A control system for a face grinder

Abstract

The application discloses an end face grinding machine control system and relates to the technical field of numerical control systems.The end face grinding machine control system comprises a data preparation and recording module, a tool type parameter dynamic modeling module, a multi-axis linkage control module, a grinding wheel pressure self-adaptive control module, a grinding wheel wear compensation module and a process parameter optimization module; multi-dimensional historical processing data is comprehensively collected and a set framework is constructed to realize automatic parameter classification storage; a NURBS curve interpolation algorithm is adopted to guarantee the processing precision and smoothness of complex shapes; real-time synchronous instructions are compensated for errors, vibration is monitored in combination with an acoustic emission sensor to improve processing stability; a fuzzy PID model and machine learning are used to dynamically adjust parameters and optimize in advance; machine vision and digital twinning are used to accurately identify wear and predict a dressing cycle; based on a hybrid model and reinforcement learning, a processing result is predicted and parameters are dynamically adjusted to perform error compensation and sensitivity analysis.

Description

Technical Field

[0001] This invention relates to the field of CNC system technology, specifically to a control system for an end face grinding machine. Background Technology

[0002] A face grinding machine is a specialized piece of machinery used for precision grinding of the end faces of cutting tools. It uses a high-speed rotating grinding wheel to precisely grind the tool's end face, effectively removing burrs, correcting flatness, and improving the tool's cutting performance and lifespan. This equipment is typically equipped with adjustable grinding parameters and a precise positioning system, adaptable to the processing needs of tools of different specifications. It is widely used in industries such as tool manufacturing, machining, and mold making, and is one of the key pieces of equipment for ensuring tool quality and improving production efficiency.

[0003] Currently, CNC end-face grinding machines require different programs to be edited for different cutting tools. Manually editing these programs is difficult and requires a certain level of skill from the operator. Publication number CN102662354A discloses a control system for an end-face grinding machine, including an industrial computer, a motion control card, and a terminal board. The industrial computer and motion control card are connected via a PCI interface, and the motion control card and terminal board are connected via a SCSI interface. The PCI interface is a 32-bit PCI bus interface. It features automatic graphical programming, capable of automatically generating machining programs based on the shape characteristics of the cutting tools and scissors, combined with simple process parameter settings. This eliminates the need for manual programming by the user, making it easy to learn and use.

[0004] The above technologies can achieve the function of automatically writing machining programs. However, in the actual machining process, it is not enough to simply edit the program. There are also various problems that exist in the equipment itself, such as errors caused by the wear of the grinding wheel, the influence of different material tools on the grinding wheel, and the vibration error of the equipment. All of these will cause errors in actual production. This still requires manual trial and error to adjust the program. Moreover, as batch processing progresses, frequent adjustments are required, which is still quite troublesome. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a control system for an end face grinding machine, which solves the problems.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a control system for an end face grinding machine, comprising:

[0007] Data preparation and recording module: comprehensively collects historical processing data, including tool parameters, material properties, environmental and equipment status, processing results, and processing error correlation data of grinding wheels under different working conditions;

[0008] Dynamic modeling module for tool shape parameters: Input tool shape parameters and generate discretized point clouds. Use NURBS curve interpolation algorithm to adjust the density of interpolation points to ensure the accuracy and smoothness of machining complex shapes;

[0009] Multi-axis linkage control module: Real-time synchronization of five-axis position commands via EtherCAT to compensate for mechanical transmission errors; combined with acoustic emission sensors to monitor vibration, automatically switching to conservative mode when the threshold is exceeded;

[0010] The grinding wheel pressure adaptive control module uses a triaxial force sensor to monitor the cutting force in real time, constructs a fuzzy PID model to dynamically adjust the gain, and uses machine learning to predict the influence of material hardness, combined with force threshold triggering for rapid compensation, to optimize parameters in advance.

[0011] Grinding wheel wear compensation module: It uses machine vision to identify wear areas, calculates wear amount and triggers automatic dressing; it predicts dressing cycle through digital twin model to reduce downtime;

[0012] Process parameter optimization module: Based on historical data, a hybrid model of XGBoost and neural network is trained to predict processing results and errors; parameters are dynamically adjusted through reinforcement learning, and error compensation and parameter sensitivity analysis are performed.

[0013] Preferably, the processing error correlation data includes:

[0014] Influence of workpiece hardness: Record the cutting force fluctuation, vibration amplitude, and final machining error of the grinding wheel when machining workpieces with different hardness.

[0015] Influence of grinding wheel speed and feed rate: Cutting temperature, grinding wheel wear rate and machined surface quality were collected under different combinations of speed and feed rate.

[0016] Effect of grinding wheel hardness: Record the cutting force distribution, grinding wheel life and machining error when machining the same material with different grinding wheel hardness;

[0017] Dynamic error compensation data: Real-time errors during the processing are collected using a laser interferometer or a coordinate measuring machine and correlated with the corresponding combination of process parameters.

[0018] Preferably, the data preparation and recording module collects historical processing data, clarifies the categories of each parameter, and constructs a set framework, setting each type of data as an independent set; when new data is input into the system, data identification and classification rules are set, and the data is automatically identified and divided into the corresponding pre-set sets based on parameter characteristics.

[0019] Preferably, the specific working content of the dynamic modeling module for the blade shape parameters includes:

[0020] (1) Parametric input: Supports importing CAD models or directly inputting tool parameters through the HMI interface, including cutting edge angle, radius of curvature, and step height. The system automatically generates discretized point cloud data of the tool profile.

[0021] (2) Dynamic weighted interpolation: The NURBS curve interpolation algorithm is adopted to adjust the density of interpolation points in real time according to the radius of curvature of the processing area. The larger the curvature, the denser the interpolation points, ensuring smooth movement of the grinding wheel when processing complex shapes.

[0022] Preferably, the multi-axis linkage control module synchronously updates the five-axis position commands of X / Y / Z / A / C axes at a 1ms cycle via EtherCAT real-time Ethernet to compensate for mechanical transmission errors; the arc-shaped blade is machined on a standard specimen, and the deviation between the actual contour and the theoretical contour is verified using a laser interferometer. The commands are adjusted according to the results to ensure that the error is ≤2μm.

[0023] The multi-axis linkage control module also features a conservative machining mode and incorporates an acoustic emission sensor to monitor machining vibration signals. When the vibration amplitude exceeds a threshold, it automatically switches to the conservative machining mode to prevent the grinding wheel from chipping.

[0024] Preferably, the working content of the grinding wheel pressure adaptive control module specifically includes:

[0025] (1) Multimodal force sensing system and dynamic PID gain adjustment: A three-dimensional force sensor is installed on the grinding wheel spindle to monitor the cutting force components Fx, Fy and Fz in real time. A multimodal fuzzy PID control model is constructed by combining the grinding wheel speed and material hardness parameters, and the PID gain is dynamically adjusted.

[0026] (2) Force threshold trigger compensation: Set the force threshold and complete the parameter adjustment within 10ms through the PLC high-speed counting module;

[0027] When processing materials with different hardness, the fluctuation range of surface roughness Ra value is statistically analyzed to ensure that Ra≤0.15μm;

[0028] (3) Material hardness prediction and pre-compensation: By combining historical machining error correlation data with machine learning models, the influence of material hardness on cutting force is predicted, and control parameters, including grinding wheel speed and feed rate, are adjusted in advance to reduce the number of dynamic adjustments during the machining process.

[0029] Preferably, the method of constructing a multimodal fusion fuzzy PID control model and dynamically adjusting the PID gain specifically includes:

[0030] (1) Definition of input variables

[0031] Cutting force components: Fx, Fy, and Fz, collected by the triaxial force sensor, represent cutting forces in different directions, respectively;

[0032] Grinding wheel speed: The speed at which the grinding wheel rotates, denoted by n;

[0033] Material hardness: The hardness value of the workpiece material, represented by H;

[0034] (2) Fuzzification: The input variables are fuzzified and divided into different fuzzy sets, with each fuzzy set corresponding to a linguistic variable;

[0035] The membership function maps specific input values ​​to the membership degrees of various fuzzy sets.

[0036] (3) Establishment of fuzzy rule base: Based on expert experience or actual processing experiments, a series of fuzzy rules are established to infer the PID gain based on the fuzzy values ​​of the input variables, namely the adjustment amount of proportional gain Kp, integral gain Ki, and derivative gain Kd.

[0037] (4) Fuzzy reasoning: Fuzzy reasoning is performed based on the fuzzy values ​​of the input variables and the fuzzy rule base; for each rule, the membership degree of the rule conclusion is obtained through fuzzy implication operation based on the membership degree of the input variables to the rule premises; then, the conclusions of all rules are fuzzy synthesized to obtain the fuzzy output of the PID gain adjustment.

[0038] (5) Defuzzification: The centroid method is used to convert the fuzzy output into a specific PID gain adjustment amount, that is, the value corresponding to the centroid of the membership function of the fuzzy output is calculated as the result of defuzzification;

[0039] (6) PID gain adjustment: Add the PID gain adjustment obtained from defuzzification to the current PID gain value to obtain a new PID gain value:

[0040] Kp new =Kp+ΔKp;

[0041] Ki new =Ki+ΔKi;

[0042] Kd new =Kd+ΔKd;

[0043] Where Kp, Ki, and Kd are the current PID gain values, Kp new Ki new Kd new This is the adjusted new PID gain value.

[0044] Preferably, the specific working content of the grinding wheel wear compensation module includes:

[0045] (1) High-precision image acquisition and processing: An industrial camera and a blue ring light source are installed on the side of the grinding wheel to acquire images of the grinding wheel end face at a frequency of 50fps; a deep learning semantic segmentation model is used to identify the wear area of ​​the grinding wheel and calculate the wear area ratio;

[0046] (2) Dynamic calculation and adjustment of wear amount: The wear amount ΔD=D is calculated by fitting the center coordinates and radius of the circle through Hough transform. 初始 -D 实时 ;D 初始 D is the initial diameter of the grinding wheel. 实时 Given the current diameter of the grinding wheel, if ΔD > 0.1 mm, trigger the grinding wheel dressing program and dynamically adjust the dressing path according to the wear area distribution;

[0047] (3) Closed-loop verification of repair effect: After repair, the image is collected again and the difference in wear before and after repair is calculated. If the difference is <0.05mm, the repair is deemed qualified; otherwise, the repair is repeated.

[0048] After continuously machining 100 cutting tools, the edge size distribution is measured to ensure that CPK ≥ 1.67; CPK is the process capability index, which is a quantitative indicator that measures the ability of the production process to produce products that meet specifications under stable conditions.

[0049] (4) Digital twin simulation optimization: The grinding wheel wear process is simulated by digital twin model to predict the dressing cycle and reduce downtime.

[0050] Preferably, the specific functions of the process parameter optimization module include:

[0051] (1) Hybrid model training and prediction: The multidimensional data is trained using a hybrid model of XGBoost and neural network. Inputs include tool shape parameters, material properties, environmental parameters, equipment status parameters, and historical error data; outputs the predicted value of the processing result and the predicted value of the real-time error.

[0052] The hybrid model is trained using at least 2000 sets of data, with a validation set R² ≥ 0.98. R² represents the proportion of the dependent variable variation explained by the regression model to the total variation. An attention mechanism is introduced to strengthen the weight of key parameters on the error.

[0053] (2) Enhance dynamic adjustment of learning, including:

[0054] Recommended initial parameters: After inputting the new tool shape and material parameters in the HMI interface, the model outputs initial process parameters based on historical data;

[0055] Dynamic adjustment: Real-time data is collected every 10 seconds during the processing, and parameters are fine-tuned through reinforcement learning algorithms to ensure processing accuracy and efficiency;

[0056] If the prediction error shows that the runout of a certain axis exceeds the standard, the feed speed of the corresponding axis will be automatically reduced by 10% and the grinding wheel angle will be finely adjusted.

[0057] If the cutting force fluctuation exceeds the threshold, the grinding wheel speed is dynamically adjusted to stabilize the cutting process.

[0058] Error compensation: Based on the real-time error prediction value, the five-axis interpolation trajectory is directly corrected through the inverse kinematics algorithm to ensure machining accuracy;

[0059] (3) Parameter sensitivity analysis and visualization: Quantify the influence of each parameter on the processing error by using SHAP value; visualize the parameter sensitivity heat map in the HMI interface to guide engineers to prioritize the optimization of high-impact parameters.

[0060] Preferably, in the hybrid model training and prediction step, a weighted average method is used to fuse the prediction results of the XGBoost model and the neural network model. The performance of the two models is evaluated on the validation set, and their respective weights are determined based on their performance. For new input data, it is input into the trained XGBoost model and the neural network model respectively to obtain the prediction results of the two models. Then, according to the predetermined weights, the prediction results of the two models are weighted and averaged to obtain the final processing result prediction value of the hybrid model.

[0061] During the processing, the current tool shape parameters, material properties, environmental parameters, and equipment status parameters are collected in real time. The real-time collected data is preprocessed in the same way as historical data, and then input into the trained hybrid model to obtain the real-time processing result prediction value. The real-time prediction value is compared with the actual processing result to calculate the real-time error prediction value, which is the difference between the prediction value and the actual value. The error prediction value is used for subsequent error compensation and dynamic adjustment of process parameters.

[0062] This invention provides a control system for an end-face grinding machine. Compared with the prior art, it has the following advantages:

[0063] 1. This end-face grinding machine control system comprehensively collects multi-dimensional historical processing data and constructs an aggregate framework to achieve automatic parameter classification and storage, providing a solid foundation for subsequent analysis. The tool shape parameter dynamic modeling module adopts the NURBS curve interpolation algorithm to ensure the processing accuracy and smoothness of complex shapes. The multi-axis linkage control module synchronizes commands in real time and compensates for errors, combined with acoustic emission sensors to monitor vibration and improve processing stability. The grinding wheel pressure adaptive control module uses a fuzzy PID model and machine learning to dynamically adjust parameters and optimize in advance. The grinding wheel wear compensation module accurately identifies wear and predicts dressing cycles through machine vision and digital twins. The process parameter optimization module, based on hybrid models and reinforcement learning, predicts processing results and dynamically adjusts parameters, performing error compensation and sensitivity analysis. The cooperation of multiple modules comprehensively improves processing accuracy, efficiency, and stability, and reduces quality costs.

[0064] 2. The control system of this end-face grinding machine supports CAD model import or HMI interface input for dynamic modeling of tool shape parameters. It automatically generates discretized point clouds and adjusts the interpolation point density in real time according to the radius of curvature, ensuring smooth processing even for complex shapes and a high degree of automation. The multi-axis linkage control module synchronously updates five-axis position commands at a 1ms cycle, compensating for mechanical transmission errors. A laser interferometer is used to verify and adjust contour deviations, ensuring processing accuracy. Simultaneously, a conservative processing mode is introduced, using an acoustic emission sensor to monitor vibration. When vibration exceeds a threshold, the mode automatically switches to reduce the feed rate, preventing grinding wheel chipping. These improvements effectively enhance the system's ability to process complex tool shapes, strengthen the stability and safety of the processing, reduce scrap rates, and improve production efficiency and product quality.

[0065] 3. The control system of this end-face grinding machine features a multimodal force sensing system and dynamic PID gain adjustment. Combining multimodal information such as cutting force, grinding wheel speed, and material hardness, it constructs a fuzzy PID control model and dynamically adjusts the PID gain to achieve precise control. A force threshold trigger compensation mechanism rapidly responds to changes in cutting force, completing parameter adjustments within 10ms to ensure machining stability. Material hardness prediction and pre-compensation functions adjust control parameters in advance through a machine learning model, reducing the number of dynamic adjustments. These improvements enable the system to automatically optimize grinding wheel pressure according to different working conditions, effectively addressing changes in material hardness, improving machining accuracy and surface quality, reducing machining errors caused by cutting force fluctuations, and enhancing overall machining efficiency and product quality.

[0066] 4. The control system of this end-face grinding machine utilizes an industrial camera and a deep learning model to accurately identify the wear area of ​​the grinding wheel and calculate the wear area percentage. It then calculates the wear amount through Hough transform, triggering an automatic dressing program and dynamically adjusting the dressing path based on the wear area distribution. Closed-loop verification of the dressing effect ensures dressing quality and avoids repeated dressing. Digital twin simulation optimization can predict the dressing cycle in advance, reducing downtime. These improvements achieve accurate monitoring and efficient compensation of grinding wheel wear, ensuring the grinding wheel is always in good working condition, improving machining accuracy and stability, reducing machining errors caused by grinding wheel wear, and enhancing production efficiency and product quality.

[0067] 5. The control system of this end-face grinding machine adopts a hybrid model of XGBoost and neural networks, combining the advantages of both to improve prediction accuracy. It inputs multi-dimensional data and outputs processing results and real-time error predictions, providing a basis for parameter adjustment. Reinforcement learning dynamic adjustment is based on initial parameter recommendations. Data is collected in real time during processing, and parameters are fine-tuned through reinforcement learning algorithms to ensure processing accuracy and efficiency, while also performing error compensation. Parameter sensitivity analysis and visualization quantify the impact of each parameter on processing errors and visualize heatmaps on the HMI interface to guide engineers in parameter optimization. These improvements achieve intelligent optimization of process parameters, improving processing accuracy, efficiency, and stability, reducing quality costs, and enhancing product competitiveness. Attached Figure Description

[0068] Figure 1 This is a system module block diagram of the present invention;

[0069] Figure 2 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0070] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0071] See Figures 1-2 This invention discloses a control system for an end face grinding machine and provides the following five technical solutions:

[0072] First implementation method: includes:

[0073] Data preparation and recording module: Comprehensively collects historical machining data, including tool parameters (edge ​​angle, radius of curvature, step height, etc.), material properties (workpiece hardness such as stainless steel HRC30-50, carbon steel HRC20-40, grinding wheel hardness such as K, L, M grades), environmental parameters, equipment status parameters (grinding wheel speed 1000-5000rpm, feed rate 0.1-2mm / s), and machining results (surface roughness Ra value, contour deviation, dimensional tolerance, edge chipping rate, etc.), and machining error correlation data of the grinding wheel under different working conditions, including:

[0074] Influence of workpiece hardness: Record the cutting force fluctuation of the grinding wheel (Fx / Fy / Fz collected by a triaxial force sensor), vibration amplitude (collected by an accelerometer), and final machining error when machining workpieces of different hardness;

[0075] Influence of grinding wheel speed and feed rate: Cutting temperature (infrared thermometer), grinding wheel wear rate (image recognition of wear amount) and machined surface quality were collected under different combinations of speed and feed rate;

[0076] Effect of grinding wheel hardness: Record the cutting force distribution, grinding wheel life and machining error when machining the same material with different grinding wheel hardness;

[0077] Dynamic error compensation data: Real-time errors (such as axial runout and radial error) during the processing are collected by a laser interferometer or a coordinate measuring machine and correlated with the corresponding process parameter combinations;

[0078] Define the categories of each parameter and construct a set framework, setting each type of data as an independent set; when new data is input into the system, set data recognition and classification rules, automatically identify the data based on parameter characteristics and classify it into the corresponding pre-defined sets. For example, determine whether the input parameters include blade length, radius of curvature, etc., to be classified into the tool type parameter set; determine whether the data is classified into the workpiece hardness influence set based on whether it involves hardness information such as stainless steel HRC30-50, carbon steel HRC20-40, etc., and so on, to achieve automatic classification and storage of parameter data;

[0079] The data preparation and recording module forms the foundation of the entire system, responsible for comprehensively collecting and organizing various types of data during the machining process. This data includes not only basic information such as tool parameters, material properties, environmental parameters, and equipment status parameters, but also crucial information such as machining results and dynamic error compensation data. By constructing an aggregate framework and an automatic classification and storage mechanism, this module provides solid data support for subsequent data analysis, model training, and parameter optimization.

[0080] Dynamic modeling module for tool shape parameters: Input tool shape parameters and generate discretized point clouds. Use NURBS curve interpolation algorithm to adjust the density of interpolation points to ensure the accuracy and smoothness of machining complex shapes;

[0081] Multi-axis linkage control module: Real-time synchronization of five-axis position commands via EtherCAT to compensate for mechanical transmission errors; combined with acoustic emission sensors to monitor vibration, automatically switching to conservative mode when the threshold is exceeded;

[0082] The grinding wheel pressure adaptive control module uses a triaxial force sensor to monitor the cutting force in real time, constructs a fuzzy PID model to dynamically adjust the gain, and uses machine learning to predict the influence of material hardness, combined with force threshold triggering for rapid compensation, to optimize parameters in advance.

[0083] Grinding wheel wear compensation module: It uses machine vision to identify wear areas, calculates wear amount and triggers automatic dressing; it predicts dressing cycle through digital twin model to reduce downtime;

[0084] Process parameter optimization module: Based on historical data, a hybrid model of XGBoost and neural network is trained to predict processing results and errors; parameters are dynamically adjusted through reinforcement learning, and error compensation and parameter sensitivity analysis are performed.

[0085] The second implementation method differs from the first in that the dynamic modeling module for blade parameters specifically includes the following functions:

[0086] (1) Parametric input: Supports importing CAD models or directly inputting tool parameters through the HMI interface, including cutting edge angle, radius of curvature, step height, etc. The system automatically generates discretized point cloud data of the tool profile;

[0087] (2) Dynamic weighted interpolation: The NURBS curve interpolation algorithm is adopted to adjust the density of interpolation points in real time according to the curvature radius of the processing area. The larger the curvature, the denser the interpolation points, ensuring that the grinding wheel moves smoothly when processing complex shapes (such as wave blades and spiral blades).

[0088] The tool profile dynamic modeling module focuses on the dynamic modeling of tool profile parameters. Through parametric input and dynamic weighted interpolation technology, it can generate accurate discretized point cloud data of the tool profile contour. This is crucial for machining complex-shaped tools, ensuring smooth wheel movement during machining, thereby improving machining accuracy and surface quality.

[0089] The multi-axis linkage control module synchronously updates the five-axis position commands of X / Y / Z / A / C axes at a 1ms cycle via EtherCAT real-time Ethernet to compensate for mechanical transmission errors (such as lead screw backlash and motor step loss); the arc-shaped cutting edge is machined on a standard test piece, and the deviation between the actual contour and the theoretical contour is verified using a laser interferometer. The commands are adjusted based on the results to ensure that the error is ≤2μm.

[0090] The multi-axis linkage control module also features a conservative machining mode and incorporates an acoustic emission sensor to monitor machining vibration signals. When the vibration amplitude exceeds a threshold, it automatically switches to the conservative machining mode (reducing the feed rate by 20%) to prevent the grinding wheel from chipping.

[0091] The multi-axis linkage control module, as the core control component of the system, synchronously updates the five-axis position commands via EtherCAT real-time Ethernet, achieving precise linkage control across multiple axes. It not only compensates for mechanical transmission errors but also adjusts commands in real-time during machining to ensure machining accuracy. Furthermore, the introduction of a conservative machining mode and the monitoring function of acoustic emission sensors further enhance the system's stability and safety.

[0092] The main difference between the third and first implementation methods is that the adaptive control module for grinding wheel pressure specifically includes the following functions:

[0093] (1) Multimodal force sensing system and dynamic PID gain adjustment: A three-dimensional force sensor is installed on the grinding wheel spindle to monitor the cutting force components Fx, Fy and Fz in real time. A multimodal fuzzy PID control model is constructed by combining parameters such as grinding wheel speed and material hardness, and the PID gain is dynamically adjusted.

[0094] (2) Force threshold trigger compensation: Set a force threshold (e.g., trigger compensation when Fz>50N), and complete parameter adjustment within 10ms through the PLC high-speed counting module;

[0095] When processing materials with different hardness (such as stainless steel and carbon steel), the fluctuation range of surface roughness Ra value is statistically analyzed to ensure that Ra≤0.15μm;

[0096] (3) Material hardness prediction and pre-compensation: By combining historical machining error correlation data with machine learning models, the influence of material hardness on cutting force is predicted, and control parameters, including grinding wheel speed and feed rate, are adjusted in advance to reduce the number of dynamic adjustments during the machining process.

[0097] The specific methods for constructing a multimodal fusion fuzzy PID control model and dynamically adjusting the PID gain include:

[0098] (1) Definition of input variables

[0099] Cutting force components: Fx, Fy, and Fz, collected by the triaxial force sensor, represent cutting forces in different directions, respectively;

[0100] Grinding wheel speed: The speed at which the grinding wheel rotates, represented by n (unit: rpm);

[0101] Material hardness: The hardness value of the workpiece material, represented by H (which can be quantified according to actual hardness standards, such as HRC, etc.);

[0102] (2) Fuzzification: The input variables are fuzzified and divided into different fuzzy sets, each corresponding to a linguistic variable; for example:

[0103] The cutting force components Fx, Fy, and Fz are divided into three fuzzy sets: "small (S)", "medium (M)", and "large (B)".

[0104] The grinding wheel speed n is divided into three fuzzy sets: "low (L)", "medium (M)" and "high (H)".

[0105] Material hardness H is divided into three fuzzy sets: "Soft", "Medium", and "Hard".

[0106] The specific input value is mapped to the membership degree of each fuzzy set through the membership degree function (such as the triangular membership degree function). For example, for the cutting force Fx, if its value is 10N, according to the set membership degree function, the membership degree of the "small (S)" fuzzy set may be 0.8, the membership degree of the "medium (M)" fuzzy set may be 0.2, and the membership degree of the "large (B)" fuzzy set may be 0.

[0107] (3) Establishment of fuzzy rule base: Based on expert experience or actual processing experiments, a series of fuzzy rules are established to infer the PID gain, i.e., the adjustment amount of proportional gain Kp, integral gain Ki, and derivative gain Kd, based on the fuzzy values ​​of the input variables; such as the following example rules:

[0108] Rule 1: If Fz is "large (B)" and n is "high (H)" and H is "hard (Hard)", then Kp increases by a large amount, Ki remains unchanged, and Kd increases by a small amount;

[0109] Explanation: When the vertical cutting force is large, the grinding wheel speed is high, and the material hardness is high, a larger proportional gain is needed to quickly respond to changes in the cutting force. At the same time, the differential gain is appropriately increased to suppress overshoot, while the integral gain is not adjusted for the time being.

[0110] Rule 2: If Fx or Fy is "Medium" and n is "Medium" and H is "Medium", then Kp increases slightly, Ki increases slightly, and Kd remains unchanged;

[0111] Explanation: When the lateral cutting force, grinding wheel speed, and material hardness are moderate, the proportional and integral gains can be appropriately increased to improve the steady-state performance of the system, while the derivative gain remains unchanged for the time being.

[0112] Rule 3: If Fz is "small (S)" and n is "low (L)" and H is "soft (Soft)", then Kp decreases slightly, Ki decreases slightly, and Kd decreases slightly.

[0113] Explanation: When the vertical cutting force is small, the grinding wheel speed is low, and the material hardness is soft, the PID gain should be appropriately reduced to avoid excessive system response and oscillation.

[0114] (4) Fuzzy reasoning: Fuzzy reasoning is performed based on the fuzzy values ​​of the input variables and the fuzzy rule base; for each rule, the membership degree of the rule conclusion is obtained through fuzzy implication operation (such as the minimum-maximum method) based on the membership degree of the input variables to the rule premises; then, the conclusions of all rules are fuzzy synthesized (such as the maximum-average method) to obtain the fuzzy output of the PID gain adjustment.

[0115] (5) Defuzzification: The centroid method is used to convert the fuzzy output into a specific PID gain adjustment amount, that is, the value corresponding to the centroid of the membership function of the fuzzy output is calculated as the result of defuzzification; for example, for the fuzzy output of Kp, a specific value ΔKp is calculated by the centroid method, which represents the adjustment amount of the proportional gain.

[0116] (6) PID gain adjustment: Add the PID gain adjustment obtained from defuzzification to the current PID gain value to obtain a new PID gain value:

[0117] Kp new =Kp+ΔKp;

[0118] Ki new =Ki+ΔKi;

[0119] Kd new =Kd+ΔKd;

[0120] Where Kp, Ki, and Kd are the current PID gain values, Kp new Ki new Kd new This is the adjusted new PID gain value.

[0121] The fuzzy PID control model based on multimodal fusion can dynamically adjust the PID gain according to multimodal information such as cutting force components, grinding wheel speed, and material hardness, thereby achieving adaptive control of the grinding wheel pressure in the end face grinding machine and improving machining accuracy and stability.

[0122] The adaptive control module for grinding wheel pressure achieves adaptive control of the grinding wheel pressure through a multimodal force sensing system and dynamic PID gain adjustment technology. It can dynamically adjust the PID gain based on multimodal information such as cutting force components, grinding wheel speed, and material hardness, thereby ensuring the stability and accuracy of the machining process. Furthermore, force threshold trigger compensation and material hardness prediction and pre-compensation functions further enhance the system's adaptability and intelligence.

[0123] The fourth implementation method differs from the first implementation method in that the specific working content of the grinding wheel wear compensation module includes:

[0124] (1) High-precision image acquisition and processing: An industrial camera and a blue ring light source are installed on the side of the grinding wheel to acquire images of the grinding wheel end face at a frequency of 50fps; a deep learning semantic segmentation model (such as U-Net) is used to identify the wear area of ​​the grinding wheel and calculate the wear area ratio;

[0125] (2) Dynamic calculation and adjustment of wear amount: The wear amount ΔD=D is calculated by fitting the center coordinates and radius of the circle through Hough transform. 初始 -D 实时 ;D 初始 D is the initial diameter of the grinding wheel. 实时 Given the current diameter of the grinding wheel, if ΔD > 0.1 mm, the grinding wheel dressing program is triggered, and the diamond dressing pen is used to automatically sharpen the wheel. The dressing path (spiral or radial dressing) is dynamically adjusted according to the wear area distribution (such as local wear or uniform wear).

[0126] (3) Closed-loop verification of repair effect: After repair, the image is collected again and the difference in wear before and after repair is calculated. If the difference is <0.05mm, the repair is deemed qualified; otherwise, the repair is repeated.

[0127] After machining 100 tools consecutively, the edge size distribution is measured to ensure CPK ≥ 1.67 (4σ level). CPK is a process capability index, a quantitative indicator that measures the ability of a production process to produce products that meet specifications under steady-state conditions. CPK is calculated based on the upper and lower limits of specifications (USL / LSL), the process mean (μ), and the standard deviation (σ), and the formula is: CPK = min( ); where USL is the upper limit of specifications, LSL is the lower limit of specifications, μ is the process mean, and σ is the process standard deviation; the larger the CPK value, the better the process capability, that is, the more stably the process can produce products that meet the specifications.

[0128] (4) Digital twin simulation optimization: The grinding wheel wear process is simulated by digital twin model to predict the dressing cycle and reduce downtime.

[0129] The grinding wheel wear compensation module focuses on monitoring and compensating for grinding wheel wear. It identifies wear areas and calculates wear amounts using high-precision image acquisition and processing technology. It dynamically adjusts the dressing path and dressing procedure based on the wear amount, ensuring timely compensation for grinding wheel wear. Furthermore, closed-loop verification of dressing effects and digital twin simulation optimization functions further improve dressing efficiency and accuracy.

[0130] The fifth implementation method differs from the first implementation method in that the specific work content of the process parameter optimization module includes:

[0131] (1) Hybrid model training and prediction: The multidimensional data is trained using a hybrid model of XGBoost and neural network. Inputs include tool shape parameters, material properties, environmental parameters, equipment status parameters (including grinding wheel hardness, rotation speed, and feed rate), and historical error data. Outputs the predicted values ​​of the processing results (surface roughness, processing time, and grinding wheel wear) and the predicted values ​​of real-time errors.

[0132] The mixture model is trained using at least 2000 datasets, with a validation set R² ≥ 0.98. R² represents the proportion of the variance in the dependent variable explained by the regression model relative to the total variance. 2 = =1− ;

[0133] SSR (Sum of Squares in Regression): The amount of variance explained by the model (the sum of squares of the differences between the predicted values ​​and the mean of the dependent variable).

[0134] SST (Total Sum of Squares): The total variation of the dependent variable (the sum of squares of the differences between the actual values ​​and the mean).

[0135] RSS (Residual Sum of Squares): The amount of variation not explained by the model (the sum of squares of the differences between the actual and predicted values).

[0136] An attention mechanism is introduced to enhance the weight of the influence of key parameters (such as the matching degree between the hardness of the grinding wheel and the hardness of the workpiece) on the error;

[0137] (2) Enhance dynamic adjustment of learning, including:

[0138] Recommended initial parameters: After inputting the new tool shape and material parameters in the HMI interface, the model outputs initial process parameters based on historical data (such as grinding wheel speed 3200rpm, feed rate 0.45mm / s, grinding wheel hardness L grade).

[0139] Dynamic adjustment: Real-time data is collected every 10 seconds during the processing, and parameters are fine-tuned through reinforcement learning algorithms (such as PPO) to ensure processing accuracy and efficiency;

[0140] If the prediction error shows that the runout of a certain axis exceeds the standard (such as Z-axis error > 1μm), the feed speed of the corresponding axis will be automatically reduced by 10% and the grinding wheel angle will be finely adjusted.

[0141] If the cutting force fluctuation exceeds the threshold (e.g., Fz standard deviation > 15N), dynamically adjust the grinding wheel speed to stabilize the cutting process;

[0142] Error compensation: Based on the real-time error prediction value, the five-axis interpolation trajectory is directly corrected through inverse kinematics algorithm (e.g., X-axis compensation ΔX=0.02mm) to ensure machining accuracy;

[0143] (3) Parameter sensitivity analysis and visualization: The influence of each parameter on the machining error is quantified by SHAP value. For example, for every 10 HRC increase in workpiece hardness, the vibration amplitude of the grinding wheel increases by 0.05 mm, and the rotation speed needs to be reduced by 500 rpm to compensate; for every 1 grade decrease in grinding wheel hardness, the feed rate needs to be reduced by 0.05 mm / s to avoid chipping; the parameter sensitivity heat map is visualized in the HMI interface to guide engineers to prioritize the optimization of high-impact parameters.

[0144] In the hybrid model training and prediction steps, a weighted average method is used to fuse the prediction results of the XGBoost model and the neural network model. The performance of the two models is evaluated on a validation set, and their respective weights are determined based on their performance. For example, if the R² value of the XGBoost model on the validation set is 0.95 and the R² value of the neural network model is 0.93, the XGBoost model can be given a higher weight, such as 0.6, and the neural network model a weight of 0.4. For new input data (including tool shape parameters, material properties, environmental parameters, equipment state parameters, etc.), the data is input into the trained XGBoost model and the neural network model respectively to obtain the prediction results of the two models. Then, according to the predetermined weights, the prediction results of the two models are weighted and averaged to obtain the final predicted value of the hybrid model's machining result (surface roughness, machining time, and grinding wheel wear).

[0145] During the processing, data such as current tool shape parameters, material properties, environmental parameters, and equipment status parameters are collected in real time. The real-time collected data is preprocessed using the same preprocessing method as historical data, and then input into the trained hybrid model to obtain the real-time processing result prediction value. The real-time prediction value is compared with the actual processing result (obtained through measuring equipment) to calculate the real-time error prediction value, which is the difference between the prediction value and the actual value. The error prediction value is used for subsequent error compensation and dynamic adjustment of process parameters.

[0146] The process parameter optimization module, as the intelligent optimization part of the system, optimizes process parameters through hybrid model training and prediction and reinforcement learning dynamic adjustment technology. It can train a hybrid model based on multidimensional data to predict processing results and real-time errors, and dynamically adjust parameters through reinforcement learning algorithms to ensure processing accuracy and efficiency. Furthermore, parameter sensitivity analysis and visualization functions provide engineers with intuitive guidance for parameter optimization.

[0147] Furthermore, any content not described in detail in this specification is existing technology known to those skilled in the art.

[0148] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0149] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A control system for an end face grinding machine, characterized in that, include: Data preparation and recording module: comprehensively collects historical processing data, including tool parameters, material properties, environmental and equipment status, processing results, and processing error correlation data of grinding wheels under different working conditions; Dynamic modeling module for tool shape parameters: Input tool shape parameters and generate discretized point clouds. Use NURBS curve interpolation algorithm to adjust the density of interpolation points to ensure the accuracy and smoothness of machining complex shapes; Multi-axis linkage control module: Real-time synchronization of five-axis position commands via EtherCAT to compensate for mechanical transmission errors; Introducing acoustic emission sensors to monitor machining vibration signals, automatically switching to conservative machining mode when the vibration amplitude exceeds the threshold to avoid grinding wheel chipping; The grinding wheel pressure adaptive control module uses a triaxial force sensor to monitor the cutting force in real time, constructs a fuzzy PID model to dynamically adjust the gain, and predicts the influence of material hardness through machine learning, combined with force threshold triggering for rapid compensation, to optimize parameters in advance. The job duties specifically include: (1) Multimodal force sensing system and dynamic PID gain adjustment: A three-dimensional force sensor is installed on the grinding wheel spindle to monitor the cutting force components Fx, Fy and Fz in real time. A multimodal fuzzy PID control model is constructed by combining the grinding wheel speed and material hardness parameters, and the PID gain is dynamically adjusted. (2) Force threshold trigger compensation: Set the force threshold and complete the parameter adjustment within 10ms through the PLC high-speed counting module; When processing materials with different hardness, the fluctuation range of surface roughness Ra value is statistically analyzed to ensure that Ra≤0.15μm; (3) Material hardness prediction and pre-compensation: By combining historical machining error correlation data with machine learning models, the influence of material hardness on cutting force is predicted, and control parameters, including grinding wheel speed and feed rate, are adjusted in advance to reduce the number of dynamic adjustments during the machining process; Grinding wheel wear compensation module: It uses machine vision to identify wear areas, calculates wear amount and triggers automatic dressing; it predicts dressing cycle through digital twin model to reduce downtime; Process parameter optimization module: Based on historical data, a hybrid model of XGBoost and neural network is trained to predict processing results and errors; parameters are dynamically adjusted through reinforcement learning, and error compensation and parameter sensitivity analysis are performed.

2. The end face grinding machine control system according to claim 1, characterized in that: The processing error correlation data includes: Influence of workpiece hardness: Record the fluctuation of the grinding wheel's cutting force, vibration amplitude, and final machining error when machining workpieces of different hardness; Influence of grinding wheel speed and feed rate: Cutting temperature, grinding wheel wear rate and machined surface quality were collected under different combinations of speed and feed rate; Influence of grinding wheel hardness: Record the cutting force distribution, grinding wheel life, and machining error when machining the same material with different grinding wheel hardness; Dynamic error compensation data: Real-time errors during the processing are collected by a laser interferometer or a coordinate measuring machine and correlated with the corresponding combination of process parameters.

3. The end face grinding machine control system according to claim 2, characterized in that: The data preparation and recording module collects historical processed data, clarifies the categories of each parameter, and constructs a set framework, setting each type of data as an independent set; when new data is input into the system, data identification and classification rules are set, and the data is automatically identified and divided into the corresponding pre-set sets based on parameter characteristics.

4. The end face grinding machine control system according to claim 1, characterized in that: The specific functions of the dynamic modeling module for blade parameters include: (1) Parametric input: Supports importing CAD models or directly inputting tool parameters through the HMI interface, including cutting edge angle, radius of curvature, and step height. The system automatically generates discretized point cloud data of the tool profile. (2) Dynamic weighted interpolation: The NURBS curve interpolation algorithm is adopted to adjust the density of interpolation points in real time according to the radius of curvature of the processing area. The larger the curvature, the denser the interpolation points, ensuring smooth movement of the grinding wheel when processing complex shapes.

5. The end face grinding machine control system according to claim 1, characterized in that: The multi-axis linkage control module synchronously updates the five-axis position commands of X / Y / Z / A / C axes at a 1ms cycle via EtherCAT real-time Ethernet to compensate for mechanical transmission errors. A circular arc blade is machined on a standard specimen, and a laser interferometer is used to verify the deviation between the actual contour and the theoretical contour. The commands are adjusted based on the results to ensure that the error is ≤2μm.

6. The end face grinding machine control system according to claim 1, characterized in that: The method of constructing a multimodal fusion fuzzy PID control model and dynamically adjusting the PID gain specifically includes: (1) Definition of input variables Cutting force components: Fx, Fy, and Fz, collected by the triaxial force sensor, represent cutting forces in different directions, respectively; Grinding wheel speed: The speed at which the grinding wheel rotates, denoted by n; Material hardness: The hardness value of the workpiece material, represented by H; (2) Fuzzification: The input variables are fuzzified and divided into different fuzzy sets, with each fuzzy set corresponding to a linguistic variable; The membership function maps specific input values ​​to the membership degrees of various fuzzy sets. (3) Establishment of fuzzy rule base: Based on expert experience or actual processing experiments, a series of fuzzy rules are established to infer the PID gain based on the fuzzy values ​​of the input variables, namely the adjustment amount of proportional gain Kp, integral gain Ki, and derivative gain Kd. (4) Fuzzy reasoning: Fuzzy reasoning is performed based on the fuzzy values ​​of the input variables and the fuzzy rule base; for each rule, the membership degree of the rule conclusion is obtained through fuzzy implication operation based on the membership degree of the input variables to the rule premises; then, the conclusions of all rules are fuzzy synthesized to obtain the fuzzy output of the PID gain adjustment. (5) Defuzzification: The centroid method is used to convert the fuzzy output into a specific PID gain adjustment amount, that is, the value corresponding to the centroid of the membership function of the fuzzy output is calculated as the result of defuzzification; (6) PID gain adjustment: Add the PID gain adjustment obtained from defuzzification to the current PID gain value to obtain a new PID gain value. Kp. new =Kp+ΔKp; Ki new =Ki+ΔKi; Kd new =Kd+ΔKd; Where Kp, Ki, and Kd are the current PID gain values, Kp new Ki new Kd new This is the adjusted new PID gain value.

7. The end face grinding machine control system according to claim 1, characterized in that: The specific working functions of the grinding wheel wear compensation module include: (1) High-precision image acquisition and processing: An industrial camera and a blue ring light source are installed on the side of the grinding wheel to acquire images of the grinding wheel end face at a frequency of 50fps; a deep learning semantic segmentation model is used to identify the wear area of ​​the grinding wheel and calculate the wear area ratio; (2) Dynamic calculation and adjustment of wear amount: The wear amount ΔD=D is calculated by fitting the center coordinates and radius of the circle through Hough transform. 初始 -D 实时 ;D 初始 D is the initial diameter of the grinding wheel. 实时 Given the current diameter of the grinding wheel, if ΔD > 0.1 mm, trigger the grinding wheel dressing program and dynamically adjust the dressing path according to the wear area distribution; (3) Closed-loop verification of repair effect: After repair, the image is collected again and the difference in wear before and after repair is calculated. If the difference is <0.05mm, the repair is deemed qualified; otherwise, the repair is repeated. After continuously machining 100 cutting tools, the edge size distribution is measured to ensure that CPK ≥ 1.67; CPK is the process capability index, which is a quantitative indicator that measures the ability of the production process to produce products that meet specifications under stable conditions. (4) Digital twin simulation optimization: The grinding wheel wear process is simulated by digital twin model to predict the dressing cycle and reduce downtime.

8. The end face grinding machine control system according to claim 1, characterized in that: The specific functions of the process parameter optimization module include: (1) Hybrid model training and prediction: The multidimensional data is trained using a hybrid model of XGBoost and neural network. Inputs include tool shape parameters, material properties, environmental parameters, equipment status parameters, and historical error data; outputs the predicted value of the processing result and the predicted value of the real-time error. The hybrid model is trained using at least 2000 sets of data, with a validation set R² ≥ 0.

98. R² represents the proportion of the variance in the dependent variable explained by the regression model to the total variance. An attention mechanism is introduced to strengthen the weight of key parameters on the error. (2) Enhance dynamic adjustment of learning, including: Recommended initial parameters: After inputting the new tool shape and material parameters in the HMI interface, the model outputs initial process parameters based on historical data; Dynamic adjustment: Real-time data is collected every 10 seconds during the processing, and parameters are fine-tuned through reinforcement learning algorithms to ensure processing accuracy and efficiency; If the prediction error shows that the runout of a certain axis exceeds the standard, the feed speed of the corresponding axis will be automatically reduced by 10% and the grinding wheel angle will be finely adjusted. If the cutting force fluctuation exceeds the threshold, the grinding wheel speed is dynamically adjusted to stabilize the cutting process. Error compensation: Based on the real-time error prediction value, the five-axis interpolation trajectory is directly corrected through the inverse kinematics algorithm to ensure machining accuracy; (3) Parameter sensitivity analysis and visualization: Quantify the influence of each parameter on the processing error by using SHAP value; visualize the parameter sensitivity heat map in the HMI interface to guide engineers to prioritize the optimization of high-impact parameters.

9. The end face grinding machine control system according to claim 8, characterized in that: In the hybrid model training and prediction steps, a weighted average method is used to fuse the prediction results of the XGBoost model and the neural network model. The performance of the two models is evaluated on the validation set, and their respective weights are determined based on their performance. For new input data, it is input into the trained XGBoost model and the neural network model respectively to obtain the prediction results of the two models. Then, according to the predetermined weights, the prediction results of the two models are weighted and averaged to obtain the final processing result prediction value of the hybrid model. During the processing, the current tool shape parameters, material properties, environmental parameters, and equipment status parameters are collected in real time. The real-time collected data is preprocessed in the same way as the historical data, and then input into the trained hybrid model to obtain the real-time processing result prediction value. The real-time predicted value is compared with the actual processing result to calculate the real-time error prediction value, which is the difference between the predicted value and the actual value. The error prediction value is used for subsequent error compensation and dynamic adjustment of process parameters.

Images