A fully automatic polishing apparatus

Abstract

This invention provides a fully automated polishing device, relating to the field of polishing equipment technology. It includes a polishing host and a pre-process AI detection unit based on the polishing host, a defect-linked adaptive polishing unit, a digital twin process optimization unit, a polishing media closed-loop management unit, and a full-station collaborative scheduling unit. The invention features a pre-process AI detection unit that integrates two-dimensional image recognition, three-dimensional morphology scanning, and multispectral assisted detection technologies before polishing. Through a deep learning model, it comprehensively detects various defects on the tile surface, such as scratches, pits, color differences, and microcracks, completing pixel-level digital grading and three-dimensional modeling of defect type, depth, range, and quantity. This generates a unique defect distribution map and grading data for each tile. This achieves digital source control through pre-detection before polishing, accurately locating all defect areas and their severity, providing precise data support for subsequent differentiated polishing operations.

Description

Technical Field

[0001] This invention relates to the field of polishing equipment technology, and in particular to a fully automatic polishing device. Background Technology

[0002] Tile polishing is a crucial step in the production of architectural ceramics. Its function is to finely process the tile surface, removing surface imperfections such as unevenness, improving surface smoothness and gloss, and enabling the tiles to meet decorative requirements. Existing tile polishing equipment typically consists of multiple polishing machines connected in series to form a production line. These machines sequentially undergo rough polishing, medium polishing, and fine polishing processes. Each polishing machine is equipped with multiple polishing heads, which grind and polish the tile surface through rotation and oscillation. Traditional polishing equipment mainly includes a frame, conveyor rollers, polishing spindle, polishing heads, lifting mechanism, and polishing fluid supply system. Tiles are sequentially transported to each polishing station via a conveyor belt, where the polishing heads continuously polish the tiles at fixed pressure, speed, and feed rate. Currently, tile polishing equipment on the market has achieved basic automated production, capable of continuous tile conveying, automatic polishing, and online cleaning. Some high-end equipment is also equipped with simple thickness detection and pressure adjustment devices, allowing for segmented setting of polishing pressure.

[0003] However, existing equipment generally adopts a "polish first, inspect later" or "uniform polishing without inspection" approach. Before polishing, surface defects in the tiles are not inspected or graded. All tiles are processed using the same polishing parameters, leading to insufficient polishing in severely defective areas and over-polishing in intact areas. This results in uneven polishing, with both under-polishing and over-polishing occurring simultaneously, severely impacting product quality consistency. Secondly, the process parameters of existing polishing equipment are mostly fixed values. Polishing pressure, speed, and grinding head angle remain constant during processing, making dynamic adjustments impossible based on the actual distribution of defects on the tile surface. This hinders precise polishing with individual parameters and processes for each tile and area, resulting in poor polishing quality stability for tiles with complex surface morphologies and diverse defect distributions. Thirdly, the process debugging of existing equipment relies heavily on manual experience. For new specifications and materials of tiles, repeated trial polishing and parameter adjustments are necessary, leading to long debugging cycles, high trial-and-error costs, and a lack of automatic compensation for time-varying factors such as abrasive wear and environmental changes. Over long-term operation, the process accuracy gradually declines. Furthermore, existing polishing slurry supply systems mostly adopt a crude open-loop supply mode, resulting in large fluctuations in parameters such as slurry concentration, temperature, and pH value. This leads to incomplete removal of abrasive particles, low utilization of the polishing slurry, and large waste discharge, increasing production costs and environmental pressure. Moreover, the fluctuations in media performance cause significant batch-to-batch variations in polishing quality. Finally, existing polishing production lines lack effective coordination between stations. When a single station malfunctions, the entire line often needs to be shut down for repairs, resulting in low production line efficiency, long downtime, and difficulty in further improving the overall equipment efficiency (OEE), thus failing to meet the demands of continuous, high-load, unmanned production. Therefore, this invention proposes a fully automated polishing device to address the problems existing in the prior art. Summary of the Invention

[0004] To address the aforementioned issues, this invention proposes a fully automated polishing device. This device is equipped with a pre-positioned AI detection unit that integrates 2D image recognition, 3D morphology scanning, and multispectral assisted detection technologies before polishing. Through a deep learning model, it comprehensively detects various defects on the tile surface, such as scratches, pits, color differences, and microcracks. It completes pixel-level digital grading and 3D modeling of defect type, depth, range, and quantity, generating a unique defect distribution map and grading data for each tile. This achieves digital source control through pre-detection before polishing, accurately locating all defect areas and their severity, providing precise data support for subsequent differentiated polishing operations.

[0005] To achieve the objectives of this invention, the following technical solution is provided: a fully automatic polishing device, comprising a polishing host and a pre-AI detection unit based on the polishing host, a defect-linked adaptive polishing unit, a digital twin process optimization unit, a polishing media closed-loop control unit, and a full-station collaborative scheduling unit. The pre-AI detection unit is located at the front end of the polishing host inlet and integrates two-dimensional image recognition and three-dimensional morphology scanning technologies to perform full-area defect detection and pixel-level digital grading of the tile surface. The defect-linked adaptive polishing unit is data-connected to the pre-AI detection unit and dynamically adjusts the polishing pressure, spindle speed, and polishing head posture of each area according to the defect grading results.

[0006] The digital twin process optimization unit is connected to the pre-AI detection unit and the defect-linked adaptive polishing unit, respectively, to construct a full-dimensional digital twin model covering substrate characteristics, abrasive wear, fluid dynamics, and equipment operating conditions, and to iteratively optimize the process parameter library; the polishing medium closed-loop management unit is connected to the polishing host to realize a closed-loop cycle of the entire life cycle of polishing fluid intelligent proportioning, real-time monitoring, multi-level purification, automatic replenishment, and state adjustment; the full-station collaborative scheduling unit is connected to each station device through a time-sensitive network to realize millisecond-level time synchronization, dynamic negotiation of task priorities, and fault self-healing flexible scheduling for all stations.

[0007] A further improvement is made in that: the pre-processing AI detection unit includes a two-dimensional imaging module, a three-dimensional topography scanning module, a multispectral auxiliary detection module, and a defect grading processing module; the defect grading processing module grades defects using a defect severity quantification model, which satisfies the following formula:

[0008]

[0009] Where S is the defect severity index; D is the defect depth; Dmax is the maximum depth threshold for this type of defect; A is the defect projection area; Atotal is the total area of ​​the tile; N is the number of this type of defect per unit area; Nref is the reference value for the number of defects per unit area; w1, w2, and w3 are the depth weight, area weight, and number weight, respectively, satisfying w1+w2+w3=1.

[0010] Further improvements include: the two-dimensional imaging module uses a linear CCD camera in conjunction with a multi-angle ring light source to acquire high-resolution images of the tile surface; the three-dimensional topography scanning module uses laser line scanning or structured light three-dimensional reconstruction technology to obtain depth information of the tile surface; the multispectral auxiliary detection module uses polarized light and hyperspectral imaging technology to identify subsurface defects and microcracks; and the defect grading processing module uses a multimodal fusion convolutional neural network to identify defect types, measure dimensions, and grade severity of the fused multimodal data.

[0011] Further improvements include: the defect-linked adaptive polishing unit comprises a zoned polishing execution module, a force control closed-loop module, and a pose adjustment module; the zoned polishing execution module divides the tile surface into multiple independently controlled polishing zones, each zone equipped with an independent polishing head group; the force control closed-loop module dynamically adjusts the polishing pressure of each zone through a pressure adaptive adjustment model, which satisfies the following formula:

[0012]

[0013] Where P(x,y) is the polishing pressure at coordinate (x,y) on the tile surface; P0 is the reference polishing pressure; k is the pressure gain coefficient; S(x,y) is the defect severity index at coordinate (x,y); f(θ) is the polishing head tilt angle correction function; and θ is the angle between the polishing head and the normal to the tile surface.

[0014] Further improvements include: the force control closed-loop module uses a high-precision piezoelectric force sensor and an electro-hydraulic servo valve to form a millisecond-level force control closed-loop system, with a force sensor sampling frequency of 1kHz and a servo valve response time of less than 5ms; the posture adjustment module includes a polishing head tilt angle adjustment mechanism and a swing trajectory adjustment mechanism, with a tilt angle adjustment range of 0-15 degrees and an angle adjustment accuracy of ±0.1 degrees; the posture adjustment module also has an edge protection function, automatically reducing polishing pressure and adjusting the tilt angle when the polishing head runs to the edge area of ​​the tile.

[0015] A further improvement is that the digital twin process optimization unit includes a virtual simulation module, a parameter optimization module, and a model update module; the virtual simulation module constructs a multi-physics coupled digital twin model; the parameter optimization module continuously updates the optimal process parameter library through an iterative optimization model of process parameters, and the iterative optimization model of process parameters adopts the following formula:

[0016]

[0017] Where Θ(n) is the process parameter vector for the nth iteration; Θ(n+1) is the process parameter vector for the (n+1)th iteration; and α is the learning rate. Θ* is the gradient vector of the objective function J with respect to the process parameters; β is the inertia coefficient; and Θ* is the historical optimal process parameter vector.

[0018] Further improvements include: the digital twin model includes a material removal model, an abrasive wear model, a thermo-mechanical coupling model, a fluid dynamics model, and a device dynamics model; the model update module compares the actual polishing detection data with the simulation prediction results after polishing a set number of tiles or at set intervals, and automatically adjusts the model parameters using a Bayesian optimization method; the digital twin process optimization unit also has a rapid adaptation function for new tile categories, using transfer learning technology to transfer existing process knowledge to new categories and quickly build an initial process parameter library.

[0019] Further improvements include: the polishing medium closed-loop control unit comprises an intelligent proportioning module, an online monitoring module, a multi-stage purification module, and an automatic replenishment module; the online monitoring module monitors the concentration, pH value, conductivity, temperature, and abrasive particle size distribution of the polishing slurry in real time; the automatic replenishment module maintains a constant polishing slurry concentration through a dynamic adjustment model, which uses the following formula:

[0020]

[0021] Where C(t) is the mass concentration of the polishing fluid at time t; C(t+Δt) is the mass concentration of the polishing fluid at time t+Δt; η is the concentration adjustment gain coefficient; Ctarget is the target concentration; γ is the concentration decay rate caused by unit volume wear; V(t) is the polishing fluid circulation flow rate at time t; Δt is the time step; and Vtotal is the total volume of the polishing fluid system.

[0022] Further improvements include: the intelligent proportioning module automatically and quantitatively proportions polishing concentrate, pure water, and functional additives according to the working conditions of each polishing process, and mixes them thoroughly to output a standardized concentration of polishing liquid suitable for each process; the multi-stage purification module adopts a three-stage purification process of centrifugal separation, magnetic separation filtration, and precision filter element to remove abrasive particles larger than 0.5μm from the polishing liquid; the polishing medium closed-loop management unit is also equipped with polishing liquid status assessment and lifespan prediction functions, establishing a polishing liquid health assessment model based on indicators such as concentration changes, pH value changes, abrasive particle size distribution, and abrasive content; and the polishing liquid temperature is controlled within the range of 22±1℃ through a heating / cooling heat exchange system.

[0023] Further improvements include: the full-station collaborative scheduling unit comprises a TSN network communication module, a status monitoring module, a fault diagnosis module, and a dynamic scheduling module; the TSN network communication module uses a time-sensitive network protocol to achieve millisecond-level time synchronization and deterministic data transmission for each station; the status monitoring module relies on the time-sensitive network to collect real-time operating parameters, operating conditions, and execution data of each station on the polishing production line, achieving full-domain, high-frequency, deterministic data collection and status aggregation of the entire line's equipment operating status; the fault diagnosis module, based on a preset equipment health benchmark model, connects to the real-time operating condition data collected by the status monitoring module, and achieves fault identification, anomaly assessment, and potential fault prediction for the entire line's equipment through data deviation comparison and intelligent analysis; the dynamic scheduling module achieves self-healing flexible scheduling through a task priority dynamic scheduling model, which uses the following formula:

[0024]

[0025] Where Pri(i,t) is the task priority of workstation i at time t; D(i,t) is the number of unprocessed tiles at workstation i at time t; Dtotal is the total number of unprocessed tiles in the entire line; T(i,t) is the expected fault recovery time of workstation i; Tmax is the maximum allowable downtime; F(i,t) is the fault status factor of workstation i; λ1, λ2, and λ3 are weighting coefficients, satisfying λ1+λ2+λ3=1.

[0026] The beneficial effects of this invention are as follows:

[0027] 1. This invention features a pre-processed AI detection unit that integrates 2D image recognition, 3D morphology scanning, and multispectral assisted detection technologies before polishing. Through a deep learning model, it comprehensively detects various defects on the tile surface, such as scratches, pits, color differences, and microcracks. It completes pixel-level digital grading and 3D modeling of defect type, depth, range, and quantity, generating a unique defect distribution map and grading data for each tile. This achieves digital source control through pre-detection before polishing, accurately locating all defect areas and their severity, providing precise data support for subsequent differentiated polishing operations. It transforms the polishing process from blind polishing to precise polishing, significantly improving the controllability and predictability of polishing quality. The multimodal fusion detection technology not only identifies visible surface defects but also detects subsurface defects and microcracks through polarized light and hyperspectral imaging. Its detection accuracy and comprehensiveness surpass traditional single-vision inspection solutions, providing a reliable quality assurance foundation for high-quality tile production.

[0028] 2. This invention features a defect-linked adaptive polishing unit. Based on defect classification data generated by the pre-mounted AI detection unit, it automatically maps the severity of defects to specific polishing process parameters, dynamically adjusting the polishing pressure, spindle speed, polishing head tilt angle, and oscillation trajectory of each area. A high-precision force control closed-loop algorithm enables millisecond-level pressure fine-tuning, achieving precise polishing with one parameter per tile and one process per area. The system can automatically match appropriate polishing posture and force according to defect depth, applying greater polishing pressure and more polishing rotations to severely defective areas, while using a light polishing or skipping strategy for intact areas. This eliminates over-polishing, under-polishing, and edge collapse problems caused by parameter fixation. The independently controlled polishing head group design ensures that polishing parameters in each area do not interfere with each other, enabling simultaneous processing of defects of varying severity on the same tile surface. This improves the polishing consistency and surface quality of tile products with complex surface morphology and diverse defect distributions, significantly improving the flatness and gloss uniformity of the tile surface after polishing.

[0029] 3. This invention sets up a digital twin process optimization unit to construct a full-dimensional digital twin model covering the characteristics of the ceramic tile substrate, abrasive wear behavior, fluid dynamics effects, and equipment operating conditions. This model is linked in real-time with the physical equipment, automatically simulating and iterating in virtual space based on actual processing data and test results, continuously updating the optimal process parameter library through intelligent optimization algorithms. This endows the polishing equipment with autonomous learning and continuous evolution capabilities, significantly shortening the process debugging cycle for new ceramic tile categories and reducing trial-and-error costs and material waste. The digital twin model can accurately simulate physical phenomena such as material removal rate, surface temperature distribution, and abrasive wear state during the polishing process, providing a scientific basis for process parameter optimization rather than relying solely on experience and guesswork. Furthermore, it can continuously compensate for the impact of time-varying factors such as abrasive wear and changes in environmental temperature and humidity based on real-time operating data, ensuring long-term process stability and making the equipment more precise with use, thus achieving the accumulation and inheritance of process knowledge.

[0030] 4. This invention features a closed-loop management unit for the polishing medium, integrating intelligent proportioning, real-time monitoring, multi-stage purification, automatic replenishment, and state adjustment functions for the polishing slurry. It dynamically maintains the concentration, pH value, temperature, and cleanliness of the polishing slurry based on the processing steps and real-time operating conditions, achieving closed-loop recycling of the polishing slurry throughout its entire lifecycle. This significantly improves the utilization rate of the polishing slurry, approaching zero waste discharge and reducing the cost of polishing slurry consumption and environmental treatment. By precisely controlling the temperature fluctuation of the polishing slurry within a set range and efficiently removing abrasive particles larger than 0.5μm through a multi-stage purification system, it ensures a highly consistent grinding environment during batch polishing, eliminating quality instability issues. Simultaneously, the online monitoring module tracks various performance indicators of the polishing slurry in real time. When the abrasive particle size distribution deviates from the normal range or the chemical composition changes, the system can promptly issue warnings and automatically adjust the replenishment strategy, ensuring the polishing slurry is always in optimal working condition, providing a medium-level guarantee for stable polishing quality.

[0031] 5. This invention sets up a full-station collaborative scheduling unit, breaking down traditional station barriers. Through time-sensitive networking protocol, it performs millisecond-level time synchronization and dynamic negotiation of task priorities for all stations, including feeding, rough polishing, intermediate polishing, fine polishing, cleaning, and inspection. The system monitors the operating status of each station 24 / 7. In the event of a single station failure, it can automatically adjust the upstream and downstream cycle times, initiate redundancy compensation, or bypass conveying without requiring a complete line shutdown, significantly improving the stability of the entire line operation and significantly reducing single-line downtime. This achieves self-healing flexible production under fault conditions. Furthermore, the deterministic communication of the TSN network ensures precise collaboration between stations, enabling multiple stations to cooperate accurately in the time dimension. Fault prediction and self-healing scheduling functions allow the equipment to provide early warnings and take preventive measures before a fault occurs, and quickly reconstruct the production process after a fault occurs, minimizing downtime losses and greatly improving the overall equipment efficiency and the flexible adaptability of the production line. Attached Figure Description

[0032] Figure 1 This is a diagram illustrating the composition of the present invention. Detailed Implementation

[0033] To enhance understanding of the present invention, the present invention will be further described in detail below with reference to embodiments. These embodiments are only used to explain the present invention and do not constitute a limitation on the scope of protection of the present invention.

[0034] Example 1

[0035] according to Figure 1 As shown in the figure, this embodiment proposes a fully automated polishing device, including a polishing host and a pre-processing AI detection unit. The pre-processing AI detection unit is located at the front end of the polishing host inlet. Before entering the polishing process, the tile first passes through this detection unit to complete a comprehensive surface defect detection and classification.

[0036] The front-end AI detection unit includes a 2D imaging module, a 3D topography scanning module, a multispectral auxiliary detection module, and a defect grading and processing module. The 2D imaging module uses two 4K resolution linear CCD cameras to acquire images of the tile surface from directly above and at a 45-degree angle, respectively. Combined with a multi-angle ring LED light source and a coaxial light source, it ensures clear imaging contrast for different types of defects. The 3D topography scanning module uses a laser line-scanning 3D reconstruction system equipped with two 405nm blue laser line-scanning sensors. It scans along the width of the tile to acquire 3D topography data, achieving a vertical resolution of 0.01mm and a depth measurement accuracy of ±1μm. The multispectral auxiliary detection module includes a polarization imaging submodule and a hyperspectral imaging submodule. The polarization imaging submodule acquires images with different polarization directions through polarizer rotation, used to identify subsurface stress and microcracks. The hyperspectral imaging submodule covers the 400-1000nm wavelength band with a spectral resolution of 5nm, used to detect color difference defects and material composition inhomogeneity.

[0037] The defect grading and processing module uses a deep learning model to fuse and analyze multimodal data. Specifically, it first uses an image registration algorithm to precisely align 2D images, 3D depth maps, polarization images, and hyperspectral images in space, forming a unified multimodal data cube. Then, a multimodal fusion convolutional neural network (MMF-CNN) is used to extract features and identify defects in the data cube. The network structure includes multiple parallel feature extraction branches, each processing data from different modalities. After fusion at a high-level feature layer, the defect type, location, and confidence level are output. The identified defect types include more than ten types such as scratches, dents, bumps, color differences, pinholes, edge chipping, and microcracks.

[0038] For each identified defect, the system further performs pixel-level size measurement and severity quantification. The defect severity quantification model satisfies the following formula:

[0039]

[0040] Wherein, S is the defect severity index, ranging from 0 to 1; D is the defect depth, in μm, measured from 3D topographic scanning data; Dmax is the maximum depth threshold for this type of defect, in μm, preset according to the tile type and quality grade; A is the defect projected area, in mm², calculated from 2D image segmentation; Atotal is the total area of ​​the tile, in mm²; N is the number of this type of defect per unit area, in units / dm²; Nref is the reference value for the number of defects per unit area, in units / dm², set according to product quality standards; w1, w2, and w3 are the depth weight, area weight, and quantity weight, respectively, their values ​​determined according to the defect type and product requirements, satisfying w1+w2+w3=1. For example, for scratch defects, w1 is 0.5, w2 is 0.3, and w3 is 0.2; for pinhole defects, w1 is 0.3, w2 is 0.2, and w3 is 0.5.

[0041] Based on the defect severity index S, the system classifies defects into five levels: S < 0.2 is minor, 0.2 ≤ S < 0.4 is moderate, 0.4 ≤ S < 0.6 is medium, 0.6 ≤ S < 0.8 is severe, and S ≥ 0.8 is extremely severe. Each level corresponds to a different polishing process strategy. Minor defects can be removed using light polishing parameters, moderate and medium defects use standard polishing parameters, while severe and extremely severe defects require increased polishing pressure, extended polishing time, or more polishing passes.

[0042] After inspection, the system generates a complete defect distribution map and grading report for each tile, including tile number, inspection time, defect type statistics, heat maps of defect distribution at each level, and 3D topographic cloud maps. This data is transmitted in real time via industrial Ethernet to the subsequent adaptive polishing unit and digital twin optimization unit, serving as the basis for polishing parameter setting and process optimization.

[0043] In this embodiment, the detection speed of the front-end AI detection unit is matched with the tile conveying speed, with a detection cycle of no more than 5 seconds per tile, and the detection accuracy reaches the pixel level. The smallest identifiable defect size is 5μm×5μm, the defect recognition accuracy is no less than 99.5%, the false detection rate is less than 0.3%, and the missed detection rate is less than 0.2%, which fully meets the online detection requirements of industrial production lines.

[0044] Example 2

[0045] according to Figure 1As shown, this embodiment proposes a fully automatic polishing device. Based on Embodiment 1, the defect-linked adaptive polishing unit includes a zoned polishing execution module, a force control closed-loop module, and a posture adjustment module. The zoned polishing execution module divides the tile surface into 8 independently controlled polishing zones along the width direction. Each zone is approximately 100mm wide (taking an 800mm wide tile as an example). Each zone is equipped with a set of independent polishing heads, and each set contains 3-5 polishing heads. Each polishing head set can be independently raised and lowered, and its pressure and speed can be independently adjusted.

[0046] The force control closed-loop module employs a high-precision force sensor and an electro-hydraulic servo valve to form a millisecond-level force control closed-loop system. Each polishing head is equipped with a piezoelectric force sensor with a range of 0-500N, a sampling frequency of 1kHz, and a force measurement accuracy of ±0.5N. The servo valve has a response time of less than 5ms and can adjust the hydraulic cylinder pressure in real time based on the feedback signal from the force sensor, thereby achieving millisecond-level precise control of the polishing pressure.

[0047] The posture adjustment module includes a polishing head tilt angle adjustment mechanism and a swing trajectory adjustment mechanism. The tilt angle adjustment mechanism uses a servo motor-driven worm gear mechanism, which allows the polishing head to be continuously tilted within the range of 0-15 degrees, with an angle adjustment accuracy of ±0.1 degrees. The swing trajectory adjustment mechanism can control the swing amplitude, swing frequency, and swing trajectory shape of the polishing head, supporting multiple modes such as linear swing, circular swing, and sinusoidal swing.

[0048] The dynamic adjustment of polishing parameters is based on defect severity distribution data. The system first maps the defect distribution map output by the pre-AI detection unit onto the polishing coordinate system, and then calculates the corresponding polishing process parameters according to the defect severity in each region. The adaptive adjustment model of polishing pressure satisfies the following formula:

[0049]

[0050] Wherein, P(x,y) is the polishing pressure at coordinate (x,y) on the tile surface, in N; P0 is the reference polishing pressure, in N, set according to the tile material and polishing process, 300-400N for rough polishing and 150-250N for fine polishing; k is the pressure gain coefficient, dimensionless, ranging from 50 to 200, determined according to the polishing head type and abrasive particle size; S(x,y) is the defect severity index at coordinate (x,y), ranging from 0 to 1; f(θ) is the polishing head tilt angle correction function, where θ is the angle between the polishing head and the normal to the tile surface, in degrees, f(θ)=cos(θ)+α·sin(2θ), where α is the edge effect compensation coefficient, ranging from 0.1 to 0.3, used to compensate for changes in effective contact area and differences in edge polishing efficiency when the polishing head is tilted.

[0051] In addition to pressure, the polishing spindle speed is also dynamically adjusted according to the severity of the defects. The speed adjustment formula is:

[0052]

[0053] Wherein, n(x,y) is the spindle speed at coordinate (x,y), in rpm; n0 is the reference speed, in rpm, which is 1500-2000 rpm for rough polishing and 2500-3000 rpm for fine polishing; Δn is the speed increment, in rpm, with a value range of 300-800 rpm.

[0054] During the polishing process, the system matches the defect distribution map with the current polishing area based on the real-time position of the tile, dynamically adjusting the pressure, speed, and tilt angle of each polishing head group. When the polishing head passes through a severely defective area, the system automatically increases the pressure and speed, increases the oscillation amplitude, and extends the effective polishing time for that area; when the polishing head passes through an area without defects or with minor defects, the system automatically reduces the pressure and speed to minimize unnecessary material removal and avoid over-polishing.

[0055] Furthermore, to address the issue of tile edge sagging, the posture adjustment module also features edge protection. When the polishing head reaches the tile edge area (within 0-30mm of the edge), the system automatically reduces the polishing pressure and adjusts the polishing head's tilt angle, shifting the pressure center of the polishing head inwards towards the tile, effectively preventing edge sagging. The pressure correction coefficient for the edge area decreases linearly with decreasing distance from the edge, dropping to 60%-70% of the normal value at the edge.

[0056] In this embodiment, the pressure adjustment response time of the adaptive polishing unit is less than 10ms, the pressure control accuracy is ±1%, the speed adjustment range is 500-3500rpm, and the tilt angle adjustment range is 0-15 degrees. It can achieve differentiated and precise polishing of different areas of the tile surface. The surface gloss uniformity after polishing can reach more than 95%, which is far better than the approximately 85% of traditional fixed parameter polishing.

[0057] Example 3

[0058] according to Figure 1 As shown, this embodiment proposes a fully automated polishing device. Based on the above embodiment, the digital twin process optimization unit includes a virtual simulation module, a parameter optimization module, and a model update module. The virtual simulation module constructs a full-dimensional digital twin model covering the characteristics of the ceramic tile substrate, abrasive wear behavior, fluid dynamics effects, and equipment operating conditions, and links with the physical polishing device in real time.

[0059] The digital twin model adopts a multi-physics coupling modeling method and includes the following sub-models: (1) Material removal model: Based on the Preston equation and the improved material removal function, it simulates the material removal rate distribution on the surface of the tile under different process parameters, and considers the influence of factors such as abrasive particle size, polishing pressure, relative speed, and polishing liquid concentration; (2) Abrasive wear model: Based on abrasive wear theory, it simulates the wear, breakage and shedding process of abrasive particles during polishing and predicts the change law of the wear of the grinding disc with polishing time; (3) Thermo-mechanical coupling model: It simulates the frictional heat generation, heat conduction and thermal deformation phenomena during polishing and calculates the temperature distribution and thermal deformation of the tile surface; (4) Fluid dynamics model: It simulates the flow state of the polishing liquid in the polishing area, including the flow velocity distribution, pressure distribution and abrasive particle concentration distribution; (5) Equipment dynamics model: It simulates the dynamic response characteristics of the polishing spindle, lifting mechanism and transmission system.

[0060] The parameter optimization module employs an intelligent optimization algorithm to automatically search for the optimal combination of process parameters in virtual space. The optimization objectives include maximizing surface gloss, optimizing material removal rate, minimizing grinding disc wear, and minimizing energy consumption, among others. A multi-objective optimization algorithm is used to solve these objectives. The iterative optimization model for process parameters satisfies the following formula:

[0061]

[0062] Where Θ(n) is the process parameter vector for the nth iteration, including parameters such as polishing pressure, spindle speed, feed rate, polishing slurry concentration, and polishing slurry flow rate; Θ(n+1) is the process parameter vector for the (n+1)th iteration; α is the learning rate, with a value range of 0.01-0.1, which controls the step size of each iteration; Θ* is the gradient vector of the objective function J with respect to the process parameters, which is obtained by digital twin simulation. The objective function J comprehensively considers factors such as surface quality, production efficiency and cost. β is the inertia coefficient, which ranges from 0.1 to 0.5. It is used to introduce the guiding role of the historical optimal parameters, accelerate convergence and prevent getting trapped in local optima. Θ* is the historical optimal process parameter vector, which is the parameter combination with the best objective function value so far.

[0063] The iterative optimization process is as follows: First, starting with the initial process parameter Θ(0), the objective function value J(Θ(0)) and gradient are calculated through digital twin simulation. Then, update the process parameters according to the above formula to obtain Θ(1), and perform simulation calculation again; repeat this process until the change in the objective function value is less than the set threshold or the maximum number of iterations is reached, and finally obtain the optimal process parameter Θ_opt.

[0064] The model update module performs online calibration and updates to the twin model based on actual processing data and inspection results. Every certain number of tiles polished (e.g., every 100 tiles) or at regular intervals (e.g., every hour), the system compares the actual polished inspection data with the simulation prediction results to calculate the model prediction error. When the error exceeds a set threshold, the system automatically adjusts model parameters, such as the material removal coefficient and wear coefficient, to ensure that the model prediction results are consistent with the actual situation. The model update employs a Bayesian optimization method, which can efficiently search for optimal model parameters in a high-dimensional parameter space.

[0065] In addition, the digital twin process optimization unit also has the function of rapid adaptation to new tile categories. When introducing new materials and specifications of tile products, the system first obtains basic data through a small number of trial polishing experiments, and then uses transfer learning technology to transfer the process knowledge of existing tile categories to the new categories, quickly builds an initial process parameter library, and then performs fine optimization through digital twin simulation, shortening the process debugging cycle of new categories from the traditional several hours to minutes.

[0066] In this embodiment, the material removal prediction accuracy of the digital twin model can reach ±5%, the surface gloss prediction accuracy can reach ±2%, the process parameter optimization iterations can usually converge after 20-50 iterations, and the process debugging time for new types of ceramic tiles does not exceed 30 minutes, which is far better than the 4-8 hours of traditional manual debugging.

[0067] Example 4

[0068] according to Figure 1 As shown, this embodiment proposes a fully automated polishing device. Based on the above embodiments, the polishing medium closed-loop management unit includes an intelligent proportioning module, an online monitoring module, a multi-stage purification module, and an automatic replenishment module. The entire system realizes a closed-loop cycle of the polishing fluid throughout its entire lifecycle, from preparation, use, purification to recycling.

[0069] The intelligent mixing module includes a stock solution storage tank, a pure water storage tank, an additive storage tank, and a mixing tank. The stock solution storage tank stores the polishing slurry concentrate, the pure water storage tank stores deionized water, and the additive storage tank stores chemical additives such as pH adjusters, dispersants, and rust inhibitors. The mixing tank is equipped with a stirring device and a weighing sensor. Based on the requirements of different polishing processes, the system automatically extracts the appropriate amount of stock solution, water, and additives from each storage tank according to a preset formula, and thoroughly mixes them in the mixing tank to prepare a working polishing slurry that meets the requirements.

[0070] The online monitoring module monitors various performance indicators of the polishing slurry in real time, including: (1) Concentration monitoring: The mass concentration of abrasive particles in the polishing slurry is measured by a refractometer or densitometer, with a measurement range of 0-50% and an accuracy of ±0.1%; (2) pH monitoring: The pH value of the polishing slurry is measured by a glass electrode pH meter, with a measurement range of 0-14 and an accuracy of ±0.01pH; (3) Conductivity monitoring: The conductivity of the polishing slurry is measured by a conductivity sensor to reflect the ion concentration in the solution; (4) Temperature monitoring: The temperature of the polishing slurry is measured by a platinum resistance temperature sensor with an accuracy of ±0.1℃; (5) Abrasive particle size distribution monitoring: The particle size distribution of abrasive particles is measured by an online laser particle size analyzer to monitor the wear and breakage of abrasive particles; (6) Turbidity monitoring: The content of suspended abrasive debris in the polishing slurry is monitored by a turbidity sensor.

[0071] The multi-stage purification module employs a three-stage purification process to efficiently remove grinding debris and impurities from the polishing slurry: The first stage is centrifugal separation, which uses the centrifugal force generated by high-speed rotation to separate denser grinding debris particles from the polishing slurry, removing particles larger than 20μm in diameter; the second stage is magnetic separation filtration, which uses a high-gradient magnetic field to adsorb and separate ferromagnetic grinding debris, making it particularly suitable for polishing slurries containing metallic impurities; the third stage is precision filter filtration, using a high-precision filter element with a pore size of 0.5μm to remove the remaining fine grinding debris particles. After three stages of purification, the removal rate of grinding debris larger than 0.5μm in the polishing slurry can reach over 99%, significantly improving the cleanliness of the polishing slurry.

[0072] The automatic replenishment module automatically replenishes the stock solution, water, and various additives based on online monitoring data and set target parameters to maintain constant polishing slurry performance. The dynamic adjustment model for polishing slurry concentration satisfies the following formula:

[0073]

[0074] Where C(t) is the mass concentration of the polishing fluid at time t, in %; C(t+Δt) is the mass concentration of the polishing fluid at time t+Δt, in %; η is the concentration adjustment gain coefficient, dimensionless, ranging from 0.1 to 0.5, which determines the speed of concentration adjustment; a larger value results in faster adjustment but also a greater risk of overshoot; Ctarget is the target concentration, in %; it is set according to the polishing process, taking 15-25% for the rough polishing stage and 5-15% for the fine polishing stage; γ is the concentration decay rate caused by unit volume wear, in %·L. -1 , reflects the concentration decrease per liter of polishing fluid per unit area of ​​tile processed, and is related to the abrasive wear rate and the amount of tile removed; V(t) is the polishing fluid circulation flow rate at time t, in L / min; Δt is the time step, in min; Vtotal is the total volume of the polishing fluid system, in L.

[0075] In addition to concentration, pH and temperature also employ a similar closed-loop control strategy. pH is adjusted automatically by adding acid or alkali solutions, with a control accuracy of ±0.05 pH. Temperature is regulated via a heating / cooling heat exchange system, stabilizing the polishing slurry temperature within the range of 22±1℃ to ensure a consistent grinding environment temperature during the polishing process.

[0076] The polishing slurry circulation system is also equipped with polishing slurry condition assessment and lifespan prediction functions. The system comprehensively considers multiple indicators such as concentration changes, pH changes, abrasive particle size distribution, and abrasive debris content to establish a polishing slurry health assessment model and evaluate the usage status of the polishing slurry in real time. When the health level falls below a set threshold, the system automatically issues a replacement warning and calculates the optimal partial replacement amount and timing to maintain best polishing performance with minimal polishing slurry consumption.

[0077] In this embodiment, the polishing slurry concentration is controlled with an accuracy of ±0.3%, the temperature is controlled with an accuracy of ±0.5℃, and the pH value is controlled with an accuracy of ±0.05pH. The polishing slurry utilization rate can reach over 95%, and the waste liquid discharge is reduced by over 90%, significantly reducing the polishing slurry consumption cost and environmental treatment cost.

[0078] Example 5

[0079] according to Figure 1 As shown, this embodiment proposes a fully automated polishing device. Based on the above embodiment, the full-station collaborative scheduling unit includes a TSN network communication module, a status monitoring module, a fault diagnosis module, and a dynamic scheduling module. This unit realizes collaborative control and self-healing scheduling of all stations, including feeding, rough polishing, intermediate polishing, fine polishing, cleaning, and inspection.

[0080] The TSN network communication module adopts the IEEE 802.1 Time-Sensitive Networking Protocol to construct a unified industrial communication network across the entire line. All workstation devices, sensors, and actuators connect to the network through TSN switches, achieving millisecond-level time synchronization and deterministic data transmission. The TSN network employs a Time-Aware Shaper (TAS) mechanism to allocate appropriate time slots to data of different priorities, ensuring that the transmission delay of control commands and status data is bounded and deterministic, with an end-to-end transmission delay not exceeding 1ms and a time synchronization accuracy better than 1μs. Compared to traditional industrial Ethernet, the TSN network provides higher communication determinism and reliability, providing a communication foundation for precise multi-workstation collaboration.

[0081] The status monitoring module collects the operating status and process parameters of each workstation in real time, including: the speed, current, vibration, and temperature of each polishing spindle; the pressure, position, and temperature of each polishing head; the speed, position, and tension of the conveying system; the flow rate, pressure, temperature, and concentration of the polishing slurry system; and the operating status and fault codes of each motor and driver. The status data sampling frequency is 100Hz, and the data sampling frequency for critical control loops can reach 1kHz. All data is transmitted to the dispatch center in real time via the TSN network.

[0082] The fault diagnosis module performs fault prediction and diagnosis based on the equipment health model. The system establishes a health status model for each key component, and by analyzing the deviation between real-time monitoring data and the normal state model, it determines the equipment's health status and predicts potential faults. The fault diagnosis employs deep learning methods, using historical fault data to train the fault diagnosis model. It can identify various common fault types such as bearing wear, motor abnormalities, hydraulic system leaks, and sensor failures, with a fault identification accuracy of no less than 95% and fault prediction lead time ranging from several hours to several days.

[0083] The dynamic scheduling module automatically adjusts production cycle time and logistics routes in the event of a single workstation failure, achieving self-healing flexible production. The task priority dynamic scheduling model satisfies the following formula:

[0084]

[0085] Where Pri(i,t) is the task priority of workstation i at time t, with a higher value indicating higher priority; D(i,t) is the number of tiles to be processed at workstation i at time t, in units of tiles; Dtotal is the total number of tiles to be processed on the entire line, in units of tiles; T(i,t) is the estimated fault recovery time of workstation i, in units of minutes, which is 0 during normal operation; Tmax is the maximum allowable downtime, in units of minutes, set according to the production plan; F(i,t) is the fault status factor of workstation i, which is 0 during normal operation, 1 during fault operation, and 0.5 during warning operation; λ1, λ2, and λ3 are weighting coefficients, ranging from 0 to 1, satisfying λ1+λ2+λ3=1, reflecting the degree of influence of the amount of tiles to be processed, the impact of faults, and the fault status on priority, respectively, with typical values ​​of λ1=0.3, λ2=0.4, and λ3=0.3.

[0086] The dynamic scheduling strategy includes the following modes: (1) Cycle adjustment mode: When a workstation experiences a minor fault or performance degradation, the system automatically reduces the conveying speed upstream of the workstation and increases the processing speed downstream. By adjusting the cycle, the production line continues to run in a reduced-speed state, avoiding a complete shutdown. (2) Redundancy compensation mode: The production line is equipped with redundant equipment at key workstations. When the main equipment fails, the system automatically switches to the backup equipment to continue production. The switching time does not exceed 30 seconds. (3) Bypass conveying mode: For non-critical process workstations, when the workstation fails, the system automatically starts the bypass conveying channel to directly convey the tiles to the next workstation, skipping the faulty workstation and ensuring continuous operation of the production line. The workstation will be reprocessed after the fault is cleared. (4) Task redistribution mode: When a workstation fails and cannot be quickly recovered, the system redistributes the task of the workstation to other workstations with similar functions. By adjusting the process parameters, other workstations can take on some of the functions of the faulty workstation.

[0087] In addition, the full-station collaborative scheduling unit also has a production planning optimization function. Based on factors such as order demand, equipment status, and raw material availability, the system automatically optimizes the production schedule, rationally arranges the production sequence and switchover timing of different specifications of tiles, and maximizes the overall output of the production line and equipment utilization.

[0088] In this embodiment, the overall stability of the production line can reach over 99.92%, the average annual downtime of a single line is reduced by over 65%, and the overall equipment efficiency (OEE) is increased from around 70% to over 90%, enabling continuous high-load unmanned production for 24 hours.

[0089] Comparative Example

[0090] The comparative example uses a traditional tile polishing production line, mainly equipped with: 8 polishing machines connected in series, each equipped with 6 fixed polishing heads; polishing pressure is pneumatically regulated, and can only be uniformly set at the level of each polishing machine, without zone adjustment; polishing speed is fixed at 1800 rpm and cannot be adjusted; polishing head angle is fixed and does not have tilt adjustment function; polishing liquid is supplied in an open-loop manner, manually prepared and continuously supplied, with waste liquid directly discharged without a circulation and purification system; each station is independently controlled, without collaborative scheduling function, and a single station failure will shut down the entire line; there is no pre-detection system, and all tiles are polished using the same process parameters.

[0091] The comparative experiment used 1000 pieces each of 800mm×800mm fully polished glazed ceramic tiles from the same batch. These tiles were polished on the fully automatic polishing equipment of this embodiment and the traditional polishing equipment of the comparative example, respectively. After polishing, the product quality, production efficiency, material costs, and operational stability were tested and statistically analyzed. The comparison results are shown in the table below:

[0092] Table 1. Performance comparison between the present invention and traditional polishing equipment.

[0093] Comparison indicators Comparative example (traditional equipment) Embodiments of the present invention Improvement / Enhancement Scope Surface gloss (GU) 85±6 92±2 Gloss increased by 8.2%, and uniformity increased by 66.7%. Surface smoothness (μm) 25±8 12±3 Smoothness improved by 52%, and fluctuation decreased by 62.5%. Missed shipment rate (%) 3.5 0.2 Reduced by 94.3% Over-throw rate (%) 5.2 0.5 Reduced by 90.4% Corner collapse rate (%) 4.8 0.3 Reduced by 93.8% Grade 1 product rate (%) 82 96 An increase of 17.1% New product debugging cycle (h) 6 0.5 Shortened by 91.7% Polishing fluid consumption (L / m²) 0.8 0.08 Reduced by 90% Waste liquid discharge (L / m²) 0.75 0.05 Reduced by 93.3% Overall line operational stability (%) 96.5 99.92 Increased by 3.54% Average annual downtime (h) 320 112 Reduce by 65% Overall Equipment Effectiveness (OEE) (%) 70 90 An increase of 28.6% Energy consumption per unit product (kWh / m²) 1.2 0.95 Reduced by 20.8%

[0094] The comparative data shows that the fully automatic tile polishing equipment of this invention is superior to traditional polishing equipment in all aspects, including surface quality, production efficiency, consumable costs, and operational stability. In particular, the improvement in quality indicators such as surface gloss uniformity, missed polishing rate, over-polishing rate, and edge collapse rate all exceed 90%, while polishing liquid consumption and waste liquid discharge are reduced by more than 90%, resulting in a significant improvement in the overall stability of the production line and the overall efficiency of the equipment.

[0095] This fully automated polishing equipment is equipped with a pre-positioned AI detection unit that integrates 2D image recognition, 3D morphology scanning, and multispectral assisted detection technologies before polishing. Through a deep learning model, it comprehensively detects various defects on the tile surface, such as scratches, pits, color differences, and microcracks. It completes pixel-level digital grading and 3D modeling of defect type, depth, range, and quantity, generating a unique defect distribution map and grading data for each tile. This achieves digital source control through pre-detection before polishing, accurately locating all defect areas and their severity, providing precise data support for subsequent differentiated polishing operations. It transforms the polishing process from blind polishing to precise polishing, significantly improving the controllability and predictability of polishing quality. The multimodal fusion detection technology can not only identify visible surface defects but also detect subsurface defects and microcracks through polarized light and hyperspectral imaging. Its detection accuracy and comprehensiveness are superior to traditional single-vision inspection solutions, providing a reliable quality assurance foundation for high-quality tile production. This invention features a defect-linked adaptive polishing unit. Based on defect classification data generated by a pre-mounted AI detection unit, it automatically maps defect severity to specific polishing process parameters, dynamically adjusting polishing pressure, spindle speed, polishing head tilt angle, and oscillation trajectory for each area. A high-precision force control closed-loop algorithm enables millisecond-level pressure fine-tuning, achieving precise polishing with one parameter per tile and one process per area. The system automatically matches appropriate polishing posture and force according to defect depth, applying greater polishing pressure and more polishing rotations to severely defective areas, while using light polishing or skipping strategies for intact areas. This eliminates over-polishing, under-polishing, and edge collapse problems caused by parameter fixation. The independently controlled polishing head group design ensures that polishing parameters in each area do not interfere with each other, enabling simultaneous processing of defects of varying severity on the same tile surface. This improves the polishing consistency and surface quality of tile products with complex surface morphology and diverse defect distributions, significantly improving the flatness and gloss uniformity of the tile surface after polishing. This invention establishes a digital twin process optimization unit, constructing a full-dimensional digital twin model encompassing the characteristics of the ceramic tile substrate, abrasive wear behavior, fluid dynamics effects, and equipment operating conditions. This model is linked in real-time with the physical equipment, automatically simulating and iterating in virtual space based on actual processing data and test results, continuously updating the optimal process parameter library through intelligent optimization algorithms. This endows the polishing equipment with autonomous learning and continuous evolution capabilities, significantly shortening the process debugging cycle for new ceramic tile categories and reducing trial-and-error costs and material waste. The digital twin model can accurately simulate physical phenomena such as material removal rate, surface temperature distribution, and abrasive wear state during the polishing process, providing a scientific basis for process parameter optimization rather than relying solely on experience and guesswork. Furthermore, it can continuously compensate for the impact of time-varying factors such as abrasive wear and changes in environmental temperature and humidity based on real-time operating data, ensuring long-term process stability and making the equipment increasingly precise with use, thus achieving the accumulation and inheritance of process knowledge.This invention establishes a closed-loop management unit for the polishing medium, integrating intelligent proportioning, real-time monitoring, multi-stage purification, automatic replenishment, and state adjustment functions for the polishing slurry. It dynamically maintains constant concentration, pH value, temperature, and cleanliness of the polishing slurry based on processing steps and real-time operating conditions, achieving closed-loop recycling of the polishing slurry throughout its entire lifecycle. This significantly improves the utilization rate of the polishing slurry, approaching zero waste discharge and reducing the cost of polishing slurry consumption and environmental treatment. By precisely controlling the temperature fluctuation of the polishing slurry within a set range and efficiently removing abrasive particles larger than 0.5μm through a multi-stage purification system, it ensures a highly consistent grinding environment during batch polishing, eliminating quality instability issues. Simultaneously, the online monitoring module tracks various performance indicators of the polishing slurry in real time. When the abrasive particle size distribution deviates from the normal range or the chemical composition changes, the system can promptly issue warnings and automatically adjust the replenishment strategy, ensuring that the polishing slurry is always in optimal working condition, providing a medium-level guarantee for stable polishing quality. This invention establishes a full-station collaborative scheduling unit, breaking down traditional station barriers. Through Time-Sensitive Network Protocol (TSN), it achieves millisecond-level time synchronization and dynamic negotiation of task priorities across all stations, including feeding, coarse polishing, medium polishing, fine polishing, cleaning, and inspection. The system monitors the operating status of each station 24 / 7. In the event of a single station failure, it can automatically adjust upstream and downstream cycle times, initiate redundancy compensation, or bypass conveying without requiring a complete line shutdown. This significantly improves the overall line stability and reduces single-line downtime, achieving self-healing flexible production under fault conditions. Furthermore, the deterministic communication of the TSN network ensures precise collaboration between stations, enabling accurate time-based coordination among multiple stations. Fault prediction and self-healing scheduling functions allow for early warning and preventative measures before faults occur, and rapid reconstruction of the production process after a fault, minimizing downtime losses and greatly improving overall equipment efficiency and the production line's adaptability.

[0096] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A full-automatic polishing device, comprising a polishing main machine and a front-end AI detection unit, a defect linkage adaptive polishing unit, a digital twin process optimization unit, a polishing medium closed-loop control unit and a full-station collaborative scheduling unit based on the polishing main machine, characterized in that: The front-end AI detection unit is located at the front end of the polishing host entrance, and integrates two-dimensional image recognition and three-dimensional morphology scanning technology to perform full-area defect detection and pixel-level digital grading of the tile surface. The defect-linked adaptive polishing unit is connected to the front-end AI detection unit and dynamically adjusts the polishing pressure, spindle speed and polishing head posture of each area according to the defect classification results. The digital twin process optimization unit is connected to the front-end AI detection unit and the defect-linked adaptive polishing unit respectively, to build a full-dimensional digital twin model covering substrate characteristics, abrasive wear, fluid dynamics and equipment conditions, and to iteratively optimize the process parameter library. The polishing medium closed-loop management unit is connected to the polishing host to realize the entire life cycle closed loop of polishing fluid intelligent proportioning, real-time monitoring, multi-level purification, automatic replenishment and status adjustment; the full-station collaborative scheduling unit is connected to each station equipment through a time-sensitive network to realize millisecond-level time synchronization of all stations, dynamic negotiation of task priorities and fault self-healing flexible scheduling.

2. The fully automatic polishing apparatus according to claim 1, characterized by: The front-end AI detection unit includes a two-dimensional imaging module, a three-dimensional topography scanning module, a multispectral auxiliary detection module, and a defect grading and processing module. The defect grading and processing module grades defects using a defect severity quantification model, which satisfies the following formula: Where S is the defect severity index; D is the defect depth; Dmax is the maximum depth threshold for this type of defect; A is the defect projection area; Atotal is the total area of ​​the tile; N is the number of this type of defect per unit area; Nref is the reference value for the number of defects per unit area; w1, w2, and w3 are the depth weight, area weight, and number weight, respectively, satisfying w1+w2+w3=1.

3. A fully automatic polishing apparatus according to claim 2, characterized in that: The two-dimensional imaging module uses a linear CCD camera in conjunction with a multi-angle ring light source to acquire high-resolution images of the tile surface; the three-dimensional topography scanning module uses laser line scanning or structured light three-dimensional reconstruction technology to obtain the depth information of the tile surface; the multispectral auxiliary detection module uses polarized light and hyperspectral imaging technology to identify subsurface defects and microcracks; the defect grading processing module uses a multimodal fusion convolutional neural network to identify defect types, measure dimensions, and grade severity of the fused multimodal data.

4. The fully automatic polishing apparatus according to claim 1, characterized by: The defect-linked adaptive polishing unit includes a zoned polishing execution module, a force control closed-loop module, and a posture adjustment module. The zoned polishing execution module divides the tile surface into multiple independently controlled polishing zones, each equipped with an independent polishing head group. The force control closed-loop module dynamically adjusts the polishing pressure of each zone through a pressure adaptive adjustment model, which satisfies the following formula: Where P(x,y) is the polishing pressure at coordinate (x,y) on the tile surface; P0 is the reference polishing pressure; k is the pressure gain coefficient; S(x,y) is the defect severity index at coordinate (x,y); f(θ) is the polishing head tilt angle correction function; and θ is the angle between the polishing head and the normal to the tile surface.

5. A fully automatic polishing apparatus according to claim 4, characterized in that: The force control closed-loop module uses a high-precision piezoelectric force sensor and an electro-hydraulic servo valve to form a millisecond-level force control closed-loop system. The force sensor sampling frequency is 1kHz, and the servo valve response time is less than 5ms. The posture adjustment module includes a polishing head tilt angle adjustment mechanism and a swing trajectory adjustment mechanism. The tilt angle adjustment range is 0-15 degrees, and the angle adjustment accuracy is ±0.1 degrees. The posture adjustment module also has an edge protection function. When the polishing head runs to the edge area of ​​the tile, it automatically reduces the polishing pressure and adjusts the tilt angle.

6. The fully automated polishing apparatus according to claim 1, wherein: The digital twin process optimization unit includes a virtual simulation module, a parameter optimization module, and a model update module. The virtual simulation module constructs a multi-physics coupled digital twin model. The parameter optimization module continuously updates the optimal process parameter library through an iterative optimization model of process parameters, which adopts the following formula: Where Θ(n) is the process parameter vector for the nth iteration; Θ(n+1) is the process parameter vector for the (n+1)th iteration; and α is the learning rate. Θ* is the gradient vector of the objective function J with respect to the process parameters; β is the inertia coefficient; and Θ* is the historical optimal process parameter vector.

7. A fully automatic polishing apparatus according to claim 6, characterized in that: The digital twin model includes a material removal model, an abrasive wear model, a thermo-mechanical coupling model, a fluid dynamics model, and a device dynamics model. The model update module compares the actual polishing detection data with the simulation prediction results after polishing a set number of tiles or at set intervals, and automatically adjusts the model parameters using a Bayesian optimization method. The digital twin process optimization unit also has a rapid adaptation function for new tile categories, using transfer learning technology to transfer existing process knowledge to new categories and quickly build an initial process parameter library.

8. The fully automated polishing apparatus according to claim 1, wherein: The polishing medium closed-loop control unit includes an intelligent proportioning module, an online monitoring module, a multi-stage purification module, and an automatic replenishment module. The online monitoring module monitors the concentration, pH value, conductivity, temperature, and abrasive particle size distribution of the polishing fluid in real time. The automatic replenishment module maintains a constant polishing fluid concentration through a dynamic adjustment model, which uses the following formula: Where C(t) is the mass concentration of the polishing fluid at time t; C(t+Δt) is the mass concentration of the polishing fluid at time t+Δt; η is the concentration adjustment gain coefficient; Ctarget is the target concentration; γ is the concentration decay rate caused by unit volume wear; V(t) is the polishing fluid circulation flow rate at time t; Δt is the time step; and Vtotal is the total volume of the polishing fluid system.

9. The fully automatic polishing equipment according to claim 8, characterized in that: The intelligent proportioning module automatically and quantitatively proportions polishing concentrate, pure water, and functional additives according to the working conditions of each polishing process of the tile, and mixes them evenly to output a standardized concentration of polishing liquid suitable for the process; the multi-stage purification module adopts a three-stage purification process of centrifugal separation, magnetic separation filtration, and precision filter to remove abrasive particles larger than 0.5μm from the polishing liquid; the polishing medium closed-loop management unit is also equipped with polishing liquid status assessment and life prediction functions, and establishes a polishing liquid health assessment model by comprehensively considering indicators such as concentration change, pH value change, abrasive particle size distribution, and abrasive content; the polishing liquid temperature is controlled within the range of 22±1℃ by a heating / cooling heat exchange system.

10. The fully automatic polishing equipment according to claim 1, characterized in that: The full-station collaborative scheduling unit includes a TSN network communication module, a status monitoring module, a fault diagnosis module, and a dynamic scheduling module. The TSN network communication module uses a time-sensitive networking protocol to achieve millisecond-level time synchronization and deterministic data transmission across all workstations. The status monitoring module relies on the time-sensitive network to collect real-time operating parameters, operating conditions, and execution data of each workstation on the polishing production line, achieving comprehensive, high-frequency, deterministic data collection and status aggregation of all equipment operating conditions. The fault diagnosis module, based on a preset equipment health benchmark model, connects to the real-time operating condition data collected by the status monitoring module, and achieves fault identification, anomaly assessment, and potential fault prediction for all equipment through data deviation comparison and intelligent analysis. The dynamic scheduling module achieves self-healing flexible scheduling through a task priority dynamic scheduling model, which uses the following formula: Where Pri(i,t) is the task priority of workstation i at time t; D(i,t) is the number of unprocessed tiles at workstation i at time t; Dtotal is the total number of unprocessed tiles in the entire line; T(i,t) is the expected fault recovery time of workstation i; Tmax is the maximum allowable downtime; F(i,t) is the fault status factor of workstation i; λ1, λ2, and λ3 are weighting coefficients, satisfying λ1+λ2+λ3=1.

Images