Multi-axis linkage engraving equipment control algorithm optimization method

Abstract

The invention provides a multi-axis linkage engraving equipment control algorithm optimization method, which belongs to the technical field of engraving equipment, and comprises the following steps of: decomposing a model into a plurality of curved surface segmentation segments according to a curvature change rate, and performing NURBS curve parameterization processing on each segment to generate control point coordinates and weight coefficient data; a dynamic weight adjustment model is adopted to dynamically calculate a path planning weight coefficient according to a tool cutting force value, a surface roughness demand value and a machining efficiency index value, a minimum spanning tree path algorithm is utilized to carry out path connection sequence planning, and a layered interpolation operation framework combining preprocessing interpolation calculation and real-time interpolation compensation is established. And the optimized interpolation coordinate data is converted into an axis control instruction sequence containing a position instruction, a speed instruction and an acceleration instruction, and the technical problem that the interpolation calculation precision and the real-time performance are difficult to balance when the multi-axis linkage carving equipment is used for machining a complex curved surface is solved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of engraving equipment, and in particular relates to a multi-axis linkage engraving equipment control algorithm optimization method. BACKGROUND

[0002] As an important equipment for modern precision machining, multi-axis linkage engraving equipment is widely used in the machining of complex curved surface parts in the fields of aerospace, automobile manufacturing, mold machining, etc. The traditional multi-axis linkage control technology mainly adopts the combination of linear interpolation and circular interpolation, processes the complex curved surface by segmenting through the preset interpolation algorithm, and realizes the coordinated motion control of each axis by using the interpolation module of the numerical control system to ensure that the tool completes the engraving machining task according to the predetermined trajectory. However, when dealing with complex curved surfaces with high curvature changes, the traditional technology needs a large number of mathematical operations to ensure the interpolation accuracy, including complex geometric transformation calculation, curve parameterization processing and real-time coordinate interpolation, etc. These intensive computing tasks put high requirements on the processing capacity of the system. At the same time, the real-time requirement in the actual machining process requires the system to complete the interpolation calculation and output control instructions within a very short time. This strict limitation on the calculation speed makes it difficult for the system to perform high-precision interpolation processing, resulting in a contradiction between interpolation calculation accuracy and real-time performance when machining complex curved surfaces. That is to say, there is a technical problem in the prior art that the interpolation calculation accuracy and real-time performance of the multi-axis linkage engraving equipment are difficult to balance when machining complex curved surfaces. SUMMARY

[0003] Therefore, the present application provides a multi-axis linkage engraving equipment control algorithm optimization method, which can solve the technical problem that the interpolation calculation accuracy and real-time performance of the multi-axis linkage engraving equipment are difficult to balance when machining complex curved surfaces in the prior art.

[0004] The application is implemented in the following manner: the application provides a multi-axis linkage engraving equipment control algorithm optimization method, which comprises the following steps: obtaining an original three-dimensional data model of an engraving object, performing surface geometry analysis on the original three-dimensional data model and calculating surface curvature information, decomposing the original three-dimensional data model into a plurality of surface segmentation segments according to the curvature change rate of the surface curvature information; performing NURBS curve parameterization processing on each surface segmentation segment to generate corresponding NURBS control point coordinate data and NURBS weight coefficient data, and establishing a surface segmentation segment parameterization database; based on the kinematic geometric constraints of a multi-axis linkage equipment, performing kinematic modeling analysis on the shaft position parameters, shaft speed parameters and shaft acceleration parameters of each motion shaft, and establishing a multi-axis linkage kinematic equation set; performing path optimization processing on the engraving path by using a dynamic weight adjustment model, the dynamic weight adjustment model dynamically calculates the path planning weight coefficient according to three input parameters, i.e., the tool cutting force value, the surface roughness requirement value and the machining efficiency index value; using a minimum spanning tree path algorithm to plan the path connection sequence of the optimized engraving path, taking all the engraving path segments as graph theory nodes and the switching distance values between the path segments as graph theory edge weights, and solving the shortest path connection scheme; establishing a multi-axis linkage interpolation operation framework, adopting a hierarchical interpolation operation strategy combining pre-processing interpolation calculation and real-time interpolation compensation, calculating rough interpolation coordinate points in the pre-processing interpolation calculation stage, and performing fine interpolation coordinate correction according to the actual machining state in the real-time interpolation compensation stage; constructing a multi-axis linkage control instruction sequence, converting the optimized interpolation coordinate data into shaft control instructions of each shaft, and the shaft control instructions comprising shaft position instructions, shaft speed instructions and shaft acceleration instructions.

[0005] The curvature change rate refers to the change degree value of the surface curvature information between adjacent coordinate points on the surface of the original three-dimensional data model, and the surface segmentation segment refers to a geometric area unit obtained by decomposing the original three-dimensional data model according to the curvature change rate.

[0006] The NURBS control point coordinate data refers to a control point position coordinate set of a non-uniform rational B-spline curve, and the NURBS weight coefficient data refers to a numerical parameter set for controlling the influence degree of the NURBS control point coordinate data.

[0007] The kinematic geometric constraints refer to the geometric position constraints and motion speed constraints that must be met by each motion shaft in the multi-axis linkage equipment during coordinated motion, and the multi-axis linkage kinematic equation set refers to a mathematical equation set describing the relationship between the shaft position parameters, shaft speed parameters and shaft acceleration parameters according to the kinematic geometric constraints.

[0008] The structure of the dynamic weight adjustment model is an encoder-decoder neural network architecture, the encoder part contains 3 fully connected calculation layers for input feature extraction, the decoder part contains 2 fully connected calculation layers to output path planning weight coefficients, and the weight fusion coefficient parameters in the model are determined by the weight adjustment calculation function according to the tool cutting force value, surface roughness requirement value and machining efficiency index value.

[0009] The training data set of the dynamic weight adjustment model includes collecting data of engraving processing tasks of different material types and different geometric complexities, recording corresponding tool cutting force values, surface roughness requirement values, machining efficiency index values and optimal path planning weight coefficient labels, and establishing a training data set containing 5000 training samples and a test data set containing 1000 test samples.

[0010] The training of the dynamic weight adjustment model includes using the mean square error loss function and the Adam optimization algorithm for neural network training, setting the learning rate to 0.001, the batch size to 32, and the training period to 200, and evaluating the model performance on the test data set every 20 training periods during the training process, and stopping training and saving the optimal model parameters when the test loss does not decrease for 5 consecutive training periods.

[0011] The weight adjustment calculation function is used to adjust the weight fusion coefficient parameters of the dynamic weight adjustment model, and the weight adjustment calculation function is calculated based on the tool cutting force value, the surface roughness requirement value and the machining efficiency index value. When the weight adjustment coefficient value belongs to different value ranges, different weight fusion coefficient parameter values are used to adjust the weight fusion coefficient parameters of the dynamic weight adjustment model.

[0012] When the weight adjustment coefficient value α∈[0, 0.25), the weight fusion coefficient parameter is set to 0.1, when the weight adjustment coefficient value α∈[0.25, 0.5), the weight fusion coefficient parameter is set to 0.3, when the weight adjustment coefficient value α∈[0.5, 0.75), the weight fusion coefficient parameter is set to 0.6, and when the weight adjustment coefficient value α∈[0.75, 1.0], the weight fusion coefficient parameter is set to 0.9.

[0013] The interpolation coordinate data refers to the final interpolation position coordinate data obtained after the hierarchical interpolation operation strategy processing, the axis control instruction refers to the instruction signal for controlling the movement of each motion axis, the axis position instruction refers to the instruction component for controlling the axis position parameter in the axis control instruction, the axis velocity instruction refers to the instruction component for controlling the axis velocity parameter in the axis control instruction, and the axis acceleration instruction refers to the instruction component for controlling the axis acceleration parameter in the axis control instruction.

[0014] wherein the minimum spanning tree path algorithm refers to a graph theory algorithm for finding a tree-like connection structure connecting all graph theory nodes with minimum total weight in a weighted connected graph, the graph theory node refers to a graph theory element representing an engraving path segment in the minimum spanning tree path algorithm, and the graph theory edge weight refers to a graph theory parameter representing a switching distance value between path segments in the minimum spanning tree path algorithm.

[0015] wherein the hierarchical interpolation operation strategy refers to an interpolation operation method dividing interpolation coordinate calculation into two calculation levels of a preprocessing interpolation calculation level and a real-time interpolation compensation level, the coarse interpolation coordinate point refers to a preliminary interpolation position coordinate calculated in the preprocessing interpolation calculation stage, and the fine interpolation coordinate correction refers to a coordinate position correction processing of the coarse interpolation coordinate point in the real-time interpolation compensation stage.

[0016] wherein the axis position parameter refers to a coordinate value describing a current spatial position of each motion axis, the axis velocity parameter refers to a value describing a current motion velocity of each motion axis, and the axis acceleration parameter refers to a value describing a current acceleration of each motion axis.

[0017] wherein the tool cutting force value refers to a value of a cutting resistance received by the tool in the engraving process, the surface roughness requirement value refers to a roughness value index of a surface quality requirement of the engraving process, the machining efficiency index value refers to a value index measuring the production efficiency of the engraving process, and the path planning weight coefficient refers to a value coefficient for balancing the relationship between the machining quality and the machining efficiency.

[0018] wherein the switching distance value refers to a spatial distance value required for the tool to move between different engraving path segments, and the shortest path connection scheme refers to an optimal order scheme connecting all engraving path segments calculated by the minimum spanning tree path algorithm.

[0019] wherein the weight fusion coefficient parameter refers to a value parameter for fusing weights of different input features in the dynamic weight adjustment model, and the weight adjustment coefficient value refers to a value result calculated by the weight adjustment calculation function for determining the weight fusion coefficient parameter.

[0020] This invention establishes a hierarchical interpolation strategy, dividing the interpolation calculation process into two levels: preprocessing interpolation calculation and real-time interpolation compensation. The preprocessing stage completes most of the complex geometric calculations and coarse interpolation coordinate calculations, while the real-time stage only performs necessary fine coordinate corrections, effectively reducing the real-time computational burden. By introducing NURBS curve parameterization and a dynamic weight adjustment model, the system can adaptively adjust the interpolation strategy according to the geometric characteristics of the surface and processing requirements, optimizing computational efficiency while ensuring processing accuracy, thus solving the problem of decreased real-time performance due to improved interpolation accuracy in traditional technologies. By employing a minimum spanning tree path algorithm for global optimization of the carving path, combined with multi-axis linkage kinematic modeling, the system can achieve high-precision interpolation calculations while meeting real-time requirements, thereby solving the technical problem of balancing interpolation accuracy and real-time performance in complex surface processing using multi-axis linkage carving equipment. Attached Figure Description

[0021] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0023] like Figure 1 The diagram shown is a flowchart of a control algorithm optimization method for a multi-axis linkage engraving equipment provided by the present invention. This method includes the following steps:

[0024] A method for optimizing the control algorithm of a multi-axis linkage engraving equipment includes the following steps:

[0025] S01. Obtain the original three-dimensional data model of the carving object, perform surface geometry analysis on the original three-dimensional data model and calculate the surface curvature information, and decompose the original three-dimensional data model into multiple surface segmentation segments according to the curvature change rate of the surface curvature information.

[0026] S02. Perform NURBS curve parameterization on each surface segment to generate corresponding NURBS control point coordinate data and NURBS weight coefficient data, and establish a surface segment parameterization database.

[0027] S03. Based on the kinematic geometric constraints of the multi-axis linkage equipment, kinematic modeling and analysis are performed on the axis position parameters, axis velocity parameters and axis acceleration parameters of each motion axis, and a set of kinematic equations for multi-axis linkage is established.

[0028] S04, the dynamic weight adjustment model is used to optimize the path of the carving path, and the dynamic weight adjustment model dynamically calculates the path planning weight coefficient according to the three input parameters of the tool cutting force value, the surface roughness requirement value and the machining efficiency index value;

[0029] S05, the minimum spanning tree path algorithm is used to plan the path connection sequence of the optimized carving path, all carving path segments are taken as graph theory nodes, and the switching distance value between the path segments is taken as the graph theory edge weight, and the shortest path connection scheme is solved;

[0030] S06, a multi-axis linkage interpolation operation framework is established, a hierarchical interpolation operation strategy combining preprocessing interpolation calculation and real-time interpolation compensation is adopted, rough interpolation coordinate points are calculated in the preprocessing interpolation calculation stage, and fine interpolation coordinate correction is carried out according to the actual machining state in the real-time interpolation compensation stage;

[0031] S07, a multi-axis linkage control instruction sequence is constructed, and the optimized interpolation coordinate data is converted into axis control instructions of each axis, the axis control instructions including axis position instructions, axis speed instructions and axis acceleration instructions.

[0032] Wherein, the curvature change rate refers to the change degree value of the curvature information of the adjacent coordinate points on the surface of the original three-dimensional data model.

[0033] Wherein, the surface segmentation segment is a geometric area unit obtained by decomposing the original three-dimensional data model according to the curvature change rate.

[0034] Wherein, the NURBS control point coordinate data is a set of control point position coordinates of a non-uniform rational B-spline curve.

[0035] Wherein, the NURBS weight coefficient data is a set of numerical parameter values for controlling the influence degree of the NURBS control point coordinate data.

[0036] Wherein, the kinematic geometric constraint refers to the geometric position constraint and motion speed constraint condition that must be met by each motion axis in the multi-axis linkage device in the coordinated motion process.

[0037] Wherein, the axis position parameter is a coordinate value describing the current spatial position of each motion axis.

[0038] Wherein, the axis speed parameter is a numerical value describing the current motion speed of each motion axis.

[0039] Wherein, the axis acceleration parameter is a numerical value describing the current acceleration of each motion axis.

[0040] Wherein, the multi-axis linkage kinematic equation set is a set of mathematical equations describing the relationship between the axis position parameter, the axis speed parameter and the axis acceleration parameter according to the kinematic geometric constraint.

[0041] The dynamic weight adjustment model is a mathematical calculation model for dynamically calculating the path planning weight coefficients according to the tool cutting force values, surface roughness requirement values and machining efficiency index values. The specific structure of the dynamic weight adjustment model is an encoder-decoder neural network architecture, the encoder part includes three fully connected calculation layers for input feature extraction, the decoder part includes two fully connected calculation layers to output the path planning weight coefficients, and the weight fusion coefficient parameters in the model are determined by calculating the weight adjustment calculation function according to the tool cutting force values, surface roughness requirement values and machining efficiency index values. The steps of establishing the training data set of the dynamic weight adjustment model specifically include collecting engraving processing task data of different material types and different geometric complexity, recording the corresponding tool cutting force values, surface roughness requirement values, machining efficiency index values and optimal path planning weight coefficient labels, establishing a training data set containing 5000 training samples and a test data set containing 1000 test samples, each training sample in the training data set contains an input feature vector and a corresponding path planning weight coefficient label, and each test sample in the test data set contains an input feature vector and a corresponding path planning weight coefficient label. The steps of training the dynamic weight adjustment model specifically include adopting the mean square error loss function and the Adam optimization algorithm to perform neural network training, setting the learning rate to 0.001, the batch processing size to 32, and the training cycle number to 200, evaluating the model performance on the test data set every 20 training cycles during the training process, stopping the training and saving the optimal model parameters when the test loss does not decrease for 5 consecutive training cycles.

[0042] The tool cutting force value is a value of a cutting resistance received by the tool in the engraving process. The surface roughness requirement value is a value index of the roughness of the surface quality required in the engraving process. The processing efficiency index value is a value index for measuring the production efficiency of the engraving process. The path planning weight coefficient is a value coefficient for balancing the relationship between the processing quality and the processing efficiency. The minimum spanning tree path algorithm is a graph theory algorithm for finding a tree-shaped connection structure with the minimum total weight connecting all graph theory nodes in a weighted connected graph. The graph theory node is a graph theory element representing the engraving path segment in the minimum spanning tree path algorithm. The graph theory edge weight is a graph theory parameter representing the switching distance value between path segments in the minimum spanning tree path algorithm. The switching distance value is a spatial distance value required for the tool to move between different engraving path segments. The shortest path connection scheme is an optimal sequence scheme for connecting all engraving path segments calculated by the minimum spanning tree path algorithm. The layered interpolation operation strategy is an interpolation operation method of dividing the interpolation coordinate calculation into two calculation levels of a preprocessing interpolation calculation layer and a real-time interpolation compensation layer. The rough interpolation coordinate point is a preliminary interpolation position coordinate calculated in the preprocessing interpolation calculation stage. The fine interpolation coordinate correction is a coordinate position correction processing of the rough interpolation coordinate point in the real-time interpolation compensation stage. The interpolation coordinate data is the final interpolation position coordinate data obtained after the layered interpolation operation strategy processing. The axis control instruction is an instruction signal for controlling the movement of each motion axis. The axis position instruction is an instruction component for controlling the axis position parameter in the axis control instruction. The axis velocity instruction is an instruction component for controlling the axis velocity parameter in the axis control instruction. The axis acceleration instruction is an instruction component for controlling the axis acceleration parameter in the axis control instruction.

[0043] The weight adjustment calculation function is used to adjust the weight fusion coefficient parameter of the dynamic weight adjustment model. The weight adjustment calculation function is calculated based on the tool cutting force value, the surface roughness requirement value, and the processing efficiency index value to obtain a weight adjustment coefficient value. When the weight adjustment coefficient value belongs to different value ranges, different weight fusion coefficient parameter values are used to adjust the weight fusion coefficient parameter of the dynamic weight adjustment model. When the weight adjustment coefficient value α∈[0, 0.25), the weight fusion coefficient parameter is set to 0.1. When the weight adjustment coefficient value α∈[0.25, 0.5), the weight fusion coefficient parameter is set to 0.3. When the weight adjustment coefficient value α∈[0.5, 0.75), the weight fusion coefficient parameter is set to 0.6. When the weight adjustment coefficient value α∈[0.75, 1.0], the weight fusion coefficient parameter is set to 0.9.

[0044] The weight fusion coefficient parameter is a value parameter for fusing different input feature weights in the dynamic weight adjustment model.

[0045] The weight adjustment coefficient value is a numerical result for determining the weight fusion coefficient parameter calculated by the weight adjustment calculation function.

[0046] The specific implementation of the above steps is described in detail below.

[0047] The specific implementation of step S01 is that the original three-dimensional data model of the carving object is first obtained by using a three-dimensional scanning device or computer-aided design software, and the model stores geometric information in the form of point cloud data or polygonal mesh. When performing surface geometry analysis on the original three-dimensional data model, the principal curvature and mean curvature of each point on the model surface are calculated by using differential geometry theory, and the curvature information of the surface is calculated by using the normal vector change between adjacent coordinate points. In the process of calculating the curvature change rate, the curvature change threshold is set to 0.15, and when the curvature difference between adjacent coordinate points exceeds the threshold, the curvature change boundary is determined. Based on the curvature change rate, the surface segmentation uses a region growing algorithm, takes a region with small curvature change as a seed point, and gradually expands to the curvature change boundary to form a surface segmentation segment with relatively uniform geometric characteristics. The purpose of this step is to decompose the complex three-dimensional model into geometric units that are easy to process, thereby providing a basis for subsequent parameterization processing and path planning.

[0048] The specific implementation of step S02 is that each surface segmentation segment is subjected to parameterization processing by using a non-uniform rational B-spline surface fitting algorithm. First, the control point grid distribution of the NURBS surface is determined by using the least square method, and the number of control points is adaptively adjusted according to the complexity of the surface, and is generally set to a control point grid of 8×8 to 16×16. The calculation of the weight coefficient uses an iterative optimization algorithm, the initial weight is set to 1.0, the fitting error is calculated, and the weight value is adjusted to make the surface fitting accuracy within 0.01 mm. When establishing the parameterization database of the surface segmentation segment, a hierarchical storage structure is used, and the NURBS parameters of each segment are independently stored, including the control point coordinate matrix, the weight coefficient matrix, and the node vector. This step uses the continuity and differentiability characteristics of the NURBS mathematical representation to provide a smooth geometric description for the multi-axis linkage device, thereby ensuring the continuity and accuracy of the machining path.

[0049] The specific implementation of step S03 is to establish a kinematic geometry constraint model based on the mechanical structure of the multi-axis linkage device. The coordinate transformation analysis is performed on the shaft position parameters of each motion axis, and the homogeneous transformation matrix is used to describe the spatial relationship between the axes. The constraints of the shaft velocity parameters are established by using the Jacobian matrix, and the tool coordinate system velocity is mapped to the joint velocity of each axis. The constraints of the shaft acceleration parameters consider the dynamics characteristics, and the maximum acceleration limit is set to 2000 mm / s 2, ensure the stability of the device movement. The establishment of multi-axis linkage kinematics equation set adopts the Lagrange mechanics principle, and converts the geometric constraints into mathematical equation set, including position constraint equation, velocity constraint equation and acceleration constraint equation. The purpose of this step is to establish the mathematical model of multi-axis movement, and to provide the kinematics basis for the subsequent path optimization and interpolation calculation.

[0050] The specific implementation of step S04 is to optimize the engraving path by constructing a dynamic weight adjustment model. The model adopts an encoder-decoder neural network architecture, and the 3 fully connected calculation layers of the encoder part contain 64, 128 and 256 neurons respectively, and use ReLU activation function for feature extraction. The 2 fully connected calculation layers of the decoder part contain 128 and 64 neurons, and the final output layer generates path planning weight coefficients. The adjustment of the weight fusion coefficient parameter is based on the weight adjustment calculation function, which considers the cutting force value, surface roughness requirement value and machining efficiency index value as three input parameters. When the weight adjustment coefficient value is in different ranges, the weight fusion coefficient parameter is adjusted accordingly to realize adaptive optimization under different machining conditions. This step realizes the intelligent association between machining parameters and path planning strategy through the nonlinear mapping ability of neural network, improves the precision and adaptability of path optimization.

[0051] The specific implementation of step S05 is to plan the connection order of the optimized engraving path by using the minimum spanning tree path algorithm. First, all the engraving path segments are regarded as nodes in graph theory, and the switching distance value between the path segments is calculated as the weight of the graph theory edge. The calculation of switching distance considers the tool lifting height and horizontal moving distance, and sets the safety height to 5mm. The minimum spanning tree algorithm is realized by Kruskal algorithm, and the connectedness is maintained by the union-find set data structure, and the edges with the smallest weight are gradually selected to construct the spanning tree. In the construction process, loop formation is avoided and all nodes are connected. The final shortest path connection scheme can minimize the tool idle time and improve the machining efficiency. This step uses the mathematical properties of graph theory algorithm to systematically solve the order optimization problem of multiple path segments, avoiding the defect that the traditional greedy algorithm may fall into local optimum.

[0052] The specific implementation of step S06 is to establish a hierarchical interpolation operation framework, and adopt a strategy combining pre-processing interpolation calculation and real-time interpolation compensation. In the pre-processing interpolation calculation stage, a look-ahead algorithm is used to analyze the geometric characteristics of the path segment, and the rough interpolation coordinate points are calculated, with the interpolation step length set to 0.1 mm and the chord height error controlled within 0.005 mm. In the real-time interpolation compensation stage, an adaptive interpolation algorithm is used to finely correct the interpolation coordinates according to the feedback information of the actual machining state. The compensation algorithm considers factors such as machine tool vibration, tool wear and material deformation, and dynamically adjusts the interpolation parameters. The hierarchical interpolation operation strategy distributes the computational load to the pre-processing and real-time stages, ensuring both interpolation accuracy and real-time requirements. This step solves the contradiction between high-precision interpolation and real-time requirements, and provides a stable and reliable motion control foundation for multi-axis linkage equipment.

[0053] The specific implementation of step S07 is to convert the optimized interpolation coordinate data into control instruction sequences for each axis. The axis position instruction maps the interpolation points in the tool coordinate system to the joint coordinates of each axis through coordinate transformation, and uses a homogeneous transformation matrix for coordinate conversion. The axis velocity instruction is calculated based on the time interval between interpolation coordinate points, and a quintic polynomial interpolation is used to ensure the smoothness of the velocity curve, with the maximum feed rate limited to 3000 mm / min. The axis acceleration instruction is obtained by numerically differentiating the velocity instruction and is filtered to reduce high-frequency noise. The generation of control instruction sequences uses a time synchronization mechanism to ensure the coordination and synchronization of the motion of each axis. This step completes the conversion from the geometric path to the control instruction, providing direct instruction input for the precise motion control of multi-axis linkage equipment.

[0054] It should be noted that the dynamic weight adjustment model adopts an encoder-decoder neural network architecture. The encoder part includes 3 fully connected calculation layers, the first layer includes 64 neurons, the second layer includes 128 neurons, and the third layer includes 256 neurons, each layer uses a ReLU activation function and batch normalization technology. The decoder part includes 2 fully connected calculation layers, the first layer includes 128 neurons, the second layer includes 64 neurons, and the final output layer outputs the path planning weight coefficient through a Sigmoid activation function. The weight fusion coefficient parameters in the network are dynamically adjusted through a weight adjustment calculation function, which calculates the weight adjustment coefficient based on the tool cutting force value, surface roughness requirement value and machining efficiency index value, and sets the corresponding weight fusion coefficient parameters according to different value ranges.

[0055] The establishment of the training data set includes the following detailed steps. First, collect engraving task data covering different material types such as aluminum alloy, stainless steel, engineering plastic, etc., and the material hardness range covers 20HRC to 60HRC. Secondly, collect three-dimensional models of different geometric complexity, including plane relief, curved surface modeling and complex three-dimensional structure, and the model complexity is quantified by the number of surface patches and the degree of curvature variation. Then record the corresponding tool cutting force value of each machining task, monitor and record the peak and average value in real time through the force sensor. The surface roughness value is set according to the product quality requirement, ranging from 0.8 μm to 6.3 μm. The machining efficiency index value is quantified by the material removal rate per unit time. The optimal path planning weight coefficient label is determined by expert experience and test optimization, and a training data set containing 5000 training samples and a test data set containing 1000 test samples are established.

[0056] The model training process uses the mean square error loss function to measure the difference between the predicted value and the true value, uses the Adam optimization algorithm for parameter update, the learning rate is set to 0.001, and the batch processing size is 32 samples. The number of training cycles is set to 200, and the model performance is evaluated on the test data set every 20 training cycles during the training process to monitor the trend of test loss. When the test loss does not decrease for 5 consecutive training cycles, the training is stopped in advance to avoid overfitting, and the model parameters at the time of the minimum test loss are saved as the optimal model.

[0057] The dynamic weight adjustment model can adaptively adjust the path planning strategy according to the change of the machining condition. Compared with the traditional fixed weight path planning method, the model can establish a complex correlation between the machining parameters and the optimization strategy through the nonlinear mapping ability of the neural network. The genetic algorithm path optimization method in the prior art needs to preset fixed fitness function weights and cannot be dynamically adjusted according to the actual machining condition, which limits the optimization effect. Although the particle swarm optimization algorithm has global search ability, its convergence speed is slow and the parameter setting depends on experience, which is difficult to adapt to complex and variable machining environments. The dynamic weight adjustment model of the present application automatically learns the optimal weight configuration under different machining conditions through deep learning technology, avoids the subjectivity and limitations of manual parameter adjustment, realizes the intelligentization and adaptability of the path optimization strategy, and significantly improves the machining quality and efficiency under different materials and geometric structures.

[0058] The key technical ideas of the present application include intelligent curved surface segmentation based on curvature variation rate, dynamic weight adjustment path optimization, layered interpolation operation strategy and minimum spanning tree path connection optimization.

[0059] The intelligent curved surface segmentation technology based on curvature rate has significant advantages over the traditional equal parameter line segmentation method. The traditional method uses uniform grid division, which cannot adapt to the changes in the geometric characteristics of the curved surface, resulting in over-subdivision in flat areas and insufficient segmentation in complex areas. The present invention calculates the curvature rate of the curved surface and adaptively adjusts the segmentation density according to the complexity of the geometric characteristics, realizing intelligent segmentation driven by geometric characteristics and improving the efficiency and accuracy of subsequent processing.

[0060] The path optimization technology with dynamic weight adjustment breaks through the limitations of traditional fixed weight optimization methods. The traditional method requires manual adjustment of weight parameters for different materials and geometric structures, and the optimization effect is heavily dependent on the operator's experience. The present invention establishes a mapping relationship between processing parameters and optimization strategies through deep learning, realizing adaptive adjustment of weight coefficients and avoiding the subjectivity of manual parameter setting, improving the universality and adaptability of the optimization algorithm.

[0061] The layered interpolation operation strategy solves the contradiction between high-precision interpolation and real-time requirements. Traditional interpolation methods either pursue high precision leading to high computational complexity or focus on real-time performance at the expense of interpolation accuracy. The present invention divides the interpolation calculation into two levels: preprocessing and real-time compensation. The preprocessing stage performs rough calculation to reduce the real-time calculation load, and the real-time stage performs fine correction based on feedback information, ensuring both interpolation accuracy and real-time requirements.

[0062] The path connection optimization technology of minimum spanning tree has global optimization characteristics compared to traditional greedy algorithms. The traditional greedy algorithm is prone to local optimal solutions and cannot guarantee the global optimal path connection order. The present invention uses the minimum spanning tree algorithm in graph theory to ensure global optimal solutions and systematically solves the connection optimization problem of multiple path segments.

[0063] The synergistic effect of these four key technology ideas forms a complete multi-axis linkage engraving optimization system. Intelligent curved surface segmentation provides a reasonable geometric basis for path planning, dynamic weight adjustment realizes the intelligentization of path optimization strategy, layered interpolation guarantees the precision and real-time performance of motion control, and minimum spanning tree optimization improves the overall processing efficiency. Compared with the single optimization strategy of existing technology, the present invention realizes the whole-process optimization from geometric modeling to motion control through the integration of multiple technologies, forms a technical synergy effect, and significantly improves the overall performance and adaptability of multi-axis linkage engraving equipment.

[0064] It should be noted that the present application also solves the technical problems of low machining efficiency caused by lack of global optimization of complex curved surface engraving path planning. Traditional path planning methods usually adopt local optimization strategy, only optimize adjacent path segments, lack of global consideration of the whole engraving task, leading to frequent long-distance movement of the tool between different curved surface regions, increasing the invalid processing time and reducing the overall processing efficiency. The present application introduces the minimum spanning tree path algorithm, takes all the engraving path segments as graph theory nodes, and the switching distance between path segments as the graph theory edge weight, realizes global path optimization by solving the shortest path connection scheme, effectively reduces the invalid movement distance of the tool, and improves the processing efficiency.

[0065] The technical problem of unstable control effect under different machining conditions caused by lack of adaptive adjustment mechanism of multi-axis linkage control parameters. Traditional control systems usually adopt fixed control parameter settings, which cannot dynamically adjust according to different material properties, geometric complexity and machining precision requirements, resulting in uneven control effect when facing diversified machining tasks, affecting the consistency of machining quality. The present application constructs a dynamic weight adjustment model, uses neural network architecture to calculate path planning weight coefficient according to tool cutting force, surface roughness requirement and machining efficiency index in real time, realizes intelligent adaptive adjustment of control parameters, so that the system can automatically optimize the control strategy according to different machining conditions, and ensures the stability and consistency of the control effect.

[0066] The technical problem of waste of computing resources caused by high computational complexity and lack of pertinence of traditional NURBS parameterization processing method. Existing technologies usually perform unified NURBS parameterization processing on the whole three-dimensional model, ignoring the difference of geometric characteristics of different curved surface regions, leading to unreasonable allocation of computing resources, and over-fine parameterization calculation in the area with gentle geometric change, while the parameterization accuracy may be insufficient in the area with severe geometric change. The present application adopts a surface segmentation strategy based on curvature change rate, performs targeted NURBS parameterization processing according to the geometric feature difference of each curved surface segment, reasonably allocates computing resources, and significantly improves the calculation efficiency while ensuring the parameterization accuracy.

[0067] Specifically, the principle of the present application is that the core principle of the present application for balancing the interpolation calculation accuracy and real-time performance lies in adopting a time separation and space layering calculation strategy. First, through surface geometry analysis and curvature information calculation, the complex three-dimensional data model is intelligently segmented according to the curvature change rate to form a plurality of surface segments with relatively uniform geometric characteristics. This segmentation method can reduce the geometric complexity within a single segment and lay a foundation for subsequent parameterization processing. Then, a mathematical description model of the surface segment is established through NURBS curve parameterization processing to convert the complex surface geometry information into parameterized data convenient for calculation. This parameterized representation not only maintains the geometric accuracy of the surface, but also significantly simplifies the mathematical complexity of the interpolation calculation. In the interpolation calculation link, the layered interpolation operation strategy processes the calculation task according to the time urgency. The preprocessing layer undertakes most of the calculation-intensive tasks, including complex geometric transformation and rough coordinate calculation. These calculations can be completed in the preprocessing stage with less time pressure. The real-time layer only needs to make local fine coordinate correction according to the current machining state, which greatly reduces the burden of real-time calculation. The dynamic weight adjustment model realizes adaptive response to different machining conditions through a neural network architecture. According to the tool cutting force, surface roughness requirement and machining efficiency index, the path planning weight is dynamically adjusted, so that the interpolation strategy can be optimized according to the actual machining requirement, and the real-time performance is maximized on the premise of ensuring the necessary accuracy. The minimum spanning tree path algorithm reduces unnecessary tool movement and axis switching through global path optimization, further reducing the complexity of real-time control, so that the system can balance high-precision interpolation and real-time response under limited computing resources.

[0068] A specific embodiment 1 of the present application is provided below, and the specific implementation of each step in embodiment 1 is described in detail as follows.

[0069] The specific implementation of step S01 is that when performing surface geometry analysis on the original three-dimensional data model, the curvature information of each point on the model surface is calculated using differential geometry theory. The principal curvature calculation formula is:

[0070]

[0071] In the formula, κ1 is the principal curvature, with a unit of m -1 ; f xx , f yy is the second-order partial derivative of the surface function f(x, y), with a unit of m -1 ; f xy is the mixed partial derivative, with a unit of m -1 ; f x , f y is the first-order partial derivative, dimensionless; x, y are surface parameter coordinates, with a unit of m. The curvature change rate calculation uses the curvature difference formula between adjacent coordinate points:

[0072] Δκ i =||κ i+1 -κ i ||;

[0073] In the formula, Δκ i The rate of change of curvature at the i-th coordinate point is expressed in meters. -1 ;κ i+1 The curvature of the (i+1)th coordinate point, in meters. -1 ;κ i Let be the curvature of the i-th coordinate point, in meters. -1 ; i is the coordinate point index, dimensionless. When the rate of change of curvature exceeds the set threshold of 0.15, it is determined as the surface segmentation boundary, realizing adaptive segmentation based on geometric features.

[0074] The specific implementation of step S02 is as follows: when performing NURBS parameterization processing on the surface segment, the non-uniform rational B-spline surface representation formula is adopted:

[0075]

[0076] In the formula, S(u, v) are the coordinates of a point on the NURBS surface, in meters; N i，p (u), N j，q (v) is a B-spline basis function, dimensionless; w i，j P is the weighting coefficient, dimensionless; i，j Here, represents the coordinates of the control points, in meters (m); m and n are the number of rows and columns in the control point grid, dimensionless; p and q are the orders of the surface in the u and v directions, dimensionless; u and v are parameter variables, both ranging from [0, 1], dimensionless. The optimization of the weighting coefficients uses the least squares fitting error function:

[0077]

[0078] In the formula, E represents the fitting error, in meters (m). 2 Q k Here are the coordinates of the k-th sampling point, in meters; K is the total number of sampling points, dimensionless; u k v k The coordinates of the parameters corresponding to the k-th sampling point are dimensionless.

[0079] The specific implementation of step S03 is that, when establishing the multi-axis linkage kinematic equations, a homogeneous transformation matrix is ​​used to describe the spatial relationship between each axis:

[0080]

[0081] In the formula, T is a homogeneous transformation matrix with a dimension of 4*4; R is a 3*3 rotation matrix, dimensionless; and t is a 3*1 translation vector, with a unit of m. The multi-axis linkage kinematics equation set is expressed as:

[0082]

[0083] In the formula, is a tool coordinate system velocity vector, with a unit of m / s; J(q) is a Jacobian matrix, with a unit of m / rad; is a joint angular velocity vector, with a unit of rad / s; and q is a joint angular position vector, with a unit of rad. The calculation formula of the Jacobian matrix is:

[0084]

[0085] In the formula, f(q) is a forward kinematics function, with an output of a tool coordinate, with a unit of m.

[0086] The specific implementation of step S04 is that the expression of the weight adjustment calculation function in the dynamic weight adjustment model is:

[0087]

[0088] In the formula, a is a weight adjustment coefficient value, dimensionless, and the value range is [0, 1]; F c is a tool cutting force value, with a unit of N; R a is a surface roughness requirement value, with a unit of μm; E f is a machining efficiency index value, with a unit of mm 3 / min; w1, w2, and w3 are weight coefficients, dimensionless, and the values are 0.4, 0.35, and 0.25 respectively. In order to ensure the dimensionality, the input parameters need to be normalized, and the normalization formula is In the formula, F c，norm , R a，norm , and E f，norm are normalized parameters, dimensionless; F c，max , F c，min are the maximum and minimum values of the cutting force; R a，max , R a，min are the maximum and minimum values of the roughness; and E f，max , E f，min are the maximum and minimum values of the efficiency. The weight fusion coefficient parameter β is determined according to the weight adjustment coefficient value, dimensionless, and when a∈[0, 0.25), β=0.1; when a∈[0.25, 0.5), β=0.3; when a∈[0.5, 0.75), β=0.6; and when a∈[0.75, 1.0], β=0.9.

[0089] The specific implementation of step S05 is the same as the foregoing, and will not be described in detail here.

[0090] The specific implementation of step S06 is that the pre-processing interpolation calculation in the layered interpolation operation framework uses a parameterized interpolation formula:

[0091] P(t) = (1-t) 3 P0+3t(1-t) 2 P1+3t 2 (1-t)P2+t 3 P3;

[0092] In the formula, P(t) is the interpolation point coordinate, with a unit of m; t is a parameter variable, with a value range of [0, 1] and being dimensionless; P0, P1, P2, and P3 are Bezier curve control point coordinates, with a unit of m. The real-time interpolation compensation uses an error compensation function:

[0093] P comp (t) = P(t) + ΔP(t);

[0094] In the formula, P comp (t) is the compensated interpolation point coordinate, with a unit of m; and ΔP(t) is a compensation vector, with a unit of m. The calculation formula of the compensation vector is:

[0095]

[0096] In the formula, e(t) is a position error, with a unit of m; K p is a proportional coefficient, with a dimensionless value, and a typical value is 0.8; K i is an integral coefficient, with a unit of s -1 , and a typical value is 0.2; K d is a differential coefficient, with a unit of s, and a typical value is 0.1; and τ is an integral variable, with a unit of s.

[0097] The specific implementation of step S07 is the same as the foregoing, and will not be described in detail here.

[0098] It needs to be explained that the formula involved in the explanation part of the weight adjustment calculation function is:

[0099] F c is obtained in an experimental manner, including the following steps: step 1: installing a three-way force sensor at the connection between the main shaft and the tool; step 2: monitoring the cutting component forces in the x, y, and z directions in real time during the cutting process; and step 3: calculating the resultant cutting force In the formula, F x , F y , and F z are the component forces in the x, y, and z directions, respectively, all with a unit of N. R aThe surface roughness is measured by a surface roughness meter, and the measurement range is 0.1-25 μm. f The material removal rate is calculated by the formula In the formula, V remove is the volume of removed material, and the unit is mm 3 , t machine is the processing time, and the unit is min.

[0100] It should be noted that the principle of the curvature change rate calculation formula is based on the quantitative description of the bending degree of the surface in differential geometry. By calculating the gradient change of the curvature between adjacent points, the mutation position of the geometric feature of the surface can be accurately identified. Compared with the traditional equal parameter segmentation method, this formula realizes adaptive segmentation driven by geometric features, avoids the problems of excessive subdivision of flat areas and insufficient segmentation of complex areas, and improves the accuracy and efficiency of subsequent path planning. The principle of the NURBS surface representation formula is based on the mathematical theory of rational B-spline. By combining control points, weight coefficients and basis functions, the surface can be accurately described. Compared with the traditional polygonal mesh representation, this formula provides a continuous and differentiable surface description, ensuring the smoothness of the multi-axis linkage machining path and eliminating the mutation and discontinuity problems caused by discrete representation. The principle of the multi-axis linkage kinematics equation set is based on the kinematics transformation theory in robotics. By establishing the mapping relationship between the tool coordinate system and the joint coordinate system through the Jacobian matrix, compared with the traditional single-axis control method, this equation set realizes the mathematical description of multi-axis coordinated motion, ensuring the synchronization and continuity of the motion of each axis in the complex surface machining process. The principle of the weight adjustment calculation function is based on the weight distribution strategy in multi-objective optimization theory. By considering the cutting force, surface quality and efficiency as three key indicators, the optimization weight of path planning is dynamically adjusted. Compared with the traditional fixed weight method, this function realizes the adaptive adjustment of machining parameters, improving the machining adaptability of different materials and geometric structures. The principle of the layered interpolation formula is based on the parametric curve interpolation theory in numerical analysis. By combining Bezier curves and PID compensation, high precision and real-time are realized. Compared with the traditional single-layer interpolation method, this formula effectively solves the contradiction between interpolation accuracy and computational complexity, ensuring the stability and accuracy of the motion control of the multi-axis linkage equipment.

[0101] In order to better understand and implement the present application, the following provides an embodiment 2 of a specific application scenario of the present application: a precise carving machining task of a jade carving work, which requires a surface roughness of Ra≤0.8 μm and a carving accuracy controlled within ±0.01 mm. The traditional multi-axis linkage carving method has the problem of difficulty in balancing the interpolation calculation accuracy and real-time when processing such complex curved jade, resulting in low carving efficiency and unsatisfactory detail performance.

[0102] The technical team first obtains the original three-dimensional data model of the jade carving work, which is generated through three-dimensional scanning and digital modeling technology, containing 245680 triangular facets and 122945 vertex coordinate data. Surface geometry analysis is performed on the original three-dimensional data model, and the surface curvature information distribution is calculated. According to the standard of setting the curvature change rate threshold to 0.35 / mm, the original three-dimensional data model is decomposed into 73 surface segmentation segments, of which the high curvature change area is decomposed into 48 segments and the low curvature change area is decomposed into 25 segments.

[0103] NURBS curve parameterization processing is performed on each surface segmentation segment to generate corresponding NURBS control point coordinate data and NURBS weight coefficient data. The high curvature change segment adopts a 5th order NURBS curve, the control point number is 16-20, and the weight coefficient range is 0.7-1.3. The low curvature change segment adopts a 3rd order NURBS curve, the control point number is 8-12, and the weight coefficient range is 0.9-1.1. A surface segmentation segment parameterization database is established, with a total capacity of 4.6MB, containing complete parameterization information of 73 segments.

[0104] Based on the kinematic geometry constraints of the five-axis linkage jade carving equipment, the axis position parameters, axis speed parameters and axis acceleration parameters of X axis, Y axis, Z axis, A axis and C axis are analyzed. The X axis stroke is 400mm, the Y axis stroke is 300mm, the Z axis stroke is 250mm, the A axis rotation range is ±90°, and the C axis rotation range is ±360°. The maximum speed of each axis is X axis 8m / min, Y axis 8m / min, Z axis 6m / min, A axis 12° / s, C axis 18° / s. The maximum acceleration of each axis is X axis 1.2m / s 2 , Y axis 1.2m / s 2 , Z axis 0.8m / s 2 , A axis 3° / s 2 , C axis 5° / s 2 . A multi-axis linkage kinematics equation set is established, containing 15 constraint equations, describing the position, speed and acceleration relationship between axes.

[0105] The dynamic weight adjustment model is used for path optimization processing of the engraving path. The dynamic weight adjustment model adopts an encoder-decoder neural network architecture. The encoder part includes 3 fully connected calculation layers with neuron numbers of 64, 32, and 16 respectively. The decoder part includes 2 fully connected calculation layers with neuron numbers of 16 and 8 respectively. According to the current engraving conditions, the tool cutting force value is 85N, the surface roughness requirement value is 0.8μm, and the processing efficiency index value is 78%. The weight adjustment coefficient value α=0.58 is calculated by the weight adjustment calculation function. The value belongs to the interval [0.5, 0.75), so the weight fusion coefficient parameter is set to 0.6. The path planning weight coefficient output by the dynamic weight adjustment model is 0.69, which is used for subsequent path optimization calculation.

[0106] The minimum spanning tree path algorithm is used to plan the path connection sequence of the optimized engraving path. The 73 curved surface segmentation path segments are taken as graph theory nodes, and the switching distance values between the path segments are calculated as graph theory edge weights, as shown in Table 1.

[0107] Table 1 Switching distance values between main path segments

[0108] Path segment number Adjacent path segment Switching distance value (mm) 1 2 6.8 2 15 18.5 3 4 9.2 4 5 12.6 15 16 14.3 25 26 21.7 35 36 16.9 48 49 19.4 62 63 13.8 72 73 8.1

[0109] Through the minimum spanning tree path algorithm, the shortest path connection scheme is obtained, and the total switching distance is 892.6mm, which is reduced by 20.6% compared with the original sequence connection of 1124.3mm.

[0110] A multi-axis linkage interpolation operation framework is established, and a hierarchical interpolation operation strategy combining preprocessing interpolation calculation and real-time interpolation compensation is adopted. In the preprocessing interpolation calculation stage, the system precalculates the interpolation coordinates of the 73 curved surface segmentation segments in the offline state, and the interpolation step is set to 0.02mm. A total of 67845 rough interpolation coordinate points are calculated. The preprocessing calculation time is 6.8s, and the occupied memory space is 26.4MB. In the real-time interpolation compensation stage, the system corrects the rough interpolation coordinate points according to the actual engraving state, and the correction range is controlled within ±0.005mm. The real-time calculation period is 0.8ms, which meets the real-time requirements of the system.

[0111] In the actual engraving process, the dynamic weight adjustment model continuously optimizes the path planning weight coefficient according to the real-time monitoring of the engraving parameters. The key parameter changes in the engraving process are shown in Table 2.

[0112] Table 2 Key parameter changes in the engraving process

[0113]

[0114] The multi-axis linkage control instruction sequence is constructed, and the optimized interpolation coordinate data is converted into axis control instructions of each axis. The axis control instructions include axis position instructions, axis speed instructions and axis acceleration instructions. The X-axis position instruction range is 0-280mm, the speed instruction range is 0-6m / min, and the acceleration instruction range is 0-1.0m / s 2 . The Y-axis position instruction range is 0-180mm, the speed instruction range is 0-6m / min, and the acceleration instruction range is 0-1.0m / s 2 . The Z-axis position instruction range is -80-80mm, the speed instruction range is 0-4m / min, and the acceleration instruction range is 0-0.6m / s 2 . The A-axis position instruction range is -60°-60°, the speed instruction range is 0-10° / s, and the acceleration instruction range is 0-2.5° / s 2 . The C-axis position instruction range is -180°-180°, the speed instruction range is 0-15° / s, and the acceleration instruction range is 0-4° / s 2 .

[0115] The whole jade carving work carving process consumes 428 minutes, the actual carving precision reaches ±0.008mm, the average value of the surface roughness is Ra=0.6μm, and the design requirements are met. The interpolation calculation of the system during the carving process is good in real-time performance, the average interpolation period is 0.6ms, the maximum interpolation period is 0.9ms, and no real-time overrun phenomenon occurs. The multi-axis linkage control precision is stable, the position control error of each axis is controlled within 0.003mm, the speed control error is controlled within 4%, and the acceleration control error is controlled within 6%.

[0116] It should be noted that the variables involved in the present application are explained in detail as shown in Table 3.

[0117] Table 3: Explanation of variables

[0118]

[0119] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A method of optimizing a control algorithm for a multi-axis linked engraving apparatus, the method comprising: The application relates to a method for generating a multi-axis linkage control instruction sequence based on a dynamic weight adjustment model. The method comprises the following steps: acquiring an original three-dimensional data model of a sculpture object, performing surface geometry analysis on the original three-dimensional data model and calculating surface curvature information, decomposing the original three-dimensional data model into a plurality of surface segmentation segments according to the curvature variation rate of the surface curvature information, performing NURBS curve parameterization processing on each surface segmentation segment to generate corresponding NURBS control point coordinate data and NURBS weight coefficient data, and establishing a surface segmentation segment parameterization database; based on the kinematic geometry constraints of a multi-axis linkage device, performing kinematic modeling analysis on the axis position parameters, the axis speed parameters and the axis acceleration parameters of each motion axis, and establishing a multi-axis linkage kinematic equation set; The method adopts a dynamic weight adjustment model to perform path optimization processing on the sculpture path, the dynamic weight adjustment model dynamically calculates path planning weight coefficients according to three input parameters, namely the tool cutting force value, the surface roughness requirement value and the machining efficiency index value; the minimum spanning tree path algorithm is used to plan the path connection sequence of the optimized sculpture path, all sculpture path segments are taken as graph theory nodes, and the switching distance values between the path segments are taken as the graph theory edge weights, so that the shortest path connection scheme is solved; a multi-axis linkage interpolation operation framework is established, a hierarchical interpolation operation strategy combining preprocessing interpolation calculation and real-time interpolation compensation is adopted, rough interpolation coordinate points are calculated in the preprocessing interpolation calculation stage, and fine interpolation coordinate correction is performed according to the actual machining state in the real-time interpolation compensation stage; a multi-axis linkage control instruction sequence is constructed, the optimized interpolation coordinate data are converted into axis control instructions of each axis, and the axis control instructions include axis position instructions, axis speed instructions and axis acceleration instructions.

2. The multi-axis gang engraving apparatus control algorithm optimization method of claim 1, wherein, The curvature variation rate refers to the variation degree value of the surface curvature information between adjacent coordinate points on the surface of the original three-dimensional data model, and the surface segmentation segment refers to a geometric area unit obtained by decomposing the original three-dimensional data model according to the curvature variation rate.

3. The multi-axis gang engraving apparatus control algorithm optimization method of claim 2, wherein, The NURBS control point coordinate data refer to a control point position coordinate set of a non-uniform rational B-spline curve, and the NURBS weight coefficient data refer to a numerical parameter set for controlling the influence degree of the NURBS control point coordinate data.

4. The multi-axis gang engraving apparatus control algorithm optimization method of claim 3, wherein, The kinematic geometry constraints refer to the geometric position constraints and motion speed constraints that must be met by each motion axis of the multi-axis linkage device during coordinated motion, and the multi-axis linkage kinematic equation set refers to a mathematical equation set describing the relationship between the axis position parameters, the axis speed parameters and the axis acceleration parameters according to the kinematic geometry constraints.

5. The multi-axis gang engraving apparatus control algorithm optimization method of claim 4, wherein, The structure of the dynamic weight adjustment model is an encoder-decoder neural network architecture, the encoder part contains three fully connected calculation layers for input feature extraction, the decoder part contains two fully connected calculation layers to output path planning weight coefficients, and the weight fusion coefficient parameters in the model are determined by calculating a weight adjustment calculation function according to the tool cutting force value, the surface roughness requirement value and the machining efficiency index value.

6. The multi-axis gang engraving apparatus control algorithm optimization method of claim 5, wherein, The step of establishing the training data set of the dynamic weight adjustment model includes collecting carving processing task data of different material types and different geometric complexities, recording corresponding tool cutting force values, surface roughness requirement values, processing efficiency index values and optimal path planning weight coefficient labels, and establishing a training data set containing 5000 training samples and a test data set containing 1000 test samples.

7. The multi-axis gang engraving apparatus control algorithm optimization method of claim 6, wherein, The step of training the dynamic weight adjustment model includes using the mean square error loss function and the Adam optimization algorithm for neural network training, setting the learning rate to 0.001, the batch processing size to 32, and the training cycle number to 200, evaluating the model performance on the test data set every 20 training cycles during the training process, stopping training and saving the optimal model parameters when the test loss does not decrease for 5 consecutive training cycles.

8. The multi-axis gang engraving apparatus control algorithm optimization method of claim 7, wherein, The weight adjustment calculation function is used to adjust the weight fusion coefficient parameters of the dynamic weight adjustment model. The weight adjustment calculation function is calculated based on the tool cutting force value, the surface roughness requirement value and the processing efficiency index value to obtain a weight adjustment coefficient value. When the weight adjustment coefficient value belongs to different value ranges, different weight fusion coefficient parameter values are used to adjust the weight fusion coefficient parameters of the dynamic weight adjustment model.

9. The multi-axis gang engraving apparatus control algorithm optimization method of claim 8, wherein, When the weight adjustment coefficient value α∈[0, 0.25), the weight fusion coefficient parameter is set to 0.1, when the weight adjustment coefficient value α∈[0.25, 0.5), the weight fusion coefficient parameter is set to 0.3, when the weight adjustment coefficient value α∈[0.5, 0.75), the weight fusion coefficient parameter is set to 0.6, and when the weight adjustment coefficient value α∈[0.75, 1.0], the weight fusion coefficient parameter is set to 0.

9.

10. The multi-axis gang engraving apparatus control algorithm optimization method of claim 9, wherein, The interpolation coordinate data refers to the final interpolation position coordinate data obtained after the hierarchical interpolation operation strategy processing, the axis control instruction refers to the instruction signal for controlling the movement of each motion axis, the axis position instruction refers to the instruction component for controlling the axis position parameter in the axis control instruction, the axis velocity instruction refers to the instruction component for controlling the axis velocity parameter in the axis control instruction, and the axis acceleration instruction refers to the instruction component for controlling the axis acceleration parameter in the axis control instruction.

Images