Adaptive flexible polishing method, system, device and medium for precious metal artware

Abstract

The application discloses a precious metal handicraft adaptive flexible polishing method, system, device and medium. The method comprises the following steps: obtaining the current polishing state data and the workpiece curved surface geometric information of the precious metal handicraft, estimating the dynamic contact area based on the workpiece curved surface geometric information and the deformation amount of the flexible polishing tool; inputting the polishing state data and the dynamic contact area into a pre-constructed mechanism and data mixed driven material removal rate prediction model to output the predicted material removal rate at the current moment; and according to the deviation between the predicted material removal rate and the preset target removal rate, calculating the adjustment amount of the polishing force and the feed speed based on a multi-target control law decoupling to control the flexible polishing device to execute the polishing action. Through the construction of the prediction model combining the advantages of physical mechanism and data driving and the dynamic contact area estimation and multi-target decoupling control, the prediction accuracy and control robustness of the complex curved surface polishing process can be improved, and the processing quality consistency can be ensured.

Description

Technical Field

[0001] This application relates to the fields of precision machining and intelligent manufacturing technology, and in particular to an adaptive flexible polishing method, system, equipment and medium for precious metal crafts. Background Technology

[0002] Precious metal crafts (such as gold, silver, platinum, and their alloys) have extremely high requirements for surface finish due to their high economic value and aesthetic attributes. Polishing, as a key process to improve the smoothness and luster of the craft's surface, directly affects the market value of the final product. However, precious metals themselves have physical properties such as good ductility, low hardness, and high thermal conductivity, and crafts often have complex curved structures (such as openwork or irregular shapes), making traditional rigid polishing processes difficult to adapt.

[0003] In existing technologies, polishing of complex curved surfaces typically employs offline programming combined with constant force control. However, this method often relies on an idealized machining model, neglecting the influence of nonlinear and time-varying factors such as polishing tool wear, batch variations in workpiece material, and changes in contact stiffness during actual machining. Particularly in the polishing of precious metals, due to the softness of the material, even slight miscontrol of the polishing force can cause surface scratches or excessive material removal, leading to irreversible scrap losses; conversely, insufficient polishing force results in low polishing efficiency, failing to meet production cycle time.

[0004] Furthermore, existing polishing control strategies mostly employ linear control algorithms such as proportional-integral-derivative (PID), which are ill-suited for handling flexible polishing systems with strong nonlinear characteristics. Although recent research has introduced artificial intelligence methods such as neural networks for modeling the polishing process, purely data-driven models lacking physical mechanism constraints often lack physical interpretability. They are prone to prediction biases in areas with insufficient training data coverage, and may even violate basic physical laws (such as the material removal mechanism described by the Preston equation), resulting in insufficient robustness of the control system.

[0005] Therefore, how to achieve precise control over the polishing process of complex curved surfaces of precious metal handicrafts is a technical problem that urgently needs to be solved. Summary of the Invention

[0006] This application provides an adaptive flexible polishing method, system, equipment, and medium for precious metal crafts, aiming to solve the problems of low accuracy of polishing process models, poor adaptability of control strategies, and easy damage to precious metal surfaces in the prior art.

[0007] To achieve the above objectives, this application proposes an adaptive flexible polishing method for precious metal crafts. The method includes the following steps: Obtain the current polishing status data of precious metal handicrafts and the workpiece surface geometry information, and estimate the dynamic contact area based on the workpiece surface geometry information and the deformation of the flexible polishing tool; The polishing state data and dynamic contact area are input into a pre-built mechanism and data hybrid driven material removal rate prediction model, which outputs the predicted material removal rate at the current moment. The material removal rate prediction model uses the polishing mechanism equation as its framework and uses a neural network to fit the nonlinear changes of the mechanism coefficients. Based on the deviation between the predicted material removal rate and the preset target removal rate, the adjustment amount of polishing force and feed speed is calculated based on the decoupling of the multi-objective control law. The flexible polishing device is controlled to perform polishing actions based on the polishing force adjustment amount and the feed speed adjustment amount, and the polishing status data is updated in real time.

[0008] In some embodiments, obtaining the current polishing status data of the precious metal craft and the workpiece surface geometry information includes: The three-dimensional contact force signal at the contact point between the polishing tool and the workpiece is acquired in real time using a force sensor. The spindle speed of the polishing tool and the feed rate of the robot's end effector are collected by an encoder. The surface roughness parameters and local surface curvature radii of the workpiece are obtained by scanning with a vision sensor as the geometric information of the workpiece surface.

[0009] In some embodiments, estimating the dynamic contact area based on the workpiece surface geometry and the deformation of the flexible polishing tool includes: Obtain the local surface curvature radius of the workpiece at the polishing trajectory point; Detect the amount of compression deformation at the tip of a flexible polishing tool; By combining the local surface curvature radius and compression deformation, the dynamic contact area between the polishing tool and the workpiece surface is calculated in real time.

[0010] In some embodiments, the process of constructing the material removal rate prediction model includes: A basic removal rate model based on the Preston principle is established, which includes initial coefficients characterizing the relationship between material removal rate and pressure and velocity. A deep neural network model is constructed, taking polishing force, polishing speed, workpiece surface curvature and polishing wheel wear state as inputs, and outputting a dynamic correction factor for the initial coefficients; By fusing the basic removal rate model with the dynamic correction factor, a dynamic prediction model for material removal rate that can reflect the nonlinear influence of process parameters is obtained.

[0011] In some embodiments, the adjustment amount for calculating the polishing force and feed rate based on the decoupling of the multi-objective control law includes: Calculate the error between the predicted material removal rate and the target removal rate, as well as the rate of change of the error; The normal contact force adjustment is calculated using a sliding mode variable structure control algorithm to eliminate the error. By introducing polishing temperature as a constraint, when the predicted polishing temperature exceeds the preset threshold, the feed rate is adjusted first to limit heat accumulation, thereby decoupling the removal rate control from the surface quality control.

[0012] In some embodiments, after outputting the predicted material removal rate at the current moment, an online model correction step is further included: The actual material removal depth on the workpiece surface is collected at a set cycle, and the measured material removal rate is calculated. Calculate the relative error between the measured material removal rate and the predicted material removal rate; When the relative error exceeds a preset threshold, the weight parameters of the neural network model are updated using the error backpropagation algorithm.

[0013] This application also provides an adaptive flexible polishing system for precious metal crafts, the system comprising: The data acquisition and estimation module is used to acquire polishing status data and workpiece surface geometry information, and to estimate the dynamic contact area; The prediction processing module is used to input the polishing state data and dynamic contact area into the mechanism and data hybrid-driven material removal rate prediction model, and output the predicted material removal rate. The control decision module is used to calculate the adjustment amount of polishing force and feed speed based on the deviation between the predicted material removal rate and the target removal rate, and decoupled from the multi-objective control law. The execution drive module is used to drive the flexible polishing device to operate according to the polishing force adjustment amount and the feed speed adjustment amount.

[0014] In some embodiments, the system further includes: The online correction module is used to collect the measured material removal rate and update the parameters of the prediction model online based on the error between the measured value and the predicted value.

[0015] This application also provides an electronic device. The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the adaptive flexible polishing method for precious metal crafts as described above.

[0016] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the adaptive flexible polishing method for precious metal crafts as described above.

[0017] This application proposes an adaptive flexible polishing method, system, equipment, and medium for precious metal crafts. The adaptive flexible polishing method for precious metal crafts includes the following steps: acquiring the current polishing state data and workpiece surface geometry information of the precious metal craft; estimating the dynamic contact area based on the workpiece surface geometry information and the deformation of the flexible polishing tool; inputting the polishing state data and dynamic contact area into a pre-constructed mechanism- and data-driven material removal rate prediction model, outputting the predicted material removal rate at the current moment; the prediction model uses the polishing mechanism equation as its framework and utilizes a neural network to fit the nonlinear changes of the mechanism coefficients; based on the deviation between the predicted material removal rate and the preset target removal rate, calculating the adjustment amounts of polishing force and feed speed based on multi-objective control law decoupling; controlling the flexible polishing device to perform polishing actions based on the polishing force adjustment amount and feed speed adjustment amount, and updating the polishing state data in real time. This application's technical solution has the following beneficial effects: (1) A dynamic contact area estimation mechanism is introduced, which calculates the contact area in real time based on the workpiece surface geometry and the deformation of the flexible polishing tool. This overcomes the defect of the traditional model that treats the contact area as a constant, and can accurately reflect the actual contact state in the polishing process of complex curved surfaces of precious metal crafts, significantly improving the adaptability of the prediction model to complex curved surfaces.

[0018] (2) A mechanism- and data-driven material removal rate prediction model was constructed, using the polishing mechanism equation as the framework and a neural network to fit the nonlinear changes of the mechanism coefficients. This not only preserves the interpretability and generalization ability of the physical model, but also has the ability of the data-driven model to accurately capture nonlinear factors, solving the problem of low prediction accuracy or poor robustness of a single model, and realizing high-precision prediction of material removal rate.

[0019] (3) A multi-objective control law is constructed. Based on the deviation between the predicted material removal rate and the target removal rate, the polishing force adjustment and feed rate adjustment are decoupled and calculated. This achieves coordinated and precise control of polishing force and feed rate, effectively eliminating the limitations of a single control variable, ensuring the consistency of material removal rate during polishing, and improving the stability of the processing process and the quality of the finished product. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort, wherein: Figure 1 This is a schematic flowchart of an embodiment of the adaptive flexible polishing method for precious metal handicrafts according to this application; Figure 2 This is a schematic diagram of a module of an adaptive flexible polishing system for precious metal crafts according to an embodiment of this application; Figure 3 This is a schematic diagram of the hardware structure of a computer device according to an embodiment of this application. Detailed Implementation

[0021] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0022] It should be noted that, unless otherwise stated or limited, all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of this application are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0023] It should also be noted that, unless otherwise stated or limited, when an element is referred to as "fixed to" or "set on" another element, it may be directly on the other element or there may be an intervening element present. When an element is referred to as "connected to" another element, it may be directly connected to the other element or there may be an intervening element present.

[0024] Furthermore, unless otherwise stated or limited, the descriptions involving "first," "second," etc., in this application are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. Additionally, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of a person skilled in the art to implement them. When the combination of technical solutions is contradictory or impossible to implement, such a combination of technical solutions should be considered non-existent and not within the scope of protection claimed in this application.

[0025] A brief description of the technical scenario to which this application is applied is provided. Polishing of precious metal crafts is typically performed on multi-axis robots or CNC machine tools, with a flexible polishing tool mounted on the end effector. Since precious metal workpieces are often irregularly shaped curved surfaces, the contact state between the polishing wheel and the workpiece constantly changes during the polishing process as the robot moves along its trajectory. This application aims to solve the above problems through a predictive control method that integrates mechanism and data.

[0026] Please see Figure 1As shown, the technical solution of this application provides an adaptive flexible polishing method for precious metal handicrafts. This method is executed by an embedded controller or an industrial control computer and mainly includes the following core steps: Step S10: Obtain the current polishing status data of the precious metal craft and the workpiece surface geometry information, and estimate the dynamic contact area based on the workpiece surface geometry information and the deformation of the flexible polishing tool.

[0027] In this step, the system's primary task is to construct a multi-source heterogeneous data fusion sensing field for the polishing process. This is not simply the acquisition of single data, but a process of fusing multi-dimensional data. Specifically, the polishing state data includes two main categories: physical quantities and motion quantities. Physical quantities are primarily acquired through a six-dimensional force sensor installed between the robot's end effector and the polishing tool. This sensor collects the three-dimensional contact force signals at the contact point between the polishing tool and the workpiece in real time at a sampling frequency of no less than 1000Hz, especially the normal contact force, which is the direct basis for calculating the material removal rate. Motion quantities are obtained through encoder feedback from the servo drive, including the spindle speed of the polishing tool and the feed speed of the robot's end effector.

[0028] Meanwhile, acquiring the workpiece's surface geometry is a prerequisite for achieving precise control. Before or during the polishing operation, the system uses a line laser sensor or structured light vision sensor to scan the workpiece surface, acquire point cloud data, and reconstruct the surface. During polishing, the system extracts the local surface curvature radius of the workpiece at the polishing trajectory points in the reconstructed surface model based on the robot's current TCP (tool center point) coordinates. For precious metal crafts, their complex hollow and undulating structures cause drastic changes in the curvature radius, which directly determines the contact pattern between the polishing wheel and the workpiece.

[0029] Estimating the dynamic contact area is a crucial intermediate calculation step. Traditional Hertzian contact theory assumes the contact body is an elastic half-space, but this assumption no longer applies in the interaction between a flexible polishing wheel and a complex curved surface. This embodiment employs a dynamic estimation strategy based on geometric fitting. First, the system reads the local surface curvature radius of the workpiece at the polishing trajectory point and simultaneously detects the compressive deformation at the end of the flexible polishing tool. The compressive deformation can be directly measured by a displacement sensor mounted on the flexible mechanism or indirectly calculated using a force-deformation relationship curve.

[0030] By combining the local surface curvature radius and compressive deformation, the system constructs a local contact geometry model in real time. Assuming the polishing wheel is an elastic cylinder, when it is pressed under a certain pressure onto a surface with a curvature radius of... When the polishing wheel is applied to the workpiece surface, a tiny contact area will form in the contact region. Using the contact deformation formula from elasticity mechanics, the functional relationship between the dynamic contact area and the radius of curvature, the amount of deformation, and the hardness of the polishing wheel can be derived. For example, when the polishing wheel is on a convex surface, the contact area decreases due to geometric divergence; conversely, on a concave surface, the contact area increases due to geometric convergence. The dynamic contact area output in real time in this step will serve as a key input feature for subsequent prediction models, thereby eliminating prediction errors caused by geometric changes.

[0031] To ensure the real-time nature and accuracy of geometric information, a dynamic point cloud update mechanism is introduced when acquiring the workpiece surface geometry. As the polishing process progresses, the line laser sensor periodically scans the processed area to acquire the latest surface topography data. The system uses the ICP (Iterative Closest Point) algorithm to register and fuse the newly acquired point cloud data with the initial model, updating the digital twin model of the workpiece surface in real time. This process corrects geometric deviations caused by initial modeling errors or workpiece clamping and positioning deviations, ensuring the reliability of the radius of curvature calculation results. Simultaneously, to address the laser speckle noise caused by the high reflectivity of precious metal surfaces, the vision processing unit employs an image preprocessing algorithm combining multi-frame averaging and morphological filtering to effectively extract edge contours and ensure the accuracy of geometric feature extraction.

[0032] In the dynamic contact area estimation stage, the system further considers the impact of polishing wheel wear. As polishing continues, the diameter of the polishing wheel decreases due to wear, and its surface morphology also changes. The system establishes a polishing wheel wear model and estimates the change in the effective radius of the polishing wheel based on the cumulative polishing time or motor current load characteristics. When calculating the contact area, a wear correction factor is introduced to correct the theoretical radius in real time, thereby ensuring the accuracy of the contact area estimation model throughout its entire lifespan.

[0033] Step S20: Input the polishing state data and dynamic contact area into the pre-built mechanism and data hybrid driven material removal rate prediction model, and output the predicted material removal rate at the current moment; the material removal rate prediction model uses the polishing mechanism equation as the skeleton and uses a neural network to fit the nonlinear changes of the mechanism coefficients.

[0034] Traditional material removal rate prediction is typically based on the Preston equation, which states that the material removal rate is proportional to the normal pressure and relative velocity. However, in actual precious metal polishing processes, the Preston coefficient is not constant; it undergoes complex nonlinear changes with factors such as polishing temperature, polishing wheel wear, workpiece surface roughness, and work hardening of the precious metal material. Using a Preston equation with fixed coefficients alone results in extremely large prediction errors. While using neural networks (such as backpropagation neural networks) can fit the nonlinear relationship, it often lacks physical constraints and is prone to producing predictions that violate physical laws in areas with scarce data.

[0035] To this end, this application constructs a prediction model driven by a hybrid mechanism and data approach. This model does not abandon the classic physical framework of the Preston equation, but rather uses it as its foundation. The model first establishes a basic removal rate model based on Preston's principle, introducing initial coefficients... Based on this, the system constructs a deep neural network model. This network does not directly output the material removal rate, but instead outputs the result based on the initial coefficients. Dynamic correction factor .

[0036] The input layer of the deep neural network receives a multi-dimensional feature vector, including polishing force, polishing speed, dynamic contact area (calculated in step one), workpiece surface curvature, polishing wheel wear state (extracted through current or vibration signal features), and ambient temperature. The network structure employs a hybrid of multi-layer fully connected layers and a Long Short-Term Memory (LSTM) network, where the LSTM layers capture temporal dependencies in the polishing process, such as the cumulative effect of polishing wheel wear. After feature extraction and nonlinear mapping, the output layer outputs a correction factor. .

[0037] Finally, the predicted material removal rate output by the prediction model is determined by the following formula: ,(1); In equation (1), To predict the material removal rate, The Preston coefficient is a benchmark based on historical data statistics. This is the dynamic correction factor output by the neural network. For normal polishing force, For dynamic contact area, The relative velocity between the polishing wheel and the workpiece is given. This architectural design allows the model to possess both the structural clarity of a physical model and the adaptive capabilities of a data-driven system. For example, when wear on the polishing wheel leads to a decrease in cutting capability, the neural network outputs a negative correction factor, thereby accurately predicting the decreasing trend of material removal rate.

[0038] To further improve the generalization ability and accuracy of the prediction model, this application deeply optimizes the structure of the deep neural network. Specifically, establishing a basic material removal rate model based on the Preston principle is the cornerstone of the entire prediction framework. The classic Preston equation states that the material removal rate is proportional to the product of polishing pressure and relative velocity. However, in the actual polishing scenario of precious metal crafts, this proportionality coefficient (i.e., the Preston coefficient) is not a constant value. It is affected by multiple factors such as changes in the exposed state of abrasive grains due to polishing wheel wear, the work hardening effect of the precious metal surface, uneven distribution of polishing slurry concentration, and fluctuations in ambient temperature as the polishing process progresses, exhibiting highly nonlinear and time-varying characteristics.

[0039] To accurately capture this nonlinear variation, this application constructs a deep neural network model to fit the dynamic evolution of the Preston coefficient. This deep neural network model employs a multi-layer feedforward neural network structure, including an input layer, multiple hidden layers, and an output layer. The feature vectors of the input layer are carefully designed to cover all key variables affecting the polishing process, specifically including: real-time polishing force obtained through a force sensor, spindle speed and feed rate read by an encoder, dynamic contact area obtained through geometric calculation, polishing wheel wear state index perceived by visual or acoustic emission sensors, and environmental temperature and humidity data. To improve the model's generalization ability and convergence speed, all input data undergoes normalization preprocessing before being fed into the network, mapping them to a specific numerical range.

[0040] The hidden layers of deep neural networks employ a combination of fully connected layers and activation functions (such as ReLU or Sigmoid) to extract higher-order nonlinear features between input features. Specifically, to account for the cumulative effects of the polishing process over time (e.g., polishing wheel wear is a gradual process), Long Short-Term Memory (LSTM) units or Gated Recurrent Units (GRUs) are introduced into the network structure, enabling the model to remember historical states and thus more accurately predict the current Preston coefficient correction factor.

[0041] The output layer of the neural network does not directly output the material removal rate, but instead outputs a dynamic correction factor for the baseline Preston coefficient. The system presets an average Preston coefficient, obtained from a large amount of historical experimental data, as a benchmark. The dynamic correction factor output by the neural network acts as an adjustment term, reflecting the degree of deviation of the current instantaneous operating condition from the average operating condition. Through this physically guided network structure design, even in areas with limited training data, the model will not produce predictions that defy common sense due to the physical support of the benchmark coefficient, significantly enhancing the system's safety and robustness.

[0042] The model training process employs a strategy combining offline training and online fine-tuning. In the offline phase, a large amount of sample data containing different combinations of process parameters, different workpiece surface features, and corresponding measured material removal rates are collected to construct a training dataset. Using the backpropagation algorithm, the mean squared error (MSE) between the predicted and measured material removal rates is used as the loss function to iteratively optimize the network weights. Simultaneously, a regularization term is added to the loss function to prevent overfitting. Furthermore, the system incorporates the concept of a Generative Adversarial Network (GAN) to expand the training set by generating simulated operating condition data, further improving the model's predictive stability under extreme conditions.

[0043] Step S30: Based on the deviation between the predicted material removal rate and the preset target removal rate, calculate the adjustment amount of polishing force and feed speed based on the decoupling of the multi-objective control law.

[0044] After obtaining the predicted material removal rate, the system enters the closed-loop control phase. The control objective is to make the actual material removal rate track the preset target removal rate trajectory. Since the material removal rate is affected by both polishing force and feed rate, and the two are coupled in physical mechanism, simple single-variable control is difficult to achieve the desired effect. This application is based on a multi-objective control law and adopts a decoupling control strategy.

[0045] First, the error and rate of change between the predicted material removal rate and the target removal rate are calculated. In the control loop, to improve the system's response speed and anti-interference capability, especially addressing modeling errors and external disturbances (such as workpiece positioning deviations) during polishing, a sliding mode variable structure control algorithm is used to calculate the normal contact force adjustment. Sliding mode control, through the design of a switching function, forces the system state to reach the sliding surface within a finite time and move along it, thereby ensuring the system's robustness.

[0046] For precious metal polishing processes, which require extremely high material removal consistency while avoiding overheating damage, traditional single-input single-output (SISO) control structures are no longer sufficient to meet control accuracy requirements. This application proposes a decoupled control strategy based on error-driven and constraint-priority approaches.

[0047] First, the core task of the control system is to eliminate the deviation between the predicted material removal rate and the target removal rate. The system calculates the error value between the predicted and target material removal rates in real time, and also calculates the rate of change of this error value over time. These two variables constitute the input state space of the controller. To address the nonlinear friction, unmodeled dynamics, and external disturbances present in the flexible polishing system, this application employs a sliding mode variable structure control algorithm to design the normal contact force control channel. Sliding mode control uses a switching function to dynamically switch the gain coefficient of the control law based on the deviation of the system state and its rate of change. When the system state is far from the sliding surface, the control law applies a large control gain, causing the system to move rapidly towards the target state; when the system state is close to the sliding surface, the control law switches to a hold state to ensure system stability. This control method not only ensures system convergence within a finite time but also exhibits strong robustness against parameter perturbations and external disturbances. The control quantity calculated by the sliding mode algorithm, after conversion, directly drives the force-controlled actuator (such as a pneumatic proportional valve or torque motor) to achieve precise adjustment of the normal contact force.

[0048] However, controlling material removal rate solely by adjusting polishing force has limitations. In polishing high-curvature areas or fine-textured regions of precious metal artifacts, excessive polishing force can lead to workpiece deformation or surface indentation. Therefore, the control strategy must incorporate feed rate as an auxiliary control variable, forming a dual-channel collaborative control. This involves decoupling and temperature constraint mechanisms. The heat generated during polishing primarily originates from frictional work and is closely related to the product of pressure and velocity. Excessively high local temperatures can cause oxidation, discoloration, and even thermal stress cracking on the precious metal surface. Therefore, the system introduces polishing temperature as a constraint. In the control loop, an infrared thermal imager is used to monitor the polishing zone temperature in real time, or a heat transfer model is used to estimate the polishing zone temperature. When the predicted polishing temperature does not exceed a preset threshold (such as the critical temperature before oxidation and discoloration of the precious metal), the control strategy prioritizes material removal rate control, maintaining the feed rate along a preset trajectory, and only adjusting the polishing force to track the target removal rate.

[0049] Once the predicted polishing temperature is detected to exceed the preset threshold, the control system immediately activates a protection strategy, prioritizing adjustments to the feed rate. At this point, the control law calculates a speed decay factor, reducing the feed rate of the robot's end effector. This reduction in feed rate directly decreases the frictional travel per unit time, effectively suppressing heat accumulation. To compensate for the decrease in material removal rate caused by the reduced speed, the control law simultaneously fine-tunes the polishing force, or accepts temporary fluctuations in material removal rate in exchange for maintaining surface quality. This method of dynamically switching control weights based on temperature constraints decouples material removal rate control from surface quality control, ensuring that while pursuing processing efficiency, product surface quality is never sacrificed.

[0050] In the specific construction of the control law, this embodiment further introduces an adaptive gain adjustment mechanism. Considering the differences in control performance requirements at different polishing stages (rough polishing, fine polishing, and mirror polishing), the system dynamically adjusts the reaching law parameters of the sliding mode controller according to the current polishing stage flag. For example, in the rough polishing stage, focusing on processing efficiency, the system sets a larger control gain to quickly eliminate errors; in the fine polishing stage, focusing on surface quality, the system reduces the control gain to suppress high-frequency jitter of the control signal and ensure the smoothness of the polishing process. Simultaneously, addressing the inherent nonlinear hysteresis characteristics of pneumatic muscles, an inverse model compensator is designed in the force control drive unit. This inverse model, based on Support Vector Machine (SVM) or lookup table method, can calculate the accurate control voltage from the target force value, effectively overcoming the hysteresis effect of the pneumatic system and significantly improving the response bandwidth and tracking accuracy of the force control system.

[0051] Step S40: Control the flexible polishing device to perform polishing actions according to the polishing force adjustment amount and the feed speed adjustment amount, and update the polishing status data in real time.

[0052] After the control commands are generated, the system converts the adjustment amounts into control signals to drive the actuators. The polishing force adjustment amount is achieved by controlling the inflation pressure of the pneumatic muscle through a proportional valve, or by controlling the spring preload through a servo motor, thereby changing the normal contact stiffness and contact force at the end of the flexible polishing tool. The feed rate adjustment amount is sent directly to the robot controller to modify the TCP's movement speed in real time.

[0053] At the command execution level, simple speed control is insufficient to overcome the dynamic defects of the mechanical transmission chain. This embodiment introduces an inertia compensation step for the actuator. The system calculates the inertial torque loss during acceleration and deceleration based on the actuator's mechanical transmission parameters (such as moment of inertia and stiffness coefficient). At the moment of initiation, the system applies an additional positive inertia compensation current pulse to the motor to preemptively compensate for the torque required for acceleration; at the moment of braking, a reverse compensation current is applied to eliminate inertia overshoot. This feedforward inertia compensation counteracts the elastic deformation of the mechanical transmission chain, ensuring accurate tracking of the command by the actual speed and eliminating dynamic lag errors at start and stop.

[0054] As polishing progresses, the system continues to run the above cycle, updating the polishing status data in real time, forming a dynamic perception-prediction-control closed loop until the polishing task of the entire workpiece is completed.

[0055] Furthermore, the system integrates an active damping control module to address potential vibrations from the high-speed rotation of flexible polishing tools. An accelerometer mounted on the polishing spindle monitors the vibration spectrum in real time. When self-excited vibrations at a specific frequency are detected, the system generates a counter-damping force by adjusting the pressure fluctuations of the pneumatic muscles or by adjusting the spindle speed to avoid the resonance zone, thereby eliminating vibration marks and ensuring a mirror-like finish on the precious metal surface. Simultaneously, the system monitors the polishing slurry supply in real time, using a flow sensor to control the opening and closing of the nozzles, ensuring uniform coverage of the processing area with the polishing slurry and preventing surface burns due to insufficient lubrication.

[0056] Considering the complexity of actual production environments, such as slight differences in hardness due to different batches of precious metal raw materials, or changes in the elastic modulus of the polishing wheel after long-term use, these factors can cause prediction drift in the offline model during long-term operation. Therefore, after step S40, the system also includes an online model correction step, giving the model the ability to self-evolve during operation. This mechanism is triggered at set intervals (e.g., every 10 seconds or after processing each feature segment) while the polishing process is underway.

[0057] First, using a high-precision laser displacement sensor or vision measurement system, in-situ measurements are performed on the surface of the freshly polished workpiece to obtain actual material removal depth data. Combined with the processing time, the measured material removal rate is calculated.

[0058] Subsequently, the system calculates the relative error between the measured material removal rate and the model-predicted material removal rate. This relative error directly reflects the current deviation of the model. If the relative error is within the allowable tolerance range (e.g., less than 5%), it indicates that the model is still accurate and no correction is needed; if the relative error exceeds a preset threshold, the error backpropagation algorithm is triggered.

[0059] During online correction, to avoid system control oscillations caused by drastic modifications to network weights, a gradient descent algorithm with a small learning rate is employed. Only the weight parameters of the last few fully connected layers of the neural network are fine-tuned, while the parameters of the front-end network layers extracting general features remain unchanged. This gradual learning approach allows the model to quickly adapt to the material properties of the current workpiece while preserving the knowledge base accumulated during offline training. The corrected model weight parameters are stored in real-time in the controller's non-volatile memory, serving as the basis for subsequent processing predictions. Through this online correction closed loop, the prediction model can consistently follow the actual state of the polishing system, forming an intelligent closed-loop control ecosystem of processing, learning, and correction simultaneously, greatly enhancing the system's adaptability to different processing tasks.

[0060] Furthermore, this embodiment also provides an adaptive flexible polishing system 100 for precious metal crafts, used to implement the methods described in the above embodiments. Please refer to... Figure 2 The system relies on industrial robots and flexible polishing workstations in terms of hardware architecture, and is divided into a data acquisition and estimation module 110, a prediction processing module 120, a control decision module 130, and an execution drive module 140 in terms of software architecture.

[0061] The data acquisition and estimation module 110, serving as the system's perception center, is responsible for multi-source information fusion. Its hardware interface connects to a six-dimensional force sensor, a servo driver encoder interface, a laser vision sensor, and an infrared thermometer. This module integrates data filtering algorithms (such as Kalman filtering) and geometric calculation algorithms. When the robot's end effector drives the polishing tool along a predetermined trajectory, this module reads the three-dimensional contact force signal in real time and acquires point cloud data of the workpiece surface from the vision system via TCP / IP or fieldbus. Based on the point cloud data, the module's internal geometric engine calculates the local surface curvature radius of the current trajectory point in real time; combining the force signal with the stiffness model of the flexible mechanism, it estimates the compression deformation, and finally calculates the dynamic contact area through the geometric model. All the above data is then packaged into a standard feature vector and sent to the prediction processing module.

[0062] The prediction processing module 120 is the core computing unit of the system. This module contains a trained mechanism- and data-driven prediction model. After receiving the feature vector, the module performs forward computation according to the topology defined by the model. The model first extracts information such as polishing force, velocity, and curvature from the feature vector, inputs it into the deep neural network submodule, and calculates the Preston coefficient correction factor. Subsequently, the correction factor is fused with the basic Preston coefficient, combined with the contact area parameter, to output the predicted material removal rate and predicted polishing temperature at the current moment. This prediction result serves as a key state variable and is input into the control decision module.

[0063] The control decision module 130 is responsible for generating the control strategy. The module internally presets the target removal rate curve and temperature threshold parameters. This module receives the output from the prediction processing module 120 in real time and calculates the error. Internally, the module runs a sliding mode variable structure control algorithm and multi-objective decision logic. Upon receiving the prediction result, the algorithm module immediately calculates the required control inputs, namely the polishing force adjustment and feed rate adjustment. During the calculation process, the module monitors the predicted temperature value in real time; once the threshold is reached, the constraint logic is immediately activated to adjust the weight of the speed command and generate the final control command word.

[0064] The execution drive module 140, acting as the system's actuator, translates control commands into physical actions. This module includes a force control drive unit and a motion control drive unit. The force control drive unit controls an electro-proportional valve via an analog output interface to adjust the air intake pressure of the pneumatic muscle, thereby changing the normal contact force at the end of the flexible polishing wheel to achieve compliant force-controlled polishing. The motion control drive unit modifies the robot's motion parameters and adjusts the feed speed via a robot controller interface. Both work together to ensure the polishing tool always applies optimally to the surface of the precious metal craft. The system also includes an online correction module, which operates independently of the real-time control loop to avoid the computationally intensive backpropagation process affecting the real-time performance of the control. This module monitors prediction errors in the background and periodically updates the model parameter file in the prediction processing module 120, enabling continuous system optimization.

[0065] Please see Figure 3 This application also provides a computer device. Figure 3 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of this application, such as... Figure 3 As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processor can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces).

[0066] In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories, if needed. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 3 Take a processor 10 as an example.

[0067] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0068] The memory 20 stores instructions executable by at least one processor 10 to cause the at least one processor 10 to perform the method shown in the above embodiments.

[0069] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0070] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0071] The computer device also includes a communication interface 30 for communicating with other devices or communication networks.

[0072] This application also provides a computer-readable storage medium. The methods described in this application can be implemented in hardware or firmware, or implemented as recordable on a storage medium, or implemented as computer code downloaded over a network, originally stored on a remote storage medium or a non-transitory machine-readable storage medium, and subsequently stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium may also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods shown in the above embodiments.

[0073] In summary, the technical solution of this application solves the problems of poor consistency and surface damage in the polishing process of precious metal crafts by organically combining a prediction model driven by mechanism and data, dynamic contact area estimation, multi-objective decoupling control and online correction mechanism, and significantly improves the processing quality and automation level.

[0074] The above description is only a part or preferred embodiment of this application. Neither the text nor the drawings should limit the scope of protection of this application. All equivalent structural transformations made using the content of this application's specification and drawings under the overall concept of this application, or direct / indirect applications in other related technical fields, are included within the scope of protection of this application.

Claims

1. A self-adapting flexible polishing method for precious metal artifacts, characterized by, Includes the following steps: Obtain the current polishing status data of precious metal handicrafts and the workpiece surface geometry information, and estimate the dynamic contact area based on the workpiece surface geometry information and the deformation of the flexible polishing tool; The polishing state data and dynamic contact area are input into a pre-built mechanism and data hybrid driven material removal rate prediction model, which outputs the predicted material removal rate at the current moment. The material removal rate prediction model uses the polishing mechanism equation as its framework and uses a neural network to fit the nonlinear changes of the mechanism coefficients. Based on the deviation between the predicted material removal rate and the preset target removal rate, the adjustment amount of polishing force and feed speed is calculated based on the decoupling of the multi-objective control law. The flexible polishing device is controlled to perform polishing actions based on the polishing force adjustment amount and the feed speed adjustment amount, and the polishing status data is updated in real time.

2. The noble metal artefact self-adapting flexible polishing method according to claim 1, characterized in that, The process of obtaining the current polishing status data and workpiece surface geometry information of the precious metal crafts includes: The three-dimensional contact force signal at the contact point between the polishing tool and the workpiece is acquired in real time using a force sensor. The spindle speed of the polishing tool and the feed rate of the robot's end effector are collected by an encoder. The surface roughness parameters and local surface curvature radii of the workpiece are obtained by scanning with a vision sensor as the geometric information of the workpiece surface.

3. The noble metal art piece adaptive flexible polishing method of claim 2, wherein, The estimation of dynamic contact area based on the workpiece surface geometry and the deformation of the flexible polishing tool includes: Obtain the local surface curvature radius of the workpiece at the polishing trajectory point; Detect the amount of compression deformation at the tip of a flexible polishing tool; By combining the local surface curvature radius and compression deformation, the dynamic contact area between the polishing tool and the workpiece surface is calculated in real time.

4. The precious metal art piece adaptive flexible polishing method of claim 1, wherein, The construction process of the material removal rate prediction model includes: A basic removal rate model based on the Preston principle is established, which includes initial coefficients characterizing the relationship between material removal rate and pressure and velocity. A deep neural network model is constructed, taking polishing force, polishing speed, workpiece surface curvature and polishing wheel wear state as inputs, and outputting a dynamic correction factor for the initial coefficients; By fusing the basic removal rate model with the dynamic correction factor, a dynamic prediction model for material removal rate that can reflect the nonlinear influence of process parameters is obtained.

5. The adaptive flexible polishing method for precious metal handicrafts according to claim 1, characterized in that, The adjustment amounts for polishing force and feed rate calculated based on multi-objective control law decoupling include: Calculate the error between the predicted material removal rate and the target removal rate, as well as the rate of change of the error; The normal contact force adjustment is calculated using a sliding mode variable structure control algorithm to eliminate the aforementioned error; By introducing polishing temperature as a constraint, when the predicted polishing temperature exceeds the preset threshold, the feed rate is adjusted first to limit heat accumulation, thereby decoupling the removal rate control from the surface quality control.

6. The adaptive flexible polishing method for precious metal crafts according to claim 4, characterized in that, After outputting the predicted material removal rate at the current moment, an online model correction step is also included: The actual material removal depth on the workpiece surface is collected at a set cycle, and the measured material removal rate is calculated. Calculate the relative error between the measured material removal rate and the predicted material removal rate; When the relative error exceeds a preset threshold, the weight parameters of the neural network model are updated using the error backpropagation algorithm.

7. An adaptive flexible polishing system for precious metal handicrafts, characterized in that, include: The data acquisition and estimation module is used to acquire polishing status data and workpiece surface geometry information, and to estimate the dynamic contact area; The prediction processing module is used to input the polishing state data and dynamic contact area into the mechanism and data hybrid-driven material removal rate prediction model, and output the predicted material removal rate. The control decision module is used to calculate the adjustment amount of polishing force and feed speed based on the deviation between the predicted material removal rate and the target removal rate, and decoupled from the multi-objective control law. The execution drive module is used to drive the flexible polishing device to operate according to the polishing force adjustment amount and the feed speed adjustment amount.

8. The adaptive flexible polishing control system for precious metal crafts according to claim 7, characterized in that, The system also includes: The online correction module is used to collect the measured material removal rate and update the parameters of the prediction model online based on the error between the measured value and the predicted value.

9. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the adaptive flexible polishing method for precious metal artifacts as described in any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the adaptive flexible polishing method for precious metal crafts as described in any one of claims 1 to 6.

Images