8k mirror surface plate adaptive grinding and polishing control method and system

Abstract

This invention provides an adaptive grinding and polishing control method and system for 8K mirror panels, relating to the field of precision machining technology. It aims to solve the problems of incomplete condition perception, simplistic control strategies, and difficulty in determining the processing endpoint in existing technologies. The method includes: during the grinding and polishing process, real-time acquisition of multi-source state information reflecting the processing status, including at least contact force information between the grinding / polishing head and the material to be processed, and surface image information of the material; based on the surface image information, real-time prediction of the surface roughness of the material to be processed; adaptive and dynamic adjustment of grinding and polishing parameters based on the predicted surface roughness and the contact force information; and determination of the processing endpoint based on the multi-source state information. This invention achieves precise control and intelligent operation of the grinding and polishing process through multi-source information closed-loop control, accurately determining the endpoint, improving processing quality and consistency, and increasing production efficiency while reducing costs.

Description

Technical Field

[0001] This invention relates to the field of precision machining technology, and in particular to an adaptive grinding and polishing control method and system for 8K mirror panels. Background Technology

[0002] Mirror-finish panels, such as 8K stainless steel mirror panels, are widely used in high-end decoration and electronic product casings due to their extremely high surface finish. The core processing step is polishing, which aims to reduce the surface roughness of the sheet material to a mirror level. Traditional polishing processes rely heavily on the experience of skilled workers, controlling polishing parameters manually or semi-automatically. This "experience-driven" model has significant drawbacks: First, the processing is difficult to standardize, resulting in poor product quality consistency; second, the processing quality depends entirely on the worker's skill level and condition, leading to high labor costs; finally, manual judgment of the processing status and endpoint is subjective and lagging, easily leading to under- or over-processing, resulting in material waste and inefficiency.

[0003] To overcome these problems, the industry has made some automation attempts, such as using vision sensors to acquire online images of the surface of the material to be processed to assess surface roughness. However, these methods usually only stay at the "monitoring" level and fail to provide a complete closed-loop control scheme from comprehensive perception and intelligent decision-making to precise execution. Specifically, existing monitoring methods often ignore other key physical information in the processing, such as the actual contact force between the polishing head and the material to be processed, and subsurface damage generated during material removal, resulting in an incomplete understanding of the processing status. Therefore, existing technologies suffer from incomplete perception of the polishing process, a single control strategy, and difficulty in accurately determining the processing endpoint, leading to difficulties in guaranteeing processing quality and efficiency. Summary of the Invention

[0004] The primary objective of this invention is to provide an adaptive grinding and polishing control method for 8K mirror panels, aiming to solve the technical problems in the prior art where the grinding and polishing process of mirror panels suffers from incomplete perception of working conditions, a single control strategy, and difficulty in accurately determining the processing endpoint, resulting in poor processing quality and efficiency.

[0005] The second objective of this invention is to provide an adaptive polishing control system for 8K mirror panels that implements the above-described method.

[0006] To achieve the above objectives, the present invention provides the following technical solution: This invention provides an adaptive polishing control method for 8K mirror panels, comprising the following steps: during the polishing process, real-time acquisition of multi-source state information reflecting the processing status, wherein the multi-source state information includes at least the contact force information between the polishing head and the material to be processed and the surface image information of the material to be processed; based on the surface image information, real-time prediction of the surface roughness of the material to be processed; based on the predicted surface roughness and the contact force information, adaptive and dynamic adjustment of polishing parameters; and based on the multi-source state information, determination of the endpoint of the processing process.

[0007] Optionally, the polishing parameters are selected from at least one of polishing pressure, polishing head rotation speed, or feed speed.

[0008] Optionally, the step of acquiring multi-source state information reflecting the processing status in real time further includes: acquiring subsurface damage signals generated during the grinding and polishing process.

[0009] Furthermore, the step of determining the end point of the processing process specifically involves: using a Kalman filter model to fuse the contact force information, the predicted surface roughness, and the subsurface damage signal to determine the end point of the processing process.

[0010] Optionally, the step of predicting the surface roughness of the material to be processed in real time specifically involves: analyzing the surface image information using a pre-trained convolutional neural network to output a real-time predicted value of the surface roughness.

[0011] Optionally, it also includes: automatically deciding when to switch polishing tools of different grit sizes based on the predicted rate of change of surface roughness.

[0012] Optionally, it further includes: using a digital twin model to perform calculations on the polishing process to obtain calculation results for guiding the adaptive dynamic adjustment of polishing parameters; and the step of adaptive dynamic adjustment of polishing parameters is also based on the calculation results.

[0013] Furthermore, it also includes: using the digital twin model to predict the processing state to obtain a predicted value; comparing the multi-source state information collected in real time with the predicted value, and correcting the digital twin model in real time according to the comparison result; wherein, the correction includes correcting the grinding disc wear parameter and / or material removal rate parameter in the digital twin model.

[0014] This invention also provides an adaptive polishing control system for 8K mirror panels, comprising: a multimodal sensing unit configured to acquire, in real time, contact force information between the polishing head and the material to be processed and surface image information of the material to be processed during the polishing process; and a controller electrically connected to the multimodal sensing unit, the controller being configured to: receive the contact force information and surface image information; predict the surface roughness of the material to be processed in real time based on the surface image information; generate and output control commands for adaptively and dynamically adjusting polishing parameters based on the predicted surface roughness and the contact force information; and determine the endpoint of the processing process based on the contact force information and surface image information, and generate and output a stop command when the endpoint is determined.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. Improved Processing Quality and Consistency: By collecting multi-source status information such as contact force and surface images in real time, and predicting surface roughness based on this information, closed-loop control of the processing process is achieved. This data-driven approach can precisely control the grinding and polishing process, effectively avoiding the subjectivity and uncertainty of manual operation, significantly improving the surface quality of mirror panels and the consistency of products between batches, thereby increasing the yield rate.

[0016] 2. Achieving Intelligent and Automated Processing: This invention transforms the traditional "experience-driven" model into a "data-driven" intelligent processing model. From process monitoring and parameter adjustment to endpoint determination, the entire process is completed automatically, significantly reducing reliance on highly skilled technicians and lowering labor costs.

[0017] 3. Improve production efficiency and reduce costs: By integrating multi-source status information to accurately determine the processing endpoint, it effectively avoids inefficiency caused by conservative processing or material damage and energy waste caused by over-processing, shortens the single-piece processing cycle, and reduces the scrap rate. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the adaptive polishing control system for mirror panels according to an embodiment of the present invention.

[0020] Figure 2 This is a schematic diagram of the core process of the adaptive polishing control method for 8K mirror panels according to an embodiment of the present invention.

[0021] Figure 3 This is a schematic diagram of closed-loop control information interaction according to an embodiment of the present invention.

[0022] Figure 4 This is a schematic flowchart of the automatic abrasive grain switching timing decision according to an embodiment of the present invention.

[0023] In the attached diagram: 10-Industrial robot; 20-Sheet material to be processed; 30-Grinding and polishing actuator; 31-Grinding and polishing head; 32-Force sensor; 33-Vision sensor; 34-Acoustic emission sensor; 40-Controller; 41-Data acquisition module; 42-CNN prediction module; 43-Adaptive control module; 44-Kalman filter module; 50-Sensing system; 60-Robot actuator. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be considered as limitations on the invention. All other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this invention. Unless otherwise defined, all technical and scientific terms used in the embodiments of this invention have the same meaning as commonly understood by those skilled in the art. The terminology used in the embodiments of this invention is for the purpose of describing the embodiments of this invention only and is not intended to limit the invention.

[0025] Before providing a further detailed description of the embodiments of the present invention, some of the nouns and terms involved in the embodiments of the present invention will be explained, and the nouns and terms involved in the embodiments of the present invention shall be interpreted as follows.

[0026] (1) Multi-source state information: refers to various physical signals that can reflect the processing state and are collected in real time during the grinding and polishing process. In this invention, this information includes at least the contact force information between the grinding and polishing head and the workpiece to be processed, collected by a force sensor, and the surface image information of the workpiece to be processed, collected by a vision sensor. As a preferred embodiment, this information may further include subsurface damage signals collected by an acoustic emission sensor.

[0027] (2) Digital twin model: refers to a high-fidelity digital model that corresponds to physical objects such as grinding and polishing equipment, the material to be processed, and the processing environment in the physical world and runs synchronously in virtual space. This model can receive real-time sensor data, perform simulation calculations on the grinding and polishing process, predict the processing status (such as theoretical grinding amount and surface morphology evolution), and can self-correct based on actual feedback from the physical world. Its output results can be used to assist or guide the decision-making of the controller.

[0028] (3) Convolutional Neural Network (CNN): refers to a deep learning model, specifically a pre-trained algorithm model for image analysis in this invention. This model can receive surface image information of the material to be processed as input, and automatically extract deep features related to surface roughness through multi-layer convolution, pooling and other operations, and output a real-time, quantitative prediction value of the current surface roughness (such as Ra value).

[0029] (4) Kalman filter model: refers to an optimal state estimation algorithm for processing time series data. In this invention, this model is used to perform dynamic data fusion and smoothing on multi-source state information (such as contact force signals, roughness prediction value sequences, subsurface damage signals, etc.) collected from different sensors and containing measurement noise, so as to obtain a more accurate estimate of the true state of the processing process, and to determine whether the processing process has reached the end point based on the estimation result and preset criteria.

[0030] Please see Figures 1 to 4 This invention provides an adaptive grinding and polishing control method and system for 8K mirror panels. This solution aims to address the problems in existing 8K mirror panel grinding and polishing processes, such as low automation, reliance on manual experience, unstable processing quality, difficulty in comprehensively sensing the processing status, and inability to accurately determine the processing endpoint. By constructing a complete closed-loop control system of "multi-source sensing - real-time evaluation - intelligent decision-making - precise termination," this invention can significantly improve the quality, efficiency, and consistency of mirror panel processing.

[0031] In a basic embodiment, refer to Figure 2 The process shown illustrates that this invention provides an adaptive grinding and polishing control method for 8K mirror panels. This method first collects multi-source state information reflecting the processing status in real time during the grinding and polishing process (step S102). Specifically, as shown... Figure 1 As shown, a polishing actuator 30 is mounted at the end of an industrial robot 10. The actuator 30 integrates a force sensor 32 and a vision sensor 33. When the actuator 30 drives its polishing head 31 to process the fixed workpiece 20, the force sensor 32 measures and outputs the contact force information between the polishing head 31 and the workpiece 20 in real time, while the vision sensor 33 (e.g., an industrial camera) simultaneously captures images of the surface of the workpiece 20 to obtain surface image information. These two types of information constitute the most basic multi-source state information and are sent to a controller 40 for processing.

[0032] Upon receiving multi-source status information, the controller 40 executes the next step, namely, predicting the surface roughness of the material to be processed in real time based on the surface image information (step S103). In a basic implementation, the image processing unit inside the controller 40 analyzes the received surface image, for example, by calculating the image's gray-level co-occurrence matrix, texture contrast, or frequency domain features, and estimates an approximate value of the surface roughness according to a preset mapping relationship. This step enables online, non-contact evaluation of key quality indicators and is a prerequisite for achieving closed-loop control.

[0033] Next, the controller 40 adaptively and dynamically adjusts the polishing parameters based on the predicted surface roughness and the contact force information (steps S104 and S105). The control logic within the controller 40 comprehensively analyzes two aspects of information: on the one hand, the predicted roughness value reflects the gap between the current processing quality and the target; on the other hand, the contact force information reflects whether the actual processing load is stable and appropriate. For example, if the controller 40 detects that the current roughness is decreasing slowly and the contact force is lower than the set value, it generates and outputs a control command to the industrial robot 10, causing it to adjust its posture to increase the pressure applied by the polishing head 31 to the workpiece 20. Conversely, if the contact force is too large and may cause scratches, it will instruct to reduce the pressure. This dynamic adjustment based on dual feedback enables the processing process to converge towards the optimal state in real time.

[0034] Finally, the method also includes determining the endpoint of the processing based on the multi-source state information (steps S106 and S107). Throughout the processing, the controller 40 continuously monitors the predicted surface roughness value and the changing trend of the contact force signal. When the controller 40 determines that the predicted roughness value has reached and stabilized below a preset target threshold (e.g., Ra 0.02μm required for mirror finish), and the contact force signal remains stable for a period of time without abnormal fluctuations, the system determines that the processing has been completed. At this time, the controller 40 generates and outputs a stop command to the industrial robot 10, causing it to stop the polishing operation (step S108), thereby completing a complete automated polishing cycle.

[0035] The aforementioned basic embodiment transforms traditional, experience-dependent grinding and polishing operations into a data-driven automated process by constructing a closed-loop control loop incorporating force-vision dual-modal perception. Its technical advantage lies in its ability to initially achieve adaptive control of the grinding and polishing process, significantly improving the stability and repeatability of processing quality compared to purely manual or semi-automatic operations, and reducing reliance on operator skills.

[0036] Furthermore, to enable the controller 40 to perform more refined control, the polishing parameters can be selected from at least one of the following: polishing pressure, the rotational speed of the polishing head 31, or the feed speed driven by the industrial robot 10. In practical applications, these three are the most critical process parameters affecting the polishing effect. For example, in the rough polishing stage, a higher polishing pressure and a lower rotational speed can be used to pursue a higher material removal rate; while in the fine polishing stage, the pressure should be reduced and the rotational speed increased to obtain a smoother surface. This method can dynamically adjust these parameters in combination, thereby achieving the optimal control strategy at different processing stages. Clearly defining these adjustable parameters makes the control objective clearer, enhancing the practicality and flexibility of the method.

[0037] In a preferred embodiment, to obtain a more comprehensive understanding of the processing status, especially at the level of internal material damage, the step of real-time acquisition of multi-source status information (step S102) may further include acquiring subsurface damage signals generated during the grinding and polishing process. For example... Figure 1 As shown, this can be achieved by additionally integrating an acoustic emission sensor 34 onto the polishing actuator 30. Acoustic emission technology is a non-destructive testing technique. When a material undergoes plastic deformation or microcrack propagation under stress, it releases energy in the form of elastic waves, which can be captured by the acoustic emission sensor 34. Therefore, by analyzing the intensity, frequency, and other characteristics of the acoustic emission signal, it is possible to determine in real time whether there is subsurface damage caused by over-processing in the workpiece 20.

[0038] As one feasible approach, the raw signal acquired by the acoustic emission sensor 34 is amplified by a preamplifier, then bandpass filtered by the controller 40, and its root mean square (RMS) value is extracted as a feature signal. The intensity of this feature signal is positively correlated with the propagation activity of microcracks inside the material. When it exceeds a preset threshold, it is determined that there is a risk of subsurface damage caused by overprocessing.

[0039] Introducing subsurface damage signals brings significant technical benefits. It provides the control system with an additional sensing dimension, enabling monitoring of the material's internal state. Traditional control methods focus only on surface gloss, potentially sacrificing the integrity of the material's internal structure. This implementation, by fusing subsurface damage signals, can maintain surface quality while avoiding residual stress and microcracks introduced by over-processing, thus producing high-quality mirror panels with both high surface quality and good structural integrity. This is particularly important for fields such as aerospace and precision molds, where stringent requirements for component fatigue life and reliability are placed.

[0040] Furthermore, based on the above-described implementation method including subsurface damage signal acquisition, the step of determining the end point of the processing process (step S106) can be specified. Specifically, the Kalman filter module 44 in the controller 40 employs a Kalman filter model to fuse the contact force information, the predicted surface roughness, and the subsurface damage signal to determine the end point of the processing process. These three signals represent the processing "force," "surface effect," and "internal damage," respectively, but all inevitably contain measurement noise. The Kalman filter model can perform optimal state estimation on these three heterogeneous, noisy time-series data, filter out random interference, and output a smooth and accurate estimate of the true state of the system (e.g., a comprehensive state vector including roughness and damage degree).

[0041] In this embodiment, the Kalman filter model is used to fuse multi-source state information to determine the processing endpoint, and its specific implementation is as follows.

[0042] I. Definitions of State Variables and Observation Variables The Kalman filter model is constructed based on a discrete linear state-space model. State variables... Characterizing the processing system at time The true state is represented by a two-dimensional state vector. .in, This represents the true value of the surface roughness of the material to be processed. This represents the characteristic value of the subsurface damage level. This characteristic value is a dimensionless parameter, ranging from 0 to 1, where 0 represents no damage and 1 represents severe damage.

[0043] Observed variables Characterizing the sensor at time The measured values ​​are obtained using a three-dimensional observation vector. .in, The contact force measurement value collected by the force sensor. The surface roughness prediction value output by the convolutional neural network. This is the characteristic value of the subsurface damage signal acquired by the acoustic emission sensor. This value is usually the root mean square value extracted after bandpass filtering.

[0044] II. State transition and observation equations of the Kalman filter model The state transition equation describes the evolution of the system state over time and is defined as follows: Among them, the state transition matrix Based on the physical laws of the processing process, it is set as follows: This means that the system assumes that roughness and damage level remain constant in the absence of external forces. The process noise follows a zero-mean Gaussian distribution, and its covariance matrix is... Obtained through offline experimental data calibration.

[0045] The observation equation describes the mapping relationship between the observed variables and the state variables, and is defined as follows: Among them, the observation matrix for The physical meaning of this matrix lies in the contact force measurement values. These do not directly correspond to state variables but serve as auxiliary information during the fusion process to correct prediction errors. This can be achieved through extended Kalman filtering or by setting up virtual observations. Roughness observations Directly reflects state variables Acoustic emission observations Directly reflects the state of damage . The observed noise follows a zero-mean Gaussian distribution, and its covariance matrix is... This was determined through sensor calibration experiments.

[0046] III. The Recursive Fusion Process of Kalman Filtering The Kalman filter model performs the following recursive steps in each control cycle. First, it performs state prediction based on the optimal estimate from the previous time step. and state transition matrix Predict the prior estimate of the state at the current moment. Simultaneously, the covariance matrix of the predicted state estimation error... .

[0047] Next, calculate the Kalman gain based on the observation matrix. Predicting the covariance matrix and observation noise covariance matrix Calculate the Kalman gain matrix .

[0048] Perform another state update. Use the current observation values. The difference between the observed and predicted values ​​is used to correct the prior state estimate, thus obtaining the posterior state estimate. Finally, update the estimated error covariance matrix. .

[0049] Through the above recursion, the Kalman filter module 44 outputs the optimal state estimate at each time step. and These values ​​represent the fused surface roughness estimate and subsurface damage estimate, respectively, effectively filtering out random noise and transient disturbances in the sensor measurements.

[0050] IV. Endpoint Determination Criteria Based on Kalman Filter Output The controller 40 uses the optimal state estimate output by the Kalman filter, combined with a preset endpoint determination criterion, to comprehensively determine whether the processing has reached its endpoint. The determination criterion specifically includes the following three conditions.

[0051] Condition 1 requires that the surface quality meets the standard. Estimated surface roughness after fusion. Requires continuity The number of sampling periods is lower than the preset target roughness threshold. For example, when Furthermore, when the sampling period corresponds to 0.5 seconds, the target roughness threshold can be set to... .

[0052] Condition 2 requires a safe subsurface damage state. The estimated subsurface damage level after fusion. Requires continuity The number of sampling periods is lower than the preset damage safety threshold. For example, when Furthermore, when the sampling period corresponds to 1 second, the damage safety threshold can be set to 0.1, and the value should show a trend of converging to the baseline level.

[0053] Condition three requires a stable contact force state. The contact force measurement value is acquired by the force sensor. The standard deviation should be below the preset fluctuation threshold within a specific time window. This indicates that the processing is smooth and free from abnormal shocks or vibrations.

[0054] When all three conditions above are met simultaneously, the controller 40 determines that the processing has reached its end point and generates a stop command output to the industrial robot 10 to terminate the polishing operation. If only the surface quality meets the standard but the estimated damage level is still higher than the safety threshold, the system will continue to perform the fine polishing process and prioritize reducing the polishing pressure in the control strategy to reduce the risk of subsurface damage. If the damage level continues to rise and exceeds the danger threshold, for example... If the system determines that there is a risk of over-processing, it will immediately terminate the current processing and issue an alarm signal.

[0055] By employing the Kalman filter model described above to fuse multi-source state information and a comprehensive judgment mechanism based on the fusion results, this system can accurately and robustly identify the processing endpoint in complex processing environments. This mechanism effectively avoids the misjudgment and missed judgment problems caused by threshold judgment from a single sensor, significantly improving the consistency of processing quality and the reliability of the production process.

[0056] The technical advantage of this fusion-based decision-making method lies in its significant improvement in the accuracy and robustness of endpoint determination. The endpoint determination criterion is no longer a simple threshold comparison of a single indicator, but rather a comprehensive judgment based on a multi-dimensional state space. For example, the system can set endpoint conditions as: "the fused roughness estimate is below the target value for N consecutive seconds" and "the fused subsurface damage signal energy falls back to the baseline level for M consecutive seconds." This combination of "AND / OR" logic ensures that processing only stops when both surface quality and internal structural integrity simultaneously meet the standards, fundamentally eliminating under-processing and over-processing phenomena, and achieving a good balance between processing time and processing quality.

[0057] In another preferred embodiment, to improve the accuracy and robustness of surface roughness prediction, the step of real-time prediction of the surface roughness of the material to be processed (step S103) can be specified as follows: using a pre-trained convolutional neural network (CNN) to analyze the surface image information to output a real-time predicted value of the surface roughness. Figure 1 As shown, the controller 40 integrates a CNN prediction module 42, which stores the trained CNN model. The training process is usually completed offline, by collecting a large number of images of board materials with different roughness levels and their corresponding precision instrument measurements (as labels) to train the CNN model to learn the complex mapping relationship from image texture to roughness values.

[0058] The advantage of using CNNs for prediction lies in its improved prediction accuracy and generalization ability compared to traditional image processing algorithms. Traditional algorithms rely on manually designed features (such as contrast and directionality), which are prone to failure when faced with complex textures, changes in lighting, or uneven material reflection. CNNs, on the other hand, can automatically learn and extract the most discriminative multi-level abstract features through their deep structure, exhibiting stronger robustness to various disturbances. For example, it can distinguish between real surface scratches and harmless light reflections. Therefore, CNN-based predictions are more stable and reliable, providing high-quality input for subsequent adaptive control and representing a crucial step in improving the performance of the entire closed-loop control system.

[0059] Furthermore, based on the above-described implementation method that can predict surface roughness in real time and accurately, this method can also include a more advanced function: automatically deciding when to switch between grinding and polishing tools of different grit sizes based on the predicted rate of change of surface roughness. This corresponds to... Figure 4 The process is illustrated. In actual multi-stage grinding and polishing processes (e.g., from coarse grinding to semi-fine grinding to fine polishing), it is crucial when to change to a finer grinding disc or sandpaper. In this embodiment, the controller 40 not only focuses on the absolute value of roughness, but also calculates its derivative with respect to time in real time, i.e., the "improvement rate" (step S203).

[0060] Reference Figure 4 At the start of processing, the system uses a coarser grinding disc for a basic grinding and polishing cycle (steps S201, S202). In the early stages of processing, surface quality improves rapidly, and the roughness improvement rate is very high. As processing progresses, the cutting capability of the current grit grinding disc reaches its limit, and the improvement in surface quality becomes increasingly slower, with the improvement rate decreasing significantly. When the controller 40 detects that this rate is lower than a preset switching threshold (step S204), it determines that the current processing stage has reached an efficiency bottleneck and automatically triggers a disc switching action, switching to the next level of finer grinding disc (step S205). This process repeats until the final fine polishing is completed using the last level of grinding disc (steps S206, S207), and then finally ends (S208).

[0061] The addition of this technological feature enables the entire multi-stage polishing process to be fully intelligent and automated. It transforms the traditional process, which relied on skilled workers to judge the timing of plate changes by "listening to sounds" and "watching sparks," into scientific decision-making based on real-time data and quantitative indicators. The technological effect is a significant reduction in total processing time, as the system always operates within its most efficient range, avoiding wasted time in inefficient phases and thus greatly improving overall production efficiency.

[0062] In another preferred embodiment, to make the control decision more forward-looking and optimal, the method can also introduce digital twin technology in the step of adaptively adjusting the polishing parameters. Specifically, the method further includes: using a digital twin model to calculate the polishing process to obtain calculation results; and the step of adaptively and dynamically adjusting the polishing parameters (step S104) is also based on the calculation results. This digital twin model is a virtual copy corresponding one-to-one with the physical polishing system (including industrial robot 10, polishing actuator 30, and the workpiece 20 to be processed, etc.), which can simulate and calculate the theoretical material removal rate, temperature distribution, and surface morphology evolution in real time according to the current polishing parameters and the state of the workpiece to be processed.

[0063] In this embodiment, the digital twin model is constructed, interacted with, and modified in the following ways: I. Methods for constructing digital twin models A digital twin model is a high-fidelity simulation model that operates in virtual space and corresponds one-to-one with a physical polishing system. Its construction includes the following steps: First, a geometric model of the physical entity is established. A three-dimensional geometric model of the industrial robot 10, the polishing actuator 30, the polishing head 31, and the sheet material 20 to be processed is constructed using 3D modeling software and imported into the simulation platform.

[0064] Secondly, configure the physical property model. Based on the actual material properties, set parameters such as the material removal rate coefficient, elastic modulus, Poisson's ratio, and hardness of the plate to be processed 20; set parameters such as the abrasive particle size, grinding disc diameter, and wear coefficient of the polishing head 31.

[0065] Next, a process model was established. A material removal rate model was constructed based on the Preston equation to describe the relationship between polishing pressure, relative velocity, and material removal amount; a contact mechanics model was established to simulate the contact force distribution between the polishing head and the plate; and a surface morphology evolution model was established to simulate the changing trend of surface roughness during processing.

[0066] Finally, through offline calibration and fitting of experimental data, the key coefficients in the above model (such as material removal rate constant, wear coefficient, friction coefficient, etc.) are identified to ensure that the model has high simulation accuracy in the initial state.

[0067] II. Data Interaction Methods Between Digital Twin Models and Physical Systems Establish a two-way data exchange channel between the digital twin model and the physical polishing system: On one hand, there is the data flow from the physical to the virtual. During the processing, the controller 40 collects contact force information from the force sensor 32, surface image information from the vision sensor 33, and subsurface damage signals from the acoustic emission sensor 34 in real time through the data acquisition module 41, and synchronously inputs this multi-source state information into the digital twin model. The model uses this data as boundary conditions to drive the simulation operation in the virtual space.

[0068] On the other hand, there is the data flow from virtual to physical. Based on current processing parameters and real-time sensor data, the digital twin model calculates the processing state within a preset time window (e.g., 0.5 to 2 seconds into the future), outputting predicted values, including: predicted material removal amount, predicted surface roughness, predicted grinding disc wear state, and predicted contact force change trend. These prediction results are transmitted to the adaptive control module 43 as feedforward references for adjusting the grinding and polishing parameters.

[0069] The data interaction frequency is dynamically set according to the signal type: the force signal and the acoustic emission signal are synchronized at a high frequency of 100Hz, and the visual signal and the roughness prediction value are synchronized at a frequency of 10Hz, to ensure that the virtual model and the physical system are closely aligned in the time dimension.

[0070] III. Real-time Correction Mechanism of Digital Twin Model To maintain the accuracy of the digital twin model over the long term, the system incorporates a self-correcting mechanism based on real-time feedback, which includes the following steps: First, a state comparison is performed. The controller 40 compares the predicted values ​​of the digital twin model for the current moment (such as theoretical contact force, theoretical material removal amount, and theoretical surface roughness) with the actual values ​​collected by the physical sensors in real time, and calculates the deviation.

[0071] Secondly, error analysis is performed. When the deviation exceeds a preset threshold (e.g., contact force deviation exceeds 5%, or roughness deviation exceeds 0.005μm) and persists for more than a set time (e.g., 1 second), the system triggers correction logic. Controller 40 identifies the source of error based on the deviation characteristics: If the actual contact force is consistently lower than the predicted value, and the polishing head is used for a longer period than the set value, then the polishing disc is considered worn. If the actual amount of material removed is lower than the predicted value, but the contact force and rotation speed are normal, it is determined that the material removal rate parameter is mismatched. If the actual surface roughness improvement rate is lower than the predicted value, it is determined to be abrasive passivation or abnormal material hardness.

[0072] Next, parameter correction is performed. Based on the above error analysis, the system automatically adjusts the key parameters in the digital twin model: Grinding disc wear parameter correction: The grinding disc wear coefficient is dynamically updated based on the cumulative deviation between the actual contact force and the predicted force, so that the model can more accurately predict the grinding disc life and removal capacity in subsequent processing. Material removal rate parameter correction: Based on the difference between the actual material removal amount and the predicted removal amount, the material removal rate constant in the Preston equation is updated online using the recursive least squares method to adapt the model to the material characteristics of the current batch. Surface morphology evolution parameter correction: Based on the difference between the actual roughness and the predicted roughness, the convergence coefficient in the surface morphology model is adjusted to improve the prediction accuracy of the processing effect.

[0073] Finally, model updates and synchronization are performed. The corrected model parameters are updated to the digital twin model in real time and take effect in subsequent simulations. Each correction is recorded in the log system of controller 40 for long-term trend analysis and predictive maintenance.

[0074] Through the aforementioned construction, interaction, and correction mechanisms, the digital twin model can maintain a high degree of consistency with the physical system throughout the entire processing lifecycle, providing accurate forward-looking guidance for adaptive control and effectively improving the robustness of the system and the stability of processing quality.

[0075] The technical advantage of introducing a digital twin model lies in providing a theoretical reference for the decision-making of the controller 40. When making decisions, the controller 40 no longer relies solely on the "after-the-fact" response from current sensor feedback, but can combine the "before-the-fact" results predicted by the digital twin. For example, the adaptive control module 43 can compare the real-time predicted roughness with the theoretical roughness calculated by the digital twin, and adjust the control strategy based on the deviation between the two. This combination of virtual and real approaches enables the control system to anticipate the effects of control commands over a future period, thereby making smoother, more stable, and closer-to-global-optimal control decisions, effectively suppressing oscillations and overshoot during the processing.

[0076] Furthermore, to ensure the long-term accuracy of the digital twin model's predictions, this method may also include real-time correction of the model itself. Specifically, the method further includes: using the digital twin model to predict the processing state to obtain predicted values; then, comparing the real-time collected multi-source state information with the predicted values, and correcting the digital twin model in real time based on the comparison results. For example, during processing, the controller 40 continuously compares the actual contact force measured by the force sensor 32 with the contact force predicted by the digital twin. If the actual force is consistently less than the predicted force, it may mean that the polishing head 31 has worn. At this time, the system will automatically adjust the parameters in the digital twin model regarding the wear of the polishing disc to realign it with the actual situation in the physical world.

[0077] This closed-loop correction mechanism for the digital twin model endows the system with the ability to learn and evolve adaptively. Its technical effect is a significant enhancement of the robustness and long-term stability of the entire control system. A static model can gradually become inaccurate due to factors such as equipment aging, environmental changes, and batch-to-batch material variations, while a self-correcting dynamic model can continuously and closely follow the changes in the physical entity, ensuring high-precision predictions and guidance throughout its entire lifecycle. This allows the method of this invention to adapt to more complex and variable real-world production environments.

[0078] In one specific implementation, the reliability of determining the end point of the processing process can be improved through more advanced algorithms, even without introducing subsurface damage signals. Specifically, the step of determining the end point of the processing process (step S106) can be further defined as follows: using a Kalman filter model to fuse the multi-source state information (mainly contact force information and roughness predicted based on surface image information) to determine the end point of the processing process. Even with only two signals, their respective measurement noise can interfere with simple threshold judgment methods. The Kalman filter model can perform weighted fusion of the two signals based on their statistical characteristics to obtain an optimal estimate that is closer to the true value than any single signal.

[0079] The technical advantage of this approach lies in providing an effective method for achieving high-precision endpoint determination even with simplified sensor configuration. By effectively suppressing and smoothing noise through the Kalman filter algorithm, premature system termination due to accidental jumps in roughness prediction values ​​or misjudgments due to instantaneous fluctuations in contact force can be avoided. This makes endpoint determination decisions more robust, ensuring final processing quality and representing an optimized solution that balances cost and performance.

[0080] like Figure 1 and Figure 3 As shown, this embodiment of the invention also provides an adaptive polishing control system for mirror panels, which serves as the physical carrier for implementing the above-described method. The system includes a multimodal sensing unit and a controller 40. In the specific physical implementation, the multimodal sensing unit is not a separate packaged device, but rather a sensing system 50 composed of multiple sensors distributed at the polishing site. This sensing system 50 is specifically integrated into the polishing actuator 30 carried by the end effector of the industrial robot 10, and includes at least a force sensor 32 for acquiring contact force information and a vision sensor 33 for acquiring surface image information.

[0081] The controller 40 can be a high-performance industrial control computer or an embedded system, which is electrically connected to the multimodal sensing unit via wired or wireless means. The controller 40 is configured to execute the core logic of the method of the present invention. Specifically, the controller 40 internally stores or runs corresponding software modules, such as... Figure 1 and Figure 3 As shown, these modules include a data acquisition module 41, a CNN prediction module 42, an adaptive control module 43, and a Kalman filter module 44.

[0082] During operation, the controller 40 first receives real-time data streams from the force sensor 32 and the vision sensor 33 via the data acquisition module 41. Next, it transmits the surface image information to the CNN prediction module 42, which, based on a pre-trained convolutional neural network model, outputs a real-time predicted value for the surface roughness of the material to be processed 20. Then, the adaptive control module 43, acting as the decision-making core, receives contact force information from the data acquisition module 41 and the roughness prediction value from the CNN prediction module 42. Based on a built-in control algorithm (e.g., PID, fuzzy control, or other appropriate control strategies incorporating a digital twin model), it generates control commands for adaptively and dynamically adjusting the polishing parameters. These commands are then output to the robot actuator 60, i.e., the industrial robot 10, to adjust its motion trajectory, posture, or speed, thereby changing the polishing pressure or feed rate.

[0083] Simultaneously, the controller 40 also performs the endpoint determination task. It inputs contact force information and surface image information (or roughness prediction values ​​generated from them) into the Kalman filter module 44. The Kalman filter module 44 fuses this information and continuously evaluates whether the processing state meets the preset endpoint conditions. Once the processing endpoint is determined, the controller 40 (specifically through the adaptive control module 43) generates and outputs a stop command, instructing the robot actuator 60 to stop all movements, thereby precisely ending the entire grinding and polishing process. This systematic design tightly integrates perception, decision-making, and execution, forming a complete and efficient automated closed-loop control system.

[0084] To better illustrate the technical effects of the present invention, a comprehensive embodiment is described below. This embodiment aims to achieve fully automated, high-precision, and high-quality unmanned processing from rough grinding to mirror polishing. The hardware system of this embodiment is as follows: Figure 1 As shown, the system includes a six-axis industrial robot 10, with a fully functional grinding and polishing actuator 30 at its end effector. This actuator integrates a high-frequency force sensor 32, a high-resolution vision sensor 33, and a wideband acoustic emission sensor 34, and is equipped with an automatic plate-changing mechanism. The sheet material 20 to be processed is a SUS304 stainless steel plate with an initial surface roughness of Ra 0.5μm. The core controller 40 is an industrial computer equipped with a high-performance GPU. Its internal control software integrates all advanced functional modules: a self-correcting, refined digital twin model; a CNN prediction module 42 trained for stainless steel surfaces; a multi-level processing strategy module capable of determining plate-changing timing; and a Kalman filter module 44 that integrates force, vision, and acoustic emission signals.

[0085] The complete workflow of this comprehensive embodiment is as follows: First, before processing begins, the entire sheet material 20 to be processed is scanned using a vision sensor 33 to establish a three-dimensional digital model of its initial surface, and the digital twin is initialized accordingly. Upon processing commences, the system automatically loads a 400# coarse grinding disc (step S201). The industrial robot 10 begins grinding and polishing according to the optimal path pre-simulated and planned by the digital twin. During processing (step S202), three sensors continuously collect multi-source status information at different frequencies (e.g., force sensor 100Hz, vision sensor 10Hz, acoustic emission sensor 1kHz) and send it to the controller 40.

[0086] The controller 40 processes these data streams in parallel internally. The CNN prediction module 42 outputs a roughness prediction value every 0.1 seconds. The adaptive control module 43 fuses and analyzes this prediction value, real-time force feedback, and the theoretical grinding amount of the digital twin at this moment, dynamically fine-tuning the grinding and polishing pressure and feed rate to achieve efficient material removal. At the same time, the controller 40 compares the measured data with the prediction value of the digital twin, continuously correcting the grinding disc wear coefficient and material hardness parameters in the twin model to keep them synchronized with the physical world. The controller 40 also calculates the roughness improvement rate in real time (step S203). When this rate drops below a preset threshold (step S204), the system determines that the efficiency of the rough grinding stage has decreased, automatically instructs the robot to switch to the 800# medium grinding disc (step S205), loads a new set of process parameters, and begins the next stage of grinding and polishing.

[0087] This process is repeated, for example, sequentially using 800# and 1500# grinding discs (step S206). During the final polishing using a 2000# fine polishing disc (step S207), the endpoint determination mechanism is activated. The Kalman filter module 44 begins deep fusion of the contact force signal, the roughness sequence output by the CNN, and the root mean square value of the acoustic emission signal. The system's endpoint conditions are: the fused roughness estimate remains stable below the target value Ra 0.02μm for 5 consecutive seconds; simultaneously, the intensity of the fused acoustic emission signal returns to a baseline level indistinguishable from the equipment's idle state; and the contact force remains stable without abrupt changes. When all conditions are met, the controller 40 issues a final stop command, and the processing ends (step S208).

[0088] This comprehensive embodiment achieves significant overall technical benefits by synergistically combining features such as multi-source sensing, CNN prediction, adaptive digital twin, multi-level policy decision-making, and multimodal fusion judgment. First, compared to a single embodiment, it boasts higher processing efficiency; through intelligent tray-changing decision-making, the total processing time can be reduced by more than 30%. Second, it delivers superior processing quality; not only can the surface finish consistently reach the 8K mirror standard, but monitoring acoustic emission signals effectively prevents subsurface damage, ensuring the structural integrity and fatigue life of the parts. Finally, the system exhibits strong intelligence and robustness; the self-correcting digital twin enables it to adapt to differences in materials from different batches and the wear and tear of the equipment itself, achieving long-term, stable, and reliable unmanned production.

[0089] The adaptive polishing control method and system for 8K mirror panels provided by this invention has broad application prospects. Its core application area is high-end manufacturing industries with extremely high requirements for surface quality. For example, in the architectural decoration industry, it can be used to produce stainless steel or copper mirror panels for high-end elevator cars, hotel lobby pillars, and curtain walls. In the consumer electronics field, it can be used for the precision polishing of metal casings or frames of products such as smartphones, tablets, and laptops to achieve specific mirror or high-gloss textures.

[0090] Furthermore, this technology is also applicable to other precision manufacturing fields. In the automotive industry, it can be used for surface treatment of components such as interior and exterior trim strips and wheel hubs in high-end cars. In the medical device field, it can be used for the final polishing of implants or tools such as artificial joints and surgical instruments that require extremely low surface roughness to reduce friction and biological reactions. In the optics and aerospace fields, it can be used to manufacture precision optical components such as metal-based mirrors.

[0091] The flexibility of this invention is reflected in the trainability and configurability of its core algorithm model. For different processing materials (such as aluminum alloys, titanium alloys, high-temperature alloys, etc.), it can quickly adapt to new applications simply by re-collecting data, training the CNN model, and calibrating the digital twin parameters. For different product shapes, new processing paths can be planned only in the robot control software. Therefore, this invention provides a highly adaptable and scalable intelligent processing platform technology that can significantly improve precision surface treatment processes in numerous industries.

Claims

1. An adaptive grinding and polishing control method for an 8K mirror panel, characterized in that, Includes the following steps: During the grinding and polishing process, multi-source status information reflecting the processing status is collected in real time. The multi-source status information includes at least the contact force information between the grinding and polishing head and the material to be processed and the surface image information of the material to be processed. Based on the surface image information, the surface roughness of the material to be processed is predicted in real time; Based on the predicted surface roughness and the contact force information, the grinding and polishing parameters are adaptively and dynamically adjusted. Based on the multi-source state information, the endpoint of the processing is determined.

2. The method according to claim 1, characterized in that, The step of real-time acquisition of multi-source state information reflecting the processing status also includes: acquiring subsurface damage signals generated during the grinding and polishing process.

3. The method according to claim 1, characterized in that, The step of real-time prediction of the surface roughness of the material to be processed specifically involves: analyzing the surface image information using a pre-trained convolutional neural network to output a real-time predicted value of the surface roughness.

4. The method according to claim 1, characterized in that, Also includes: A digital twin model is used to perform calculations on the grinding and polishing process to obtain calculation results that guide the adaptive dynamic adjustment of grinding and polishing parameters; and the step of adaptive dynamic adjustment of grinding and polishing parameters is also based on the calculation results.

5. The method according to claim 1, characterized in that, The specific steps for determining the end point of the processing process are as follows: using a Kalman filter model to fuse the multi-source state information in order to determine the end point of the processing process.

6. The method according to claim 2, characterized in that, The specific steps for determining the end point of the processing process are as follows: using a Kalman filter model to fuse the contact force information, the predicted surface roughness, and the subsurface damage signal to determine the end point of the processing process.

7. The method according to claim 4, characterized in that, It also includes: using the digital twin model to predict the processing state to obtain a predicted value; comparing the multi-source state information collected in real time with the predicted value, and correcting the digital twin model in real time according to the comparison result; wherein, the correction includes correcting the grinding disc wear parameter and / or material removal rate parameter in the digital twin model.

8. The method according to claim 1, characterized in that, Also includes: Based on the predicted rate of change in surface roughness, the system automatically decides when to switch between different grit sizes of polishing tools.

9. The method according to claim 1, characterized in that, The polishing parameters are selected from at least one of polishing pressure, polishing head rotation speed, or feed speed.

10. An adaptive polishing control system for an 8K mirror panel, used to implement the method described in any one of claims 1 to 9, characterized in that, include: A multimodal sensing unit is configured to collect, in real time, contact force information between the polishing head and the workpiece and surface image information of the workpiece during the polishing process; A controller, electrically connected to the multimodal sensing unit, is configured to: Receive the contact force information and surface image information; Based on the surface image information, the surface roughness of the material to be processed is predicted in real time; Based on the predicted surface roughness and the contact force information, control commands are generated and output for adaptively and dynamically adjusting the polishing parameters. Based on the contact force information and surface image information, the endpoint of the processing is determined, and a stop command is generated and output when the endpoint is determined.

Images