Artificial Intelligence-Based CNC Machining Path Control System

Abstract

This invention discloses an artificial intelligence-based CNC machining path control system, specifically relating to the field of CNC machining. It addresses the continuity and stability issues in curvature abrupt change regions during toolpath planning. By employing a curvature constraint feature matrix and a curvature adaptive convolutional network, it accurately captures the global and local geometric characteristics of the surface, generating machining trajectories that satisfy differential geometric continuity, effectively avoiding path breakage in curvature abrupt change regions. Simultaneously, it introduces energy scaling criteria and curvature manifold constraints to dynamically repair the machining trajectory, balancing geometric repair and physical stability, reducing cutting force abrupt changes and stress concentration, thereby enhancing machining reliability. Furthermore, through finite element simulation on a digital twin platform and synchronous updates of servo system parameters, it achieves dynamic closed-loop matching between control commands and machining states, further optimizing control efficiency and surface quality. Thus, it achieves synergistic optimization of high precision, high stability, and high efficiency in complex surface machining.

Description

Technical Field

[0001] This invention relates to the field of CNC machining, and more specifically, to an artificial intelligence-based CNC machining path control system. Background Technology

[0002] In the machining of free-form surfaces in aerospace components, precision molds, and other fields, the workpiece surface often exhibits abrupt curvature characteristics of non-uniform rational B-spline surfaces (NURBS). This type of geometry requires the tool path to strictly adhere to the differential topological properties of the surface to ensure the continuity of the cutting trajectory in the normal and tangent directions. Existing deep learning-based path planning models (such as PointNet++) rely on discrete point cloud data to construct the tool contact point sequence. Their core drawback is that the point cloud feature extraction network captures geometric information through local neighborhood aggregation, making it difficult to effectively characterize the global differential geometric properties of the implicit surface (such as principal curvature distribution and Gaussian curvature gradient). When the machined object exceeds the surface types covered by the training set, the path coordinate sequence output by the model exhibits topological breaks near curvature extrema, resulting in a misalignment of the dynamic contact relationship between the tool envelope and the workpiece surface.

[0003] The essence of this problem stems from the limitations of artificial intelligence models in understanding the laws of geometric deformation. Existing methods simplify surface machining into a coordinate regression task of discrete points, ignoring the inherent correlation between the tool trajectory and the continuous surface manifold. When the tool travels to a region of abrupt curvature change, the local geometric features extracted by the point cloud network cannot reflect the overall topological constraints of the surface, leading to connection errors between adjacent cutting passes in the generated path. These errors manifest in two failure modes in physical machining: first, residual unremoved material forms surface protrusions, forcing the addition of subsequent finishing processes; second, overcutting causes subsurface damage to the workpiece, leading to stress concentration and reduced fatigue life. Especially in sensitive scenarios such as thin-walled parts and heterogeneous material composite structures, these defects can directly lead to functional failure of the workpiece, forcing the machining process to be interrupted and resulting in the scrapping of high-value materials.

[0004] To address the aforementioned problems, a technical solution is provided. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide an artificial intelligence-based CNC machining path control system. This system accurately captures the global and local geometric characteristics of the surface through a curvature constraint feature matrix and a curvature adaptive convolutional network, generating machining trajectories that satisfy differential geometric continuity, effectively avoiding path breakage in curvature abrupt change regions. Simultaneously, by introducing energy scaling criteria and curvature manifold constraints, the machining trajectory is dynamically repaired, balancing geometric repair and physical stability, reducing cutting force abrupt changes and stress concentration, thereby enhancing machining reliability. Furthermore, through finite element simulation and synchronous updating of servo system parameters on a digital twin platform, dynamic closed-loop matching of control commands and machining states is achieved, further optimizing control efficiency and surface quality. Thus, high precision, high stability, and high efficiency are synergistically optimized in complex surface machining, providing technical support for high-quality manufacturing in aerospace, precision molds, and other fields, thereby solving the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] Surface feature extraction module: Based on the surface geometric properties of the workpiece model, obtain the principal curvature direction field and Gaussian curvature distribution, and fuse curvature change gradient parameters and topological connectivity parameters to generate curvature constraint feature matrix;

[0008] Trajectory generation network module: Converts the curvature constraint feature matrix into heat map channels, concatenates it with the original point cloud data in the spatial dimension, and inputs it into a pre-trained convolutional network to generate the initial processing trajectory;

[0009] The trajectory effect analysis module: Based on the difference in residual height between adjacent cutting lines of the initial machining trajectory, the gradient of the abrupt change in cutting force is back-mapped, and the local surface strain energy density is calculated by combining the curvature continuity deviation, thus constructing an energy scaling criterion;

[0010] Trajectory Dynamic Repair Module: When the energy scaling criterion exceeds the material's plastic deformation criticality, a transition spline curve is generated based on the curvature constraint feature matrix. The geometric parameters and physical stability of the path are repaired through the relationship between curvature decay rate and energy dissipation.

[0011] Simulation optimization control module: Import the repaired path into the digital twin platform for finite element simulation to verify the cutting stress distribution and dynamic cutting force fluctuation range; select the optimized trajectory that meets the preset stress constraints, and update the trajectory parameters and the acceleration feedforward coefficient of the servo system synchronously.

[0012] In a preferred embodiment, the surface feature extraction module includes the following:

[0013] The implicit representation of the surface is extracted from the computer-aided design model of the workpiece and parameterized as a function dependent on two parameter coordinates. Based on the implicit representation of the surface, the first and second rates of change of the surface in the two parameter coordinate directions, as well as the normal direction vector of the surface, are calculated. The first and second sets of basic geometric quantities are calculated using the tangent vector and normal direction vector of the surface, and the curvature feature matrix is ​​constructed. The principal curvature and principal curvature direction in the curvature constraint feature matrix are determined through feature analysis, namely the maximum degree of curvature, the minimum degree of curvature and their corresponding principal curvature directions.

[0014] In a preferred embodiment, the surface feature extraction module further includes the following:

[0015] The Gaussian curvature is calculated by multiplying the maximum and minimum curvature, and a Gaussian curvature distribution is generated on the parametric grid. The rate of change of the Gaussian curvature in the two parametric coordinate directions is calculated to form the curvature change gradient parameter. The topological property values ​​and the number of connected components are calculated according to the topological structure of the surface to determine the topological connectivity parameter. For each sampling point on the surface, a feature vector containing the main curvature direction, Gaussian curvature, curvature change gradient parameter and topological connectivity parameter is constructed, and all feature vectors are arranged according to the parametric grid to form a curvature constraint feature matrix.

[0016] In a preferred embodiment, the trajectory generation network module includes the following:

[0017] Normalize each feature dimension of the curvature constraint feature matrix to generate heatmap channels; map the heatmap channels to the coordinate positions of the original point cloud data through bilinear interpolation to generate an interpolation feature matrix, and concatenate it with the original point cloud data in the spatial dimension to form comprehensive input data; input the comprehensive input data into a pre-trained convolutional network, train it by minimizing the geometric loss function, and generate an initial processing trajectory that satisfies the constraints of normal orthogonality and tangent continuity.

[0018] In a preferred embodiment, the geometric loss function is constructed as follows:

[0019] A geometric loss function is constructed, which includes normal vector orthogonality constraints and tangent vector continuity constraints. By calculating the dot product of the path tangent direction and the surface normal vector, as well as the deviation of the cosine of the angle between the tangent directions of adjacent path points and 1, the differential geometric continuity of the path is ensured.

[0020] In a preferred embodiment, the trajectory effect analysis module includes the following:

[0021] The residual height difference between adjacent cutting rows in the normal direction of the surface is calculated based on the initial machining trajectory. The ratio of the residual height difference to the average distance between adjacent cutting rows is used to back-map the abrupt change gradient of the cutting force. The Euclidean norm of the curvature eigenvectors between adjacent points is calculated based on the curvature constraint feature matrix and the initial machining trajectory, and then the global curvature continuity deviation is calculated. The local surface strain energy density is calculated by the sum of the weighted terms of the square of the abrupt change gradient of the cutting force and the weighted terms of the square of the global curvature continuity deviation. The maximum value in the local surface strain energy density sequence is extracted as the energy scaling criterion.

[0022] In a preferred embodiment, the trajectory dynamic repair module includes the following:

[0023] The system determines whether the energy scaling criterion exceeds the critical value for plastic deformation of the material. Excessive regions are identified by comparing the energy scaling criterion with the critical value. A transition spline curve is generated based on the curvature constraint feature matrix. The maximum and minimum curvature degrees, their corresponding main curvature directions, and the Gaussian curvature distribution are extracted from the curvature constraint feature matrix. Cubic B-spline interpolation is used to construct the transition spline curve, ensuring that boundary conditions and curvature continuity requirements are met. The geometric parameters and physical stability of the repair path are dynamically adapted. The curvature decay rate and energy dissipation function of the transition spline curve are calculated, and control points are adjusted to minimize the difference between the local surface strain energy density and the energy dissipation function.

[0024] In a preferred embodiment, the simulation optimization control module includes the following:

[0025] The repaired machining trajectory is discretized into three-dimensional coordinate points and imported into a digital twin platform to establish a toolpath model. The cutting stress distribution and dynamic cutting force fluctuation range on the workpiece surface are analyzed through finite element simulation. The maximum and minimum principal stresses on the workpiece surface, as well as the peak and valley values ​​of the dynamic cutting force, are calculated. The optimized trajectory is selected according to the preset stress constraints. It is checked whether the maximum principal stress exceeds the preset threshold and whether the cutting force fluctuation range is within the allowable range. If not, the trajectory is adjusted and the simulation is repeated. The speed and acceleration parameters of the tool movement are extracted from the optimized machining trajectory, and the acceleration feedforward coefficient is calculated. The speed, acceleration, and acceleration feedforward coefficient are updated to the control command stream of the servo system to achieve dynamic matching between the control command and the machining state.

[0026] The technical effects and advantages of the artificial intelligence-based CNC machining path control system of this invention are as follows:

[0027] By employing curvature constraint feature matrices and curvature adaptive convolutional networks, the global and local geometric characteristics of curved surfaces are accurately captured, generating machining trajectories that satisfy differential geometric continuity and effectively avoiding path breakage issues in curvature abrupt change regions. Simultaneously, energy scaling criteria and curvature manifold constraints are introduced to dynamically repair the machining trajectory, balancing geometric repair and physical stability, reducing cutting force abrupt changes and stress concentration, thereby enhancing machining reliability. Furthermore, through finite element simulation and servo system parameter synchronization updates via a digital twin platform, dynamic closed-loop matching of control commands and machining states is achieved, further optimizing control efficiency and surface quality. This results in the synergistic optimization of high precision, high stability, and high efficiency in complex surface machining, providing technical support for high-quality manufacturing in aerospace, precision mold, and other fields. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of the structure of the artificial intelligence-based numerical control machining path control system of the present invention.

[0029] Figure 2 This is a flowchart illustrating the surface feature extraction module of the CNC machining path control system based on artificial intelligence of the present invention. Detailed Implementation

[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] Example 1: Figure 1 The present invention provides an artificial intelligence-based numerical control machining path control system, comprising:

[0032] Surface feature extraction module: Based on the surface geometric properties of the workpiece model, obtain the principal curvature direction field and Gaussian curvature distribution, and fuse curvature change gradient parameters and topological connectivity parameters to generate curvature constraint feature matrix;

[0033] Trajectory generation network module: Converts the curvature constraint feature matrix into heat map channels, concatenates it with the original point cloud data in the spatial dimension, and inputs it into a pre-trained convolutional network to generate the initial processing trajectory;

[0034] The trajectory effect analysis module: Based on the difference in residual height between adjacent cutting lines of the initial machining trajectory, the gradient of the abrupt change in cutting force is back-mapped, and the local surface strain energy density is calculated by combining the curvature continuity deviation, thus constructing an energy scaling criterion;

[0035] Trajectory Dynamic Repair Module: When the energy scaling criterion exceeds the material's plastic deformation criticality, a transition spline curve is generated based on the curvature constraint feature matrix. The geometric parameters and physical stability of the path are repaired through the relationship between curvature decay rate and energy dissipation.

[0036] Simulation optimization control module: Import the repaired path into the digital twin platform for finite element simulation to verify the cutting stress distribution and dynamic cutting force fluctuation range; select the optimized trajectory that meets the preset stress constraints, and update the trajectory parameters and the acceleration feedforward coefficient of the servo system synchronously.

[0037] In the machining of free-form surfaces in aerospace components, precision molds, and other fields, the workpiece surface typically exhibits non-uniform rational B-spline surfaces (NURBS), whose curvature characteristics may abruptly change in local regions. To ensure the continuity of the toolpath in the normal and tangent directions, accurate geometric constraint information needs to be extracted based on the implicit surface differential geometry attributes of the workpiece CAD model. Addressing the challenge that existing path planning models struggle to effectively characterize the global differential geometry of curvature abrupt change regions, the surface feature extraction module generates a curvature constraint feature matrix. This matrix provides a geometric constraint benchmark for the curvature adaptive convolutional network of the trajectory generation network module, preventing path breaks and machining defects.

[0038] like Figure 2 As shown, the surface feature extraction module includes the following:

[0039] S1.1, Extract implicit surface differential geometric properties:

[0040] The implicit representation of the surface is obtained from the computer-aided design model of the workpiece and parameterized as a function dependent on two parameter coordinates, which represent two directions on the surface. The rate of change of the surface in the first parameter coordinate direction is calculated to obtain the tangent vector in that direction; similarly, the rate of change of the surface in the second parameter coordinate direction is calculated to obtain the tangent vector in the other direction. Then, the quadratic rate of change of the surface in the first parameter coordinate direction, the mixed rate of change in the two parameter coordinate directions, and the quadratic rate of change in the second parameter coordinate direction are calculated. These results are used for subsequent analysis of the bending characteristics of the surface. Finally, by performing a cross product operation on the two tangent vectors and normalizing their lengths, the normal direction vector at each point on the surface is calculated, representing the perpendicular direction of the surface in a local region.

[0041] By calculating the first and second rates of change of the surface in two parametric coordinate directions, and the resulting normal direction vector, the geometric characteristics of the surface in local regions can be fully captured. This information provides the basic data for subsequent analysis of the surface's curvature and direction, ensuring that toolpath planning can adapt to the shape changes of the surface in small regions, thereby improving machining accuracy and path continuity.

[0042] S1.2, Calculate the principal curvature and the direction of principal curvature:

[0043] Based on the two tangent vectors obtained in the first step, the first set of fundamental geometric quantities of the surface is calculated. Specifically, the dot product of the first tangent vector with itself, the dot product of the first tangent vector with the second tangent vector, and the dot product of the second tangent vector with itself are calculated. These dot product results constitute a metric tensor describing the local distance characteristics of the surface. Next, using the quadratic rate of change and the normal direction vector from the first step, the second set of fundamental geometric quantities is calculated, namely, the quadratic rate of change of the first parameter coordinate direction, the mixed rate of change of the two parameter coordinate directions, and the dot product of the quadratic rate of change of the second parameter coordinate direction with the normal direction vector. These results reflect the curvature change of the surface in the normal direction. Then, by multiplying the inverse of the metric tensor with the matrix formed by the second set of fundamental geometric quantities, a curvature feature matrix is ​​constructed, and feature analysis is performed on this matrix to obtain the principal curvature and principal curvature directions of the surface at each point. The principal curvature includes the maximum and minimum curvature, and the corresponding two principal curvature directions, i.e., the principal curvature directions. Finally, these local directions are continuously organized into a global field to obtain the principal curvature direction field. For example, the following calculation process can be used to obtain it:

[0044] Calculate the first fundamental form:

[0045]

[0046] in , These represent the components of the surface metric tensor, and the first tangent vector. and the second tangent vector Indicates the surface in and Tangent vector in the direction.

[0047] Calculate the second fundamental form:

[0048]

[0049] in , , It reflects the curvature change of the surface in the normal direction.

[0050] Construct the curvature feature matrix:

[0051]

[0052] right Eigenvalues ​​and eigenvectors are obtained by performing eigendecomposition. The eigenvalues ​​correspond to the principal curvatures of the surface, namely the maximum and minimum curvatures, while the eigenvectors correspond to the principal curvature directions, namely the two main curvature directions corresponding to the maximum and minimum curvatures.

[0053] The maximum and minimum degree of curvature, along with their corresponding primary curvature directions, are core indicators describing the local geometric characteristics of a surface, accurately reflecting the degree of curvature in different directions. This information provides crucial data for toolpath planning, enabling the tool orientation to be adjusted according to the surface's primary curvature direction, thereby avoiding path interruptions or machining defects and improving the quality of the machined surface.

[0054] S1.3, Calculate the Gaussian curvature distribution:

[0055] By multiplying the maximum curvature calculated in the second step by the minimum curvature, a scalar measure called Gaussian curvature is defined to represent the overall characteristics of the local geometric properties of the surface. Next, the Gaussian curvature of each grid point is assigned a value on the grid formed by the surface parametric coordinates, generating a distribution set containing the Gaussian curvature values ​​of all sampled points.

[0056] Gaussian curvature, as a comprehensive measure of the local geometric properties of a surface, reflects its overall topological properties. By analyzing the distribution of Gaussian curvature, special regions on the surface can be identified, such as saddle-shaped regions, convex regions, or concave regions. This information is crucial for toolpath planning, helping to avoid or specially handle these regions during the planning process, thereby ensuring the stability and consistency of the machining path.

[0057] S1.4, Generate curvature change gradient parameters:

[0058] Based on the Gaussian curvature obtained in the third step, the rate of change of the Gaussian curvature in the first parameter coordinate direction and the rate of change in the second parameter coordinate direction are calculated respectively, forming the gradient vector of the Gaussian curvature. Specifically, by calculating the difference of the Gaussian curvature between adjacent grid points with small step sizes in the parameter coordinates, this is used to approximate the rate of change of the Gaussian curvature in the two directions. Then, the square root of the sum of the squares of the rates of change in the two directions is used to define a scalar parameter, called the curvature change gradient parameter, to represent the drasticness of the change in the Gaussian curvature. For example, it can be obtained using the following calculation process:

[0059] Calculate the gradient of the Gaussian curvature:

[0060]

[0061] in This represents Gaussian curvature.

[0062] Approximate calculation using the finite difference method:

[0063]

[0064] in and Let the step size be a parameter. Define the curvature change gradient parameters:

[0065]

[0066] It indicates the degree of drastic change in curvature.

[0067] Curvature change gradient parameters can effectively capture regions where Gaussian curvature changes abruptly in a surface, which are often challenging areas in toolpath planning. By quantifying the severity of curvature changes, a basis can be provided for toolpath planning, enabling the planning algorithm to adopt special strategies in regions of curvature abrupt changes, avoiding toolpath breakage due to curvature discontinuities, thereby improving machining reliability and surface quality.

[0068] S1.5, Extract topological connectivity parameters:

[0069] Based on the surface's topology, a topological characteristic value is calculated. This value is obtained by subtracting the number of edges from the number of vertices and then adding the number of faces, resulting in a characteristic number that reflects the overall topological properties of the surface. This characteristic number is used to determine the surface's topological type, and the total number of independent parts within the surface is further calculated as a topological connectivity parameter. The topological connectivity parameter reflects the overall connectivity and structural complexity of the surface, which is crucial for understanding its global characteristics. In toolpath planning, considering surface connectivity ensures that the path covers all machining areas while avoiding path discontinuities caused by surface separation, thereby improving the completeness of path planning and machining efficiency.

[0070] S1.6, Fuse and generate curvature constraint feature matrix:

[0071] For each sampling point on the surface parametric coordinate grid, a feature vector containing various geometric information is constructed. Specifically, this feature vector includes the two principal curvature directions calculated by the surface feature extraction module, the Gaussian curvature calculated in step S1.3, the curvature change gradient parameters calculated in step S1.4, and the topological connectivity parameters calculated in step S1.5. Then, the feature vectors of all sampling points are arranged in the row and column order of the parametric grid to form a multi-dimensional feature matrix, called the curvature constraint feature matrix. Its dimension is determined by the number of rows and columns of the grid and the number of components of the feature vector.

[0072] Rationale and Benefits: The curvature constraint feature matrix encodes the curvature continuity and geometric abrupt changes of a surface in the form of a multidimensional tensor by integrating local geometric properties (such as dominant curvature directions and Gaussian curvature) and global properties (such as topological connectivity). This comprehensive geometric description provides a complete constraint benchmark for subsequent toolpath planning, ensuring that the path maintains continuity in regions of curvature abrupt changes while adapting to the overall structure of the surface, thereby reducing machining defects and improving machining accuracy.

[0073] The curvature constraint feature matrix encodes the global curvature continuity and local geometric abrupt changes of the surface in tensor form, providing a geometric constraint benchmark for the trajectory generation network module. Based on the curvature constraint feature matrix and combined with the original point cloud data, the trajectory generation network module generates an initial processing trajectory that satisfies the constraints of normal orthogonality and tangent continuity, laying the foundation for the back-mapping and optimization of the trajectory effect analysis module.

[0074] The trajectory generation network module includes the following:

[0075] S2.1, Transformation of the curvature constraint characteristic matrix:

[0076] First, the curvature constraint feature matrix generated by the surface feature extraction module is transformed. The curvature constraint feature matrix is ​​a multidimensional tensor containing the global curvature continuity and local geometric abrupt changes of the surface. During the transformation, normalization is performed on each feature dimension of the curvature constraint feature matrix. Specifically, for each feature dimension, the minimum and maximum values ​​of all data in that dimension are first calculated. Then, for each feature value, the following operation is performed: the minimum value is subtracted from the feature value, and the result is divided by the difference between the maximum and minimum values, thus mapping the feature value to the interval between 0 and 1. After normalization, the curvature constraint feature matrix is ​​converted into multiple heatmap channels, each corresponding to a feature dimension, reflecting the relative intensity distribution of the surface's geometric properties.

[0077] The purpose of converting the curvature constraint feature matrix into heatmap channels is to present the complex geometric features of the surface in a standardized form, facilitating subsequent processing and learning by the network model. Normalization ensures that data from different feature dimensions have a uniform numerical range, avoiding model training instability caused by differences in feature value ranges, while preserving the relative change information of the surface curvature distribution.

[0078] S2.2, stitching with the original point cloud data:

[0079] The heatmap channel is stitched and fused with the original point cloud data. The original point cloud data contains the coordinates of discrete points on the surface. The stitching process first uses bilinear interpolation to map the feature values ​​in the heatmap channel to the coordinate positions in the original point cloud data. Specifically, for each point coordinate in the point cloud data, its corresponding position is found in the heatmap channel, and the interpolated feature value of that point is calculated using the weighted average of its four nearest neighbor feature values. The weights are determined based on the distance between the point coordinates and the positions of the neighboring feature values. After interpolation, an interpolation feature matrix with the same number of points as the original point cloud data is generated. Then, the three-dimensional coordinates of the original point cloud data and the interpolation feature matrix are merged in space to form a comprehensive input data that includes point cloud coordinates and corresponding curvature features.

[0080] By stitching together the interpolated feature matrices from the original point cloud data and the heatmap channels, comprehensive input information can be provided for the subsequent network model. This information fusion method simultaneously incorporates the discrete geometric location and continuous geometric characteristics of the surface, enabling the network model to comprehensively consider the local point locations and global curvature distribution of the surface during the learning process. Bilinear interpolation ensures the continuity and smoothness of features in the point cloud space, thereby improving the adaptability of the generated toolpath to the surface geometry and ensuring the accuracy and completeness of path planning.

[0081] S2.3, Design of Geometric Loss Function

[0082] A geometric loss function is constructed to guide the curvature adaptive convolutional network in generating toolpath coordinate sequences, ensuring that the path satisfies the normal vector orthogonality and tangent vector continuity constraints. The geometric loss function consists of two parts: normal vector orthogonality and tangent vector continuity. The normal vector orthogonality constraint is calculated by, for each point on the path, calculating the dot product of the path tangent direction and the surface normal vector, and summing the squares of all dot products. The goal is to make the dot product close to zero, thus ensuring that the path tangent direction is perpendicular to the surface normal vector. The tangent vector continuity constraint is calculated by, for each pair of adjacent points on the path, calculating the cosine of the angle between the tangent directions of the two points and its deviation from 1, and summing the deviations of all adjacent point pairs. The goal is to smooth the change in tangent direction. The final geometric loss function is a weighted sum of the normal vector orthogonality and tangent vector continuity constraints, with weight coefficients used to adjust the relative importance of the two constraints.

[0083] The geometric loss function is constructed directly to address the differential geometric continuity requirements of the toolpath. It ensures the alignment of the tool orientation with the surface normal through orthogonal constraints on the normal vector, and ensures a smooth transition of the path in regions of abrupt curvature changes through continuous constraints on the cutting vector. This loss function effectively avoids path breaks or discontinuities, improving the accuracy and stability of path planning.

[0084] S2.4, Network Training and Path Generation:

[0085] The comprehensive input data is fed into a pre-trained convolutional network based on curvature. The network is trained by minimizing the geometric loss function to generate a toolpath coordinate sequence. The training process includes the following steps: the convolutional network generates an initial path coordinate sequence based on the input data, calculates the value of the geometric loss function, adjusts the network parameters through backpropagation, and repeats this process until the geometric loss function converges. After training, the network outputs an initial machining trajectory that satisfies the constraints of orthogonal normals and continuous tangents, serving as the path reference for subsequent steps.

[0086] By training the network to generate initial machining trajectories, it can automatically learn the geometric properties of the surface and generate paths that meet the requirements of differential geometric continuity. Compared with traditional manual path design methods, this significantly improves efficiency and consistency. The guidance of the geometric loss function ensures high-quality paths and reduces the complexity of subsequent optimization and adjustments.

[0087] The pre-trained convolutional network consists of multiple convolutional modules and fully connected layers, with its core feature being the dynamic generation of convolutional kernel weights. Specifically, a weight generation network is constructed, taking the interpolated feature matrix as input and calculating it through a multi-layer neural network to output convolutional kernel weights suitable for the convolutional modules. These dynamically generated convolutional kernel weights are adjusted according to the local curvature features of the surface, enabling the convolutional operation to adapt to the geometric characteristics of regions with abrupt curvature changes. The network gradually generates a preliminary feature representation of the toolpath through layer-by-layer convolution and feature extraction.

[0088] The trajectory generation network module, based on the curvature constraint feature matrix provided by the surface feature extraction module, converts it into a heatmap channel and concatenates it with the original point cloud data. This heatmap is then input into a convolutional network, where a dynamic convolutional kernel and a geometric loss function are used to generate the initial processing trajectory. This initial trajectory maintains differential geometric continuity in regions of abrupt curvature changes, providing a high-quality path benchmark for the subsequent trajectory effect analysis module's back-mapping and optimization, ensuring the stability of the processing and surface quality.

[0089] In freeform surface machining, toolpath planning must ensure that the machining trajectory satisfies the differential geometric continuity of the surface while maintaining physical stability during the machining process. The trajectory generation network module, based on the curvature constraint feature matrix generated by the surface feature extraction module, generates an initial machining trajectory through a convolutional network. This trajectory satisfies the constraints of normal orthogonality and tangent continuity. However, in regions of abrupt curvature change, the initial machining trajectory may induce abrupt changes in cutting force, affecting machining stability. The trajectory effect analysis module analyzes the differences in residual height between adjacent cutting lines and the deviation in curvature continuity of the initial machining trajectory to construct an energy scaling criterion. This provides a basis for path repair by the trajectory dynamic repair module, ensuring physical stability during the machining process.

[0090] The trajectory effect analysis module includes the following:

[0091] S3.1, Calculate the difference in residual height between adjacent cutting lines:

[0092] The initial machining trajectory generated by the trajectory generation network module is first divided into cutting lines, with each line representing a continuous cutting path. Next, within each pair of adjacent cutting lines, pairs of points with the same parameterized position are selected, and the residual height difference between these point pairs in the normal direction of the surface is calculated. Specifically, for each pair of points, the normal vector of the surface at one point is first determined, then the position vector difference between the two points is calculated, and finally, the residual height difference in the normal direction is obtained by taking the absolute value of the dot product of the normal vector and the position vector difference. This residual height difference represents the amount of material remaining in the normal direction after machining, reflecting the geometric deviation of the machining trajectory. For each pair of adjacent cutting lines, the average residual height difference of all corresponding point pairs is calculated to obtain the average residual height difference of that pair of adjacent cutting lines.

[0093] The purpose of calculating the difference in residual height between adjacent cutting passes is to quantify the geometric inconsistencies in the machining trajectory, especially in areas with significant curvature changes, where differences in residual height can lead to unevenness on the machined surface. By analyzing these differences, regions in the machining trajectory that may cause physical instability can be identified, providing fundamental data for analyzing changes in cutting forces. This method ensures effective control over the geometric accuracy of the machining trajectory.

[0094] S3.2, Reverse mapping of the abrupt change in cutting force gradient:

[0095] Based on the calculation results of the difference in residual height between adjacent cutting passes, the cutting force abrupt change gradient is further calculated. The cutting force abrupt change gradient is defined as the ratio of the average difference in residual height between adjacent cutting passes to the average distance between adjacent cutting passes. The specific calculation process is as follows: First, the Euclidean distance between corresponding points in each pair of adjacent cutting passes is averaged to obtain the average distance between adjacent cutting passes. Then, the average difference in residual height between adjacent cutting passes is divided by the average distance to obtain the cutting force abrupt change gradient. This gradient represents the rate of change of cutting force between adjacent cutting passes, reflecting the degree of dynamic fluctuation of force during the machining process.

[0096] The calculation of the abrupt change gradient of cutting force aims to quantify the physical instability during machining, especially in regions with significant differences in residual height, where cutting forces may change rapidly, leading to machining instability. By inversely mapping the abrupt change gradient of cutting force, a correlation can be established between geometric deviations and physical and mechanical effects, providing a basis for energy analysis. This method enhances the understanding of the physical characteristics of the machining process, helps prevent machining defects, and improves machining stability.

[0097] S3.3, Calculate curvature continuity deviation:

[0098] Based on the curvature constraint feature matrix generated by the surface feature extraction module and the initial processing trajectory from the trajectory generation network module, the curvature continuity deviation is calculated. Specifically, for each point on the initial processing trajectory, the corresponding curvature feature vector is extracted from the curvature constraint feature matrix, reflecting the local curvature attribute at that point. Next, the difference between the curvature feature vectors of adjacent points is calculated to obtain the change vector of the curvature feature. Then, the Euclidean norm of this change vector is calculated as the curvature continuity deviation between adjacent points. Finally, the average curvature continuity deviation of all adjacent point pairs is taken to obtain the global curvature continuity deviation, representing the overall curvature consistency of the processing trajectory.

[0099] The purpose of calculating curvature continuity deviation is to assess the geometric consistency of the machining trajectory in regions of abrupt curvature changes, especially in areas with drastic curvature variations, where large deviations can lead to an uneven machining trajectory. Quantifying curvature continuity deviation provides a basis for assessing geometric discontinuities in energy analysis. This method ensures effective control of the geometric continuity of the machining trajectory, contributing to improved surface uniformity and quality.

[0100] S3.4, Calculate the local surface strain energy density:

[0101] First, the abrupt change gradient of cutting force and the global curvature continuity deviation are standardized to a value between 0 and 1. Based on the transformed abrupt change gradient of cutting force and global curvature continuity deviation, the local surface strain energy density is calculated. The local surface strain energy density is defined as the sum of the weighted terms of the square of the abrupt change gradient of cutting force and the weighted terms of the square of the global curvature continuity deviation. The specific calculation process is as follows: First, the square of the abrupt change gradient of cutting force is calculated, representing the energy accumulation of local physical instability; then, the square of the global curvature continuity deviation is calculated, representing the influence of geometrical abrupt changes on machining stability; finally, the two energy components are added together using weighting coefficients to obtain the local surface strain energy density. The weighting coefficients are calibrated according to material properties and machining conditions to adjust the proportion of different factors contributing to the energy. For example, the following calculation method is used:

[0102] Define local surface strain energy density The effects of abrupt changes in coupled cutting force and deviations in curvature continuity:

[0103]

[0104] , : Weighting coefficients, which respectively adjust the contribution of the abrupt change in cutting force gradient and the curvature continuity deviation to energy, and need to be calibrated according to material properties and processing conditions.

[0105] The square term of the gradient of the abrupt change in cutting force represents the energy accumulation of local physical instability.

[0106] The global curvature continuity deviation square term represents the impact of geometric abrupt changes on machining stability.

[0107] The calculation of local surface strain energy density couples geometric deviations and physical and mechanical effects to comprehensively evaluate the stability during machining. By mapping abrupt changes in cutting force and deviations in curvature continuity to a unified energy scale, the stability of the machining trajectory can be analyzed more comprehensively.

[0108] S3.5, Constructing the energy scaling criterion:

[0109] Based on the calculation results of local surface strain energy density, an energy scaling criterion is constructed. The energy scaling criterion is defined as the maximum value in the local surface strain energy density sequence, reflecting the energy level of the most unstable region during processing. It is used to assess whether the processing trajectory may lead to plastic deformation or other physical instabilities in the material, providing a basis for decision-making in subsequent path repair.

[0110] The purpose of constructing the energy scaling criterion is to provide a clear judgment standard for path repair by quantifying the maximum instability in the processing. By comparing the energy scaling criterion with the material's critical value for plastic deformation, it is possible to determine whether the processing trajectory needs to be adjusted, thereby preventing processing defects and improving processing quality.

[0111] The trajectory effect analysis module, based on the initial machining trajectory of the trajectory generation network module and the curvature constraint feature matrix of the surface feature extraction module, constructs an energy scaling criterion by calculating the residual height difference between adjacent cutting lines, reverse mapping of the cutting force abrupt change gradient, calculation of curvature continuity deviation, and local surface strain energy density. This criterion provides a clear stability assessment basis for the path repair of the trajectory dynamic repair module, ensuring that the geometric continuity and physical stability during the machining process are effectively controlled.

[0112] The trajectory effect analysis module constructs an energy scaling criterion by analyzing the differences in residual height between adjacent cutting lines and the deviation in curvature continuity of the initial machining trajectory. This criterion is used to evaluate the stability of the machining trajectory. When the energy scaling criterion exceeds the critical value for plastic deformation of the material, it indicates that the machining trajectory may induce irreversible deformation of the material, requiring path repair. Based on this, the trajectory dynamic repair module generates transition spline curves according to the curvature constraint feature matrix for areas where the energy scaling exceeds the standard. Through dynamic adaptation of the relationship between curvature decay rate and energy dissipation, it optimizes the geometric parameters and physical stability of the repair path, providing a high-quality repaired path for the finite element simulation and optimization of the simulation optimization control module.

[0113] The trajectory dynamic repair module includes the following:

[0114] S4.1, Determine whether the energy scale exceeds the material's critical plastic deformation limit:

[0115] The trajectory dynamic repair module first determines whether there are areas with stability risks in the machining trajectory based on the energy scaling criterion generated by the trajectory effect analysis module and a pre-determined material plastic deformation critical value. The judgment process involves comparing the energy scaling criterion generated by the trajectory effect analysis module with the material plastic deformation critical value. When the energy scaling criterion exceeds the material plastic deformation critical value, the corresponding area is identified as an out-of-range area, requiring subsequent path repair. The material plastic deformation critical value is determined through mechanical property experiments on the workpiece material and represents the energy limit level at which irreversible deformation occurs during processing. The marking of out-of-range areas provides a clear range for path repair.

[0116] By comparing energy scaling criteria with the critical value of material plastic deformation, regions in the machining trajectory that may trigger plastic deformation can be accurately identified. This method sets clear initiation conditions for path repair, ensuring that repair operations are performed on specific problems and avoiding unnecessary processing of areas that do not require adjustment, thereby improving the efficiency of computational resource utilization and the stability of the machining process.

[0117] S4.2, Generate transition spline curves based on curvature constraint feature matrix constraints:

[0118] After identifying the out-of-specification area, the trajectory dynamic repair module generates a transition spline curve based on the curvature constraint feature matrix to repair the geometric defects of the machining trajectory. The generation process is as follows: First, the principal curvature direction field and Gaussian curvature distribution of the out-of-specification area are extracted from the curvature constraint feature matrix as geometric constraints for the transition spline curve. Next, between adjacent cutting lines in the out-of-specification area, interpolation is performed along the principal curvature direction field—that is, the main curvature directions corresponding to the maximum and minimum curvature degrees—to construct a transition spline curve connecting the starting points of adjacent cutting lines. The construction method employs cubic B-spline interpolation technology to ensure that the curve has at least second-order geometric continuity and satisfies boundary conditions and curvature continuity requirements. Boundary conditions stipulate that the starting and ending points of the transition spline curve coincide with the starting and ending points of adjacent cutting lines in the out-of-specification area, respectively; curvature continuity requires that the curvature change of the transition spline curve matches the local Gaussian curvature distribution of the surface, with the matching error controlled within the tolerance range specified by the machining accuracy.

[0119] Transition spline curves are generated based on curvature constraint feature matrices to ensure that the repair path maintains consistency with the geometric properties of the freeform surface, avoiding machining defects caused by abrupt path changes. Cubic B-spline interpolation technology ensures the smoothness and continuity of the curves, reducing vibration and stress concentration issues caused by path discontinuities during machining. Simultaneously, by matching the local Gaussian curvature distribution of the surface, the repair path achieves high compatibility with the surface characteristics, improving the accuracy and quality of the machined surface.

[0120] S4.3, Dynamically adapting the geometric parameters and physical stability of the repair path:

[0121] After generating the transition spline curve, the geometric parameters and physical stability of the repair path are optimized through dynamic adaptation. The adaptation process is as follows: First, the rate of change of the curvature of the transition spline curve with respect to the parameters is calculated, defined as the curvature decay rate, which represents how fast the curve curvature changes. Next, an energy dissipation function is introduced to describe the energy release at different locations along the repair path. Specifically, the square of the curvature decay rate is multiplied by a dissipation coefficient, which is determined experimentally through material properties and processing conditions. Then, the control points of the transition spline curve are adjusted to minimize the difference between the local surface strain energy density and the energy dissipation function. The optimization objective is to match the energy dissipation distribution of the repair path with the strain energy density of the out-of-range region, reducing the impact of sudden energy changes, while maintaining the boundary conditions and curvature continuity requirements unchanged.

[0122] By dynamically adapting the geometric parameters and physical stability of the repair path, the physical properties during processing can be optimized while maintaining geometric continuity. Adjusting the curvature decay rate and energy dissipation relationship ensures that the repair path effectively releases energy in areas exceeding tolerances, reducing the risk of stress concentration and material deformation during processing. This method achieves dual optimization of geometric repair and physical stability, improving the reliability of the processing and surface quality, and providing a stable path foundation for subsequent processing.

[0123] The trajectory dynamic repair module identifies out-of-range regions by determining whether the energy scale exceeds the material's plastic deformation critical value. It then generates transition spline curves based on curvature constraint feature matrices and dynamically adapts and optimizes the geometric parameters and physical stability of the repair path, ultimately generating the repaired machining trajectory. The repaired machining trajectory achieves dual protection of geometric continuity and physical stability in regions of abrupt curvature changes, providing a high-quality input foundation for subsequent finite element simulations and optimizations.

[0124] The trajectory dynamic repair module generates transition spline curves for areas where the energy scaling exceeds the limit, and outputs the repaired machining trajectory, which is optimized in terms of geometric continuity and physical stability. The simulation optimization control module takes over the repaired machining trajectory from the trajectory dynamic repair module, performs finite element simulation verification through a digital twin platform, and further optimizes the trajectory to meet the requirements of machining stability and surface quality.

[0125] The simulation optimization control module includes the following:

[0126] S5.1, import the repaired machining trajectory into the digital twin platform.

[0127] The repaired machining trajectory first needs to be discretized. Specifically, the continuous parametric curve output by the trajectory dynamic repair module is decomposed into a series of discrete three-dimensional coordinate points. These three-dimensional coordinate points are generated through uniform sampling or adaptive sampling based on curvature changes to ensure that the point set can completely represent the geometric features of the machining trajectory. The discretized three-dimensional coordinate points are imported into the digital twin platform as geometric input to establish a digital model of the toolpath. The digital twin platform then generates a virtual representation of the machining trajectory based on these coordinate points, laying the foundation for subsequent simulation analysis.

[0128] Reasons and benefits: Decomposing continuous machining trajectories into discrete three-dimensional coordinate points facilitates geometric modeling and simulation calculations on the digital twin platform. Discretization transforms complex curve trajectories into computer-processable point sets, ensuring accurate and easily operable input data for simulation analysis.

[0129] S5.2, perform finite element simulation to verify the distribution of cutting stress and the range of dynamic cutting force fluctuation:

[0130] In the digital twin platform, finite element analysis tools are used to create mesh models of the workpiece and cutting tool. The specific process includes dividing the workpiece geometry into multiple small units to form a mesh structure, and similar meshing is performed on the cutting tool model. Based on the discretized machining trajectory, the behavior of the cutting tool moving along the trajectory and gradually removing material during machining is simulated. During this process, the maximum and minimum principal stresses of each mesh unit on the workpiece surface are calculated to generate a cutting stress distribution map, which characterizes the stress state of the workpiece surface during machining. Furthermore, using a cutting force physical model, the dynamic cutting force experienced by the cutting tool during machining is calculated over time, and the maximum and minimum values ​​of the cutting force are recorded to determine the fluctuation range of the dynamic cutting force.

[0131] Finite element method (FEM) simulation can accurately simulate material removal behavior during machining, analyze the cutting stress distribution on the workpiece surface and the dynamic cutting force changes experienced by the tool. It can identify stress concentration areas or problems caused by excessive cutting force fluctuations in the machining trajectory. Simulation verifies the performance of predicting the machining trajectory before actual machining, helping to avoid potential machining defects and ensuring the stability of the machining process and the achievement of workpiece surface quality standards.

[0132] S5.3, Filter the optimized trajectory that meets the preset stress constraints:

[0133] Based on the cutting stress distribution map and dynamic cutting force fluctuation range generated by finite element simulation, constraint checks are performed. Specifically, it checks whether the maximum principal stress of some mesh elements in the cutting stress distribution map exceeds a preset threshold, and simultaneously assesses whether the dynamic cutting force fluctuation range is within a predefined allowable range. If the maximum principal stress exceeds the threshold or the cutting force fluctuation range exceeds the allowable range, the repaired machining trajectory is locally adjusted, for example, by modifying the trajectory curvature or tool feed rate, to generate a new machining trajectory. Subsequently, the above finite element simulation process is repeated until both the cutting stress distribution and the dynamic cutting force fluctuation range meet the preset constraint conditions, and finally, the optimized machining trajectory is output.

[0134] Optimized trajectories that meet preset stress constraints are selected to ensure that the maximum principal stress on the workpiece surface does not exceed the material's yield strength during machining, thus avoiding material deformation or damage due to excessive stress. Simultaneously, by controlling the fluctuation range of dynamic cutting forces within acceptable limits, vibration and instability during machining are reduced, thereby improving machining accuracy and workpiece surface quality. This iterative optimization method refines the machining trajectory in a virtual environment, reducing risks in actual machining.

[0135] S5.4 updates the optimized trajectory parameters and the servo system's acceleration feedforward coefficients synchronously:

[0136] The velocity and acceleration parameters of the tool motion are extracted from the optimized machining trajectory. Specifically, the velocity of the tool in each segment is calculated by analyzing the distance and time interval between discrete points in the machining trajectory, and the acceleration is further calculated by the change of velocity over time. Based on the dynamic response model of the servo system, the acceleration feedforward coefficient is determined, which is calculated from the ratio of acceleration to velocity. Subsequently, the extracted velocity, acceleration, and calculated acceleration feedforward coefficient are updated in the control command stream of the servo system, enabling dynamic matching between the control commands and the motion state of the machining trajectory.

[0137] Synchronously updating and optimizing the speed, acceleration parameters, and acceleration feedforward coefficients of the machining trajectory improves the control performance of the servo system, reduces tracking errors during tool movement, and enhances the execution accuracy of the machining trajectory. By dynamically matching control commands with the machining state, the servo system can respond to changes in the toolpath in real time, optimizing the stability of the machining process. This closed-loop coordination method enhances the response speed of the machining equipment, providing a guarantee for high-precision machining.

[0138] The simulation optimization control module verifies the cutting stress distribution and dynamic cutting force fluctuation range of the repaired machining trajectory through finite element simulation on a digital twin platform, ensuring that it meets preset constraints. Based on this, the optimized machining trajectory is selected, and the trajectory's velocity and acceleration parameters are updated synchronously with the servo system's acceleration feedforward coefficient, achieving dynamic closed-loop matching between the control command flow and the physical machining state. Through the comprehensive processing of simulation verification and parameter optimization, the stability of the freeform surface machining process is ensured, while simultaneously improving the workpiece surface quality, providing reliable technical support for high-precision machining.

[0139] It should be noted that the system of the present invention can be deployed on the device itself to realize embedded applications, or it can run on a PC or other terminal with a user interface, thereby meeting a variety of hardware environments and usage requirements.

[0140] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0141] It should be noted that, in this document, the use of relational terms such as "first" and "second" is merely to distinguish one entity or operation from another, and does not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0142] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A numerical control machining path control system based on artificial intelligence, characterized in that, include: The surface feature extraction module extracts the implicit representation of the surface from the computer-aided design model of the workpiece and parameterizes it as a function dependent on two parameter coordinates, which represent two directions on the surface. It calculates the rate of change of the surface in the first parameter coordinate direction, obtaining the tangent vector in that direction; it also calculates the rate of change of the surface in the second parameter coordinate direction, obtaining the tangent vector thereafter; and calculates the quadratic rate of change of the surface in both directions. Based on the tangent vectors in the first and second parameter coordinate directions, it calculates the normal direction vector at each point on the surface. It then calculates the first set of basic geometric quantities using these vectors. Finally, it calculates the second set of basic geometric quantities using the two quadratic rates of change and the normal direction vector in the aforementioned parameter coordinate directions. Finally, it multiplies the inverse of the first set of basic geometric quantities with the matrix formed by the second set of basic geometric quantities to construct the curvature feature matrix. The principal curvatures and principal curvature directions in the curvature feature matrix are determined by feature analysis. The principal curvatures include the maximum curvature and the minimum curvature, and the principal curvature directions include the main curvature directions corresponding to the maximum curvature and the main curvature directions corresponding to the minimum curvature. The Gaussian curvature is calculated by multiplying the maximum curvature and the minimum curvature, and a Gaussian curvature distribution is generated on the parameter grid. Calculate the rate of change of Gaussian curvature in the two parametric coordinate directions to form the curvature change gradient parameter; calculate the topological property values ​​and the number of connected components based on the topological structure of the surface to determine the topological connectivity parameter; for each sampling point on the surface, construct a feature vector containing the main bending direction, Gaussian curvature, curvature change gradient parameter and topological connectivity parameter, and arrange all feature vectors according to the parameter grid to form a curvature constraint feature matrix; The trajectory generation network module converts the curvature constraint feature matrix into a heatmap channel, concatenates it with the original point cloud data in the spatial dimension, and then inputs it into a pre-trained convolutional network to generate the initial processing trajectory; the original point cloud data contains the coordinate information of discrete points on the curved surface. Trajectory effect analysis module: Based on the difference in residual height between adjacent cutting lines of the initial machining trajectory, the cutting force abrupt gradient is back-mapped, and the local surface strain energy density is calculated by combining the curvature continuity deviation. The maximum value in the local surface strain energy density sequence is extracted as the energy scaling criterion. Trajectory Dynamic Repair Module: When the energy scaling criterion exceeds the critical value of material plastic deformation, a transition spline curve is generated based on the curvature constraint feature matrix, and the geometric parameters and physical stability of the path are repaired by the relationship between curvature decay rate and energy dissipation. Simulation optimization control module: Import the repaired path into the digital twin platform for finite element simulation to verify the cutting stress distribution and dynamic cutting force fluctuation range; select the optimized trajectory that meets the preset stress constraints, and update the trajectory parameters and the acceleration feedforward coefficient of the servo system synchronously.

2. The CNC machining path control system based on artificial intelligence according to claim 1, characterized in that, The trajectory generation network module includes the following: Normalize each feature dimension of the curvature constraint feature matrix to generate heatmap channels; map the heatmap channels to the coordinate positions of the original point cloud data through bilinear interpolation to generate an interpolation feature matrix, and concatenate it with the original point cloud data in the spatial dimension to form comprehensive input data; input the comprehensive input data into a pre-trained convolutional network, train it by minimizing the geometric loss function, and generate an initial processing trajectory that satisfies the constraints of normal orthogonality and tangent continuity.

3. The artificial intelligence-based CNC machining path control system according to claim 2, characterized in that, The geometric loss function is constructed as follows: A geometric loss function is constructed, which includes normal vector orthogonality constraints and tangent vector continuity constraints. By calculating the dot product of the path tangent direction and the surface normal vector, as well as the cosine of the angle between the tangent directions of adjacent path points and the deviation from 1, the path is ensured to satisfy differential geometric continuity.

4. The artificial intelligence-based CNC machining path control system according to claim 3, characterized in that, The trajectory effect analysis module includes the following: The residual height difference between adjacent cutting rows in the normal direction of the surface is calculated based on the initial machining trajectory. The ratio of the residual height difference to the average distance between adjacent cutting rows is used to back-map the abrupt change gradient of the cutting force. The Euclidean norm of the curvature eigenvectors between adjacent points is calculated based on the curvature constraint feature matrix and the initial machining trajectory, and then the global curvature continuity deviation is calculated. The local surface strain energy density is calculated by the sum of the weighted terms of the square of the abrupt change gradient of the cutting force and the weighted terms of the square of the global curvature continuity deviation. The maximum value in the local surface strain energy density sequence is extracted as the energy scaling criterion.

5. The artificial intelligence-based CNC machining path control system according to claim 4, characterized in that, The trajectory dynamic repair module includes the following: The system determines whether the energy scaling criterion exceeds the critical value for plastic deformation of the material. Excessive regions are identified by comparing the energy scaling criterion with the critical value. A transition spline curve is generated based on the curvature constraint feature matrix. The maximum and minimum curvature degrees, their corresponding main curvature directions, and the Gaussian curvature distribution are extracted from the curvature constraint feature matrix. Cubic B-spline interpolation is used to construct the transition spline curve, ensuring that boundary conditions and curvature continuity requirements are met. The geometric parameters and physical stability of the repair path are dynamically adapted. The curvature decay rate and energy dissipation function of the transition spline curve are calculated, and control points are adjusted to minimize the difference between the local surface strain energy density and the energy dissipation function.

6. The artificial intelligence-based CNC machining path control system according to claim 5, characterized in that, The simulation optimization control module includes the following: The repaired machining trajectory is discretized into three-dimensional coordinate points and imported into a digital twin platform to establish a toolpath model. The cutting stress distribution and dynamic cutting force fluctuation range on the workpiece surface are analyzed through finite element simulation. The maximum and minimum principal stresses on the workpiece surface, as well as the peak and valley values ​​of the dynamic cutting force, are calculated. The optimized trajectory is selected according to the preset stress constraints. It is checked whether the maximum principal stress exceeds the preset threshold and whether the cutting force fluctuation range is within the allowable range. If not, the trajectory is adjusted and the simulation is repeated. The speed and acceleration parameters of the tool movement are extracted from the optimized machining trajectory, and the acceleration feedforward coefficient is calculated. The speed, acceleration, and acceleration feedforward coefficient are updated to the control command stream of the servo system to achieve dynamic matching between the control command and the machining state.

Images