanical impedance spectroscopy excitation acquisition and contact interface mechanical fingerprint extraction; Step S200: Identification of multiphysics coupling parameters and decoupling of strain and temperature; Step S300: Cross-domain heterogeneous information fusion state estimation; Step S400: Reduced-order digital twin driven feedforward and feedback composite control; Step S500: Bayesian self-tuning optimization of control parameters based on the probabilistic surrogate model.

6. The high-precision VR lens barrel processing device according to claim 5, characterized in that, The mechanical electrical impedance spectroscopy excitation acquisition and contact interface mechanical fingerprint extraction described in step S100 specifically include: S101: The embedded real-time signal processing and control unit drives each of the PZT piezoelectric ceramic patches to apply a linear frequency-modulated sweep voltage signal within a preset frequency range, synchronously collects the voltage and current response signals of each of the PZT piezoelectric ceramic patches, and calculates the mechanical impedance spectrum Z(ω) of each patch based on Fourier transform. S102: Perform multi-order resonance feature extraction on the electromechanical impedance spectrum Z(ω), extracting the offset of the first m order resonance frequencies relative to the free-state reference spectrum. The change in damping ratio corresponding to each modal order and the change in Q factor at the anti-resonance frequency ; S103: Based on the coupling relationship between the one-dimensional piezoelectric constitutive equation and the structural admittance, an electromechanical coupling analysis model is established, including the boundary conditions of the contact interface between the clamping plate (6) and the mirror tube wall, using the offset mentioned in step S102. and change Using the observation data of the inverse problem, the equivalent contact stiffness between the clamping plate (6) and the mirror tube wall is solved by the inverse eigenvalue inversion algorithm. and equivalent contact damping The equivalent contact stiffness, equivalent contact damping, and the characteristic parameters of each order described in step S102 are combined to form a mechanical fingerprint vector of the contact interface. Output to step S200.

7. The high-precision VR lens barrel processing device according to claim 6, characterized in that, The multiphysics coupling parameter identification and strain / temperature decoupling described in step S200 specifically include: S201: The embedded real-time signal processing and control unit synchronously receives the reflected wavelength signals of each grating output by the fiber Bragg grating strain sensor array, and performs strain component analysis at the same measurement point based on the dual-wavelength sensing matrix method. With temperature component Linear decoupling and separation are performed to obtain the pure mechanical micro-strain distribution field of each inner support plate (10) along the generatrix direction. and temperature gradient distribution ; S202: Using the purely mechanical micro-strain distribution field described in step S201 As input, an integral equation relating the arc deformation of the inner support plate (10) to the normal pressure on the contact surface is established based on the elastic theory of bending beams. The integral equation is then solved stably using the Tikhonov regularization method, and the distribution of normal pressure on the contact surface between the inner support plate (10) and the inner wall of the mirror tube is obtained by inversion. ; S203: The contact interface mechanical fingerprint vector output in step S103 is... The normal pressure distribution on the contact surface described in step S202 The values ​​taken at each discrete measurement point are concatenated and spliced ​​to construct a multi-physics contact state feature vector. Simultaneously, the temperature gradient distribution described in step S201 will be... The thermally induced additional strain compensation was calculated by converting the coefficient of linear expansion. Output to step S300.

8. The high-precision VR lens barrel processing device according to claim 7, characterized in that, The cross-domain heterogeneous information fusion state estimation mentioned in step S300 specifically includes: S301: The embedded real-time signal processing and control unit acquires the radial runout timing signal of the outer wall of the lens barrel output by the differential eddy current micro-displacement sensor in real time at a preset sampling frequency. And subtract the thermally induced additional strain compensation amount mentioned in step S203. By mapping the thermally induced displacement component to the radial direction using the thin-shell geometry, the net dynamic radial runout signal after eliminating thermal effects is obtained. ; S302: Establish a dynamic model of the thin-walled shell of the mirror tube based on lumped parameters as the state transition equation, and inject the equivalent contact stiffness and equivalent contact damping described in step S103 as time-varying system parameters into the stiffness matrix and damping matrix of the state transition equation, and take the contact surface normal pressure distribution described in step S202 as the external excitation input of the system. S303: Using the net dynamic radial runout signal described in step S301 as the observation, and the mechanical parameters in the multi-physics contact state feature vector described in step S203 as auxiliary observation constraints, an improved capacitive Kalman filter algorithm is used to perform nonlinear propagation and multi-source information fusion on the state transition equation described in step S302 through capacitive point transformation, outputting the comprehensive state vector of the clamping system. To step S400, where The estimated circumferential deformation distribution of the lens barrel. For deformation rate distribution, For the estimated distribution of normal pressure at the contact surface, and The equivalent contact stiffness and damping are identified online.

9. A high-precision VR lens barrel processing device according to claim 8, characterized in that, The feedforward and feedback composite control of the reduced-order digital twin drive mentioned in step S400 specifically includes: S401: The embedded real-time signal processing and control unit pre-stores a reduced-order model based on intrinsic orthogonal decomposition. The reduced-order model generates a deformation field snapshot matrix by performing parameterized finite element simulation on the lens barrel under different clamping force and internal support force combinations in the offline stage, and extracts the first r-order dominant deformation mode basis vectors using intrinsic orthogonal decomposition. And construct the reduced-order basis matrix During online operation, the integrated state vector described in step S303 is used. In Using real-time loading boundary conditions and online-identified equivalent contact stiffness and damping as contact interface parameters, the generalized coordinate vector is solved in a reduced-order subspace. ,pass Reconstruct the full-field deformation prediction distribution of the lens barrel, and combine the fused and estimated circumferential deformation distribution of the lens barrel described in step S303 with... The residuals at the corresponding measuring points are used to update the projection coefficients online using the recursive least squares method, outputting the corrected full-field deformation prediction. and deformation trend gradient field ; S402: Using the corrected full-field deformation prediction and deformation trend gradient field described in step S401 as the input to the feedforward channel, the theoretical clamping force increment required to bring the deformation field to zero is calculated through the inverse mapping of the reduced-order model. Using the deviation between the circumferential deformation distribution of the lens barrel described in step S303 and the target zero-deformation reference as the input of the feedback channel, a sliding mode control strategy based on the exponential decay approach law is adopted. The sliding mode surface is constructed with the deformation deviation and its rate of change, and the increment of the feedback correction force is calculated. ; S403: The theoretical clamping force increment described in step S402 is superimposed with the feedback correction force increment, and converted into an optimal correction torque command through the transmission ratio of the clamping mechanism between the clamping plate (6) and the inner support plate (10). The optimal correction torque command is output to the drive device (4) to achieve real-time fine adjustment of the rotation speed of the double-segment threaded shaft (5), so that the clamping force of the clamping plate (6) and the supporting force of the inner support plate (10) dynamically tend to the optimal balance during the entire processing process. At the same time, the time sequence of the optimal correction torque command is recorded and output to step S500.

10. A high-precision VR lens barrel processing device according to claim 9, characterized in that, The Bayesian self-tuning optimization of control parameters based on the probabilistic surrogate model described in step S500 specifically includes: S501: Using the time series of the optimal corrected torque command described in step S403, the historical sequence data of the comprehensive state vector within the sliding time window described in step S303, and the projection coefficient update residual generated during the online correction of the reduced-order model described in step S401 as the feature input set, a comprehensive quality evaluation index for the processing process is defined. ,in The standard deviation of the deformation distribution is calculated based on the circumferential deformation distribution of the lens barrel after fusion estimation. The maximum absolute value of the corrected full-field deformation prediction described in step S401. As described in step S403 Time series variance Preset weighting coefficients; S502: Using the sparse Gaussian process regression algorithm, with the control parameter space formed by the sliding mode control reaching law parameters and feedforward gain coefficients described in step S402 as input and the comprehensive quality evaluation index Q1 of the processing process described in step S501 as output, a probabilistic proxy model between the control parameter space and the processing quality is established to obtain the posterior mean prediction and uncertainty estimate of the comprehensive quality evaluation index Q1 with respect to each control parameter. S503: Based on the posterior distribution of the probabilistic surrogate model described in step S502, the expected improvement acquisition function is used to search for the candidate parameter combination that maximizes the expected improvement in the control parameter space after each processing cycle. The optimal candidate parameters obtained by the search are fed back to the sliding mode control strategy and feedforward gain described in step S402 to realize cross-batch Bayesian self-tuning optimization closed loop, so as to automatically converge to the optimal clamping control strategy when processing different batches and different wall thickness specifications of lens barrels.

Images