Multi-parameter cooperative control method and device for clamping system of precision boring and milling machine

Abstract

The present application relates to the field of automation control technology, specifically to a multi-parameter collaborative control method and device for a precision boring and milling machine clamping system, which acquires and processes workpiece images through an industrial camera, extracts visual features, and fuses them with process parameters of the workpiece and geometric distribution parameters of the key area of the workpiece obtained from a process database to generate a comprehensive state vector; based on the vector, a target clamping force is output by a pre-trained MLP decision model; in the main clamping stage, vibration, temperature and pressure signals are synchronously acquired, multi-modal disturbance features are extracted and an observation vector is constructed, which is input to a pre-set fuzzy rule base for reasoning to obtain a compensation activation factor; finally, current correction instructions are dynamically generated according to the clamping force deviation and the compensation factor to drive an electro-hydraulic proportional valve to achieve precise compensation. The present application significantly improves the stability and processing reliability of the clamping system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of automation control, in particular to a multi-parameter cooperative control method and device of a precision boring and milling machine clamping system. BACKGROUND

[0002] The precision boring and milling machine is widely used for machining high-hardness and high-value workpieces such as die steel and titanium alloy. The cutting force is huge, and the machining precision requirement reaches the micron level. Once the workpiece micron-level slip is caused by clamping failure, catastrophic tool collision and workpiece scrapping are easily triggered, causing significant economic losses. The current clamping control system has technical bottlenecks: 1) Clamping parameters are mostly dependent on operator experience setting, lacking quantitative basis, resulting in poor consistency of clamping quality, frequently wavering between the two extremes of "insufficient clamping force causing workpiece displacement" and "excessive clamping causing workpiece deformation"; 2) The existing clamping system generally uses single variable feedback control, only monitoring clamping force or position deviation, and cannot comprehensively perceive multi-dimensional state information such as vibration and thermal deformation. When facing sudden changes in cutting force or external disturbances, the system has incomplete perception and slow response, and the control precision is insufficient; 3) The traditional control strategy is fixed and difficult to adapt to actual changes such as geometric deviation between workpiece batches, fixture wear or hydraulic fluctuation. The adaptation to new difficult-to-machine materials relies on trial and error, greatly increasing the machining cost and production preparation period, resulting in low comprehensive utilization rate of the equipment. The above defects are the fundamental reasons for the unstable quality, high scrap rate and difficult breakthrough in production efficiency in the current precision boring and milling machine machining field. Therefore, there is an urgent need for a clamping control method that integrates multi-source perception, intelligent decision-making and dynamic cooperation to realize accurate and adaptive adjustment of clamping force and improve the stability and reliability of precision machining. SUMMARY

[0003] The present application provides a multi-parameter cooperative control method and device of a precision boring and milling machine clamping system, which solves the technical problems of unstable machining quality, high scrap rate and low production efficiency of the precision boring and milling machine caused by the technical bottlenecks of the existing clamping control system.

[0004] In view of the above problems, the present application provides a multi-parameter cooperative control method of a precision boring and milling machine clamping system, which comprises:

[0005] The workpiece is pre-clamped and fixed on the fixture, and the workpiece surface image is collected by an industrial camera. After the collected image is processed by SIFT feature matching algorithm and illumination normalization, a composite image for feature extraction is formed;

[0006] Based on composite images, visual features of workpiece surface structure and texture distribution are extracted through a pre-trained deep convolutional network, and process parameters of the workpiece are obtained from the process database. At the same time, the geometric distribution parameters of the pre-labeled key areas of the workpiece in the image are detected. The above visual features, process parameters and geometric distribution parameters are normalized and weighted and fused to generate a comprehensive state vector for clamping decision. The process parameters include at least: material mechanical parameters and cutting process parameters.

[0007] The integrated state vector is input into the pre-trained MLP decision model, and the target clamping force is output. The MLP model learns its decision behavior from the clamping decision strategy network trained based on the DDPG algorithm through the knowledge distillation method. The main clamping stage is the working stage in which the clamping system applies and maintains the target clamping force during the stable machining process of the boring and milling machine.

[0008] Based on the obtained target clamping force, the controller generates a corresponding target current command according to the preset pressure-current mapping relationship. This command drives the electro-hydraulic proportional valve to adjust the output force of the hydraulic cylinder, thereby driving the clamping mechanism to complete the workpiece clamping.

[0009] During the main clamping stage, vibration signals, temperature signals, and hydraulic oil pressure signals of the clamping system are simultaneously acquired. Using a preset time window as the unit, the vibration and temperature signals are extracted in the time domain to obtain the vibration time domain features and temperature time domain features. The average value of the hydraulic oil pressure within the window is calculated. Based on the known effective piston area, the hydraulic oil pressure is converted into the current output force of the hydraulic cylinder. The current output force is input into the calibrated force transmission model to obtain the total effective clamping force on the workpiece. The clamping force deviation between the target clamping force and the total effective clamping force is calculated.

[0010] The extracted vibration time-domain features and temperature time-domain features are combined with the clamping force deviation to construct a normalized current multimodal disturbance observation vector, which is in the form of [vibration RMS, vibration kurtosis, hydraulic oil temperature change rate, clamping temperature change rate, clamping force deviation].

[0011] The current multimodal disturbance observation vector is input into the pre-constructed fuzzy rule model. After fuzzification, rule activation, implication operation and maximum value aggregation, a synthetic fuzzy set of hydraulic compensation factor is generated. The centroid method is used to defuzzify the hydraulic compensation factor in the domain [0,1]. The geometric center of the synthetic fuzzy set is calculated and the calculation result is used as the output value of the hydraulic compensation factor. The fuzzy rule model is constructed by fuzzy decision tree.

[0012] Based on the clamping force deviation between the target clamping force and the total effective clamping force, and the output value of the hydraulic compensation factor, the current correction value of the electro-hydraulic proportional valve is calculated and superimposed on the target current command. Together, they serve as the control signal for the electro-hydraulic proportional valve, thereby dynamically adjusting the valve core opening to generate the required hydraulic oil pressure and maintain a stable clamping force during machining.

[0013] ,

[0014] in, This is the current correction value. For clamping force deviation, This is the output value of the hydraulic compensation factor. This represents the slope of the pressure-current characteristic curve of the electro-hydraulic proportional valve.

[0015] This application also provides a multi-parameter collaborative control device for a precision boring and milling machine clamping system, the device comprising:

[0016] The composite image module is used to pre-clamp and fix the workpiece on the fixture, and to acquire images of the workpiece surface through an industrial camera. The acquired images are processed by SIFT feature matching algorithm and illumination normalization to form a composite image for feature extraction.

[0017] The state vector module is used to extract visual features of the workpiece surface structure and texture distribution based on the composite image through a pre-trained deep convolutional network, obtain the workpiece's process parameters from the process database, and detect the geometric distribution parameters of the pre-labeled key areas of the workpiece in the image. The above visual features, process parameters and geometric distribution parameters are normalized and weighted and fused to generate a comprehensive state vector for clamping decision. The process parameters include at least: material mechanical parameters and cutting process parameters.

[0018] The MLP model module is used to input the comprehensive state vector into the pre-trained MLP decision model and output the target clamping force. The MLP model learns its decision behavior from the clamping decision strategy network trained based on the DDPG algorithm through the knowledge distillation method. The main clamping stage is the working stage in which the clamping system applies and maintains the target clamping force during the stable machining process of the boring and milling machine.

[0019] The optimal clamping module is used to generate a corresponding target current command based on the obtained target clamping force and the controller according to the preset pressure-current mapping relationship. This command drives the electro-hydraulic proportional valve to adjust the output force of the hydraulic cylinder, thereby driving the clamping mechanism to complete the workpiece clamping.

[0020] The disturbance information acquisition module is used to simultaneously acquire vibration signals, temperature signals, and hydraulic oil pressure signals of the clamping system during the main clamping stage. It extracts time-domain features of the vibration and temperature signals in preset time windows to obtain vibration time-domain features and temperature time-domain features. It also calculates the average value of the hydraulic oil pressure within the window. Based on the known effective piston area, it converts the hydraulic oil pressure into the current output force of the hydraulic cylinder. The current output force is input into the calibrated force transmission model to obtain the total effective clamping force on the workpiece. It also calculates the clamping force deviation between the target clamping force and the total effective clamping force.

[0021] The current disturbance observation vector module combines the extracted vibration time-domain features and temperature time-domain features with the clamping force deviation to construct a normalized current multimodal disturbance observation vector, which takes the form of [vibration RMS, vibration kurtosis, hydraulic oil temperature change rate, clamping temperature change rate, clamping force deviation].

[0022] The hydraulic compensation factor module is used to input the current multimodal disturbance observation vector into a pre-constructed fuzzy rule model. After fuzzification, rule activation, implication operation and maximum value aggregation, a synthetic fuzzy set of hydraulic compensation factors is generated. The centroid method is used to defuzzify the hydraulic compensation factor in the domain [0,1]. The geometric center of the synthetic fuzzy set is calculated and the calculation result is used as the output value of the hydraulic compensation factor. The fuzzy rule model is constructed by a fuzzy decision tree.

[0023] The compensation force module calculates the current correction value for the electro-hydraulic proportional valve based on the clamping force deviation between the target clamping force and the total effective clamping force, as well as the output value of the hydraulic compensation factor. This correction value is then superimposed on the target current command, serving as the control signal for the electro-hydraulic proportional valve. This dynamically adjusts the valve core opening to generate the required hydraulic oil pressure, maintaining a stable clamping force during machining.

[0024] ,

[0025] in, This is the current correction value. For clamping force deviation, This is the output value of the hydraulic compensation factor. This represents the slope of the pressure-current characteristic curve of the electro-hydraulic proportional valve.

[0026] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0027] This application provides a multi-parameter collaborative control method and apparatus for a clamping system of a precision boring and milling machine. Industrial cameras capture real-time images of the workpiece after clamping, which are then stitched together and processed with lighting to obtain a surface view of the workpiece. The camera captures the current, actual clamping state. Random factors that may change with each clamping cannot be obtained from a fixed database, but directly affect the distribution and effect of the clamping force. This step ensures a real-time perception foundation for subsequent intelligent decision-making. A comprehensive state vector is fused, allowing the decision-making system to "see" the entire scene and make more global and forward-looking decisions. Using an MLP model, the target clamping force is output based on the comprehensive state vector, transforming the complex decision problem into a simple, normalized strategy selection problem. The target clamping force is converted into a current command to drive the hydraulic system to perform clamping, completing the closed loop from decision signal to physical execution. Multimodal disturbance monitoring and feature extraction enable real-time diagnosis of the machining process, used to capture dynamic disturbances that suddenly occur during machining but are not anticipated in the intelligent stage of the MLP model. Real-time compensation decision-making based on fuzzy rules, with the fuzzy rule base acting as a fast-reaction arc, can respond to sudden disturbances in milliseconds, determining whether compensation is needed and how strong the willingness to compensate is. The compensation intention u is multiplied by the force deviation ΔF to obtain the final compensation current value. Here, the fuzzy intelligent system modulates the PID output according to the global state, avoiding the problems of "sluggishness" or "overshooting" of traditional PID under complex disturbances.

[0028] In summary, this invention fundamentally improves the clamping quality and reliability of the clamping system for precision boring and milling machines. Deep integration of multi-source information enables full-dimensional state perception during operation, allowing for earlier identification of potential risks rather than passively waiting for deviations to occur. The adoption of an offline / online dual-level decision-making mechanism enhances the response speed to disturbances, shortens the production preparation cycle, and improves production efficiency. Attached Figure Description

[0029] Figure 1 This is a flowchart illustrating a multi-parameter collaborative control method for a precision boring and milling machine clamping system provided in an embodiment of this application.

[0030] Figure 2 This is a schematic diagram of the structure of a multi-parameter collaborative control device for a precision boring and milling machine clamping system provided in an embodiment of this application.

[0031] Explanation of reference numerals in the attached figures: Composite image module 10, State vector module 20, MLP model module 30, Optimal clamping module 40, Disturbance information acquisition module 50, Current disturbance observation vector module 60, Hydraulic compensation factor module 70, Compensation force module 80. Detailed Implementation

[0032] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application.

[0033] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description of this application will be provided in conjunction with the accompanying drawings. The described embodiments should not be considered as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0034] This application provides a multi-parameter collaborative control method for a precision boring and milling machine clamping system, such as... Figure 1 As shown, the method includes:

[0035] S100: The workpiece is pre-clamped and fixed on the fixture. An industrial camera is used to acquire images of the workpiece surface. The acquired images are processed by SIFT feature matching algorithm and illumination normalization to form a composite image for feature extraction.

[0036] Specifically, in this embodiment, images are acquired after the workpiece is pre-clamped. Two industrial cameras (optional 5-megapixel, 2448×2048 resolution) are installed at a fixed distance of 300±5mm from the worktable reference surface. The two cameras are positioned above the workpiece clamping area, with the mechanical coordinate system of the workpiece clamping area (such as the worktable coordinate system) as a reference to ensure consistency between image acquisition and actual processing position. The optical axis of the first camera forms a 45° angle with the Z-axis in the XZ plane, which can completely capture the side features of the workpiece in the positive X direction. The optical axis of the second camera forms a 45° angle with the Z-axis in the YZ plane, which can completely capture the side features of the workpiece in the positive Y direction. The intersection point of the optical axes of the two cameras is aligned with the geometric center of the workpiece clamping area (such as the center of the workpiece or the center of the fixture), ensuring that the coverage of the multi-view images overlaps and the field of view completely covers the maximum clamping size of the workpiece and the contact surface of the fixture. The workpiece is temporarily fixed to the worktable with a preload of 10-20N using a pneumatic clamp and the workpiece is stabilized. Then, dual cameras are simultaneously triggered to acquire multi-view images of the workpiece surface.

[0037] Furthermore, a composite image for feature extraction is formed, including:

[0038] Two industrial cameras are positioned above the workpiece clamping area. With the mechanical coordinate system of the workpiece clamping area as a reference, the optical axis of the first camera is located in the XZ plane and forms a 45° angle with the Z-axis, and the optical axis of the second camera is located in the YZ plane and forms a 45° angle with the Z-axis. The intersection point of the optical axes of the two cameras is aligned with the geometric center of the workpiece clamping area to obtain multi-angle images of the workpiece surface.

[0039] Image A is the surface image of the workpiece acquired by the first camera, and image B is the surface image of the workpiece acquired by the second camera. Key points of images A and B are extracted by SIFT algorithm and initial matching is performed. Then, the optimal homography matrix is ​​estimated from the initial matching point pairs by RANSAC algorithm combined with homography model, and mismatched points are eliminated. Perspective transformation is performed on the images using this matrix to obtain the stitched composite image.

[0040] The stitched composite image is subjected to illumination normalization processing, converting it from the RGB color space to the HSV color space, and separating the hue, saturation, and lightness channels. The CLAHE algorithm is applied to the lightness channel to suppress the influence of uneven illumination. Then, the highlight area is further compressed pixel by pixel through a nonlinear mapping method to reduce the interference of specular reflection from the metal surface of the workpiece. The processed lightness channel is merged with the original hue and saturation channels and converted back to the RGB color space to obtain the normalized composite image used for subsequent feature extraction.

[0041] Specifically, in this implementation, two industrial cameras are positioned above the workpiece clamping area. With the workpiece pre-clamped on the worktable, the first camera acquires image A of the workpiece surface, and the second camera acquires image B of the workpiece surface. Keypoint detection is performed on images A and B using the SIFT algorithm, which locates significant feature positions with scale invariance and rotation invariance in the images. A 128-dimensional feature vector is generated for each detected keypoint. This feature vector, called a descriptor, is used to characterize the gradient distribution characteristics of the local region around the keypoint, so that similar local structures in different images have similar descriptors. Subsequently, based on the nearest neighbor distance ratio method, the similarity between the descriptors of each keypoint in image A and the descriptors of all keypoints in image B is measured. When the distance ratio between the nearest neighbor and the second nearest neighbor is less than a set threshold, it is determined to be a valid match, thus obtaining a set of initial matching point pairs. Subsequently, the RANSAC algorithm combined with a homography model is employed. The RANSAC algorithm repeatedly selects a subset of points from the initial set of matching point pairs, uses these points to calculate a hypothetical homography matrix, and uses this matrix to test all other matching points. Based on a set threshold (e.g., reprojection error), it determines which points are inliers conforming to the model and which are outliers. While eliminating mismatched points, the optimal homography matrix is ​​determined; that is, the homography matrix with the most inliers is considered the optimal estimate. This method effectively eliminates mismatched points, ensuring the accuracy and robustness of the homography matrix. When performing perspective transformation on the image using this optimal homography matrix H, image A is used as a reference, and this matrix needs to be applied to image B (the image to be transformed), based on the homography relationship X. B =HX AFirst, the coordinates of the four corners of image B—top left, top right, bottom left, and bottom right—are calculated and transformed using the homography matrix H. These transformed points define a minimum bounding rectangle, the size of which determines the final output canvas size, i.e., the overall framework of the stitched image. For each target pixel coordinate (x', y') in the output canvas, it is transformed into homogeneous coordinate form. And use the inverse matrix of the homography matrix To calculate its corresponding position (x, y) in image A, matrix operations are used to find the mapping position of each pixel on the canvas in the original image. When (x, y) is within the range of image A, bilinear interpolation is used to estimate the pixel value at non-integer coordinates based on the color or grayscale values ​​of the four nearest neighbor pixels; if (x, y) exceeds the boundary of image A, zero values ​​are filled in as background. After completing the remapping of image B to the canvas, image A is superimposed on the same canvas, and distance weighting formula I is used in the overlapping area of ​​the two images. out =(1-λ)·I A +λ·I B Perform pixel fusion, where I out It represents the final color or grayscale value of a pixel at a specific location in the overlapping region of the output image. A I represents the color or grayscale value of image A at that pixel location. B λ represents the color or grayscale value of image B at the same pixel location. λ is a weighting coefficient between 0 and 1, representing the normalized distance from the current pixel to the boundary of the overlapping region. The closer a pixel is to image B, the more likely it is to adopt the color value of image B; conversely, the further away it is, the more likely it is to adopt the color value of image A, ensuring a smooth transition between the two images in the overlapping region, thus generating a seamlessly stitched composite image. For the stitched composite image, the cvtColor function from the OpenCV library is first used to convert the stitched RGB image to the HSV color space. This conversion process automatically decomposes the image into three independent channels: hue (H), saturation (S), and lightness (V), facilitating independent processing of the lightness component later. Next, the Limiting Contrast Adaptive Histogram Equalization (CLAHE) algorithm is applied to the brightness V channel. The image is divided into 8×8 local grid regions, and a grayscale histogram is calculated independently for each grid, followed by histogram equalization. A contrast limit threshold of 2.0 is set, and histogram regions exceeding this threshold are cropped. The cropped energy is then redistributed evenly to other regions, effectively suppressing overexposed or underexposed areas and improving illumination uniformity. Subsequently, gamma correction is applied to the CLAHE-processed V channel to further compress the brightness of highlight areas and reduce the impact of specular reflection from metallic surfaces. The correction process calculates a non-linear mapping pixel by pixel: V out =255×(V in / 255) γ V in V is the input pixel value. out Setting γ=0.8 helps reduce the dynamic range of bright areas to optimize output pixel values. After processing, OpenCV's merge function is used to re-merge the corrected V channel with the original unmodified H and S channels, forming a new three-channel HSV image. This image is then converted back to the RGB color space using the cvtColor function, resulting in a normalized composite image with uniform illumination and faithful color reproduction. This workflow achieves complete processing from multi-view image acquisition, registration, and stitching to illumination normalization, significantly improving the visual consistency and feature stability of images under different lighting and viewing angle conditions, providing high-quality input for subsequent feature extraction.

[0042] S200: Based on the composite image, visual features of the workpiece surface structure and texture distribution are extracted through a pre-trained deep convolutional network, and the process parameters of the workpiece are obtained from the process database. At the same time, the geometric distribution parameters of the pre-labeled key areas of the workpiece in the image are detected. The above visual features, process parameters and geometric distribution parameters are normalized and weighted and fused to generate a comprehensive state vector for clamping decision. The process parameters include at least the material mechanical parameters and cutting process parameters.

[0043] Specifically, the system constructs a comprehensive state vector for clamping decisions based on the generated composite image and knowledge from the process database. In this embodiment, a pre-trained deep convolutional neural network, such as the lightweight MobileNetV3 model, is first used to extract features from the composite image, outputting a set of high-dimensional vectors to characterize the structural distribution and texture characteristics of the workpiece surface. MobileNetV3 is trained under supervision on a large dataset of workpiece images, enabling it to focus on visual patterns related to clamping stability. The workpiece surface structural distribution extracted by the pre-trained MobileNetV3 includes at least: edges, pits, scratches, and positioning reference surfaces (such as grinding planes and calibration grooves) and the contour of the area to be processed (such as the boundary of the surface to be milled). The extracted workpiece surface texture characteristics include at least: roughness, texture direction, and machining marks (residual features from previous processes). Simultaneously, the system retrieves material parameters (such as elastic modulus, stiffness coefficient, and friction coefficient) and cutting process parameters (such as spindle speed, feed rate, depth of cut, and tool type) related to the current workpiece from the process database. The system detects the geometric distribution parameters of pre-labeled key structural feature regions of the workpiece in the composite image. "Pre-calibration" refers to the process of registering and marking functionally or structurally significant areas (such as positioning holes, clamping clearance grooves, and high-precision machined surfaces) in the image coordinate system based on CAD models or process drawings during the workpiece design phase or system initialization. Finally, the visual feature vectors, process parameters, and geometric distribution parameters of key structural areas are normalized. Visual feature vectors can be standardized using Z-score, process parameters can be standardized using Min-Max or Z-score, and geometric distribution parameters can be standardized using Min-Max to eliminate dimensional differences. The preprocessed multi-source features are input into a hierarchical fusion network to generate a comprehensive state vector for clamping decisions. First, the high-dimensional visual features are compressed and reduced in dimensionality through a nonlinear fully connected network to extract their compact representation. Then, according to preset weight coefficients, the dimensionality-reduced visual features, material and process parameters, and geometric distribution parameters are multiplied by corresponding fusion weight coefficients. The weighted dimensionality-reduced visual features, material and process parameters, and geometric distribution parameters are then concatenated into vectors in a predetermined order. The fusion weights reflect the relative importance of each parameter in the current machining scenario. For high-precision machining scenarios of precision boring and milling machines, the weight of geometric distribution parameters can be increased to strengthen the guiding role of the positional accuracy of key structural areas in clamping layout. The weight coefficient allocation can be preset based on historical data. Finally, a fully connected network is used to perform a nonlinear transformation on the concatenated result to output a unified comprehensive state vector.

[0044] Furthermore, the geometric distribution parameters of the pre-calibrated key areas of the workpiece in the image are detected, including:

[0045] Based on the CAD model of the workpiece, the functionally significant areas are defined as key areas, and their theoretical positions, dimensions and contour shapes in the CAD model design coordinate system are recorded.

[0046] Map the contours of key regions in the design coordinate system to the image coordinate system of the composite image to generate the corresponding binary template;

[0047] The NCC matching algorithm is used to search for the best matching position of the template in the composite image and to initially locate the key regions of the image.

[0048] Perform Canny edge detection on key regions of the image to extract the contour of the workpiece on the composite image;

[0049] The geometric transformation relationship between the designed contour and the detected contour is fitted using the least squares method. Based on this transformation relationship, the actual center coordinates, size, and contour shape of the key region in the composite image are determined and used as geometric distribution parameters for subsequent state characterization.

[0050] Specifically, the CAD model of the workpiece is analyzed to identify key functional areas, including positioning holes, clamping clearance grooves, thin-walled areas, and high-precision machined surfaces. The position (coordinates), dimensions (length, width, diameter, etc.), and contour shape (circle, rectangle, irregular shape, etc.) of each key area are recorded in the workpiece's CAD model coordinate system. Based on the obtained calibration results, the contours of the key areas in the design coordinate system are transformed to the image coordinate system, and corresponding binary templates are generated for subsequent matching processes. First, a projection transformation relationship from the design coordinate system to the image coordinate system is established using pre-calibrated camera intrinsic and extrinsic parameters. The camera intrinsic parameters include focal length, principal point coordinates, and distortion coefficients, while the extrinsic parameters are the rotation matrix and translation vector of the design coordinate system relative to the camera coordinate system. Next, the boundary vertex coordinates of key regions in the CAD model are extracted, converted to homogeneous coordinate form, and multiplied by the projection matrix for perspective transformation, obtaining the initial projection coordinates of the key region contour in the image plane. Then, the initial projection coordinates are nonlinearly corrected according to the lens distortion model to eliminate the effects of radial and tangential distortion. Finally, the corrected contour coordinates are rasterized to generate a binary template image with the same resolution as the composite image. In the template image, the pixel value inside the contour is 1, and the pixel value outside the contour is 0, completing the accurate mapping from design space to image space. This process can be implemented using OpenCV's `projectPoints` and `fillPoly` functions, ensuring the accuracy of the template's geometric features remains even under varying lighting conditions and partial occlusion. After acquiring the composite image, the NCC (Normalized Cross-Correlation) matching algorithm is used to search for the optimal matching position of the previously generated binary template in the acquired composite image to preliminarily determine the approximate location of the key region. Canny edge detection is then performed on the preliminarily located matching region to extract the true contour of the workpiece under actual imaging conditions and obtain a set of detected contour points. The design contour in the CAD model is transformed according to the preliminary positioning results to make it close to the detected contour, thereby establishing a correspondence between the design contour points and the detected contour points within a local area. Based on the corresponding point pairs, the least squares method is used to solve for the geometric transformation parameters that minimize the matching error, including translation, rotation angle, and scaling factor, to characterize the overall pose deviation of the key region. Furthermore, the degree of local shape deviation can be evaluated by analyzing the contour matching residual or fitting a non-rigid deformation model. Based on the detected contour point set, the actual center coordinates, size, and contour shape of the key region in the image coordinate system are directly fitted and calculated, and its relative spatial relationship with the surrounding structure is determined by combining the transformation parameters, thus comprehensively characterizing the geometric distribution characteristics of the workpiece in the clamping state.

[0051] S300: Used to input the comprehensive state vector into the pre-trained MLP decision model and output the target clamping force. The MLP model learns the decision behavior of the clamping decision strategy network trained based on the DDPG algorithm through the knowledge distillation method. The target clamping force generated by the strategy network is used as the decision output, and the target clamping force output by the strategy is guaranteed to be within a safe range. The main clamping stage refers to the working stage in which the clamping system applies and maintains the target clamping force during the stable machining process of the boring and milling machine.

[0052] Specifically, the MLP model is trained before system deployment. During actual machining, only forward inference is performed, without affecting the machining cycle time. The MLP model architecture adopts a three-layer fully connected structure. The number of nodes in the input layer is the same as the dimension of the integrated state vector. The hidden layer can use the ReLU activation function with a negative slope coefficient of 0.01. The output layer has one node, corresponding to the target clamping force. In the process database of intelligent manufacturing or CNC machining systems, a process record is created for each combination of "workpiece material + machining process," including at least the workpiece material, machining process, tool type, cutting parameters, minimum safe clamping force, and maximum safe clamping force. During simulation, the corresponding […] is automatically called based on the workpiece material type and cutting process code. , The parameters ensure that the clamping force of the target is within a safe and effective range.

[0053] Furthermore, the MLP model learns its decision-making behavior from the clamped decision policy network trained based on the DDPG algorithm through knowledge distillation, including:

[0054] A digital twin simulation environment for the clamping system of a boring and milling machine is constructed. This simulation environment includes a geometric data layer and a functional model layer. The geometric data layer includes the workpiece CAD geometric model and the fixture CAD geometric model. The functional model layer includes an assembly analysis module, a clamping mechanism dynamics model, a workpiece mechanical response model, a material contact characteristic model, a virtual vision perception module, and an intelligent clamping decision model.

[0055] The assembly analysis module constructs a fixture-workpiece assembly by connecting the workpiece CAD geometric model and the fixture CAD geometric model through assembly constraints. It then identifies clamping contact points on the assembly using a geometric contact detection algorithm, assigns numbers to each point, extracts the 3D coordinates and surface normal vectors of each contact point, and assigns force transmission path numbers to each contact point based on preset fixture functional topology information. Finally, it generates the structural configuration parameters for the clamping system from this contact point information.

[0056] A dynamic model of the clamping mechanism simulates the dynamic response characteristics of a clamping system composed of a hydraulic cylinder and an electro-hydraulic proportional pressure valve during the main clamping phase. Taking the received target clamping force command as input, the model outputs the real-time total clamping force generated by the hydraulic cylinder piston rod. Based on the force transmission path numbers provided by the assembly analysis module, and according to the finite element analysis of the fixture-workpiece assembly, the equivalent stiffness coefficients of each force transmission path are determined. The real-time total clamping force is distributed to each path according to the stiffness ratio, constituting the clamping force and its distribution. Simultaneously, based on the real-time total clamping force and the piston speed of the hydraulic cylinder, the instantaneous power value of the clamping system is calculated, and time integration is performed during the main clamping phase to estimate the energy consumption of the clamping system during the main clamping process.

[0057] The workpiece mechanical response model is based on the clamping force distribution on the fixture-workpiece assembly. A finite element model is constructed, and the clamping force is applied as a surface load to the corresponding contact points on the workpiece. At the same time, fixed constraints are applied in the assembly constraint relationship area. The stress distribution and elastic deformation field of the workpiece under the clamping force are obtained by analyzing the finite element model, and the maximum local deformation of the workpiece is extracted.

[0058] A material contact characteristic model stores the friction coefficients at the contact points between the workpiece and the fixture materials. During simulation, the maximum total static friction resistance of the clamping system in the sliding direction is calculated, and the estimated total cutting force in the sliding direction is obtained. The two are compared to obtain the actual safety factor, which is used to assess the risk of workpiece slippage. The estimated total cutting force comes from an external cutting force prediction model, which is called before each clamping decision. The cutting process parameters of the workpiece are retrieved from the process database to estimate the total cutting force of the boring and milling machine.

[0059] The virtual visual perception module is constructed within a digital twin simulation environment to generate visual features consistent with those extracted in reality. It is used in the training phase of the clamping decision strategy to learn the ability to respond to changes in the workpiece surface condition.

[0060] The intelligent clamping decision model is built based on the DDPG algorithm. The process of training the reinforcement learning policy network using DDPG is as follows:

[0061] S1: Randomly initialize the parameters of the Actor policy network and the Critic evaluation network, and initialize their corresponding target networks;

[0062] S2: For each time step (t), the current integrated state vector S(t) is input to the Actor policy network. Based on this, the policy network outputs the target clamping force F(t). Before outputting the target clamping force, the minimum and maximum allowable safe clamping forces under the current operating conditions are retrieved from the process database and compared and constrained to ensure that the generated target clamping force command is always within the safe clamping force window. The target clamping force is then converted into a target clamping force command and input into the digital twin simulation environment.

[0063] S3: Based on this instruction, the digital twin simulation environment runs a simulation step. After the simulation ends, it generates the next comprehensive state vector S(t+1). At the same time, based on the simulation results, it calculates the value of the multi-objective reward function and obtains the reward value R(t) of S(t). The simulation results include the maximum local deformation of the workpiece, the actual safety factor, and the energy consumption of the clamping system.

[0064] S4: Store the quadruple (S(t), F(t), R(t), S(t+1)) generated in this interaction as an experience sample in the experience replay buffer.

[0065] S5: Randomly sample mini-batch quadruplets from the experience replay buffer, and first update the parameters of the Critic evaluation network to...

[0066] To minimize its prediction error, the policy gradient of the Actor parameters is calculated based on the updated Critic network, guiding the direction of the Actor policy network parameter updates. This allows the policy to be gradually optimized to improve long-term cumulative rewards. Simultaneously, the target network parameters of both the Actor policy network and the Critic evaluation network are updated to improve training stability.

[0067] S6: Using S(t+1) as the new current integrated state vector, repeat steps S2-S5 until the reinforcement learning policy network is established.

[0068] The network converges, resulting in a well-trained clamping decision-making strategy network.

[0069] The MLP model is obtained by transferring the decision-making ability of the clamped decision policy network trained and converged by the DDPG algorithm.

[0070] Specifically, this embodiment constructs a digital twin simulation environment for the clamping system of a boring and milling machine to simulate the mechanical response of the workpiece under different clamping forces, supporting offline training for clamping decisions. This simulation environment includes a geometric data layer and a functional model layer. The geometric data layer includes the workpiece CAD geometric model and the fixture CAD geometric model. The functional model layer includes an assembly analysis module, a clamping mechanism dynamics model, a workpiece mechanical response model, a material contact characteristic model, a virtual vision perception module, and an intelligent clamping decision model. These modules and models are interconnected through clearly defined data interfaces, forming a complete closed-loop system from geometric modeling, physical simulation, state perception to intelligent decision-making. The functions of each model and module in the functional model layer are as follows:

[0071] The assembly analysis module first imports the workpiece CAD geometric model and the fixture CAD geometric model into the simulation environment, and then completes the rigid body assembly in virtual space according to the assembly constraints in the process design, such as datum surface mating, locating pin-hole mating, and stop alignment. Then, an alignment bounding box is established on the fixture-workpiece assembly to quickly determine whether spatial overlap exists. Further analysis is performed on the mesh patches corresponding to the overlapping areas, including constructing an Oriented Bounding Box (OBB) for each mesh patch and performing collision detection between patch pairs. For OBB intersecting patch pairs, their minimum Euclidean distance is calculated, and candidate patch pairs with a minimum Euclidean distance less than a preset contact distance threshold (0.1 mm) are retained. The angle between the surface normal vectors of the candidate faces is analyzed, and only patch pairs with an angle less than a preset threshold (e.g., 10°) are retained to ensure they are potential face-to-face contacts. A set of candidate faces for potential contact is selected, and candidate contact areas are filtered through this hierarchical collision detection strategy. Next, a face-to-face nearest point search algorithm is used in the candidate contact areas to identify the contact area between the fixture end face and the workpiece clamped surface. The coordinates of the contact center point and its surface normal vector are extracted, and each contact center point, i.e., the contact point, is assigned a contact point. First, all normal vectors are uniformly normalized to unit vectors, ensuring their direction points inward (i.e., the direction of pressure). Next, based on preset fixture functional topology information, such as hydraulic cylinder-gripper drive relationships and lever transmission chains, a force transmission path number is assigned to each contact point. This number identifies the actuator branch to which the contact point belongs. For example, all contact points on grippers driven by the same hydraulic cylinder share the same path number, and contact points at different positioning levels (primary positioning, auxiliary positioning) have different path priority numbers. The preset fixture functional topology information can be retrieved from the fixture design database by calling a predefined force transmission topology table or manually annotated in the simulation configuration interface. Finally, the identified contact point information is structured to generate a structural configuration parameter file for the clamping system, including the contact point number, three-dimensional coordinates of the contact point, unit normal vector of the contact point, the fixture component to which it belongs, and the force transmission path number.

[0072] The clamping mechanism dynamics model is used to simulate the dynamic response characteristics of the hydraulic cylinder and electro-hydraulic proportional pressure valve in the clamping system during the main clamping phase. Taking the received target clamping force command as input, it simulates the pressure-flow characteristics of the hydraulic system and the dynamics of the actuator to output the real-time total clamping force generated by the hydraulic cylinder. That is, based on the force transmission path number provided by the assembly analysis module, all contact points are divided into several force transmission subsystems. Based on the finite element analysis of the fixture-workpiece assembly, the equivalent stiffness coefficient of each force transmission path is determined, and the output real-time total clamping force is distributed to each path according to the stiffness ratio.

[0073] ;

[0074] in, For the first The clamping force received by the force transmission path at time t. For the first The equivalent stiffness of the force transmission path. For the sum of stiffness of all force transmission paths, The total clamping force at time t output by the dynamic model is weighted according to the local stiffness of the contact point within each path, based on the force distribution weights. Based on this, Decomposed into normal forces at each contact point:

[0075] ;

[0076] in, Let be the normal force at the i-th contact point at time t. In the first In the path, the local force distribution weighting coefficient at contact point i is used to ultimately obtain the time-varying directional force at each contact point on the workpiece surface. The sequence constitutes the clamping force distribution. Here, the target clamping force command is converted into a control signal u through the controller model, with a value range of [0, 1], corresponding to a voltage of 0-10V or a current of 4-20mA in a real system. The target clamping force output by this model... To estimate the energy consumption of the main clamping process, the instantaneous power value P(t) = F(t)·v(t) is integrated over the main clamping time interval. As a proxy for the energy consumption of the hydraulic system, v(t) represents the piston speed of the cylinder, and dt represents the time increment of the integral. This dynamic model provides high-fidelity physical constraints for the offline training of the clamping decision strategy, ensuring that the clamping process conforms to the physical limitations of the actual actuator.

[0077] The workpiece mechanical response model, based on the geometric and contact relationship of the fixture-workpiece assembly, retrieves the workpiece material mechanical parameters from the process database and receives the clamping force distribution output by the clamping mechanism dynamic model. This response model constructs a static finite element model that considers contact nonlinearity, applies the clamping force as a surface load to the corresponding contact area, and applies fixed constraints to the contact area of ​​the positioning element, thereby accurately calculating the stress distribution and elastic deformation field of the workpiece under the action of clamping force, and outputting the maximum local deformation as a key basis for evaluating workpiece deformation control.

[0078] The material contact characteristic model stores the friction coefficients at the contact points when the workpiece and fixture materials are paired. During the clamping system simulation, it calls upon the contact point information from the structural configuration parameters and the real-time total clamping force output from the clamping mechanism dynamics model. Based on Coulomb's law of friction, it calculates the clamping force at each contact point i under the current target clamping force. The maximum static friction force is determined, and then the direction of the total cutting force vector is determined based on the current cutting process and tool path. This allows us to determine the most likely direction of workpiece slippage. For example, if the total cutting force vector is along the positive X-axis of the airport coordinate system, then the slippage risk direction is along the negative X-axis. Based on this, the maximum total static friction resistance of the clamping system in the slippage direction is obtained, as follows:

[0079] For the i-th contact point, the maximum static friction force is :

[0080] , i=1,2,3,...,n;

[0081] in, Let be the maximum static friction force at the i-th contact point at time t. Let be the coefficient of friction at the i-th contact point where the workpiece and the fixture material are mated. Let be the normal force at the i-th contact point at time t. Project the maximum static friction force vector of all contact points onto the anti-slip direction, and sum the magnitudes of these projected forces to obtain the maximum total static friction resistance that the entire clamping system can provide in that slip direction. Then, the estimated tangential force is projected onto the determined slip direction to obtain the magnitude of its component in that direction. Calculate the actual safety factor, η = / It is used to assess the risk of workpiece slippage.

[0082] A virtual visual perception module is used during the training phase of clamping decision strategies to learn the ability to respond to changes in the workpiece surface state. Built within a digital twin simulation environment, the virtual visual perception module generates visual features consistent with the actual deployment phase. Based on the fixture-workpiece assembly, and combined with pre-set material surface statistical parameters from the process database, such as surface roughness, texture direction, and previous machining marks, as well as common workpiece defect patterns like edges, pits, scratches, and positioning reference surfaces, two virtual industrial cameras are configured with positions and parameters identical to the actual deployment system. A physical rendering engine is used to generate synthetic images containing the following perturbations: illumination changes, surface disturbances, and imaging noise. The generated synthetic images visually and semantically simulate workpiece surface images captured by real industrial cameras. The generated synthetic images are input into a pre-trained deep convolutional network identical to that used in the actual deployment phase to extract high-dimensional feature vectors of the workpiece surface structure and texture distribution. These feature vectors serve as "simulated visual features" in the construction of the training state vectors.

[0083] An external integrated cutting force prediction model is invoked before each clamping decision. It first predicts the tangential force on the workpiece being machined on the precision boring and milling machine using the mechanical power method, specifically as follows: The cutting process parameters for the current workpiece, including the axial depth of cut 'a', are retrieved from the process database. p (mm), radial depth of cut a e (mm), feed rate V f (mm / min), spindle speed N (rpm), number of tool teeth Z, tool diameter D (mm), and unit cutting energy K corresponding to the workpiece material. c (N / mm²) (e.g., 800–1100 N / mm² for aluminum alloys, 2500–4000 N / mm² for titanium alloys), calculate the tangential force during workpiece machining. The material removal rate (MRR) is calculated as follows: MRR = a p ·a e ·V f (mm³ / min), calculate the cutting power P c :P c =K c ·MRR / 60(W), calculate the cutting speed V: V=π·D·N / 60000(m / s), calculate the tangential cutting force : =P c / V(N), then calculate the total cutting force for machining the workpiece. =K· (N), K are empirical coefficients, and the calculated Magnification is used to estimate the total cutting force during precision milling. The total cutting force is a three-dimensional vector, including tangential force, radial force and axial force. The coefficient K is not a fixed value. It depends on the tool geometry, cutting type and workpiece material, etc. The range of K value is [1.5, 2.5]. This estimated main cutting force serves as the benchmark input for workpiece slip risk assessment.

[0084] In the aforementioned simulation environment, a reinforcement learning training framework is constructed, defining the comprehensive state vector S, action space, and reward mechanism. The comprehensive state vector S includes visual feature samples, geometric distribution parameter samples, material mechanics parameter samples, and cutting process parameter samples, derived from the DDPG algorithm's observations of the current state of the digital twin simulation environment. This reflects the instantaneous state of the twin environment. The sample parameters are normalized and weighted to form a comprehensive state vector S with n dimensions (n≥100), comprehensively representing the key factors affecting the clamping effect under the current working condition. A multi-objective reward function is designed. ,in, As a workpiece deformation penalty term, the finite element analysis results of the workpiece geometric model are used, and the negative value of the maximum local deformation is taken as the reward value. , where k d Δd is the deformation penalty coefficient. max To maximize local deformation, ensure that excessive deformation is suppressed first during the strategy learning process; As a reward for the energy consumption of the clamping system, the negative value of the integral of the actual clamping force output from the dynamic model of the clamping mechanism is used as a penalty. , where k e The energy consumption penalty coefficient guides the strategy to reduce resource usage while meeting constraints, including: the clamping force is sufficient to resist the estimated cutting force to prevent workpiece slippage, the maximum workpiece deformation does not exceed the process allowable value to avoid machining errors, and the clamping force is within the physical realization range of the clamping system. As a penalty for slippage risk, a negative reward is introduced if the actual safety factor η is less than a preset safety threshold (e.g., 1.5) to prevent relative slippage during processing; weighting , , Adjust dynamically according to the required machining accuracy.

[0085] Based on the above settings, the Deep Deterministic Policy Gradient (DDPG) algorithm is used to train the reinforcement learning policy network. The training process relies on the constructed digital twin simulation environment of the boring and milling machine clamping system. The reinforcement learning policy network receives the comprehensive state vector S as input and outputs the target clamping force. During the training process, the controller model converts the target clamping force into a control signal and inputs it into the digital twin simulation environment. Based on the assembly analysis module, the material mechanical property model, and the clamping mechanism dynamic model, the stress distribution of the workpiece under the target clamping force is simulated through finite element analysis to obtain the maximum local maximum deformation. The calculated workpiece slippage risk is called, and the calculated hydraulic system energy consumption is also called. Subsequently, the system calculates the comprehensive reward value R of this decision according to the preset multi-objective reward function. The comprehensive reward value R of this decision and the next comprehensive state vector are fed back to the DDPG algorithm to form the feedback signal of reinforcement learning. The training process is as follows: Construct the Actor network (policy network) and Critic network (value function network), and configure the corresponding target networks (TargetActor and TargetCritic). Network parameters are randomly initialized. Establish an experience replay buffer to store the integrated state vector of the DDPG algorithm interacting with the twin digital environment. In each training round, DDPG observes the current integrated state vector S(t). The Actor network outputs the action F(t), which, after mapping, is input into the simulation environment to trigger the clamping process simulation. The simulation environment returns an immediate reward R. tThe new integrated state vector S(t+1) is used to store the quadruple (S(t), F(t), R(t), S(t+1)) into the experience replay buffer. The current integrated state vector S(t) is the observation of the current state of the digital twin simulation environment by the DDPG algorithm at time step t, i.e., the t-th decision time. Its data is all derived from and reflects the instantaneous state of the twin environment in real time. The vector is constructed and fused from the twin environment through the following process: Visual feature samples: Based on the jig-workpiece assembly geometric model in the twin environment, combined with the material surface properties preset in the process database, a synthetic image representing the current environmental state is generated through the virtual visual perception module and physical rendering engine. This image is input into the pre-trained MobileNetV3 network to extract the feature vectors of the workpiece surface structure and texture distribution; Based on the jig-workpiece assembly geometric model in the twin environment, the spatial coordinates, dimensional tolerances, shape and other geometric parameters of the key positioning area of ​​the workpiece are extracted; According to the workpiece and jig material types defined in the twin environment, the corresponding material properties are called from the process database, including elastic modulus, Poisson's ratio, density, and the static friction coefficient of the interface at each contact point, to form material mechanical parameter samples; According to the current simulated machining process, the cutting parameters, including ap, ae, Vf, N, Z, D and Kc, are called from the process database to form cutting process parameter samples. The above four types of feature vectors are concatenated and the concatenated vector is normalized. In each training iteration, the system randomly selects a batch of historical interaction data from the experience replay buffer for updating network parameters. First, the Critic network (value function network) is optimized: based on the current state and the action taken, the Critic network evaluates the expected long-term reward value of the action; simultaneously, it uses the target network to predict the optimal action reward in the next state, and combines this with the current immediate reward to calculate a target reward value. The Critic network continuously adjusts its internal parameters by comparing the difference between its own prediction value and the target reward value, gradually reducing this prediction bias and thus improving its evaluation accuracy. Subsequently, the Actor network (policy network) is updated: based on the Critic network's evaluation feedback on the action, the Actor network adjusts its policy parameters along the direction that maximizes the expected reward. The system calculates the sensitivity of the reward value to the action and the dependence of the action on the policy parameters, updating the weights of the Actor network through gradient ascent, making it more inclined to output clamping forces that yield higher rewards in similar states. To improve the stability of the training process, the system adopts a target network mechanism, which sets up independent TargetActor and TargetCritic networks. By slowly tracking, the network gradually approaches the latest parameters of the main network at a small scale, thereby effectively suppressing oscillations and divergences during the learning process.As the number of training rounds increases, the decision-making performance of the policy network tends to stabilize and reach a convergent state, resulting in the final clamping decision policy network.

[0086] Furthermore, the MLP model is obtained by transferring the decision-making ability of the clamped decision policy network trained and converged by the DDPG algorithm, including:

[0087] The trained clamping decision policy network is run in the constructed digital twin simulation environment to generate various comprehensive state vectors S. i The target clamping force F output by the clamping decision strategy network in this state is collected. i ;

[0088] The obtained state-action pairs (S) i F i ), constructing a distillation dataset D={(S i F i The state vectors are preprocessed by normalizing them.

[0089] The MLP model structure is designed based on the dimension of the comprehensive state vector. The number of neurons in the input layer is consistent with the dimension of the comprehensive state vector. The hidden layer uses the ReLU activation function, and the output layer is a single-neuron linear unit used to predict the target clamping force.

[0090] Using the distillation dataset D as the training sample, a supervised learning approach is adopted to train the MLP model to mimic the policy decision-making ability of DDPG by minimizing the mean square error between the MLP prediction output and the clamping decision policy network output.

[0091] The trained MLP model is compressed using INT8 quantization, and the compressed lightweight MLP model is deployed to the industrial control computer of the boring and milling machine.

[0092] Specifically, a digital twin-driven synthetic data distillation method is employed. A clamping decision strategy network (the teacher model), trained and converged using the DDPG algorithm, is run in a digital twin simulation environment. This network iterates through the geometric types of the fixture-workpiece assembly and combines them with preset material mechanics parameters and cutting process parameters from the process database. This generates a comprehensive state vector covering multiple machining scenarios within the simulation environment. The output of the clamping decision strategy network under each state is then collected to form state-action pairs (S...). i F i Construct a distillation dataset D={(S) containing M samples. i F iLet |i=1,2,...,M}. All state vectors are preprocessed by normalization to ensure consistent input distribution. A multilayer perceptron (MLP) is constructed as the student model. A supervised learning method is adopted, using the distillation dataset D as the training sample. The goal is to minimize the mean squared error between the MLP's predicted output and the DDPG teacher model's output. The MLP model is trained to mimic the policy decision-making ability of DDPG. The trained MLP model is compressed using INT8 quantization, converting floating-point weights and biases into 8-bit integer representations. While maintaining decision accuracy, this significantly reduces the model's storage space and inference computation. The compressed lightweight model is deployed to the machine tool industrial control computer, and a state vector input interface is established and connected to the output port of the electro-hydraulic proportional pressure valve controller.

[0093] S400: Based on the obtained target clamping force, the controller generates a corresponding target current command according to the preset pressure-current mapping relationship. This command drives the electro-hydraulic proportional valve to adjust the output force of the hydraulic cylinder, thereby driving the clamping mechanism to complete the workpiece clamping.

[0094] Specifically, this step is implemented to convert the target clamping force output in the preceding steps into actual executable mechanical control commands. This can be achieved by configuring an independent industrial controller for the clamping system, which is a high-performance PLC or industrial PC, and interacts with the CNC controller of the precision boring and milling machine through a standard communication interface. The controller of the clamping system receives the target clamping force and, combined with the known piston area of ​​the cylinder, calculates the required target hydraulic oil pressure. Based on the target hydraulic oil pressure, it uses a lookup table method to find the corresponding target current in a preset pressure-current mapping relationship. The target current value is usually in the range of 0-2A or 4-20mA. The controller writes the calculated target current into the register of the AO module. The AO module converts it into a real current control signal output. The output terminal of the AO module is connected to the solenoid coil of the electro-hydraulic proportional valve, driving the valve core to move and adjust its valve opening, thereby controlling the oil pressure and flow rate delivered by the hydraulic system to the hydraulic cylinder, driving the clamping assembly to complete the reliable clamping of the workpiece. The clamping assembly is a hydraulically driven linear motion unit, including a hydraulic cylinder, piston rod, and clamping plate or claw assembly linked with it, used to convert hydraulic energy into mechanical clamping force applied to the workpiece. The target clamping force is converted into control commands for the electro-hydraulic proportional valve. This can be found in the product manual, which usually provides detailed pressure-current characteristic curves. Alternatively, a lookup table method can be used. During the testing phase of the electro-hydraulic proportional valve, different currents are applied to the valve, and its output pressure is measured to obtain a set of data points (Ii, Pi). Using curve fitting algorithms such as the least squares method, this set of data points is fitted into a continuous pressure-current characteristic curve function P=f(I). This continuous function is then discretized to generate a pressure-current mapping lookup table. The clamping system must maintain a stable clamping force throughout the entire cutting process and must not loosen. Therefore, the main clamping stage refers to the entire process from the completion of clamping and the start of cutting on the precision boring and milling machine to the end of machining. During this stage, the clamping system is under dynamic load and requires real-time monitoring and compensation. Vibration signals are obtained by installing miniature vibration sensors near the clamping body or pressure plate, or by using the vibration monitoring module built into the precision boring and milling machine; temperature signals are obtained by placing temperature sensors at the hydraulic oil inlet / valve block oil circuit and near the workpiece clamping area; and the output force of the hydraulic cylinder is obtained in real time by installing pressure sensors at the outlet of the electro-hydraulic proportional valve (near the actuator side) or the hydraulic cylinder inlet to ensure that the measured pressure is the actual pressure acting on the cylinder.

[0095] S500: During the main clamping stage, vibration signals, temperature signals, and hydraulic oil pressure signals of the clamping system are collected simultaneously. Using a preset time window as the unit, the vibration and temperature signals are extracted in the time domain to obtain the vibration time domain characteristics and temperature time domain characteristics. The average value of the hydraulic oil pressure within the window is calculated. Based on the known effective piston area, the hydraulic oil pressure is converted into the current output force of the hydraulic cylinder. The current output force is input into the calibrated force transmission model to obtain the total effective clamping force on the workpiece. The clamping force deviation between the target clamping force and the total effective clamping force is calculated.

[0096] Specifically, clamping force deviation is a result-oriented indicator, only indicating that the current total effective clamping force deviates from the target clamping force, but it cannot reveal the cause and nature of the deviation. Vibration and temperature signals are process-oriented or causal indicators, containing rich information about the source and characteristics of disturbances. Vibration signals are extremely sensitive to high-frequency, instantaneous disturbances, cutting impacts, and spindle vibrations. Temperature signals show significant reactions to low-frequency, slow disturbances (such as fixture thermal deformation, hydraulic oil viscosity changes, valve body thermal drift, etc.). Clamping force deviation must be analyzed synchronously with vibration and temperature signals to comprehensively diagnose the "health status" of the system, thereby achieving intelligent compensation decisions. In most cases, the clamping system of precision boring and milling machines typically adopts a single-cylinder driven multi-point linkage clamping mechanism. The core structure is a centralized power source (a single hydraulic cylinder) that transmits force synchronously and proportionally to multiple clamping points through a mechanical force distribution mechanism. The most common form of force distribution mechanism is the lever-type force distribution mechanism, which is the most frequently used structure. The output force of the hydraulic cylinder is distributed to two or four clamping points according to the lever ratio through one or more rigid levers. The force loss to be compensated mainly comes from the friction and deformation of the force distribution mechanism. The total effective clamping force estimated by the model is the actual effective clamping force acting on the workpiece.

[0097] Furthermore, the calibrated force transmission models include:

[0098] Obtain the material, size, and hardness information of the workpiece, process calibration blocks with the same properties, and install force sensors at the bottom of the calibration blocks;

[0099] The calibration block is installed into the fixture, and the controller sends N levels of current signals. At each level, the hydraulic oil pressure P and the total effective clamping force monitored by the force sensor are simultaneously acquired. The hydraulic oil pressure P, combined with the known effective area A of the cylinder piston, is used to calculate the current output force of the hydraulic cylinder in real time. The data is collected M times at each level, and the average value is taken to obtain a dataset of N points. j = 1, 2, ..., N, where N ≥ 5 and M > 1;

[0100] A linear model is directly fitted between the current output force and the total effective clamping force of the hydraulic cylinder to establish an overall force transmission model. = + j=1,2,…,N,

[0101] The parameters are obtained using the least squares method. and , The total mechanical efficiency of the clamping system, <1,

[0102] This refers to the total force loss of the clamping system;

[0103] Will and Packaging is a force transmission model. Solidify model parameters, and The force transmission model is written to the controller's non-volatile memory and marked as a calibrated model.

[0104] Specifically, this embodiment determines specific parameters in a known model structure through physical experimental measurement. First, a calibration fixture is developed to obtain the workpiece's material properties, dimensions, and hardness information. A monolithic calibration block with identical properties is then manufactured, allowing one calibration block to serve multiple production lines, thus reducing costs. A high-precision force sensor is installed between the calibration block and the machine tool table (i.e., the bottom of the workpiece). The total clamping force applied by the fixture through the pressure head is equal to the support force of the table on the workpiece. The force sensor measures this support force. For clamping forces with diverse directions, such as in 5-sided machining, mechanically, it represents the sum of all clamping forces applied to the workpiece by the fixture system, resisting the sum of cutting forces in any direction. Because the purpose of calibration is to anchor reality, the task of calibration is to establish a mapping, ensuring that when the controller commands a certain number of Newtons of hydraulic cylinder pressure, it knows and ensures that the actual number of Newtons of normal force acts on the workpiece. The total clamping force is the foundation of the spatial force system. Regardless of the direction of the cutting force, the workpiece resists slippage primarily through the frictional force on the contact surface of the fixture. The magnitude of friction is proportional to the normal force (i.e., normal clamping force) on the contact surface. Therefore, the target clamping force is essentially a total, composite normal clamping force, and what the single sensor measures at the bottom is precisely this total, composite normal support force; their physical essence is the same. Calibration compensates for "system internal friction," not "machining conditions." This internal friction is independent of the machining direction. Whether machining on three or five sides, the inherent force transmission efficiency of the system remains essentially unchanged. Therefore, using a unidirectional calibration to measure and compensate for this inherent, unidirectional internal friction is reasonable. Secondly, multi-level calibration is performed. The calibration fixture is installed in the actual fixture, and the clamping system controller sends N levels of current signals, i.e., N different current signals, N≥5. The design principle for the N levels of current signals is to follow a uniform distribution principle or a boundary point principle within the current signal value range. At each level, the hydraulic cylinder pressure P and the total effective clamping force measured by the force sensor are simultaneously collected. The hydraulic cylinder pressure P, combined with the preset cylinder piston area A, is used to calculate the current output of the hydraulic cylinder in real time. =P×A, repeat each level M times (M>1), take the average, reduce noise, and obtain a dataset of N points. Let j = 1, 2, ..., N. Then, a force transmission model is established, directly fitting a linear model between the cylinder output force and the total effective clamping force. = + Using the least squares method, find an optimal pair. and The value of makes the dependent variable in this linear model... With independent variable To minimize the sum of squared errors, a set of optimal parameters is obtained. and , The total mechanical efficiency of the system, <1 represents the overall transmission efficiency from the hydraulic cylinder output force to the effective force of the workpiece. This represents the total force loss of the system, primarily indicating the static friction force that needs to be overcome. Next, we will... and Packaging is a force transmission model. Finally, the model parameters are solidified. and Write the data to the controller's non-volatile memory, mark it as a calibrated model, and call it during the runtime phase.

[0105] S600: Combine the extracted vibration time-domain features and temperature time-domain features with the clamping force deviation to construct a normalized current multimodal disturbance observation vector, which takes the form of [vibration RMS, vibration kurtosis, hydraulic oil temperature change rate, clamping temperature change rate, clamping force deviation].

[0106] Specifically, during the main clamping phase, a unified high-precision clock is used as the synchronization reference to achieve multi-rate synchronous acquisition of hydraulic oil pressure, clamping system vibration, hydraulic oil temperature, and fixture temperature signals. All signals are accurately timestamped during acquisition and aligned based on a unified clock source to ensure strict synchronization of physical quantities on the time axis in subsequent analyses. For the hydraulic oil pressure signal, with a 100ms analysis period, the arithmetic mean of the most recent 100ms sampled data within this period is calculated and converted into the current output force of the hydraulic cylinder, thereby obtaining the clamping force deviation. Vibration signal processing: with a 100ms analysis period, a sliding time window method is used to extract the latest continuous 1s of raw vibration data each time, and its root mean square (RMS) value and kurtosis are calculated. For the temperature signal, the same time window method is used, and linear fitting is employed to perform linear regression on the hydraulic oil temperature signal and the fixture temperature signal respectively. The resulting slopes are the hydraulic oil temperature change rate and the fixture temperature change rate, respectively. For the hydraulic oil pressure signal, the output force at the last moment within the preset time window is calculated. The Min-Max normalization method is used to map each feature to the [0, 1] interval, resulting in a normalized multimodal perturbation observation vector [vibration RMS, vibration kurtosis, hydraulic oil temperature change rate, fixture temperature change rate, clamping force deviation]. Given the large dynamic range and potential extreme values ​​of the vibration kurtosis feature, direct min-max normalization would lead to a loss of effective feature discrimination. Therefore, preprocessing is necessary before normalization. First, a natural logarithmic transformation, K' = ln(K + 1), is applied to the original vibration kurtosis value to map it to a reasonable linear interval. Then, the processed vibration kurtosis feature K' is combined with other features for min-max normalization.

[0107] S700: Input the current multimodal disturbance observation vector into the pre-constructed fuzzy rule model, and generate a synthetic fuzzy set of hydraulic compensation factors through fuzzification, rule activation, implication operation and maximum value aggregation. Defuzzify the hydraulic compensation factor in the domain [0,1] using the centroid method, calculate the geometric center of the synthetic fuzzy set, and use the calculation result as the output value of the hydraulic compensation factor. The fuzzy rule model is constructed by a fuzzy decision tree.

[0108] Furthermore, pre-built fuzzy rule models include:

[0109] Historical operating time series data of a precision boring and milling machine is acquired, covering normal and various fault states of the equipment. The historical data is used to extract features such as vibration RMS, vibration kurtosis, hydraulic oil temperature change rate, fixture temperature change rate, and hydraulic oil pressure window mean value. Based on the hydraulic oil window mean value and the known effective piston area, the hydraulic cylinder output force is calculated and regarded as the total effective clamping force. The clamping force deviation is obtained by subtracting it from the set clamping force.

[0110] The sample values ​​of the five features obtained, namely vibration RMS, vibration kurtosis, hydraulic oil temperature change rate, fixture temperature change rate, and clamping force deviation, are combined in a fixed order to form a single multimodal disturbance observation sample vector. After completing the above operation on all historical data, all the obtained sample vectors are normalized to obtain a multimodal disturbance observation sample vector set for training. At the same time, experts label the corresponding hydraulic compensation factor value for each multimodal disturbance observation sample vector in the sample vector set according to the severity of the fault and the compensation requirements. The label value range is [0, 1]. The set composed of all sample vectors and their hydraulic compensation factor values ​​constitutes the training dataset.

[0111] Define a linguistic value for each feature in the multimodal disturbance observation sample vector, which includes low, medium, and high. At the same time, define a linguistic value for the hydraulic compensation factor, which includes zero, small, medium, and large. Configure a corresponding membership function for each of the above linguistic values. The membership function adopts a triangular or trapezoidal distribution. Initialize the parameters of each membership function to generate the corresponding initial fuzzy set for each of the above linguistic values, and preset the global parameters of the fuzzy decision tree.

[0112] The process of constructing a tree structure using fuzzy decision trees:

[0113] S1: Starting from the root node, iterate through the five features of the multimodal perturbation observation sample vector. For each feature, iterate through each of its linguistic values, and use each feature-linguistic value combination as a candidate splitting condition to be evaluated.

[0114] S2: For each candidate split condition, extract the sample values ​​of all sample vectors in the training dataset on that feature, and substitute each sample value into the membership function corresponding to that linguistic value to calculate the membership degree belonging to that linguistic value. Based on the membership degrees calculated from the sample values, calculate the fuzzy information gain ratio of that candidate split condition.

[0115] S3: At the current node, compare the fuzzy information gain ratios of all candidate splitting conditions, select the feature-language value combination with the largest gain ratio as the optimal splitting condition for the current node, and generate two child nodes: "belonging" and "not belonging". For each sample vector and its hydraulic compensation factor value in the training dataset, calculate its membership degree with the linguistic value of the optimal splitting condition based on its data on the feature specified by the optimal splitting condition. The sample vector and its hydraulic compensation factor value, carrying the weight of the membership degree, enter the "belonging" child node. At the same time, the sample vector and its hydraulic compensation factor value, carrying the weight of 1 minus the membership degree, enter the "not belonging" child node. The sample vector and its hydraulic compensation factor value remain unchanged during the allocation process, forming the training dataset of each child node. The weights carried are used in all subsequent recursive calculations. Take each non-leaf child node as the new current node and recursively repeat the above splitting process until the preset stopping condition is met. Nodes that fail to continue splitting become leaf nodes, and a fuzzy decision tree structure with no assigned output is constructed.

[0116] S4: For each leaf node in the fuzzy decision tree structure, based on all sample vectors in that node and the cumulative path weight from the root node to that node, the membership degree of each linguistic value of the hydraulic compensation factor is weighted and aggregated. The linguistic value with the highest aggregated membership degree is selected as the output of that leaf node, thus completing the construction of the fuzzy decision tree.

[0117] S5: Based on the completed fuzzy decision tree, traverse each path from the root node to the leaf node, connect the splitting conditions (feature-language value combinations) of all non-leaf nodes on the path using the AND logical operator to form the antecedent (IF part) of the rule, and use the output language value assigned to the leaf node as the consequent (THEN part) of the rule to form a complete IF-THEN fuzzy rule. The rules corresponding to all paths together constitute the initial fuzzy rule set.

[0118] This rule set combines the linguistic values ​​of all features in the multimodal disturbance observation sample vector, the linguistic values ​​of the hydraulic compensation factor, and their corresponding membership functions and initial parameters to form an initial fuzzy rule knowledge base;

[0119] S6: Optimize the initial fuzzy rule base, encapsulate and store the optimized fuzzy rule set and the optimal parameters of the membership function corresponding to each language value in a structured data format to form an optimized fuzzy rule knowledge base. By introducing a fuzzy inference engine and combining it with this knowledge base, a fuzzy rule model that can be embedded in the control system is formed.

[0120] Specifically, the constructed fuzzy rule model is used to solve the compensation force problem. The compensation control objective of the hydraulic clamping system is to maintain the stability of the clamping force. The fuzzy rule base is a set of fuzzy inference rules in the form of "IF-THEN", which describes the fuzzy mapping relationship between the input five-dimensional normalized features and the output hydraulic compensation factor u.

[0121] In the data acquisition and preprocessing stage, historical operating time series data of the precision boring and milling machine is acquired. This data covers all important states of the equipment, including normal state and various fault states. Normal state includes stable operating data under different workpieces and different cutting parameters. Various fault states include direct disturbance sources such as tool wear, bearing failure, insufficient guide rail lubrication, hydraulic system leakage and insufficient clamping force. The acquired historical operating time series data includes at least the original time series data such as clamping system vibration, hydraulic oil temperature, fixture temperature and hydraulic oil pressure. For the acquired historical operating time series data, consistent with the sliding time window method mentioned in the S600 implementation, the vibration RMS, vibration kurtosis, hydraulic oil temperature change rate, and fixture temperature change rate are calculated for the data within each window, and the average hydraulic oil pressure within that window is also calculated. Based on this average value and the known effective piston area, the output force of the hydraulic oil is calculated and used as the total effective clamping force for that time period. The difference between this and the set clamping force is then calculated to obtain the clamping force deviation. The feature values ​​of the five characteristics—vibration RMS, vibration kurtosis, hydraulic oil temperature change rate, fixture temperature change rate, and clamping force deviation—are combined into a five-dimensional feature vector, which serves as the multimodal disturbance observation sample corresponding to that time window. By traversing the entire time series through the sliding window, a set of multimodal disturbance observation sample vectors is generated. The clamping force set by the system comes from human experience and can be a fixed parameter preset in the machining program (G-code), based on drawing requirements and historical experience. To label the output hydraulic compensation factor u for each sample vector, the labeling rules are as follows: for sample vectors in normal state, u=0, indicating that no compensation is needed; for sample vectors in fault state, experts assign a value between 0 and 1 according to the severity of the fault and the compensation requirements. The normalized multimodal disturbance observation sample vectors and the corresponding hydraulic compensation factor u values ​​are then used to form a training dataset.

[0122] In the initialization phase of the fuzzy rule model, linguistic values ​​{low, medium, high} are defined for each sample feature in the sample vector, and the hydraulic compensation factor u is defined as {zero, small, medium, large} linguistic values. The sample vector has 5 features, each with 3 linguistic values ​​and 1 hydraulic compensation factor. Four linguistic values ​​are preset, representing the aforementioned 15 feature linguistic values ​​and 4 hydraulic compensation factors u.

[0123] Each node is initialized with a separate membership function, typically a triangular or trapezoidal function. The membership function parameters are usually initialized using a uniform distribution strategy to cover the entire theoretical domain and provide a reasonable starting point for subsequent optimization. Global parameters of the fuzzy decision tree are preset, including maximum depth, minimum number of samples per node, minimum sum of split weights, minimum gain required for splitting, maximum number of leaf nodes, and fuzzy aggregation operators. Termination conditions for fuzzy decision tree learning are set, including: reaching the preset maximum depth, all samples belonging to the same class (or very small output variance), and fuzzy information gain ratio being less than a threshold.

[0124] In the fuzzy decision tree learning phase, a fuzzy decision tree (such as the fuzzy ID3 algorithm) is used to construct the tree structure and generate candidate splitting conditions. Based on the training dataset, starting from the root node, each feature Xj and each linguistic value Ak (such as the fuzzy linguistic values ​​low, medium, and high for vibration sample RMS) are traversed to generate each feature-linguistic value combination (Xj, Ak), such as (vibration RMS, high), (vibration RMS, medium), (vibration RMS, low), (hydraulic oil temperature change rate, high), ..., each combination represents a potential splitting scheme; the candidate splitting conditions are evaluated. For each candidate splitting condition (Xj, Ak), such as (vibration RMS, high), the following calculations are performed: a) Calculate the membership degree: extract the specific numerical values ​​of all samples in the dataset on feature Xj, substitute each numerical value into the membership function corresponding to the linguistic value Ak, and calculate its membership degree to Ak. For example, calculate the degree to which each vibration RMS value belongs to "high"; b) Calculate the fuzzy information gain ratio: Based on the category distribution of all samples at the current node, calculate the fuzzy information gain ratio that would result from fuzzily dividing the samples into two child nodes ("belongs to Ak" and "does not belong to Ak") using "the degree to which the sample belongs to Ak" as the splitting criterion. Select the optimal splitting condition by comparing the fuzzy information gain ratios calculated from all candidate splitting conditions (Xj, Ak). Select the feature-linguistic value combination with the highest gain ratio as the optimal splitting condition for the current node. For example, (hydraulic oil temperature change rate, high) has the highest gain ratio and is the splitting rule for the current node. Perform fuzzy splitting, simultaneously assigning each training sample of the current node to two new child nodes: the weight of a sample entering the "belongs to" child node is equal to its membership degree of the hydraulic oil temperature change rate value belonging to "high". The weight of a sample entering the "does not belong to" child node is equal to 1 minus the above membership degree. All original feature values ​​of the samples remain unchanged during this process; only their weights are reassigned. The newly generated "belongs to" and "does not belong to" child nodes are used as the new current nodes. This process is recursively repeated to generate candidate splitting conditions until fuzzy splitting is performed. The optimal splitting condition is searched for each new node until any of the preset stopping conditions from the initialization phase are met, forming leaf nodes. At this point, an unpruned fuzzy decision tree is initially constructed. The calculation of the fuzzy information gain ratio is a quantitative calculation process based on fuzzy set theory and information entropy, aiming to evaluate and select the most discriminative splitting condition from numerous candidate conditions. During the construction of the fuzzy decision tree, to evaluate candidate splitting conditions and select the optimal splitting rule, the entire process for each condition to be evaluated (e.g., "vibration RMS is high") is as follows: Calculate the sample membership degree. Based on the membership function corresponding to the linguistic value, calculate the degree to which each sample in the parent node satisfies the condition, i.e., its membership degree μS. 高(Xj) is used to quantify the degree to which a sample conforms to the fuzzy concept of "high vibration"; fuzzy splitting (soft assignment) is performed. This soft assignment process does not change any original values ​​of the sample vector and the hydraulic compensation factor; it simply calculates two new weight values ​​for each sample to represent the degree of belonging or contribution of the sample to the newly generated "belongs to" and "does not belong to" child nodes, with weights μS. 高 (Xj) Assign the sample to the "belongs to" branch (yes) with a weight of 1-μS. 高 (Xj) Assigning samples to the "Does not belong" branch (No) achieves soft partitioning, allowing samples to partially belong to multiple branches simultaneously, reflecting the modeling ability of fuzzy logic for uncertainty and transitional states. Calculating the weighted fuzzy entropy of child nodes involves calculating the fuzzy entropy of each of the two child nodes separately, measuring the uncertainty of their output category (e.g., the linguistic value of the hydraulic compensation factor). Then, using the proportion of the total weight received by each child node to the total weight of the parent node as a weighting coefficient, the fuzzy entropy of the two child nodes is weighted and summed to obtain the expected fuzzy entropy under this splitting method. Calculating the fuzzy information gain involves subtracting the expected fuzzy entropy from the fuzzy entropy of the parent node to obtain the fuzzy information gain for this splitting condition. This value reflects the effectiveness of the condition in reducing output uncertainty; a larger gain indicates stronger classification ability. Calculating split information is introduced to avoid the algorithm from generating too many or unevenly distributed branches; it is defined as the information entropy based on the weight proportion of each child node. The more uniform the weight distribution, the larger the splitting information value, indicating a higher data partitioning complexity resulting from the split. The fuzzy information gain ratio is calculated by dividing the fuzzy information gain by the splitting information value. This ratio comprehensively considers the classification effect and partitioning complexity of the split, effectively balancing the gain magnitude and the model's generalization ability. Finally, among all candidate splitting conditions, the one with the largest fuzzy information gain ratio is selected as the optimal splitting rule for the current node, used to generate the decision tree branch.

[0125] For each leaf node of the fuzzy decision tree, based on all training samples reaching that node, the following steps are performed to determine its output linguistic value: First, calculate the cumulative path weight of each sample from the root node to the leaf node and normalize it; second, for each candidate linguistic value of the hydraulic compensation factor, calculate its weighted aggregate membership degree at that node, which is the sum of the products of each sample's membership degree to that linguistic value and its normalized weight; finally, select the linguistic value with the highest aggregate membership degree as the final output of that leaf node, completing the construction of the fuzzy decision tree. The specific output is as follows: Obtain the path weight of each sample, calculate the cumulative path weight of each sample vector from the root node to the leaf node, which is the product of all splitting conditional membership degrees on the path; normalize the weights, normalizing the path weights of all samples so that their sum is 1; calculate the aggregate membership degree of each linguistic value, for each candidate linguistic value C of the hydraulic compensation factor... kCalculate its weighted aggregate membership degree at that node. , ,in, The value of the hydraulic compensation factor for the i-th sample vector. For the linguistic value C k membership degree is the normalized path weight of the i-th sample vector, and N is the number of sample vectors; the language value with the highest membership degree is selected as the output, and the language value with the largest aggregate membership degree is selected as the final output of the leaf node.

[0126] Traversing each root-to-leaf path of the final decision tree, extracting a fuzzy rule generates a fuzzy rule IF ... THEN ..., where the splitting conditions on the path are connected by "AND" to form the rule antecedent (IF part); the output linguistic value of the leaf node is determined by the average membership degree of the node's samples to each output fuzzy set, and the highest one is selected as the rule consequent (THEN part) to form the initial rule base. For example, to deal with typical vibration disturbances, IF vibration energy IS high AND vibration impact IS high THEN hydraulic compensation factor IS large.

[0127] In the initial rule base, rule base optimization and parameter tuning are performed. Rule pruning is carried out by removing rules with fewer than a threshold of covered samples or confidence scores below a set value to prevent overfitting. Rule merging is performed by weighting rules with the same antecedent but different consequents based on the membership degree of the consequent or selecting the maximum membership degree to resolve conflicts. Membership function parameters are optimized by using gradient descent or genetic algorithms to minimize the prediction error of the training data. The parameters of all membership functions can be uniformly represented before global fine-tuning. The optimization process keeps the rule structure unchanged and only adjusts the shape of the membership functions to improve the model's accuracy and robustness.

[0128] Furthermore, the synthetic fuzzy set of hydraulic compensation factors includes:

[0129] Substitute each feature value in the current multimodal perturbation observation vector into the membership function of its corresponding {low, medium, high} linguistic value, and calculate the membership degree of the feature value on each linguistic value;

[0130] Traverse each rule in the optimized fuzzy rule set. If its antecedent is a single condition, then the membership degree of that condition is the activation strength of the rule. If the antecedent contains multiple conditions connected by AND, take the minimum value of the membership degree of each condition as the activation strength of the rule.

[0131] For each rule with activation strength > 0, obtain the membership function corresponding to the consequent linguistic value. On the entire universe of discourse of the hydraulic compensation factor, perform min implication or multiplicative implication on the membership function to obtain the output fuzzy set of the activated rule, where the entire universe of discourse of the hydraulic compensation factor is [0, 1].

[0132] The output fuzzy sets of all activated rules are aggregated at their maximum values ​​over the universe of discourse to obtain the synthetic fuzzy set of hydraulic compensation factors.

[0133] Specifically, based on a pre-set fuzzy rule base, the activation intensity of the input variables is calculated. The output fuzzy sets of each rule are generated through implication operations, and then a synthesis operation is performed using the maximum value method to obtain the synthesized fuzzy set of u. The fuzzy inference process is as follows: In the fuzzification step, each feature value in the current disturbance observation vector is substituted into its corresponding membership function to calculate the membership degree of that feature value to each fuzzy linguistic value. Input features include: clamping system vibration RMS, vibration kurtosis, hydraulic oil temperature change rate, clamp temperature change rate, and clamping force deviation. The fuzzy linguistic value for each feature is {low, medium, high}. In the rule activation intensity calculation step, each IF-THEN rule in the fuzzy rule base (e.g., "IF vibration = high AND pressure = medium THEN u = large") is traversed. For the current rule, the membership degrees corresponding to each fuzzy condition (e.g., "vibration = high", "pressure = medium") in its antecedent (IF part) are extracted, and the activation intensity α of the rule is calculated using the minimum operation. r If α r If the value is greater than 0, the rule is activated and participates in subsequent reasoning; otherwise, it is skipped. An implication operation is performed to generate a fuzzy subset of the rule output. For each activated rule, its activation strength is compared with the fuzzy set of the consequent corresponding to the linguistic value (e.g., "large") specified by the consequent of the rule. The consequent fuzzy set is the membership function configured for the linguistic value (e.g., {zero, small, medium, large}) defined for the output variable u (hydraulic compensation factor) during system initialization, and its initial parameters have been set. The implication operation modifies the membership function point-by-point with α in the output universe of discourse [0, 1]. rTake the minimum value to achieve truncation and obtain the output fuzzy subset of this rule; perform maximum synthesis on the output fuzzy subsets of all activated rules in the domain [0, 1]. For each point in the output domain, take the maximum membership degree of all rules at this point to form the final synthetic fuzzy set, which comprehensively reflects the fuzzy inference results of all activated rules on the hydraulic compensation factor. All consequent fuzzy sets have been artificially preset and initialized at the initial stage of system construction (for example, initialize "medium" as a trapezoidal function centered at 0.5), and the initialization is completed synchronously with the membership function of the input feature; during subsequent training, the parameters of these consequent fuzzy sets (such as vertices, base widths) can be optimized through data-driven methods (such as gradient descent, genetic algorithms) to improve the inference accuracy; the optimized consequent fuzzy sets are saved as part of the fuzzy rule base for real-time compensation decision-making in the online inference stage. After fuzzy inference, a synthetic fuzzy set of the hydraulic compensation factor is obtained. To convert it into an accurate control signal that can be used in the hydraulic system, defuzzification is performed. Specifically, the centroid defuzzification method is used to discretely sample at a preset step size (such as 0.01) in the domain [0, 1] of the output variable, and calculate the geometric center u of the synthetic fuzzy set. The formula is:

[0134] ,

[0135] where, is the k-th sampling point, is the membership degree of the synthetic fuzzy set at , and N is the total number of sampling points. The finally obtained accurate value , with a value range of [0, 1] as the final output of the hydraulic compensation factor, is used to adjust the opening of the proportional valve or servo valve in the hydraulic compensation circuit to achieve dynamic compensation of the clamping force.

[0136] S800: According to the clamping force deviation between the target clamping force and the total effective clamping force and the output value of the hydraulic compensation factor, calculate the current correction value of the electro-hydraulic proportional valve, and superimpose it on the target current command, which together serve as the control signal of the electro-hydraulic proportional valve, thereby dynamically adjusting the spool opening to generate the required hydraulic oil pressure and maintaining a stable clamping force during processing in the clamping system.

[0137] ,

[0138] where, is the current correction value, is the clamping force deviation, is the output value of the hydraulic compensation factor, [[ID=三十三]] is the slope of the pressure-current characteristic curve of the electro-hydraulic proportional valve.

[0139] Specifically, this embodiment uses a simple method to solve a complex problem by generating an additional "compensation force" to counteract the disturbance. This compensation force can be directly generated by the hydraulic system. The clamping force deviation ΔF is the comprehensive final manifestation of various internal and external disturbances. Whether it is the change in viscosity characteristics caused by changes in hydraulic oil temperature, the thermal deformation caused by changes in fixture temperature, or disturbances not reflected in the modeled force such as workpiece positioning surface errors and friction fluctuations, their final effect will be concentrated in the deviation of the clamping force from its target value. Therefore, by observing ΔF, generalized disturbances, including thermal effects, can be indirectly captured. Based on the clamping force deviation ΔF and the hydraulic compensation factor u, a compensation force to counteract the disturbance is generated and converted into a current command for the electro-hydraulic proportional valve. The specific implementation steps are as follows: Compensation force calculation: The clamping force deviation ΔF is regarded as the disturbance force that needs to be counteracted. It is multiplied by the hydraulic compensation factor u to obtain a desired compensation force. , =(△F) u, as a smart weight (between 0 and 1), determines the degree of compensation required for ΔF under the current operating condition. u=1 indicates 100% compensation, and u=0 indicates no compensation. Current command conversion: the output force of the electro-hydraulic proportional valve is proportional to the input current. This proportional relationship is determined by the valve's pressure-current curve, and its slope (proportional coefficient) is... The unit is N / mA or kN / mA. Therefore, in order to generate the desired compensating force... The required current correction command ΔI is calculated by the following formula: ,parameter This is a fixed, known parameter of the electro-hydraulic proportional valve. Detailed pressure-current characteristic curves are usually provided in the product manual, or it can be calibrated experimentally. For example, apply a known current signal I to the proportional valve, calculate the output force F, and then perform a linear fit on the data points (I, F). The slope of the fitted line is the current characteristic curve. The calculated current correction command ΔI is superimposed on the original control command of the system, i.e., the target current command, and together they serve as the final control signal of the electro-hydraulic proportional valve, thereby dynamically adjusting the valve core opening, generating the required compensation force, and stabilizing the clamping force.

[0140] Based on the multi-parameter collaborative control method for a precision boring and milling machine clamping system described in the foregoing embodiments, and using the same inventive concept, this application also provides a multi-parameter collaborative control device for a precision boring and milling machine clamping system. Please refer to the appendix. Figure 2 The device includes:

[0141] Composite image module 10 is used to pre-clamp and fix the workpiece on the fixture, and to acquire the workpiece surface image through an industrial camera. The acquired image is processed by SIFT feature matching algorithm and illumination normalization to form a composite image for feature extraction.

[0142] The state vector module 20 is used to extract visual features of the workpiece surface structure and texture distribution based on the composite image through a pre-trained deep convolutional network, obtain the process parameters of the workpiece from the process database, and detect the geometric distribution parameters of the pre-labeled key areas of the workpiece in the image. The above visual features, process parameters and geometric distribution parameters are normalized and weighted and fused to generate a comprehensive state vector for clamping decision. The process parameters include at least: material mechanical parameters and cutting process parameters.

[0143] MLP Model Module 30: This module is used to input the comprehensive state vector into the pre-trained MLP decision model and output the target clamping force. The MLP model learns the decision behavior of the clamping decision strategy network trained based on the DDPG algorithm through the knowledge distillation method. The target clamping force generated by the strategy network is used as the decision output, and the target clamping force output by the strategy is guaranteed to be within a safe range. The main clamping stage refers to the working stage in which the clamping system applies and maintains the target clamping force during the stable machining process of the boring and milling machine.

[0144] Optimal clamping module 40: Based on the obtained target clamping force, the controller generates a corresponding target current command according to the preset pressure-current mapping relationship. This command drives the electro-hydraulic proportional valve to adjust the output force of the hydraulic cylinder, thereby driving the clamping mechanism to complete the workpiece clamping.

[0145] The disturbance information acquisition module 50 is used to simultaneously acquire the vibration signal, temperature signal and hydraulic oil pressure signal of the clamping system during the main clamping stage. It extracts the time-domain features of the vibration signal and temperature signal in a preset time window to obtain the vibration time-domain features and temperature time-domain features. It also calculates the average value of the hydraulic oil pressure within the window. Based on the known effective piston area, it converts the hydraulic oil pressure into the current output force of the hydraulic cylinder. The current output force is input into the calibrated force transmission model to obtain the total effective clamping force on the workpiece. It also calculates the clamping force deviation between the target clamping force and the total effective clamping force.

[0146] The current disturbance observation vector module 60 is used to combine the extracted vibration time-domain features and temperature time-domain features with the clamping force deviation to construct a normalized current multimodal disturbance observation vector, which is in the form of [vibration RMS, vibration kurtosis, hydraulic oil temperature change rate, clamping temperature change rate, clamping force deviation].

[0147] Hydraulic compensation factor module 70 is used to input the current multimodal disturbance observation vector into a pre-constructed fuzzy rule model. After fuzzification, rule activation, implication operation and maximum value aggregation, a synthetic fuzzy set of hydraulic compensation factors is generated. The centroid method is used to defuzzify the hydraulic compensation factor in the domain [0,1]. The geometric center of the synthetic fuzzy set is calculated and the calculation result is used as the output value of the hydraulic compensation factor. The fuzzy rule model is constructed by a fuzzy decision tree.

[0148] The compensation force module 80 is used to calculate the current correction value of the electro-hydraulic proportional valve based on the clamping force deviation between the target clamping force and the total effective clamping force, as well as the output value of the hydraulic compensation factor. This correction value is then superimposed on the target current command to serve as the control signal for the electro-hydraulic proportional valve. This dynamically adjusts the valve core opening to generate the required hydraulic oil pressure and maintain a stable clamping force during machining.

[0149] ,

[0150] in, This is the current correction value. For clamping force deviation, This is the output value of the hydraulic compensation factor. .

[0151] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A multi-parameter collaborative control method for a clamping system of a precision boring and milling machine, characterized in that, include: The workpiece is pre-clamped and fixed on the fixture. An industrial camera is used to acquire images of the workpiece surface. The acquired images are processed by SIFT feature matching algorithm and illumination normalization to form a composite image for feature extraction. Based on composite images, visual features of workpiece surface structure and texture distribution are extracted through a pre-trained deep convolutional network, and process parameters of the workpiece are obtained from the process database. At the same time, the geometric distribution parameters of the pre-labeled key areas of the workpiece in the image are detected. The above visual features, process parameters and geometric distribution parameters are normalized and weighted and fused to generate a comprehensive state vector for clamping decision. The process parameters include at least: material mechanical parameters and cutting process parameters. The integrated state vector is input into the pre-trained MLP decision model, and the target clamping force is output. The MLP model learns its decision behavior from the clamping decision strategy network trained based on the DDPG algorithm through the knowledge distillation method. The main clamping stage is the working stage in which the clamping system applies and maintains the target clamping force during the stable machining process of the boring and milling machine. Based on the obtained target clamping force, the controller generates a corresponding target current command according to the preset pressure-current mapping relationship. This command drives the electro-hydraulic proportional valve to adjust the output force of the hydraulic cylinder, thereby driving the clamping mechanism to complete the workpiece clamping. During the main clamping stage, vibration signals, temperature signals, and hydraulic oil pressure signals of the clamping system are simultaneously acquired. Using a preset time window as the unit, the vibration and temperature signals are extracted in the time domain to obtain the vibration time domain features and temperature time domain features. The average value of the hydraulic oil pressure within the window is calculated. Based on the known effective piston area, the hydraulic oil pressure is converted into the current output force of the hydraulic cylinder. The current output force is input into the calibrated force transmission model to obtain the total effective clamping force on the workpiece. The clamping force deviation between the target clamping force and the total effective clamping force is calculated. The extracted vibration time-domain features and temperature time-domain features are combined with the clamping force deviation to construct a normalized current multimodal disturbance observation vector, which is in the form of [vibration RMS, vibration kurtosis, hydraulic oil temperature change rate, clamping temperature change rate, clamping force deviation]. The current multimodal disturbance observation vector is input into the pre-constructed fuzzy rule model. After fuzzification, rule activation, implication operation and maximum value aggregation, a synthetic fuzzy set of hydraulic compensation factor is generated. The centroid method is used to defuzzify the hydraulic compensation factor in the domain [0,1]. The geometric center of the synthetic fuzzy set is calculated and the calculation result is used as the output value of the hydraulic compensation factor. The fuzzy rule model is constructed by fuzzy decision tree. Based on the clamping force deviation between the target clamping force and the total effective clamping force, and the output value of the hydraulic compensation factor, the current correction value of the electro-hydraulic proportional valve is calculated and superimposed on the target current command. Together, they serve as the control signal for the electro-hydraulic proportional valve, thereby dynamically adjusting the valve core opening to generate the required hydraulic oil pressure and maintain a stable clamping force during machining. ， in, This is the current correction value. For clamping force deviation, This is the output value of the hydraulic compensation factor. For electricity The slope of the pressure-current characteristic curve of the proportional valve.

2. The multi-parameter collaborative control method for a precision boring and milling machine clamping system as described in claim 1, characterized in that, Forming a composite image for feature extraction includes: Two industrial cameras are positioned above the workpiece clamping area. With the mechanical coordinate system of the workpiece clamping area as a reference, the optical axis of the first camera is located in the XZ plane and forms a 45° angle with the Z-axis, and the optical axis of the second camera is located in the YZ plane and forms a 45° angle with the Z-axis. The intersection point of the optical axes of the two cameras is aligned with the geometric center of the workpiece clamping area to obtain multi-angle images of the workpiece surface. Image A is the surface image of the workpiece acquired by the first camera, and image B is the surface image of the workpiece acquired by the second camera. Key points of images A and B are extracted by SIFT algorithm and initial matching is performed. Then, the optimal homography matrix is ​​estimated from the initial matching point pairs by RANSAC algorithm combined with homography model, and mismatched points are eliminated. Perspective transformation is performed on the images using this matrix to obtain the stitched composite image. The stitched composite image is subjected to illumination normalization processing, converting it from the RGB color space to the HSV color space, and separating the hue, saturation, and lightness channels. The CLAHE algorithm is applied to the lightness channel to suppress the influence of uneven illumination. Then, the highlight area is further compressed pixel by pixel through a nonlinear mapping method to reduce the interference of specular reflection from the metal surface of the workpiece. The processed lightness channel is merged with the original hue and saturation channels and converted back to the RGB color space to obtain the normalized composite image used for subsequent feature extraction.

3. The multi-parameter collaborative control method for a precision boring and milling machine clamping system as described in claim 1, characterized in that, MLP models learn decision-making behavior from a clamped decision policy network trained based on the DDPG algorithm using knowledge distillation, including: A digital twin simulation environment for the clamping system of a boring and milling machine is constructed. This simulation environment includes a geometric data layer and a functional model layer. The geometric data layer includes the workpiece CAD geometric model and the fixture CAD geometric model. The functional model layer includes an assembly analysis module, a clamping mechanism dynamics model, a workpiece mechanical response model, a material contact characteristic model, a virtual vision perception module, and an intelligent clamping decision model. The assembly analysis module constructs a fixture-workpiece assembly by connecting the workpiece CAD geometric model and the fixture CAD geometric model through assembly constraints. It then identifies clamping contact points on the assembly using a geometric contact detection algorithm, assigns numbers to each point, extracts the 3D coordinates and surface normal vectors of each contact point, and assigns force transmission path numbers to each contact point based on preset fixture functional topology information. Finally, it generates the structural configuration parameters for the clamping system from this contact point information. A dynamic model of the clamping mechanism simulates the dynamic response characteristics of a clamping system composed of a hydraulic cylinder and an electro-hydraulic proportional pressure valve during the main clamping phase. Taking the received target clamping force command as input, the model outputs the real-time total clamping force generated by the hydraulic cylinder piston rod. Based on the force transmission path numbers provided by the assembly analysis module, and according to the finite element analysis of the fixture-workpiece assembly, the equivalent stiffness coefficients of each force transmission path are determined. The real-time total clamping force is distributed to each path according to the stiffness ratio, constituting the clamping force and its distribution. Simultaneously, based on the real-time total clamping force and the piston speed of the hydraulic cylinder, the instantaneous power value of the clamping system is calculated, and time integration is performed during the main clamping phase to estimate the energy consumption of the clamping system during the main clamping process. The workpiece mechanical response model is based on the clamping force distribution on the fixture-workpiece assembly. A finite element model is constructed, and the clamping force is applied as a surface load to the corresponding contact points on the workpiece. At the same time, fixed constraints are applied in the assembly constraint relationship area. The stress distribution and elastic deformation field of the workpiece under the clamping force are obtained by analyzing the finite element model, and the maximum local deformation of the workpiece is extracted. A material contact characteristic model stores the friction coefficients at the contact points between the workpiece and the fixture materials. During simulation, the maximum total static friction resistance of the clamping system in the sliding direction is calculated, and the estimated total cutting force in the sliding direction is obtained. The two are compared to obtain the actual safety factor, which is used to assess the risk of workpiece slippage. The estimated total cutting force comes from an external cutting force prediction model, which is called before each clamping decision. The cutting process parameters of the workpiece are retrieved from the process database to estimate the total cutting force of the boring and milling machine. The virtual visual perception module is constructed within a digital twin simulation environment to generate visual features consistent with those extracted in reality. It is used in the training phase of the clamping decision strategy to learn the ability to respond to changes in the workpiece surface condition. The intelligent clamping decision model is built based on the DDPG algorithm. The process of training the reinforcement learning policy network using DDPG is as follows: S1: Randomly initialize the parameters of the Actor policy network and the Critic evaluation network, and initialize their corresponding target networks; S2: For each time step (t), the current integrated state vector S(t) is input to the Actor policy network. Based on this, the policy network outputs the target clamping force F(t). Before outputting the target clamping force, the minimum and maximum allowable safe clamping forces under the current operating conditions are retrieved from the process database for comparison and constraint processing to ensure that the generated target clamping force is always within the safe clamping force window. The target clamping force is then converted into a target clamping force command and input into the digital twin simulation environment. S3: Based on this instruction, the digital twin simulation environment runs a simulation step. After the simulation ends, it generates the next comprehensive state vector S(t+1). At the same time, based on the simulation results, it calculates the value of the multi-objective reward function and obtains the reward value R(t) of S(t). The simulation results include the maximum local deformation of the workpiece, the actual safety factor, and the energy consumption of the clamping system. S4: Store the quadruple (S(t), F(t), R(t), S(t+1)) generated in this interaction as an experience sample in the experience replay buffer. S5: Randomly sample mini-batch quadruplets from the experience replay buffer. First, update the parameters of the Critic evaluation network to minimize its prediction error. Based on the updated Critic network, calculate the policy gradient of the Actor parameters to guide the direction of updating the Actor policy network parameters, gradually optimizing the policy to improve long-term cumulative rewards. Simultaneously update the target network parameters of both the Actor policy network and the Critic evaluation network to improve training stability. S6: Using S(t+1) as the new current integrated state vector, repeat steps S2-S5 until the reinforcement learning policy network receives the result. Convergence yields a well-trained clamping decision-making strategy network; The MLP model is obtained by transferring the decision-making ability of the clamped decision policy network trained and converged by the DDPG algorithm.

4. The multi-parameter collaborative control method for a precision boring and milling machine clamping system as described in claim 3, characterized in that, The MLP model is obtained by transferring the decision-making ability of the clamped decision policy network trained and converged by the DDPG algorithm, including: The trained clamping decision policy network is run in the constructed digital twin simulation environment to generate various comprehensive state vectors S. i The target clamping force F output by the clamping decision strategy network in this state is collected. i ; The obtained state-action pairs (S) i F i ), constructing a distillation dataset D={(S i F i The state vectors are preprocessed by normalizing them. The MLP model structure is designed based on the dimension of the comprehensive state vector. The number of neurons in the input layer is consistent with the dimension of the comprehensive state vector. The hidden layer uses the ReLU activation function, and the output layer is a single-neuron linear unit used to predict the target clamping force. Using the distillation dataset D as the training sample, a supervised learning approach is adopted to train the MLP model to mimic the policy decision-making ability of DDPG by minimizing the mean square error between the MLP prediction output and the clamping decision policy network output. The trained MLP model is compressed using INT8 quantization, and the compressed lightweight MLP model is deployed to the industrial control computer of the boring and milling machine.

5. The multi-parameter collaborative control method for a precision boring and milling machine clamping system as described in claim 1, characterized in that, The calibrated force transmission models include: Obtain the material, size, and hardness information of the workpiece, process calibration blocks with the same properties, and install force sensors at the bottom of the calibration blocks; The calibration block is installed into the fixture, and the controller sends N levels of current signals. At each level, the hydraulic oil pressure P and the total effective clamping force monitored by the force sensor are simultaneously acquired. The hydraulic oil pressure P, combined with the known effective area A of the cylinder piston, is used to calculate the current output force of the hydraulic cylinder in real time. The data is collected M times at each level, and the average value is taken to obtain a dataset of N points. ; A linear model is directly fitted between the current output force and the total effective clamping force of the hydraulic cylinder to establish an overall force transmission model. j=1,2,…,N, The parameters are obtained using the least squares method. and , The total mechanical efficiency of the clamping system, < 1, This refers to the total force loss of the clamping system; Will and Packaging is a force transmission model. Solidify model parameters, and Write the force transmission model to the controller's non-volatile memory and mark it as a calibrated model.

6. The multi-parameter collaborative control method for a precision boring and milling machine clamping system as described in claim 1, characterized in that, Pre-built fuzzy rule models include: Historical operating time series data of a precision boring and milling machine is acquired, covering normal and various fault states of the equipment. The historical data is used to extract features such as vibration RMS, vibration kurtosis, hydraulic oil temperature change rate, fixture temperature change rate, and hydraulic oil pressure window mean value. Based on the hydraulic oil window mean value and the known effective piston area, the hydraulic cylinder output force is calculated and regarded as the total effective clamping force. The clamping force deviation is obtained by subtracting it from the set clamping force. The sample values ​​of the five features obtained, namely vibration RMS, vibration kurtosis, hydraulic oil temperature change rate, fixture temperature change rate, and clamping force deviation, are combined in a fixed order to form a single multimodal disturbance observation sample vector. After completing the above operation on all historical data, all the obtained sample vectors are normalized to obtain a multimodal disturbance observation sample vector set for training. At the same time, experts label the corresponding hydraulic compensation factor value for each multimodal disturbance observation sample vector in the sample vector set according to the severity of the fault and the compensation requirements. The label value range is [0, 1]. The set composed of all sample vectors and their hydraulic compensation factor values ​​constitutes the training dataset. A linguistic value is defined for each feature in the multimodal disturbance observation sample vector, which includes low, medium, and high. At the same time, a linguistic value is defined for the hydraulic compensation factor, which includes zero, small, medium, and large. Each linguistic value of each feature and each linguistic value of the hydraulic compensation factor is configured with a corresponding membership function. The membership function adopts a triangular or trapezoidal distribution. The parameters of each membership function are initialized to generate the corresponding initial fuzzy set for each linguistic value. The global parameters of the fuzzy decision tree are preset. The process of constructing a tree structure using fuzzy decision trees: S1: Starting from the root node, iterate through the five features of the multimodal perturbation observation sample vector. For each feature, iterate through each of its linguistic values, and use each feature-linguistic value combination as a candidate splitting condition to be evaluated. S2: For each candidate split condition, extract the sample values ​​of all sample vectors in the training dataset on that feature, and substitute each sample value into the membership function corresponding to that linguistic value to calculate the membership degree belonging to that linguistic value. Based on the membership degrees calculated from the sample values, calculate the fuzzy information gain ratio of that candidate split condition. S3: At the current node, compare the fuzzy information gain ratios of all candidate splitting conditions, select the feature-language value combination with the largest gain ratio as the optimal splitting condition for the current node, and generate two child nodes: "belonging" and "not belonging". For each sample vector and its hydraulic compensation factor value in the training dataset, calculate its membership degree with the linguistic value of the optimal splitting condition based on its data on the feature specified by the optimal splitting condition. The sample vector and its hydraulic compensation factor value, carrying the weight of the membership degree, enter the "belonging" child node. At the same time, the sample vector and its hydraulic compensation factor value, carrying the weight of 1 minus the membership degree, enter the "not belonging" child node. The sample vector and its hydraulic compensation factor value remain unchanged during the allocation process, forming the training dataset of each child node. The weights carried are used in all subsequent recursive calculations. Take each non-leaf child node as the new current node and recursively repeat the above splitting process until the preset stopping condition is met. Nodes that fail to continue splitting become leaf nodes, and a fuzzy decision tree structure with no assigned output is constructed. S4: For each leaf node in the fuzzy decision tree structure, based on all sample vectors in that node and the cumulative path weight from the root node to that node, the membership degree of each linguistic value of the hydraulic compensation factor is weighted and aggregated. The linguistic value with the highest aggregated membership degree is selected as the output of that leaf node, thus completing the construction of the fuzzy decision tree. S5: Based on the completed fuzzy decision tree, traverse each path from the root node to the leaf node, connect the splitting conditions (feature-language value combinations) of all non-leaf nodes on the path using the AND logical operator to form the antecedent (IF part) of the rule, and use the output language value assigned to the leaf node as the consequent (THEN part) of the rule to form a complete IF-THEN fuzzy rule. The rules corresponding to all paths together constitute the initial fuzzy rule set. This rule set combines the linguistic values ​​of all features in the multimodal disturbance observation sample vector, the linguistic values ​​of the hydraulic compensation factor, and their corresponding membership functions and initial parameters to form an initial fuzzy rule knowledge base; S6: Optimize the initial fuzzy rule base, encapsulate and store the optimized fuzzy rule set and the optimal parameters of the membership function corresponding to each language value in a structured data format to form an optimized fuzzy rule knowledge base. By introducing a fuzzy inference engine and combining it with this knowledge base, a fuzzy rule model that can be embedded in the control system is formed.

7. The multi-parameter collaborative control method for a precision boring and milling machine clamping system as described in claim 6, characterized in that, The composite fuzzy set of hydraulic compensation factors includes: Substitute each feature value in the current multimodal perturbation observation vector into the membership function of its corresponding {low, medium, high} linguistic value, and calculate the membership degree of the feature value on each linguistic value; Traverse each rule in the optimized fuzzy rule set. If its antecedent is a single condition, then the membership degree of that condition is the activation strength of the rule. If the antecedent contains multiple conditions connected by AND, take the minimum value of the membership degree of each condition as the activation strength of the rule. For each rule with activation strength > 0, obtain the membership function corresponding to the consequent linguistic value. On the entire universe of discourse of the hydraulic compensation factor, perform min implication or multiplicative implication on the membership function to obtain the output fuzzy set of the activated rule, where the entire universe of discourse of the hydraulic compensation factor is [0, 1]. The output fuzzy sets of all activated rules are aggregated at their maximum values ​​over the universe of discourse to obtain the synthetic fuzzy set of hydraulic compensation factors.

8. A multi-parameter collaborative control device for a clamping system of a precision boring and milling machine, characterized in that, The device includes: The composite image module is used to pre-clamp and fix the workpiece on the fixture, and to acquire images of the workpiece surface through an industrial camera. The acquired images are processed by SIFT feature matching algorithm and illumination normalization to form a composite image for feature extraction. The state vector module is used to extract visual features of the workpiece surface structure and texture distribution based on the composite image through a pre-trained deep convolutional network, obtain the workpiece's process parameters from the process database, and detect the geometric distribution parameters of the pre-labeled key areas of the workpiece in the image. The above visual features, process parameters and geometric distribution parameters are normalized and weighted and fused to generate a comprehensive state vector for clamping decision. The process parameters include at least: material mechanical parameters and cutting process parameters. The MLP model module is used to input the comprehensive state vector into the pre-trained MLP decision model and output the target clamping force. The MLP model learns its decision behavior from the clamping decision strategy network trained based on the DDPG algorithm through the knowledge distillation method. The main clamping stage is the working stage in which the clamping system applies and maintains the target clamping force during the stable machining process of the boring and milling machine. The optimal clamping module is used to generate a corresponding target current command based on the obtained target clamping force and the controller according to the preset pressure-current mapping relationship. This command drives the electro-hydraulic proportional valve to adjust the output force of the hydraulic cylinder, thereby driving the clamping mechanism to complete the workpiece clamping. The disturbance information acquisition module is used to simultaneously acquire vibration signals, temperature signals, and hydraulic oil pressure signals of the clamping system during the main clamping stage. It extracts time-domain features of the vibration and temperature signals in preset time windows to obtain vibration time-domain features and temperature time-domain features. It also calculates the average value of the hydraulic oil pressure within the window. Based on the known effective piston area, it converts the hydraulic oil pressure into the current output force of the hydraulic cylinder. The current output force is input into the calibrated force transmission model to obtain the total effective clamping force on the workpiece. It also calculates the clamping force deviation between the target clamping force and the total effective clamping force. The current disturbance observation vector module combines the extracted vibration time-domain features and temperature time-domain features with the clamping force deviation to construct a normalized current multimodal disturbance observation vector, which takes the form of [vibration RMS, vibration kurtosis, hydraulic oil temperature change rate, clamping temperature change rate, clamping force deviation]. The hydraulic compensation factor module is used to input the current multimodal disturbance observation vector into a pre-constructed fuzzy rule model. After fuzzification, rule activation, implication operation and maximum value aggregation, a synthetic fuzzy set of hydraulic compensation factors is generated. The centroid method is used to defuzzify the hydraulic compensation factor in the domain [0,1]. The geometric center of the synthetic fuzzy set is calculated and the calculation result is used as the output value of the hydraulic compensation factor. The fuzzy rule model is constructed by a fuzzy decision tree. The compensation force module calculates the current correction value for the electro-hydraulic proportional valve based on the clamping force deviation between the target clamping force and the total effective clamping force, as well as the output value of the hydraulic compensation factor. This correction value is then superimposed on the target current command, serving as the control signal for the electro-hydraulic proportional valve. This dynamically adjusts the valve core opening to generate the required hydraulic oil pressure, maintaining a stable clamping force during machining. ， in, This is the current correction value. For clamping force deviation, This is the output value of the hydraulic compensation factor. Electro-hydraulic ratio Example: The slope of the pressure-current characteristic curve of the valve.

Images