Intelligent control system and method for sapphire substrate polishing

Abstract

The invention relates to the technical field of machining, in particular to an intelligent control system and method for sapphire substrate polishing, and the method comprises the following steps: S1, surface topography data acquisition: collecting surface roughness data and a three-dimensional topography distribution diagram of a sapphire substrate in real time; s2, generating characteristic parameters: extracting the surface waviness, the local recess depth, the bulge density, the lattice distortion index and the residual stress distribution gradient as the characteristic parameters; s3, dynamic parameter adjustment: generating a dynamic adjustment instruction of the polishing pressure, the rotating speed of the polishing disk and the supply rate of the polishing solution; s4, distribution optimization of the polishing solution: realizing uniform coverage of the polishing solution on the surface of the substrate; s5, stress real-time monitoring: generating a stress abnormity early warning signal; s6, outputting a closed-loop control instruction: feeding back a stress abnormity early warning signal to the adaptive control model, and iteratively correcting the dynamic adjustment instruction; according to the method, the risks of stress concentration and fatigue damage can be reflected in real time, and the safety and high efficiency of the polishing process are ensured.

Description

Technical Field

[0001] The present invention relates to the field of mechanical processing technology, and in particular to an intelligent control system and method for polishing a sapphire substrate. Background Art

[0002] Sapphire substrates are widely used in high-tech fields such as semiconductors, optoelectronics, and blue-light displays due to their excellent mechanical strength, chemical stability, and good transparency. In the manufacturing process of sapphire substrates, polishing is one of the key steps that directly affects the surface quality of the substrate and the accuracy of subsequent processes. Traditional polishing methods mainly rely on experience and manual operation, and it is difficult to adjust polishing parameters in real time, resulting in low efficiency and unstable surface quality during the polishing process. Therefore, how to achieve intelligent, efficient, and precise polishing of sapphire substrates has become a research hotspot in the current technology field.

[0003] Most sapphire substrate polishing methods in the existing technology rely on manual adjustment or preset parameters for polishing, and lack real-time monitoring and dynamic adjustment mechanisms. In the traditional polishing process, the distribution of the polishing liquid often cannot be accurately controlled, resulting in local over-polishing or uneven polishing, which in turn affects the surface quality. At the same time, the lack of a real-time stress monitoring system makes it impossible to detect the generation of microcracks and stress concentration in a timely manner, which can easily lead to damage or fatigue of the substrate. Existing control methods often ignore the dynamic relationship between surface morphology changes and polishing parameters, resulting in the polishing process being unable to adapt to real-time changes in the substrate surface morphology, reducing the efficiency and quality of the polishing process. Summary of the Invention

[0004] The present invention provides an intelligent control system and method for polishing sapphire substrates, which realizes precise and efficient intelligent polishing, can significantly improve the precision, efficiency and safety of the polishing process, and ensure high-quality processing of sapphire substrates.

[0005] A sapphire substrate polishing intelligent control method comprises the following steps: S1, surface topography data acquisition: The surface roughness data and three-dimensional topography distribution map of the sapphire substrate are collected in real time through a laser triangulation displacement sensor; S2, feature parameter generation: Based on the surface roughness data and 3D topography distribution map, surface waviness, local depression depth, protrusion density, lattice distortion index and residual stress distribution gradient are extracted as feature parameters; S3, dynamic parameter adjustment: input the characteristic parameters into the adaptive control model to generate dynamic adjustment instructions for polishing pressure, polishing disk speed and polishing liquid supply rate; S4, polishing liquid distribution optimization: According to the dynamic adjustment instructions, the polishing liquid injection angle and flow rate are adjusted through the multi-channel solenoid valve array to achieve uniform coverage of the polishing liquid on the substrate surface; S5, real-time stress monitoring: using acoustic emission sensors to monitor the acoustic emission signals generated by micro-cracks on the substrate surface during polishing, and generating stress abnormality warning signals; S6, closed-loop control command output: Feedback the stress abnormality warning signal to the adaptive control model, iteratively correct the dynamic adjustment command until polishing is completed.

[0006] Optionally, the surface topography data acquisition in S1 includes: S11, parameter initialization: set the axial measurement accuracy of the laser triangulation displacement sensor , sampling frequency , and based on the initial thickness of the sapphire substrate As the distance measurement center, initialize the sensor measurement interval to ,in, is the maximum thickness fluctuation tolerance that may occur during the polishing process; S12, dynamic scanning path planning: Divide the substrate surface into grid coordinate points , a spiral progressive scanning path is generated according to the defect probability distribution, and its path spacing The conditions are met ,in, is the basic path spacing, is the defect weight coefficient, Indicates the historical defect probability of the coordinate point; S13, displacement data acquisition: Move the laser triangulation displacement sensor probe along the scanning path to collect height data at each grid point to form a two-dimensional height matrix

[0007] S14, surface roughness calculation: Perform spatial frequency analysis and topography evaluation to calculate surface roughness and root mean square gradient

[0008] S15, 3D shape reconstruction: 3D shape surface fitting: 2D height matrix of the acquisition Perform quadratic polynomial surface fitting to construct a continuous and smooth three-dimensional topography function; S16, Data Verification and Transmission: Verify the deviation of the 3D shape reconstruction results and calculate the local area data deviation ,when The second scan of the corresponding area is triggered at this time. After verification, the surface roughness data is output and three-dimensional morphology distribution map .

[0009] Optionally, the generating of the characteristic parameters in S2 includes: S21, Surface waviness extraction: 3D topography distribution map Perform Gaussian low-pass filtering to preserve the spatial wavelength The ripple component of the wave and calculate the height of the ripple; S22, local depression depth extraction: setting depression determination threshold , extract satisfaction The continuous area of ​​​​is taken as the concave area, where is the average surface height, for each concave area , calculate its maximum depression depth , and extract the deepest one among all regions as the global maximum concave depth; S23, convex density extraction: setting the convexity determination threshold , extract satisfaction The continuous area is regarded as the convex area, and the unit area Count the number of raised areas , calculate the bump density ; S24, Lattice distortion index extraction: three-dimensional morphology distribution map Perform fast Fourier transform to obtain frequency domain energy distribution , extract low frequency With high frequency The energy ratio of the part is used as the lattice distortion index ; S25, residual stress distribution gradient extraction: based on acoustic emission signal data The spatial distribution of residual stress intensity is constructed to approximate the residual stress intensity distribution map , and calculate the global residual stress distribution gradient .

[0010] Optionally, the dynamic parameter adjustment in S3 includes: S31, adaptive model reasoning: The extracted characteristic parameters are input into the adaptive control model, and the initial adjustment parameters are calculated using the deep reinforcement learning network, including the initial value of the polishing pressure, the initial value of the polishing disk speed, and the initial value of the polishing liquid supply rate; S32, constraint correction: correct the generated initial adjustment parameters, output the corrected adjustment parameters and generate dynamic adjustment instructions.

[0011] Optionally, the adaptive model reasoning in S31 includes: S311, feature parameter normalization: performing normalization processing on the input feature parameters; S312, reinforcement learning network inference: input the normalized feature parameters into the deep reinforcement learning network and calculate the normalized values ​​of the initial adjustment parameters; S313, parameter denormalization: performing parameter denormalization processing on the calculated normalized initial adjustment parameters, and outputting the initial value of the polishing pressure, the initial value of the polishing disk rotation speed, and the initial value of the polishing liquid supply rate.

[0012] Optionally, the constraint modification in S32 includes: S321, polishing pressure gradient constraint: Calculate the change rate between the current polishing pressure and the pressure of the previous cycle. If the change rate exceeds the set maximum allowable value (5% / s), adjust the current pressure value according to the safety gradient through the limit correction strategy. If the change rate is less than the maximum allowable value, maintain the current pressure value. S322, dynamic de-weighting of the polishing disk speed: When the protrusion density exceeds a set threshold (15%), the polishing disk speed is dynamically adjusted using a de-weighting coefficient. The de-weighting coefficient is adjusted according to the change in protrusion density. Finally, the speed value adjusted according to the de-weighting coefficient is output; S323, polishing liquid supply rate saturation protection: the polishing liquid supply rate initial value Restricted to the preset safety range; S324, dynamic adjustment instruction package: the corrected polishing pressure , polishing disc speed and polishing liquid supply rate Encapsulated as structured control instructions ,in, The current timestamp.

[0013] Optionally, the polishing liquid distribution optimization in S4 includes: S41, dynamic partition mapping of polishing area: According to the three-dimensional topography distribution map, the substrate surface is divided into rectangular areas, based on each area Priority weight , calculate the required baseline flow for each area ; S42, multi-channel solenoid valve flow control: according to the base flow required for each area , by controlling the valve port opening of the multi-channel solenoid valve to adjust the polishing liquid flow in each area, the valve port opening of the solenoid valve is proportional to the flow reference value, and the opening difference is limited; S43, adaptive adjustment of spray angle: calculate the spray angle according to the gradient of the surface height of each area, and adjust the spray angle according to the surface slope. If the calculated angle exceeds the predetermined range, , then make restriction adjustments; S44, real-time coverage verification and feedback: Obtaining a thermal map of polishing fluid coverage using an infrared thermal imager , calculate the coverage uniformity index ,like , generate coverage compensation instructions and feed them back to dynamic parameter adjustment, triggering secondary parameter optimization.

[0014] Optionally, the real-time stress monitoring in S5 includes: S51, Acoustic Emission Signal Acquisition: Embed acoustic emission sensors in a circular array on the edge of the polishing plate, with each acoustic emission sensor installed at a 30° inclination angle to the substrate surface; S52, micro-crack feature extraction: Detect the burst waveform in the acoustic emission signal with an amplitude more than 3 times that of the baseline noise, extract its rise time (the time required for the signal to rise from the baseline to the maximum value, reflecting the crack development speed), energy integral (the total energy of the signal, representing the intensity of the crack) and duration as characteristic parameters (the time the signal maintains a high amplitude state, affecting the degree of damage of the crack), and determine the crack type based on the characteristic parameters. and , it is judged as a brittle crack. If and , determined to be a ductile crack; S53, stress anomaly warning generation: Based on the extracted microcrack features, a stress anomaly warning is generated. If the number of brittle crack events per unit time is greater than or equal to 3, it is judged as high-risk stress concentration. If the energy integral of the ductile crack exceeds , it is judged as potential fatigue damage. According to the judgment result, two-level warning signals are issued, including the first-level warning triggering the reduction of the polishing disk speed and the second-level warning marking the abnormal area and feeding back to the adaptive control model.

[0015] Optionally, the closed-loop control instruction output in S6 includes: S61, Warning Signal Priority Analysis: Receives the first or second level warning signal from stress monitoring and extracts the stress risk level, abnormal area coordinates, and historical polishing parameters. Based on the warning level, the weight of the current polishing pressure is adjusted. When high-risk stress concentration is detected, the current polishing pressure is reduced by 60%-80%, and the abnormal area is marked as a no-polishing zone, suspending polishing in that area. When potential fatigue risk arises, the adaptive control model is made more sensitive to the surface topography of the abnormal area, raising the processing priority of that area and triggering a reduction in the local polishing intensity. S62, dynamically issuing a correction command: The corrected polishing pressure, polishing disk speed, and polishing liquid supply rate are smoothed through a first-order low-pass filter. In addition, the timestamp of the dynamic adjustment command is checked to see if it matches the current state of the device. If the delay exceeds 200ms, the old dynamic adjustment command is discarded and recalculated; S63, polishing termination judgment: when the surface roughness reaches When there is no new warning signal for 5 minutes, when the cumulative polishing time reaches 120 minutes and the roughness decrease trend is less than 10 minutes in a row Or when the number of iterations of the adaptive control model exceeds 50 and the convergence amplitude of the feedback parameter is less than 5% for three consecutive times, the polishing completion signal is output and all actuators are turned off, where, is the target roughness threshold.

[0016] A sapphire substrate polishing intelligent control system, used to implement the above-mentioned sapphire substrate polishing intelligent control method, includes the following modules: Surface topography data acquisition module: uses a laser triangulation sensor to collect real-time surface roughness data and three-dimensional topography distribution maps of sapphire substrates; Feature parameter generation module: Based on the collected surface roughness data and 3D topography distribution map, feature parameters are extracted, including surface waviness, local depression depth, protrusion density, lattice distortion index, and residual stress distribution gradient; Dynamic parameter adjustment module: This module inputs the extracted characteristic parameters into the adaptive control model to generate corresponding dynamic adjustment instructions for polishing pressure, polishing disk speed, and polishing liquid supply rate to optimize the polishing process. Polishing liquid distribution optimization module: According to dynamic adjustment instructions, the multi-channel solenoid valve array adjusts the spray angle and flow rate of the polishing liquid to achieve uniform coverage of the polishing liquid on the substrate surface; Real-time stress monitoring module: uses acoustic emission sensors to monitor the generation of microcracks on the substrate surface during polishing and generates stress abnormality warning signals; Closed-loop control module: Feedback the stress abnormality warning signal to the adaptive control model, and iteratively correct the dynamic adjustment instructions based on the feedback information until polishing is completed.

[0017] Beneficial effects of the present invention: The present invention uses a laser triangulation displacement sensor to collect surface roughness data and three-dimensional morphology distribution maps in real time, and based on this, generates characteristic parameters such as surface waviness, local depression depth and protrusion density. It can accurately describe the morphological characteristics of the substrate surface, and dynamically adjust parameters through a deep reinforcement learning model to generate control instructions such as polishing pressure, polishing disk speed and polishing liquid supply rate, ensuring precise control of the sapphire substrate surface morphology during the polishing process and achieving optimization of surface quality.

[0018] The present invention controls the injection angle and flow rate of the polishing liquid through a multi-channel solenoid valve array to ensure that the polishing liquid can evenly cover the substrate surface, thereby improving the polishing effect. At the same time, an acoustic emission sensor is used to monitor the generation of microcracks on the substrate surface and generate stress abnormality warning signals based on the characteristics of the cracks. This can reflect the risks of stress concentration and fatigue damage in real time, ensuring the safety and efficiency of the polishing process.

[0019] The present invention uses a closed-loop control mechanism to feed back stress anomaly warning signals to an adaptive control model, dynamically correcting adjustment instructions during the polishing process to ensure continuous optimization of polishing parameters throughout the polishing process. At the same time, dual termination conditions are set, including when the surface roughness reaches the set standard and no new warning signals are generated for a certain period of time, or when the accumulated polishing time reaches the process upper limit. Through these termination judgments, the timing of polishing completion can be accurately determined, and the actuator can be shut down when appropriate to avoid over-polishing or damage to the substrate. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0021] Figure 1 Schematic diagram of the control method flow in an embodiment of the present invention; Figure 2 Schematic diagram of system function modules according to an embodiment of the present invention. DETAILED DESCRIPTION

[0022] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments. It is also noted that, to provide a more detailed description, the following embodiments are best and preferred embodiments, and those skilled in the art may employ alternative methods for implementing certain known technologies. Furthermore, the accompanying drawings are intended only to provide a more detailed description of the embodiments and are not intended to limit the present invention.

[0023] It should be noted that references in the specification to "one embodiment," "an embodiment," "exemplary embodiments," "some embodiments," etc. indicate that the described embodiments may include specific features, structures, or characteristics, but not necessarily every embodiment will include such specific features, structures, or characteristics. Furthermore, when specific features, structures, or characteristics are described in conjunction with an embodiment, it is within the knowledge of persons skilled in the relevant art to implement such features, structures, or characteristics in conjunction with other embodiments (whether or not explicitly described).

[0024] In general, terms can be understood, at least in part, from their use in context. For example, depending at least in part on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular sense, or can be used to describe a combination of features, structures, or characteristics in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but can instead, depending at least in part on the context, allow for the presence of other factors that are not necessarily explicitly described.

[0025] like Figure 1 As shown, a sapphire substrate polishing intelligent control method includes the following steps: S1, surface topography data acquisition: The surface roughness data and three-dimensional topography distribution map of the sapphire substrate are collected in real time through a laser triangulation displacement sensor; S2, feature parameter generation: Based on the surface roughness data and 3D topography distribution map, surface waviness, local depression depth, protrusion density, lattice distortion index and residual stress distribution gradient are extracted as feature parameters; S3, dynamic parameter adjustment: input the characteristic parameters into the adaptive control model to generate dynamic adjustment instructions for polishing pressure, polishing disk speed and polishing liquid supply rate; S4, polishing liquid distribution optimization: According to the dynamic adjustment instructions, the polishing liquid injection angle and flow rate are adjusted through the multi-channel solenoid valve array to achieve uniform coverage of the polishing liquid on the substrate surface; S5, real-time stress monitoring: using acoustic emission sensors to monitor the acoustic emission signals generated by micro-cracks on the substrate surface during polishing, and generating stress abnormality warning signals; S6, closed-loop control command output: Feedback the stress abnormality warning signal to the adaptive control model, iteratively correct the dynamic adjustment command until polishing is completed.

[0026] Surface topography data acquisition in S1 includes: S11, parameter initialization: set the axial measurement accuracy of the laser triangulation displacement sensor , sampling frequency , and based on the initial thickness of the sapphire substrate As the distance measurement center, initialize the sensor measurement interval to ,in, is the maximum thickness fluctuation tolerance that may occur during the polishing process; S12, dynamic scanning path planning: Divide the substrate surface into grid coordinate points , a spiral progressive scanning path is generated according to the defect probability distribution, and its path spacing The conditions are met ,in, is the basic path spacing, is the defect weight coefficient, Indicates the historical defect probability of the coordinate point; S13, displacement data acquisition: Move the laser triangulation displacement sensor probe along the scanning path to collect height data at each grid point to form a two-dimensional height matrix ; S14, surface roughness calculation: Perform spatial frequency analysis and topography evaluation to calculate surface roughness and root mean square gradient , expressed as: ; ; in, is the surface height function, is the sampling length, is the average height, are discrete sampling points, is the number of sampling points, is the surface gradient; S15, 3D surface fitting: for the collected 2D height matrix Perform quadratic polynomial surface fitting to construct a continuous and smooth three-dimensional topography function, which can be expressed as: ; in, is the fitting coefficient, is the fitted three-dimensional morphology, 、 are independent variables 、 The power of 、 are the coordinates of the grid points on the substrate surface in the horizontal and vertical directions respectively; S16, Data Verification and Transmission: Verify the deviation of the 3D shape reconstruction results and calculate the local area data deviation ,when The second scan of the corresponding area is triggered at this time. After verification, the surface roughness data is output and three-dimensional morphology distribution map , expressed as: ; in, is the original roughness, is the roughness after fitting.

[0027] The feature parameter generation in S2 includes: S21, Surface waviness extraction: 3D topography distribution map Perform Gaussian low-pass filtering to preserve the spatial wavelength The ripple component of , and calculate the ripple height, expressed as: ; in, represents a two-dimensional convolution operation, The standard deviation is The two-dimensional Gaussian kernel function, , is the minimum ripple wavelength, take , is the height value distribution of the waviness component in the three-dimensional surface topography after Gaussian low-pass filtering; ; in, is the two-dimensional space coordinate offset in the Gaussian kernel function; ; in, is the maximum peak-to-valley height of the surface waviness, is the maximum height of the waviness, is the minimum height of the waviness; S22, local depression depth extraction: setting depression determination threshold , extract satisfaction The continuous area of ​​​​is taken as the concave area, where is the average surface height, for each concave area , calculate its maximum depression depth , and extract the deepest one among all regions as the global maximum concave depth, expressed as: ; ; S23, convex density extraction: setting the convexity determination threshold , extract satisfaction The continuous area is regarded as the convex area, and the unit area Count the number of raised areas , calculate the bump density , expressed as: ; in, is the protrusion density, which represents the percentage of protrusions per unit area. is the total surface area of ​​the substrate; S24, Lattice distortion index extraction: three-dimensional morphology distribution map Perform fast Fourier transform to obtain frequency domain energy distribution , extract low frequency With high frequency The energy ratio of the part is used as the lattice distortion index , expressed as: ; in, Expressed as low-frequency energy, Expressed as high-frequency energy; ; in, is the grayscale / height value of the spatial domain image, 、 is the spatial frequency coordinate, 、 is the two-dimensional coordinate in the spatial domain; S25, residual stress distribution gradient extraction: based on acoustic emission signal data The spatial distribution of residual stress intensity is constructed to approximate the residual stress intensity distribution map , and calculate the global residual stress distribution gradient , expressed as: ; in, is the cumulative intensity of acoustic emission events per unit time, is the acoustic-stress response proportionality coefficient; .

[0028] Dynamic parameter adjustments in S3 include: S31, adaptive model reasoning: The extracted characteristic parameters are input into the adaptive control model, and the initial adjustment parameters are calculated using the deep reinforcement learning network, including the initial value of the polishing pressure, the initial value of the polishing disk speed, and the initial value of the polishing liquid supply rate; S32, constraint correction: correct the generated initial adjustment parameters, output the corrected adjustment parameters and generate dynamic adjustment instructions.

[0029] Adaptive model inference in S31 includes: S311, feature parameter normalization: normalize the input feature parameters to eliminate dimensional differences, expressed as: ; ; ; ; ; in, 、 、 、 、 They are the normalized surface waviness, local depression depth, protrusion density, lattice distortion index, and residual stress distribution gradient, respectively. 、 are the minimum and maximum values ​​of the surface waviness, is the minimum reference value set, is the maximum reference value set, An upper limit value is set empirically to represent an extreme surface stress gradient; S312, reinforcement learning network inference: Input the normalized feature parameters into the deep reinforcement learning network and calculate the normalized values ​​of the initial adjustment parameters, expressed as: Hidden layer state calculation: ; in, is the input vector, is the weight from the input layer to the hidden layer, is the hidden layer bias; Output layer calculation: ; ; ; in, For the Sigmoid activation function, the output value range is , 、 、 is the weight from the hidden layer to the output layer, 、 、 is the bias of the output layer, 、 、 are the normalized initial values ​​of polishing pressure, polishing disc speed, and polishing liquid supply rate; S313, parameter denormalization: perform parameter denormalization processing on the calculated normalized initial adjustment parameters, and output the initial value of polishing pressure, initial value of polishing disk speed, and initial value of polishing liquid supply rate, which are expressed as: ; in, is the initial value of polishing pressure, is the minimum polishing pressure (set to 0.05MPa), is the maximum value of polishing pressure (set to 0.3MPa); ; in, is the initial value of the polishing disc speed, is the minimum value of the polishing disc speed (set to 50rpm), is the maximum speed of the polishing disc (set to 600 rpm); ; in, is the initial value of the polishing liquid supply rate, is the maximum value of the polishing liquid supply rate (set to ).

[0030] Constraint modifications in S32 include: S321, Polishing Pressure Gradient Constraint: Calculate the rate of change between the current polishing pressure and the pressure of the previous cycle. If the rate of change exceeds the set maximum allowable value (5% / s), adjust the current pressure value according to the safety gradient through the limit correction strategy to ensure a smooth polishing process. If the rate of change is less than the maximum allowable value, maintain the current pressure value. Specifically, Change rate calculation: calculate the initial value of the current polishing pressure Relative to the previous period value Rate of change: ; in, is the initial value of polishing pressure in the current cycle, is the polishing pressure value of the previous cycle, is the time interval; Limit correction: If , then reset the pressure value according to the safety gradient, that is, ,otherwise, ;in, The maximum limit of polishing pressure change rate (set to ), is a symbolic function, indicating The sign of , which takes the value of -1 or 1; S322, dynamic de-weighting of polishing disk speed: When the protrusion density exceeds the set threshold (15%), the polishing disk speed is dynamically adjusted using a de-weighting coefficient. The de-weighting coefficient is adjusted according to the change in protrusion density to ensure that over-polishing is not caused in the protrusion area. Finally, the speed value adjusted according to the de-weighting coefficient is output, including: Bump density correlation correction: When the bump density When the initial value of the speed Apply a downgrade factor , expressed as: ; in, is the current bump density, is the convex density threshold (set to ), is the weight reduction coefficient, used to adjust the speed, is the initial value of the polishing disc speed; Speed ​​correction: ; in, is the corrected polishing disc speed, is the initial polishing disc speed, is the weight reduction coefficient; S323, polishing liquid supply rate saturation protection: the polishing liquid supply rate initial value Limited to the preset safety range, expressed as: ; in, The upper limit of the supply rate (set to ), is the polishing liquid supply rate after correction; S324, dynamic adjustment instruction package: the corrected polishing pressure , polishing disc speed and polishing liquid supply rate Encapsulated as structured control instructions ,in, The current timestamp.

[0031] Slurry distribution optimization in S4 includes: S41, dynamic partition mapping of polishing area: According to the three-dimensional topography distribution map, the substrate surface is divided into rectangular areas, based on each area Priority weight , calculate the required baseline flow for each area , expressed as: ; ; in, For the region The standard deviation of the inner surface height, For the region The priority weight reflects the surface fluctuation of the area. The smaller the weight value, the higher the priority. For the region Baseline flow rate; S42, multi-channel solenoid valve flow control: according to the base flow required for each area The polishing liquid flow in each area is adjusted by controlling the valve opening of the multi-channel solenoid valve. The valve opening of the solenoid valve is proportional to the flow reference value, and the opening difference is limited to avoid excessive flow changes and ensure smooth and uniform flow distribution. Specifically, it includes: Introducing temperature compensation: In each solenoid valve channel Integrated temperature sensor , used to monitor the temperature change of the liquid in real time, considering the influence of temperature on the viscosity and flow rate of the polishing liquid, introducing a temperature compensation mechanism to correct the original liquid supply volume and obtain the temperature-corrected channel liquid supply volume , expressed as: ; in, is the temperature compensation coefficient, Process reference temperature; Valve opening calculation: Each solenoid valve channel Corresponding coverage area set , the valve opening Channel liquid supply volume corrected by temperature Mapping to opening range , expressed as: ; in, is the flow attenuation coefficient, set to 0.9, Electromagnetic channel The maximum flow rate, Electromagnetic channel The opening degree; Opening gradient limit: opening difference between adjacent channels Not exceed , otherwise a smooth transition is performed, which is expressed as: ; in, It is a sign function with a value of -1 or 1, indicating the direction of the opening adjustment. is the opening of the solenoid valve after adjustment; S43, adaptive adjustment of spray angle: calculate the spray angle according to the gradient of the surface height of each area, and adjust the spray angle according to the surface slope. If the calculated angle exceeds the predetermined range, , then limit adjustments are made to ensure that the spray angle is always within the effective range, thereby ensuring uniform distribution of the polishing liquid, including: Spray angle calculation: based on area The average height gradient , calculate the injection angle , expressed as: ; in, For the region The average height gradient, which indicates the magnitude of the surface slope, is the damping coefficient, set to 0.5; Angle execution constraint: actual injection angle Restricted to In the range, it is expressed as: ; S44, real-time coverage verification and feedback: Obtaining a thermal map of polishing fluid coverage using an infrared thermal imager , calculate the coverage uniformity index ,like , generate coverage compensation instructions and feed them back to dynamic parameter adjustment, triggering secondary parameter optimization, Indicates uneven coverage and requires readjustment of the spray flow and angle to ensure uniform polishing of the surface. Coverage uniformity index Expressed as: ; ; ; in, is the thermal value after centrifugal force correction, is the centrifugal force factor, is the radial distance from the current point to the center of the polishing disk, is the polishing disc radius, is the centrifugal compensation coefficient, and the recommended value range is 0.2-0.5. is the average thermal value of the entire heat map.

[0032] Real-time stress monitoring in S5 includes: S51, Acoustic Emission Signal Acquisition: Embed acoustic emission sensors in a circular array on the edge of the polishing plate. Each acoustic emission sensor is installed at a 30° angle to the substrate surface to ensure directional reception of microcrack acoustic emission signals. S52, micro-crack feature extraction: Detect the burst waveform in the acoustic emission signal with an amplitude more than 3 times that of the baseline noise, extract its rise time (the time required for the signal to rise from the baseline to the maximum value, reflecting the crack development speed), energy integral (the total energy of the signal, representing the intensity of the crack) and duration as characteristic parameters (the time the signal maintains a high amplitude state, affecting the degree of damage of the crack), and determine the crack type based on the characteristic parameters. and , it is judged as a brittle crack. If and , determined to be a ductile crack; S53, stress anomaly warning generation: Based on the extracted microcrack features, a stress anomaly warning is generated. If the number of brittle crack events per unit time is greater than or equal to 3, it is judged as high-risk stress concentration. If the energy integral of the ductile crack exceeds , it is judged as potential fatigue damage. According to the judgment result, two-level warning signals are issued, including the first-level warning triggering the reduction of the polishing disk speed and the second-level warning marking the abnormal area and feeding back to the adaptive control model.

[0033] The closed-loop control command outputs in S6 include: S61, Warning Signal Priority Analysis: Receives the first or second level warning signal from stress monitoring and extracts the stress risk level, abnormal area coordinates, and historical polishing parameters. Based on the warning level, the weight of the current polishing pressure is adjusted. When high-risk stress concentration is detected, the current polishing pressure is reduced by 60%-80%, and the abnormal area is marked as a no-polishing zone, suspending polishing in that area. When potential fatigue risk arises, the adaptive control model is made more sensitive to the surface topography of the abnormal area, raising the processing priority of that area and triggering a reduction in the local polishing intensity. S62, dynamically issuing the revised command: The revised polishing pressure, polishing disk speed, and polishing liquid supply rate are smoothed through a first-order low-pass filter to ensure smooth command changes. In addition, the timestamp of the dynamic adjustment command is checked to see if it matches the current state of the device. If the delay exceeds 200ms, the old dynamic adjustment command is discarded and recalculated to ensure real-time and accuracy. S63, polishing termination judgment: when the surface roughness reaches When there is no new warning signal for 5 minutes, when the cumulative polishing time reaches 120 minutes and the roughness decrease trend is less than 10 minutes in a row Or when the number of iterations of the adaptive control model exceeds 50 and the convergence amplitude of the feedback parameter is less than 5% for three consecutive times, the polishing completion signal is output and all actuators are turned off, where, is the target roughness threshold, expressed as: ; in, is the hardness perception weight coefficient (the fitting regression analysis of the roughness of the processing end point of sapphire samples at different hardness levels, ranging from , taking a typical value of 20), is the benchmark lower limit compensation item, is the Vickers hardness of the current sapphire material.

[0034] like Figure 2 As shown, a sapphire substrate polishing intelligent control system is used to implement the above-mentioned sapphire substrate polishing intelligent control method, including the following modules: Surface topography data acquisition module: uses a laser triangulation sensor to collect real-time surface roughness data and three-dimensional topography distribution maps of sapphire substrates; Feature parameter generation module: Based on the collected surface roughness data and 3D topography distribution map, feature parameters are extracted, including surface waviness, local depression depth, protrusion density, lattice distortion index, and residual stress distribution gradient; Dynamic parameter adjustment module: This module inputs the extracted characteristic parameters into the adaptive control model to generate corresponding dynamic adjustment instructions for polishing pressure, polishing disk speed, and polishing liquid supply rate to optimize the polishing process. Polishing liquid distribution optimization module: According to dynamic adjustment instructions, the multi-channel solenoid valve array adjusts the spray angle and flow rate of the polishing liquid to achieve uniform coverage of the polishing liquid on the substrate surface; Real-time stress monitoring module: uses acoustic emission sensors to monitor the generation of microcracks on the substrate surface during polishing and generates stress abnormality warning signals; Closed-loop control module: Feedback the stress abnormality warning signal to the adaptive control model, and iteratively correct the dynamic adjustment instructions based on the feedback information until polishing is completed.

[0035] The laser triangulation displacement sensor, solenoid valve, and acoustic emission sensor involved in this invention are all connected to the system control platform through standard industrial communication protocols to achieve functions such as data acquisition, parameter distribution, and status monitoring. The system integration control module has a protocol parsing and device compatibility middle layer and supports the following communication protocol modes: Laser triangulation displacement sensor communication protocol: uses high-speed serial communication (such as RS485 or USBCDC) and command interaction based on ModbusRTU or SCPI (standard command set) protocol to realize parameter configuration such as data reading, resolution setting, and exposure control; Solenoid valve array communication protocol: Communicates with the main control unit via standard CAN bus or Modbus RTU protocol. Each channel has a unique number and can accept PWM or opening control commands. It supports real-time feedback of the opening status and is suitable for multi-channel liquid distribution scenarios. Acoustic emission sensor communication protocol: Use Ethernet TCP / IP or serial port protocol (such as RS232) to transmit high-frequency sampling data packets. The system processes burst signals through asynchronous sampling buffer and edge feature extraction algorithm to achieve crack identification and stress warning. Some models support UDP broadcast output, and the system completes non-blocking reception through a multi-threaded monitoring mechanism.

[0036] The present invention encompasses any alternatives, modifications, equivalents, and solutions that fall within the spirit and scope of the present invention. To provide a thorough understanding of the present invention, specific details are described in detail below in connection with the preferred embodiments of the present invention, but those skilled in the art will be able to fully understand the present invention without these detailed descriptions. Furthermore, to avoid unnecessary confusion regarding the essence of the present invention, well-known methods, processes, procedures, components, and circuits have not been described in detail.

[0037] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A sapphire substrate polishing intelligent control method, characterized in that: The following steps are involved: S1, surface topography data acquisition: The surface roughness data and three-dimensional topography distribution map of the sapphire substrate are collected in real time using a laser triangulation displacement sensor; S2, feature parameter generation: Based on the surface roughness data and 3D topography distribution map, surface waviness, local depression depth, protrusion density, lattice distortion index and residual stress distribution gradient are extracted as feature parameters; S3, dynamic parameter adjustment: input the characteristic parameters into the adaptive control model to generate dynamic adjustment instructions for polishing pressure, polishing disk speed and polishing liquid supply rate; S4, polishing liquid distribution optimization: According to the dynamic adjustment instructions, the polishing liquid injection angle and flow rate are adjusted through the multi-channel solenoid valve array to achieve uniform coverage of the polishing liquid on the substrate surface; S5, real-time stress monitoring: using acoustic emission sensors to monitor the acoustic emission signals generated by micro-cracks on the substrate surface during polishing, and generating stress abnormality warning signals; S6, closed-loop control command output: Feedback the stress abnormality warning signal to the adaptive control model, iteratively correct the dynamic adjustment command until polishing is completed.

2. The intelligent control method for sapphire substrate polishing according to claim 1, characterized in that: The surface topography data acquisition in S1 includes: S11, parameter initialization: set the axial measurement accuracy of the laser triangulation displacement sensor , sampling frequency , and based on the initial thickness of the sapphire substrate As the distance measurement center, initialize the sensor measurement interval to ,in, is the maximum thickness fluctuation tolerance that may occur during the polishing process; S12, dynamic scanning path planning: Divide the substrate surface into grid coordinate points , a spiral progressive scanning path is generated according to the defect probability distribution, and its path spacing The conditions are met ,in, is the basic path spacing, is the defect weight coefficient, Indicates the historical defect probability of the coordinate point; S13, displacement data acquisition: Move the laser triangulation displacement sensor probe along the scanning path to collect height data at each grid point to form a two-dimensional height matrix ; S14, surface roughness calculation: Perform spatial frequency analysis and topography evaluation to calculate surface roughness and the root mean square gradient ; S15, 3D shape reconstruction: 3D shape surface fitting: 2D height matrix of the acquisition Perform quadratic polynomial surface fitting to construct a continuous and smooth three-dimensional topography function; S16, Data Verification and Transmission: Verify the deviation of the 3D shape reconstruction results and calculate the local area data deviation ,when The second scan of the corresponding area is triggered at this time. After verification, the surface roughness data is output and three-dimensional morphology distribution map .

3. The intelligent control method for sapphire substrate polishing according to claim 1, characterized in that: The characteristic parameter generation in S2 includes: S21, Surface waviness extraction: 3D topography distribution map Perform Gaussian low-pass filtering to preserve the spatial wavelength The ripple component of the wave and calculate the height of the ripple; S22, local depression depth extraction: setting depression determination threshold , extract satisfaction The continuous area of ​​​​is taken as the concave area, where is the average surface height, for each concave area , calculate its maximum depression depth , and extract the deepest one among all regions as the global maximum concave depth; S23, convex density extraction: setting the convexity determination threshold , extract satisfaction The continuous area is regarded as the convex area, and the unit area Count the number of raised areas , calculate the bump density ; S24, Lattice distortion index extraction: three-dimensional morphology distribution map Perform fast Fourier transform to obtain frequency domain energy distribution , extract the energy ratio of low frequency to high frequency parts as the lattice distortion index ; S25, residual stress distribution gradient extraction: based on acoustic emission signal data The spatial distribution of residual stress intensity is constructed to approximate the residual stress intensity distribution map , and calculate the global residual stress distribution gradient .

4. The intelligent control method for sapphire substrate polishing according to claim 3, characterized in that: The dynamic parameter adjustment in S3 includes: S31, adaptive model reasoning: The extracted characteristic parameters are input into the adaptive control model, and the initial adjustment parameters are calculated using the deep reinforcement learning network, including the initial value of the polishing pressure, the initial value of the polishing disk speed, and the initial value of the polishing liquid supply rate; S32, constraint correction: correct the generated initial adjustment parameters, output the corrected adjustment parameters and generate dynamic adjustment instructions.

5. The intelligent control method for sapphire substrate polishing according to claim 4, characterized in that: The adaptive model reasoning in S31 includes: S311, feature parameter normalization: performing normalization processing on the input feature parameters; S312, reinforcement learning network inference: input the normalized feature parameters into the deep reinforcement learning network and calculate the normalized values ​​of the initial adjustment parameters; S313, parameter denormalization: performing parameter denormalization processing on the calculated normalized initial adjustment parameters, and outputting the initial value of polishing pressure, the initial value of polishing disk rotation speed, and the initial value of polishing liquid supply rate.

6. The intelligent control method for sapphire substrate polishing according to claim 5, characterized in that: The constraint modification in S32 includes: S321, polishing pressure gradient constraint: Calculate the change rate between the current polishing pressure and the pressure of the previous cycle. If the change rate exceeds the set maximum allowable value, adjust the current pressure value according to the safety gradient through the limit correction strategy. If the change rate is less than the maximum allowable value, maintain the current pressure value. S322, dynamic de-weighting of the polishing disk speed: When the protrusion density exceeds a set threshold, the polishing disk speed is dynamically adjusted using a de-weighting coefficient. The de-weighting coefficient is adjusted according to the change in protrusion density. Finally, the speed value adjusted by the de-weighting coefficient is output; S323, polishing liquid supply rate saturation protection: the polishing liquid supply rate initial value Restricted to the preset safety range; S324, dynamic adjustment instruction package: the corrected polishing pressure , polishing disc speed and polishing liquid supply rate Encapsulated as structured control instructions ,in, The current timestamp.

7. The intelligent control method for sapphire substrate polishing according to claim 6, characterized in that: The polishing liquid distribution optimization in S4 includes: S41, dynamic partition mapping of polishing area: According to the three-dimensional topography distribution map, the substrate surface is divided into rectangular areas, based on each area Priority weight , calculate the required baseline flow for each area ; S42, multi-channel solenoid valve flow control: according to the base flow required for each area , by controlling the valve port opening of the multi-channel solenoid valve to adjust the polishing liquid flow in each area, the valve port opening of the solenoid valve is proportional to the flow reference value, and the opening difference is limited; S43, adaptive adjustment of spray angle: calculate the spray angle according to the gradient of the surface height of each area, and adjust the spray angle according to the surface slope. If the calculated angle exceeds the predetermined range, , then make restriction adjustments; S44, real-time coverage verification and feedback: Obtaining a thermal map of polishing fluid coverage using an infrared thermal imager , calculate the coverage uniformity index ,like , generate coverage compensation instructions and feed them back to dynamic parameter adjustment, triggering secondary parameter optimization.

8. The intelligent control method for sapphire substrate polishing according to claim 7, characterized in that: The real-time stress monitoring in S5 includes: S51, Acoustic Emission Signal Acquisition: Embed acoustic emission sensors in a circular array on the edge of the polishing plate, with each acoustic emission sensor installed at a 30° inclination angle to the substrate surface; S52, micro-crack feature extraction: detect the burst waveform in the acoustic emission signal with an amplitude more than 3 times that of the baseline noise, extract its rise time, energy integral and duration as characteristic parameters, and determine the crack type based on the characteristic parameters. and , it is judged as a brittle crack. If and , determined to be a ductile crack; S53, stress anomaly warning generation: Based on the extracted microcrack features, a stress anomaly warning is generated. If the number of brittle crack events per unit time is greater than or equal to 3, it is judged as high-risk stress concentration. If the energy integral of the ductile crack exceeds , it is judged as potential fatigue damage. According to the judgment result, two-level warning signals are issued, including the first-level warning triggering the reduction of the polishing disk speed and the second-level warning marking the abnormal area and feeding back to the adaptive control model.

9. The intelligent control method for sapphire substrate polishing according to claim 8, characterized in that: The closed-loop control instruction output in S6 includes: S61, Warning Signal Priority Analysis: Receives the first or second level warning signal from stress monitoring and extracts the stress risk level, abnormal area coordinates, and historical polishing parameters. Based on the warning level, the weight of the current polishing pressure is adjusted. When high-risk stress concentration is detected, the current polishing pressure is reduced by 60%-80%, and the abnormal area is marked as a no-polishing zone, suspending polishing in that area. When potential fatigue risk arises, the adaptive control model is made more sensitive to the surface topography of the abnormal area, raising the processing priority of that area and triggering a reduction in the local polishing intensity. S62, dynamically issuing a correction command: The corrected polishing pressure, polishing disk speed, and polishing liquid supply rate are smoothed through a first-order low-pass filter. In addition, the timestamp of the dynamic adjustment command is checked to see if it matches the current state of the device. If the delay exceeds 200ms, the old dynamic adjustment command is discarded and recalculated; S63, polishing termination judgment: when the surface roughness reaches When there is no new warning signal for 5 minutes, when the cumulative polishing time reaches 120 minutes and the roughness decreases less than 10 minutes in a row Or when the number of iterations of the adaptive control model exceeds 50 and the convergence amplitude of the feedback parameter is less than 5% for three consecutive times, the polishing completion signal is output and all actuators are turned off, where, is the target roughness threshold.

10. A sapphire substrate polishing intelligent control system, used to implement a sapphire substrate polishing intelligent control method according to any one of claims 1 to 9, characterized in that: Includes the following modules: Surface topography data acquisition module: uses a laser triangulation sensor to collect real-time surface roughness data and three-dimensional topography distribution maps of sapphire substrates; Feature parameter generation module: Based on the collected surface roughness data and 3D topography distribution map, feature parameters are extracted, including surface waviness, local depression depth, protrusion density, lattice distortion index, and residual stress distribution gradient; Dynamic parameter adjustment module: This module inputs the extracted characteristic parameters into the adaptive control model to generate corresponding dynamic adjustment instructions for polishing pressure, polishing disk speed, and polishing liquid supply rate to optimize the polishing process. Polishing liquid distribution optimization module: According to dynamic adjustment instructions, the multi-channel solenoid valve array adjusts the spray angle and flow rate of the polishing liquid to achieve uniform coverage of the polishing liquid on the substrate surface; Real-time stress monitoring module: uses acoustic emission sensors to monitor the generation of microcracks on the substrate surface during polishing and generates stress abnormality warning signals; Closed-loop control module: Feedback the stress abnormality warning signal to the adaptive control model, and iteratively correct the dynamic adjustment instructions based on the feedback information until polishing is completed.

Images