Automatic leveling and grinding method

Abstract

The invention discloses an automatic leveling and grinding method which comprises the following steps: mounting a vacuum adsorption carrying table on a six-axis positioning table, sampling four corners of the carrying table for multiple times by using a laser range finder, and averaging to obtain a real-time height; the control unit calculates the height difference and then drives the piezoelectric actuator to stretch out and draw back, the carrying table is perpendicular to the axis of the grinding main shaft, and repeated measurement and compensation can be conducted according to needs. Before grinding, edge warping is compensated through edge height measurement and radial basis network prediction; in the grinding process, the workpiece edge height or the chamfering angle is monitored in real time, and the spindle falling distance is dynamically corrected. In the whole process, the carrier does not need to be disassembled or a platform face is milled, model changing is fast, manual dependence is low, the carrier is not damaged in the leveling process, and the device is suitable for precise edge grinding of high-end workpieces such as wafers and glass substrates.

Description

Technical Field

[0001] This invention relates to an automatic leveling and grinding method. Background Technology

[0002] With the increasing demands for processing precision in high-end manufacturing fields such as semiconductor wafers and flexible display panels, the flatness and perpendicularity of workpiece edge grinding have become critical factors affecting the geometric accuracy of the final product. Existing technologies typically employ vacuum adsorption platforms for workpiece positioning and adapt to multi-variety, small-batch production by changing different sized platforms. However, this approach suffers from the following key drawbacks:

[0003] Low changeover efficiency:

[0004] When changing product sizes, the entire specialized carrier platform must be disassembled and reinstalled. The platform surface is then repeatedly milled flat with a milling cutter to ensure its perpendicularity to the grinding spindle. This entire process takes several hours, with equipment downtime accounting for more than 15% of daily operating time, severely restricting production cycle time.

[0005] Poor accuracy and stability:

[0006] Frequent disassembly and assembly cause the assembly reference of the vehicle and equipment base to shift. In addition, manual height setting relies on experience and is prone to random errors. Multiple milling and leveling will also cause the platform thickness to decrease year by year, shortening its lifespan and further increasing the risk of processing errors.

[0007] High dependence on human intervention:

[0008] The milling and height setting process requires skilled operators to repeatedly measure and adjust using tools such as micrometers. This not only involves high labor intensity but also easily leads to over-milling or height setting deviations, resulting in scrapped workpieces.

[0009] To address these issues, the industry has proposed integrating a laser ranging unit into the grinding mechanism to quickly acquire the platform's attitude through four-point height measurement, and then using a piezoelectric six-axis positioning stage for real-time leveling. However, this approach remains an open-loop "single measurement - one compensation" model, failing to consider error sources such as spindle rotation resonance, servo clock beat frequency, piezoelectric hysteresis nonlinearity, edge warping, and low-frequency vibration. This results in residual errors at the micrometer level after leveling, failing to meet the mass production requirements of ±0.2μm chamfer width and ±0.05° angle at the wafer level. Therefore, there is an urgent need for an automated leveling and grinding method that integrates resonance suppression, hysteresis compensation, edge prediction, and vibration feedforward at the algorithm level to further shorten changeover time, improve accuracy and stability, and reduce manual reliance. Summary of the Invention

[0010] The present invention provides an automatic leveling and grinding method to solve the problems existing in the prior art.

[0011] The technical solutions adopted in this invention are as follows:

[0012] An automatic leveling and grinding method includes the following steps:

[0013] a) Mount the vacuum adsorption stage onto the six-axis positioning stage and keep it stationary;

[0014] b) The laser rangefinder fixed to the grinding mechanism is used to continuously sample each of the four corners of the vacuum adsorption stage multiple times, and the arithmetic mean is taken as the real-time height of the corner, corresponding to: LA1, LA2, LA3, LA4;

[0015] c) The control unit calculates the minimum value Z among LA1 to LA4. min And obtain the height difference ZA at each angle. i =LA i -Zmin, where i = 1, 2, 3, 4;

[0016] d) The control unit will control the height difference ZA i This is converted into a drive signal, which drives the corresponding actuator of the six-axis positioning stage to extend and retract, making the upper surface of the vacuum adsorption platform perpendicular to the spindle axis of the grinding mechanism. The measurement and driving process is repeated until Σ|ZA i | Less than the preset error threshold;

[0017] e) Place the workpiece on a vacuum adsorption stage and fix it in place by vacuum adsorption;

[0018] f) Using a laser rangefinder to measure the height Z of the workpiece's upper surface. n The control unit presses Z. n -Z m Calculate the spindle drop distance of the grinding mechanism, where Z m The distance between the laser rangefinder and the fixed reference point on the grinding cutter head in the grinding mechanism;

[0019] g) The spindle feeds according to the drop distance and performs grinding along the edge of the workpiece.

[0020] Furthermore, in step d), the threshold is set to ≤0.5μm;

[0021] The extension resolution of any actuator of the six-axis positioning stage is ≤1nm;

[0022] Each corner is sampled M ≥ 2 times, with M ranging from 2 to 100.

[0023] Further, in step d), the drive signal is the open-loop voltage pulse of each actuator of the six-axis positioning stage, with a pulse amplitude of V. i =K p ×ZA i K pThis is the calibrated piezoelectric voltage-displacement gain, in units of V / μm.

[0024] Furthermore, during the grinding process in step g), the laser rangefinder monitors the workpiece edge height in real time at a frequency of 1 kHz and feeds back the change to the control unit to dynamically correct the spindle drop distance.

[0025] Furthermore, between steps c) and d), the control unit performs a two-period phase decorrelation filter on the LA1–LA4 sequences:

[0026] a) Real-time height LA ​​of the corresponding corner in the vacuum adsorption stage i Sequence window length L = lcm(T) r ,T s Construct a cyclic matrix C;

[0027] b) Perform a Discrete Fourier Transform (DFT) on the row vectors of the cyclic matrix C to obtain the frequency domain spectrum S(k), and retain the components S′(k) with phase difference |Δφ(k)| < π / 4;

[0028] c) Obtain the filtered height by inverse transformation and with Alternative to LA i CalculateZA i The control unit pads S'(k) with zeros to its original length and then performs a normalized IDFT to obtain the result with LA. i Equal length

[0029] Furthermore, step d) employs a bidirectional iteration of reverse and forward directions, including:

[0030] Reverse contraction: by height difference | ZA i The actuator of the descending-order drive six-axis positioning stage has a displacement ΔL. inv =-|ZA i |;

[0031] Sequential elongation: Actuator displacement ΔL of the six-axis positioning stage seq =+0.5×|ZA i |;

[0032] Criterion: If If a bidirectional path is selected, it will be retained; otherwise, a unidirectional shrinkage will be performed.

[0033] Furthermore, after step e), the control unit collects the height He at k points equidistantly along the edge of the workpiece. j and with four corners ZA i As input, the edge height error Δe is predicted using a radial basis function network. j The network weight is w i (i=1,2,3,4), then:

[0034]

[0035] φ(r)=exp(-r 2 / σ 2 ), where r = |He j -LA i |, σ>0 is the base width; edge compensation amount ZAe j =Δe j ×k, where k is the mapping coefficient, the control unit will ZAe j Superimposed on the corresponding nearest angle ZA i And immediately re-execute steps b) through d) to update the LA. i Recalculate ZA i The second leveling was completed.

[0036] Furthermore, in step g), during the grinding process, the actual chamfer angle θ is obtained using a quantum dot fluorescence probe. f If |θ f -θ target If |>0.05°, then according to the sensitivity matrix Calculate the infinitesimal ZA, ΔZA = S + ×(θ target -θ f ), S + As the pseudo-inverse of S, it drives the corresponding two actuators to complete the angle closed loop.

[0037] Furthermore, at the moment of vacuum release, the control unit monitors the pressure at a frequency of ≥20kHz. When |dP / dt|>1kPa / ms, it immediately freezes the target displacement value output to the actuator for 0.5ms. The closed loop is restored after |dP / dt|<0.1kPa / ms.

[0038] The present invention has the following beneficial effects:

[0039] By replacing overall disassembly with milling using distance measurement-leveling, attitude correction can be completed in a single clamping, reducing equipment downtime and auxiliary time. The same six-axis positioning stage is compatible with workpieces of different sizes, eliminating the need for frequent changes to dedicated carriers and reducing spare parts inventory and management costs. Leveling and height setting are automatically executed by the control unit according to a set procedure, reducing operational differences caused by experience-based judgment and manual measurement. Through repeated measurement-drive closed-loop, attitude deviations can be gradually reduced within the same clamping, avoiding the accumulation of residual errors from one-time compensation. The leveling process relies only on the actuator's slight extension and retraction, without changing the stage thickness, thus extending the carrier's lifespan. During the grinding stage, edge height or chamfer angle is continuously read, and deviations can be corrected immediately, reducing the risk of scrap. Attached Figure Description

[0040] Figure 1 This is a structural diagram of the grinding system corresponding to the present invention. Detailed Implementation

[0041] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for explaining the present invention and are not intended to limit the present invention.

[0042] like Figure 1 The automatic leveling and grinding system corresponding to this invention includes:

[0043] Vacuum adsorption stage 1: Used to rigidly fix the workpiece to be ground. Vacuum adsorption holes are opened on the surface to stably clamp the workpiece through negative pressure.

[0044] Six-axis positioning stage 2: Adopting the Stewart platform layout, it is equipped with six piezoelectric actuators at the bottom. The displacement resolution of each piezoelectric actuator is ≤1nm. The six piezoelectric actuators are used to drive the stage to achieve micron-level attitude leveling.

[0045] Laser rangefinder: There are 4 sets (such as laser triangulation sensors with a resolution of 0.1nm), which are respectively aligned with the four corners of the platform. The ranging signal of the laser rangefinder is transmitted to the control unit via a high-speed AD module.

[0046] Control unit: An industrial controller equipped with a real-time operating system (such as a multi-axis motion controller based on ARM Cortex-A9), which has high-speed data processing and independent drive capability for 6 actuators.

[0047] Grinding spindle: Equipped with grinding head 3, it performs grinding action along the edge of the workpiece, and the vertical displacement accuracy of the spindle reaches 0.1μm.

[0048] Quantum dot fluorescence probe: excitation wavelength 405nm, emission wavelength 520nm, used for real-time monitoring of grinding angle.

[0049] Pressure sensor: sampling frequency ≥20kHz, used for monitoring pressure fluctuations during vacuum release.

[0050] This invention discloses an automatic leveling and grinding method, comprising the following steps:

[0051] Step a: Installation and initialization.

[0052] The vacuum adsorption stage 1 is mounted on the bearing surface of the six-axis positioning stage; the six piezoelectric actuators are electrically connected to the control unit through a 6-channel high-voltage amplifier.

[0053] Laser rangefinder zero-point calibration: Using the initial bearing surface of the six-axis positioning stage as a reference, control the laser rangefinder to emit laser light, record the measured value at this time and set it as the zero point (reference height = 0), with a calibration error ≤ 0.05μm;

[0054] Piezoelectric actuator gain calibration: Input DC voltages of 1V, 2V, ..., 10V individually to each actuator, measure the corresponding extension and retraction displacements using a laser rangefinder, and fit the voltage-displacement gain K to obtain the result. p The calibration data is stored in the control unit parameter library.

[0055] Step b: Corner height sampling and filtering.

[0056] The control unit sends a sampling command to the laser rangefinder, which continuously samples the four corners of the platform multiple times at a frequency of 10kHz, and takes the arithmetic mean as the real-time height of the corner, corresponding to: LA1, LA2, LA3, LA4.

[0057] To eliminate the interference of spindle resonance and servo beat frequency on the sampled data, the control unit performs filtering processing on LA1 to LA4 respectively, as follows:

[0058] Windowing and Circular Matrix Construction:

[0059] (1) The principal axis resonance period T was measured by spectrum analysis. r =2ms, sampling period T s =0.1ms, calculate the window length L = lcm(T) r ,T s ) / T s =20 (lcm is the least common multiple, that is, 20 sampling points are taken to form a filter window).

[0060] (2) For the original height sequence LA of each corner i Apply a Hanning window to (i = 1, 2, 3, 4) with window function coefficients h(n) = 0.5 - 0.5cos(2πn / L) (n = 0, 1, ..., L-1), to obtain the windowed sequence LA′. i =LA i ·h(n).

[0061] (3) Using windowed sequence LA′ i Given row vectors, construct a cyclic matrix C to ensure the periodicity of frequency domain processing.

[0062] Frequency domain filtering:

[0063] (1) Perform a Discrete Fourier Transform (DFT) on each row vector of the circular matrix C to obtain the frequency domain spectrum S(k), (k is the frequency point index).

[0064] (2) Calculate the phase difference at each frequency point Preserving phase difference The frequency domain components are then processed, and interference components corresponding to spindle resonance and servo beat frequency are removed to obtain the filtered frequency domain data S'(k).

[0065] Inverse Transformation and Data Recovery:

[0066] Pad S'(k) with zeros to the length of the original sampling sequence, perform a normalized inverse discrete Fourier transform (IDFT) to obtain the result similar to LA. i The height after filtering of equal length is denoted as... The mathematical expression is:

[0067]

[0068] `padzero` is a zero-padding operation to ensure the data length matches the original height. In subsequent calculations, it will be used as... Alternative to LA i CalculateZA i .

[0069] Step c: Derivation of the height of the six-axis positioning stage support point and calculation of the height difference.

[0070] (1) Derivation of the location and height of the 6 support points:

[0071] The vacuum adsorption stage is rectangular (taking 300mm×300mm as an example). The positions of the 6 support points (Spt1~Spt6) and their associated corners (used for height derivation) are shown in Table 1.

[0072] Table 1 shows the relationship between the six support points in the six-axis positioning stage and the four corners of the vacuum adsorption platform.

[0073]

[0074] Since the horizontal distance from each support point to its two associated corners is equal, the support point height = (filtered height of associated corner 1 + filtered height of associated corner 2) / 2, as shown in the following formula:

[0075] Spt1:

[0076] Spt2:

[0077] Spt3:

[0078] Spt4:

[0079] Spt5:

[0080] Spt6:

[0081] (2) Calculation of height difference.

[0082] Extract the minimum value Z of the height of the 6 support points. min =min(LSpt1, LSpt2...LSpt6);

[0083] Calculate the height difference between each support point:

[0084]

[0085] Step d: The platform is automatically leveled.

[0086] (1) Leveling accuracy criterion: Judgment by the control unit Check if it is ≤0.5μm. If not, perform leveling.

[0087] (2) Actuator drive signal generation: The control unit will generate the height difference ZA at the four corners. i By spatially weighted mapping to six support points, the height difference ZA corresponding to each support point is obtained. map(j) (j = 1, 2, ..., 6).

[0088] Mapping rules (based on the spatial relationship between support points and corners):

[0089] If the j-th support point is close to two corners (e.g., Spt1 is close to corners 1 and 2), then the height difference of this support point is the average of the height differences of the two corners, that is:

[0090]

[0091] After mapping is completed, the control unit generates 6 open-loop voltage pulses according to the following formula:

[0092] V j =K p ×ZA map(j) (j = 1, 2, ..., 6).

[0093] Among them, ZA map(j) K represents the corner height difference associated with the j-th support point (e.g., Spt1 is associated with ZA1 and ZA2, and the average value is taken). p =10V / μm is the voltage-displacement gain of the piezoelectric actuator. The voltage is amplified by a high-voltage amplifier and then drives the corresponding actuator to extend or retract.

[0094] (3) Reverse-forward bidirectional iterative leveling:

[0095] Reverse contraction: Calculate the absolute value of the height difference associated with the 6 support points |ZA map(j) | Sorted from largest to smallest, driving the corresponding actuator to contract, displacement ΔL inv,j =-|ZA map(j) (Rapidly reduce large errors).

[0096] Sequential elongation: Sort the height differences by absolute value from smallest to largest, and drive the corresponding actuator to elongate. Displacement:

[0097] ΔLseq ,j =+0.5×|ZA map(j) (Fine correction of small errors).

[0098] Iteration criterion: If after iteration If the change is ≤0.1μm, the bidirectional path of "reverse contraction + sequential elongation" is retained; otherwise, only unidirectional contraction is performed until... Leveling complete.

[0099] Step e: Workpiece adsorption and edge compensation.

[0100] (1) Vacuum adsorption of workpiece: Place the workpiece to be ground in the center of the vacuum adsorption stage, start the vacuum pump in the control unit, and make the gauge pressure of the adsorption hole of the stage ≤ -95kPa to fix the workpiece.

[0101] (1) Edge error compensation and secondary leveling:

[0102] Edge height sampling: The control unit drives the laser rangefinder to collect the height at k=8 points at equal intervals along the edge of the workpiece, denoted as H. e1 ~H e8 (The included angle between two adjacent points is 45°, covering the entire edge of the workpiece).

[0103] Edge error calculation: based on the filtered height of the four corners. As input, the height error Δej (j=1,2,…,8) of each edge point is predicted using a radial basis function network, as follows:

[0104]

[0105] Among them, w i (i = 1, 2, 3, 4) are the network weights (i.e., radial basis network weights, which are obtained by training the “edge warping - corner height” sample set, with a training error ≤ 0.05 μm);

[0106] φ(r) is the radial basis function, taking the Gaussian function φ(r) = exp(-r 2 / σ 2 (σ = 0.5µm as the base width, optimized through cross-validation);

[0107] Where r = |He j -LA i |, where σ>0 is the base width;

[0108] Edge compensation amount ZAe j =Δe j×k, where k is the mapping coefficient (usually taken as 0.8), and ZAe j Superimposed on the corresponding nearest angle ZA i Then, immediately return to steps b to d and re-execute "laser sampling → filtering → height difference calculation → leveling" to eliminate the effect of workpiece edge warping.

[0109] Step f: Grinding spindle height setting.

[0110] (1) The laser rangefinder measures the average height of the upper surface of the workpiece, denoted as Z. n ;

[0111] (2) Call the parameter Z stored in the control unit m (Vertical distance from the laser rangefinder installation position to the fixed reference point of the grinding head, calibration accuracy ≤ 0.05μm);

[0112] (3) The control unit calculates the falling distance of the grinding spindle according to the formula: ΔZ=Zn-Zm;

[0113] (4) Drive the grinding spindle to move down ΔZ in the vertical direction so that the distance between the bottom surface of the grinding head and the upper surface of the workpiece reaches the target grinding depth (the target depth is 0.5mm in this embodiment).

[0114] Step g: Spindle grinding and process control

[0115] (1) Grinding execution: The control unit starts the grinding spindle (3000r / min) and drives the spindle to feed along the edge of the workpiece in a clockwise circular trajectory (feed speed 5mm / s) to execute the grinding action.

[0116] (2) Real-time height monitoring: The laser rangefinder collects the height of the workpiece edge in real time at a frequency of 1kHz and feeds back the height change ΔH to the control unit; if |ΔH|>0.1um (such as workpiece micro-deformation or equipment vibration), the control unit immediately corrects the spindle drop distance ΔZ'=ΔZ-ΔH to ensure uniform grinding depth.

[0117] (3) Angle closed-loop correction: The quantum dot fluorescence probe acquires the actual angle θ of the workpiece grinding chamfer in real time. f If |θ f -θ target |>0.05°(θ target (For the target chamfer angle), calculate the angle deviation Δθ = θ target -θ f .

[0118] Based on the sensitivity matrix, then according to the sensitivity matrix Calculate the infinitesimal ZA, ΔZA = S + ×Δθ,S +As the pseudo-inverse of S, it drives the corresponding two actuators to complete the angle closed loop.

[0119] The sensitivity matrix S was constructed using a physical calibration method: Four sets of standard samples with known height differences (0.2 μm, 0.4 μm, 0.6 μm, 0.8 μm) were used to measure the corresponding chamfer angle deviations. The S matrix elements were fitted to obtain values ​​ranging from 0.03° / μm to 0.04° / μm (e.g., S...). 11 =0.032° / μm, S 21 =0.038° / μm).

[0120] The control unit maps ΔZA to the corresponding two actuators, driving them to extend and retract to correct the stage posture, so that the actual chamfer angle returns to the target range.

[0121] Step h: Vacuum release and baseline protection.

[0122] After grinding is completed, the control unit drives the grinding spindle to return to the initial position (away from the workpiece surface) and turns off the vacuum pump to release the negative pressure of the vacuum adsorption stage;

[0123] The pressure sensor monitors the pressure change rate dP / dt of the negative pressure pipeline in real time at a frequency of ≥20kHz.

[0124] If |dP / dt|>1kPa / ms (the sudden pressure change during vacuum release can easily cause stage vibration), the control unit immediately freezes the output to the target displacement value of the 6 piezoelectric actuators for 0.5ms.

[0125] Once the air pressure sensor detects that |dP / dt| < 0.1 kPa / ms (air pressure stabilizes), the control unit resumes closed-loop control of the actuator; the workpiece that has been ground is removed manually or by a robotic arm, completing one grinding cycle.

[0126] The above description is only a preferred embodiment of the present invention. It should be noted that those skilled in the art can make several improvements without departing from the principle of the present invention, and these improvements should also be considered within the scope of protection of the present invention.

Claims

1. An automatic leveling and grinding method, characterized in that: Includes the following steps: a) Mount the vacuum adsorption stage onto the six-axis positioning stage and keep it stationary; b) The laser rangefinder fixed to the grinding mechanism is used to continuously sample each of the four corners of the vacuum adsorption stage multiple times, and the arithmetic mean is taken as the real-time height of the corner, corresponding to: LA1, LA2, LA3, LA4; c) The control unit calculates the minimum value Z among LA1 to LA4. min And obtain the height difference ZA at each angle. i =LA i -Zmin, where i = 1, 2, 3, 4; d) The control unit will control the height difference ZA i This is converted into a drive signal, which drives the corresponding actuator of the six-axis positioning stage to extend and retract, making the upper surface of the vacuum adsorption platform perpendicular to the spindle axis of the grinding mechanism. The measurement and driving process is repeated until Σ|ZA i | Less than the preset error threshold; e) Place the workpiece on a vacuum adsorption stage and fix it in place by vacuum adsorption; f) Using a laser rangefinder to measure the height Z of the workpiece's upper surface. n The control unit presses Z. n -Z m Calculate the spindle drop distance of the grinding mechanism, where Z m The distance between the laser rangefinder and the fixed reference point on the grinding cutter head in the grinding mechanism; g) The spindle feeds according to the drop distance and performs grinding along the edge of the workpiece.

2. The automatic leveling and grinding method as described in claim 1, characterized in that: In step d), the threshold is set to ≤0.5μm; The extension resolution of any actuator on the six-axis positioning stage is ≤1nm.

3. The automatic leveling and grinding method as described in claim 1, characterized in that: In step d), the drive signal is the open-loop voltage pulse of each actuator of the six-axis positioning stage, with a pulse amplitude of V. i =K p ×ZA i K p This is the calibrated piezoelectric voltage-displacement gain, in units of V / μm.

4. The automatic leveling and grinding method as described in claim 1, characterized in that: During the grinding process in step g), the laser rangefinder monitors the workpiece edge height in real time at a frequency of 1 kHz and feeds back the changes to the control unit to dynamically correct the spindle drop distance.

5. The automatic leveling and grinding method as described in claim 1, characterized in that: Between steps c) and d), the control unit performs a two-period phase decorrelation filter on the LA1–LA4 sequences: a) Real-time height LA ​​of the corresponding corner in the vacuum adsorption stage i Sequence window length L = lcm(T) r ,T s Construct a cyclic matrix C; b) Perform a Discrete Fourier Transform (DFT) on the row vectors of the cyclic matrix C to obtain the frequency domain spectrum S(k), and retain the components S′(k) with phase difference |Δφ(k)| < π / 4; c) Obtain the filtered height by inverse transformation and with Alternative to LA i CalculateZA i The control unit pads S'(k) with zeros to its original length and then performs a normalized IDFT to obtain the result with LA. i Equal length 6. The automatic leveling and grinding method as described in claim 1, characterized in that: Step d) employs a bidirectional iteration of reverse and forward directions, including: Reverse contraction: by height difference | ZA i The actuator of the descending-order drive six-axis positioning stage has a displacement ΔL. inv =-|ZA i |; Sequential elongation: Actuator displacement ΔL of the six-axis positioning stage seq =+0.5×|ZA i |; Criterion: If If a bidirectional path is selected, it will be retained; otherwise, a unidirectional shrinkage will be performed.

7. The automatic leveling and grinding method as described in claim 1, characterized in that: After step e), the control unit collects the height He at k points equidistantly along the edge of the workpiece. j and with four corners ZA i As input, the edge height error Δe is predicted using a radial basis function network. j， The network weight is w i (i=1,2,3,4), then: φ(r)=exp(-r 2 / σ 2 ), where r = |He j -LA i |, σ>0 is the base width; edge compensation amount ZAe j =Δe j ×k, where k is the mapping coefficient, the control unit will ZAe j Superimposed on the corresponding nearest angle ZA i And immediately re-execute steps b) through d) to update the LA. i Recalculate ZA i The second leveling was completed.

8. The automatic leveling and grinding method as described in claim 1, characterized in that: Step g) During grinding, the actual chamfer angle θ is obtained using a quantum dot fluorescence probe. f If |θ f -θ target If |>0.05°, then according to the sensitivity matrix Calculate the infinitesimal ZA, ΔZA = S + ×(θ target -θ f ), S + As the pseudo-inverse of S, it drives the corresponding two actuators to complete the angle closed loop.

9. The automatic leveling and grinding method as described in claim 1, characterized in that: At the moment of vacuum release, the control unit monitors the pressure at a frequency of ≥20kHz. When |dP / dt|>1kPa / ms, it immediately freezes the target displacement value output to the actuator for 0.5ms. The closed loop is restored after |dP / dt|<0.1kPa / ms.

Images