A method and system for planning a polishing path

Abstract

This invention relates to a method and system for planning grinding paths. The method includes: acquiring three-dimensional geometric shape data and surface physical property data of a target workpiece, and generating a physical feature fusion dataset through spatiotemporal registration; based on the physical feature fusion dataset, dividing the surface of the target workpiece to be ground into multiple processing sub-regions, and dividing at least one attitude-stabilized layer into each processing sub-region; using a preset path planning mechanism, generating an initial grinding path for each attitude-stabilized layer, wherein the initial grinding path consists of multiple path segments, forming an initial path segment set; based on each path segment in the initial path segment set, generating an optimal path segment execution sequence using a preset global temporal optimization mechanism; generating the final grinding operation instruction according to the optimal path segment execution sequence, and controlling the grinding equipment to execute it. This improves batch consistency, process stability, and overall efficiency of the processing quality.

Description

Technical Field

[0001] This invention relates to the field of automated processing technology, and specifically to a method and system for planning grinding paths. Background Technology

[0002] In the precision manufacturing of complex workpieces such as metal components and molds, grinding is a crucial final process used to remove burrs, reduce surface roughness, and achieve the desired finish. With the improvement of industrial automation, replacing manual labor with robots or CNC equipment has become a clear trend, and one of the core technologies lies in how to automatically generate efficient and precise grinding paths.

[0003] However, real-world industrial grinding scenarios present multi-dimensional and comprehensive challenges to automated systems. First, the workpiece may exhibit variations in local material properties due to material batches, welding processes, or heat treatment procedures, directly impacting material removal rates and surface finish. Second, grinding tools inevitably wear down during high-speed rotation and continuous contact, causing dynamic changes in their geometry and cutting performance over time. Furthermore, mechanical vibrations resulting from the combined effects of multi-axis motion systems, tool-workpiece contact dynamics, and environmental factors are key physical phenomena affecting machining accuracy, surface finish, and equipment stability.

[0004] Existing path planning technologies based on geometric models focus primarily on achieving spatial trajectory coverage. However, when faced with complex working conditions involving the coupling of multiple physics fields such as material properties, tool states, and process dynamics, the robustness, adaptability, and consistency of the generated paths with the final process results still need improvement. How to enable automated grinding systems to respond more intelligently to multi-source, time-varying factors in actual processing, generating processing paths that are not only geometrically correct but also physically superior, is a technological direction that continues to be explored in this field. Summary of the Invention

[0005] This invention addresses the technical problems existing in the prior art by integrating multi-source real-time sensing, constructing a dynamic dual-loop compensation and global spatiotemporal collaborative optimization mechanism, thereby achieving integrated adaptive control of the grinding process of complex workpieces, fundamentally improving batch consistency, process stability and overall efficiency of processing quality.

[0006] The technical solution of this invention to solve the above-mentioned technical problems is as follows: A method for planning a grinding path, comprising: S101, acquire the three-dimensional geometric shape data and surface physical property data of the target workpiece, and generate a physical feature fusion dataset through spatiotemporal registration; S102, based on the physical feature fusion dataset, divide the surface of the target workpiece to be polished into multiple processing sub-regions, and divide at least one attitude-stabilizing sheet into each processing sub-region; S103, using a preset path planning mechanism, an initial polishing path is generated for each attitude-stabilized layer. The initial polishing path consists of multiple path segments, forming an initial path segment set. S104: Based on each path segment in the initial path segment set, the optimal path segment execution sequence is generated using a preset global timing optimization mechanism; S105 generates the final grinding operation instruction based on the optimal path segment execution sequence and controls the grinding equipment to execute it. Preferably, in S101, the physical feature fusion dataset is represented as follows: , Let i be the coordinates of the i-th point. Let N be the hardness value at the i-th point, and N be the number of point clouds.

[0007] Preferably, dividing the workpiece surface to be ground into multiple processing sub-regions specifically includes: Based on the hardness and coordinate information in the physical feature fusion dataset D, the surface to be polished is divided into M spatially connected processing sub-regions with uniform material, and the average hardness value of each processing sub-region is calculated. .

[0008] Preferably, dividing each processing sub-region into at least one attitude-stabilizing layer specifically includes: For each processing sub-region Extract the set of unit normal vectors corresponding to its surface point set, and divide it into groups using a preset clustering algorithm. There are 3 normal vector clusters, each cluster defining an attitude-stabilized sheet and its representative grinding attitude.

[0009] Preferably, the preset path planning mechanism specifically includes: S201, based on each attitude stabilization layer, plans a theoretical path of tool contact points that continuously covers the layer. The theoretical path is defined by a series of ordered theoretical path points. S202, based on the current tool wear and the average hardness value of the current machining sub-region, the current radial compensation amount is obtained using a preset dynamic geometric compensation model; S203, For each theoretical path point in the theoretical path, apply a radial compensation amount along its corresponding surface normal direction to obtain the actual execution path point; S204, define a straight-line motion segment between adjacent actual execution path points as a path segment, and combine all path segments of all layers to form the initial path segment set. J represents the total number of segments.

[0010] Preferably, each path segment, in addition to being bound to start and end point coordinate information, is also associated with its corresponding processing sub-region, attitude stabilization layer, representative attitude, and based on... Recommended process parameters.

[0011] Preferably, the preset global timing optimization mechanism specifically includes: S301: Based on each path segment, extract the dynamic fingerprint features of the segment, input them into the pre-trained process vibration sensitivity prediction model, and output the vibration process sensitivity index, a scalar that characterizes the risk of vibration excitation. S302, Optimization Problem: Define the execution sequence of the path segment as follows ,in It is the index of the m-th executed segment in the sequence within the entire set, aiming to find an optimal path segment execution sequence. To minimize a cost function used to evaluate the quality of sequences. ; S303, during the optimization process, process constraints are defined to minimize the cost function value. The optimization cost function is solved, and the optimal path segment execution sequence is output. .

[0012] Preferably, the fragment dynamic fingerprint features include: geometric motion features, process load features, system interaction features, and context features.

[0013] Preferably, the pre-trained process vibration sensitivity prediction model is obtained in the following way: A1. Collect a large number of historical polishing data records in the same current scenario. Based on each path segment of each historical polishing data record, generate the corresponding segment dynamic fingerprint feature and label it. The label content is set to the vibration process sensitivity index that the segment actually corresponds to. A2. Using the dynamic fingerprint features of all segments after labeling as the training dataset, train the pre-selected neural network structure, continuously optimize the model parameters, and obtain the final process vibration sensitivity prediction model.

[0014] This application also provides a planning system for refining paths, including: an acquisition module, a partitioning module, and an optimization module; The acquisition module is used to acquire the three-dimensional geometric shape data and surface physical property data of the target workpiece, and generate a physical feature fusion dataset through spatiotemporal registration; The partitioning module is used to divide the surface of the target workpiece to be polished into multiple processing sub-regions based on the physical feature fusion dataset, and to partition at least one attitude stabilizing layer for each processing sub-region; using a preset path planning mechanism, an initial polishing path is generated for each attitude stabilizing layer, and the initial polishing path is composed of multiple path segments to form an initial path segment set. The optimization module is used to generate an optimal path segment execution sequence based on each path segment in the initial path segment set using a preset global timing optimization mechanism; generate the final grinding operation instruction based on the optimal path segment execution sequence; and control the grinding equipment to execute it.

[0015] The beneficial effects of this invention are: By synchronously collecting geometric and hardness data, a "physical digital twin" of the workpiece is constructed, giving it not only shape but also the "intrinsic properties" of the material. Based on this, it can intelligently identify the natural machining zones formed on the workpiece surface due to material differences and plan a preliminary scheme for each zone that balances efficiency (stability of the flask) and geometric accuracy (theoretical path). Furthermore, a dynamic dual-loop compensation mechanism is introduced: the inner loop is for real-time geometric compensation for tool wear, ensuring that the cutting depth does not decrease over time; the outer loop is based on data-driven vibration sensitivity prediction and global timing optimization, which actively "smooths out" dynamic load fluctuations during the machining process by intelligently scheduling the execution order of each path segment, thus suppressing vibration accumulation from the time dimension. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a method for planning a polishing path according to an embodiment of the present invention. Figure 2 This is a schematic diagram of a grinding path planning system according to an embodiment of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0018] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0019] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0020] Current mainstream automated grinding path generation methods primarily rely on 3D computer-aided design (CAD) models of the workpiece. These methods typically use offline programming software to plan the tool's motion trajectory in a virtual environment based on the model's geometry. However, such methods based on purely geometric models exhibit significant limitations in practical applications. First, they cannot perceive or address the inherent differences between the actual workpiece and the theoretical model, such as uneven material hardness caused by welding, casting, or heat treatment processes. When using fixed grinding parameters (such as pressure and feed rate) to process these areas with varying hardness, it is easy to over-grind soft areas or under-grind hard areas, severely affecting the consistency of finished product quality. Second, the continuous wear of grinding tools (such as grinding wheels) during operation is a dynamic factor that cannot be ignored. Wear alters the effective size and cutting performance of the tool, which static preset paths cannot compensate for, leading to a gradual deterioration of the machining effect over time and failing to guarantee stability in mass production.

[0021] A deeper problem is that existing technologies almost completely ignore the impact of the dynamic characteristics of the machining process itself on the path execution effect. Complex paths are usually composed of a large number of short motion segments. Even if each segment is independently optimized in terms of geometry and process parameters, if the execution sequence of these segments is improper, such as executing multiple segments requiring high speed, high acceleration, or drastic changes in direction consecutively, it is very easy to induce harmful vibrations in the machine tool-workpiece-tool system. This vibration not only causes the actual motion trajectory of the tool to deviate from the command trajectory, resulting in significant tracking errors, but also forms difficult-to-eliminate vibration marks on the workpiece surface. Furthermore, the accumulation of vibration energy may exacerbate tool wear and even cause equipment failure. Existing methods treat path planning as a purely "spatial coverage" problem, rather than optimizing it as a dynamic process of "spatiotemporal coupling." Therefore, when faced with complex curved workpieces with high precision and high quality requirements, its machining effect and process stability are difficult to achieve as expected.

[0022] Based on this, the present invention provides a method for planning grinding paths to improve the quality consistency, process stability and overall efficiency of automated grinding of complex workpieces.

[0023] Example 1 Figure 1 This is a flowchart illustrating a grinding path planning method according to an embodiment of the present invention.

[0024] like Figure 1 As shown, a method for planning a polishing path includes the following steps: S101: Acquire the three-dimensional geometric shape data and surface physical property data of the target workpiece, and generate a physical feature fusion dataset through spatiotemporal registration.

[0025] Specifically, step S101 includes: controlling a 3D scanning device to acquire 3D point cloud data of the workpiece surface. As three-dimensional geometric topography data, a three-dimensional geometric model of the workpiece surface is reconstructed; a micro-area hardness detection device (such as a portable Leeb hardness tester probe, driven by a robotic arm) is simultaneously controlled to collect Vickers hardness values ​​corresponding to the point cloud coordinates. As surface physical property data, a coordinate-hardness mapping is established; a physical feature fusion dataset is generated through spatiotemporal registration (registration and alignment of timestamps and spatial transformation matrices). ,in, Let N be the three-dimensional coordinates of the point in the machine tool coordinate system, and N be the number of point clouds. Let i be the coordinates of the i-th point. Let be the hardness value at the i-th point.

[0026] For example, spatiotemporal registration can be achieved through the following steps: based on the kinematic model of the robotic arm or a visual calibration plate, determine a fixed transformation matrix T between the coordinate system of the hardness probe and the coordinate system of the 3D scanner; perform scanning and measurement under synchronous clock control to ensure data timestamp consistency; for each hardness measurement point, use the transformation matrix T to transform its coordinates to the scanner coordinate system, thereby completing the coordinate-hardness mapping.

[0027] S102, based on the physical feature fusion dataset, divides the surface of the target workpiece to be polished into multiple processing sub-regions, and divides at least one attitude-stabilizing layer into each processing sub-region.

[0028] In some embodiments, the workpiece surface to be polished is divided into multiple processing sub-regions. Specifically, this includes: performing adaptive region segmentation based on the hardness and coordinate information in the physical feature fusion dataset D; and using a region growing algorithm that considers spatial constraints to divide the surface to be polished into M processing sub-regions that are uniform in material and spatially connected. And calculate the average hardness value of each processed sub-region. .

[0029] Specifically, taking any unclassified point in dataset D as a seed point, a new region is created, and the region growing algorithm is started: traversing all points in the spatial neighborhood of the seed point (e.g., points whose Euclidean distance value is less than a preset distance threshold), if the hardness difference between the neighboring points and the seed point satisfies... And the spatial distance satisfies If a data point is found to be a sub-region, it is assigned to the current region; this process is iterated until all data points are divided, ultimately generating M processing sub-regions. For each processing sub-region Calculate the average hardness value of all its points. .in, The preset hardness uniformity threshold (e.g., 20 HV). H is a predefined spatial proximity threshold (e.g., 5 mm), where H is the hardness value of the neighboring points. This represents the hardness value of the seed point. The coordinates of the neighboring points, These are the coordinates of the seed point. It should be noted that the hardness uniformity threshold and spatial proximity threshold are set based on actual conditions and expert experience. They are determined by process experts based on the workpiece material fluctuation range, processing accuracy requirements, and equipment performance, either through experience or experimental testing.

[0030] In some embodiments, at least one attitude-stabilizing layer is defined for each processing sub-region, specifically including: for each processing sub-region Extract the set of unit normal vectors corresponding to the surface point set from its three-dimensional geometric model. The pre-defined clustering algorithm (DBSCAN clustering algorithm) is used to divide it into groups. A cluster of normal vectors (exemplarily, setting an angle tolerance). (e.g., 10°), directions similar (angle less than) The normal vectors of the cells are clustered into one class, and each cluster defines an attitude-stabilized sheet. ( (representing the number of the attitude-stabilized lamella in the sub-region) and its representative grinding attitude. (Usually, the mean or principal direction of this type of normal vector is taken, representing an approximately fixed tool posture that can be used when grinding within this layer.) It should be noted that during processing within this layer, the grinding tool can maintain this representative posture or make minor adjustments in its vicinity, thereby significantly reducing unnecessary drastic changes in the global posture.

[0031] In other embodiments, the representative grinding posture can also be set as: the weighted average direction of all normal vectors in the corresponding posture-stabilized layer, with the weight being the reciprocal of the curvature of each point, so as to better fit the main surface area.

[0032] Therefore, the continuous attitude change pattern within a sub-region is transformed into a finite number of trained sequences representing the optimal contact attitude, greatly reducing the adjustment frequency of the attitude servo mechanism.

[0033] S103, using a preset path planning mechanism, an initial polishing path is generated for each remaining attitude-stabilized layer to be polished. The initial polishing path consists of multiple path segments, forming an initial path segment set.

[0034] In some embodiments, the preset path planning mechanism specifically includes: S201, based on each attitude-stabilized layer, uses an algorithm with equal residual height or equal step distance on its parametric surface (3D surface) to plan a theoretical path of tool contact points that continuously covers the layer. The theoretical path consists of a series of ordered theoretical path points (coordinate points). Define , where N is the number of path points in the theoretical path.

[0035] S202, real-time monitoring of the current thickness W(t) of the grinding tool (such as a grinding wheel) (t is time), its initial thickness is... (Factory settings determined), based on current tool wear ( ) and the material of the current processing sub-region (in terms of average hardness value) (Characteristics), using a preset dynamic geometric compensation model, the current radial compensation amount is obtained:

[0036] in, This is the current radial compensation amount. Indicates the current tool wear level. This is a nonlinear compensation function related to material properties and wear, used to compensate for the nonlinear decrease in cutting efficiency on materials of different hardnesses after wear-induced changes in the tool's cutting edge state. The parameters of this function were also obtained by fitting experimental data. For example, this function can be set as follows: , This represents the average hardness value of the current processed sub-region. This serves as a reference value for hardness, used for normalization (set by expert experience based on the actual workpiece scenario). , , The calibration coefficient is obtained by performing a series of controlled wear experiments on typical materials using specific tools on specific machine tools, measuring the actual grinding depth at different wear stages on test blocks of different hardness, and then fitting the data. The preset linear compensation coefficient can be calibrated through process experiments and is used to directly offset the reduction in depth of cut caused by radial wear of the tool. This is a material-dependent nonlinear compensation gain used to compensate for the nonlinear effect of decreased cutting performance on materials of different hardness after tool wear causes changes in the cutting edge state. The wear effect attenuation constant is used to describe the nonlinear characteristics of the impact on machining results in the early and late stages of wear. This model embeds both tool state (wear amount) and workpiece properties (regional hardness) into the path correction logic.

[0037] As an example, , , The parameters were determined through the following calibration experiments: using new tools and tools with known wear, fixed-point grinding experiments were conducted on standard test blocks of different hardness, and the actual grinding depth was measured; a system was established based on ( Let H be the input dataset and depth deviation be the output dataset; linear regression is used to determine... And use polynomial or exponential functions to fit the residual data to determine , .

[0038] It should be noted that the current thickness W(t) can be measured directly, or it can be determined by establishing a tool wear function with grinding time as the variable: , For the initial thickness, The wear coefficient for tool-workpiece material pairing. and These are the real-time grinding force and feed rate, respectively.

[0039] S203, for each theoretical path point in the theoretical path Along its corresponding surface normal direction (or the attitude represented by the sheet) Apply radial compensation in the direction of To obtain the actual execution path point .

[0040] Actual execution path point The new three-dimensional coordinates are calculated by superimposing a scalar compensation amount along the surface normal (in the opposite direction) at that point. This process ensures that even if the tool wears, the predetermined cutting geometry can be maintained by adjusting the path point position in real time. This is a key geometric correction step in achieving adaptive machining stability. In three-dimensional space, the compensation must be along the direction perpendicular to the workpiece surface (i.e., the normal); otherwise, it will change the lateral position of the tool, leading to overcutting or undercutting. After compensation, the actual contact point of the tool will penetrate deeper into the workpiece surface than the theoretical point (assuming that wear has shortened the tool length), thus ensuring a consistent depth of cut.

[0041] S204, adjacent actual execution path points and A linear motion segment (or a tiny motion unit based on the device interpolation characteristics) is defined as a path segment. All path fragments of all layers are combined to form the initial path fragment set. J represents the total number of segments.

[0042] Specifically, each path segment In addition to binding the start and end point coordinates, it is also associated with the processing sub-region to which it belongs. Attitude-stabilized layers , representing posture and based on Recommended process parameters (determined based on a preset material-process parameter matching strategy, and formulated by professionals or experts, such as feed rate) Spindle speed ).

[0043] For example, based on the regional average hardness value Query the pre-built process knowledge base to obtain the basic process parameter set. (These are feed rate, grinding force, and spindle speed, respectively).

[0044] S104: Based on each path segment in the initial path segment set, the optimal path segment execution sequence is generated using a preset global timing optimization mechanism, wherein the timing optimization aims to suppress the overall vibration excitation during the grinding process.

[0045] In some embodiments, the preset global timing optimization mechanism (used to change the execution order) specifically includes: S301, based on each path segment Extracting dynamic fingerprint features from fragments The input is fed into the pre-trained process vibration sensitivity prediction model, and the output is the vibration process sensitivity index. , a scalar that characterizes the risk of vibration excitation.

[0046] In some embodiments, the fragment dynamic fingerprint features include: geometric motion features, process load features, system interaction features, and context features.

[0047] Specifically: Geometric motion features include: segment start and end coordinates (describing the segment's spatial location), segment length, curvature, average normal vector of the surface on which the segment is located (obtained directly from the path segment description generated by S103), and planned average feed rate and maximum acceleration (derived from the numerical control (NC) instructions or motion planner output for the segment, determined by the process database or adaptive algorithm based on the area hardness). At least one or more of the following: (determined) The process load characteristics include at least one or more of the following: the average hardness value of the corresponding machining sub-region, the preset reference grinding force, and the spindle speed (recommended process parameters can be obtained by consulting the process knowledge base); System interaction features include: the relative wear and tear of the tool at the current moment. The segment motion-mode angle (the angle between the segment's motion direction and the system's principal mode direction at that location, calculated from the system modal database) can be understood as: the angle θ between the motion direction vector and the known dominant mode direction of the system (machine tool) near that spatial location. It can also be understood as the correlation with the possible principal mode direction of the machine tool in that region (a rough mode shape information of the machine tool-workpiece system needs to be obtained beforehand through modal experiments). The mode shape direction can be obtained through previous modal tests. For example, by querying a pre-stored spatial distribution map of system modal shapes, the dominant mode direction at that location can be obtained based on the segment's spatial position. And calculate the direction of motion of the segment. and The included angle .

[0048] Contextual features include: the segment's order within its attitude-stabilized layer (determined by the generation order), and the angle of change in motion direction relative to the previous segment (in the initial order). This helps to capture sequence-dependent vibrational effects.

[0049] It should be noted that the system modal database is obtained through experimental modal analysis. The specific steps are as follows: a standard test workpiece is installed on the machine tool, and a hammer or vibrator is used to excite the system at multiple measuring points. Simultaneously, response signals are collected using accelerometers placed on the spindle and the workpiece. Frequency response function fitting or parameter identification methods (such as PolyMAX) are used to identify the system's natural frequencies, damping ratios, and mode shapes at the sensor measuring points within the 0-500Hz range. Finally, the mode shape data of the discrete measuring points are expanded into a three-dimensional mode shape field database covering the entire machining space using an interpolation algorithm. This invention does not elaborate further or limit these steps.

[0050] In some embodiments, the pre-trained process vibration sensitivity prediction model is obtained as follows: A1. Collect a large number of historical polishing data records in the same current scenario. Based on each path segment of each historical polishing data record, generate corresponding segment dynamic fingerprint features and label them.

[0051] Specifically, the label content is set to the vibration process sensitivity index corresponding to the actual segment. For example, this index can be calculated as a scalar value by processing the collected vibration acceleration signal as the vibration sensitivity label for the sample segment. An effective labeling method is to calculate the dominant mode energy ratio in the frequency domain, with the following steps: The spectrum is obtained by performing a Fast Fourier Transform (FFT) on the signal; Identify the first N natural frequencies (e.g., N=3) of the machine tool-workpiece system in the low-frequency range (e.g., 0-500Hz). ; Calculate each natural frequency separately A narrow band nearby (such as) Vibrational energy within) ; Calculate the total energy of the entire analysis frequency band. ; The tag value is: The ratio It directly quantifies the intensity of harmful resonances that are triggered in the system when the segment is executed; the higher the ratio, the greater the risk.

[0052] In other embodiments, the label content can also combine the frequency domain dominant mode energy ratio with the actual surface polishing results of the corresponding segment processing area, such as the measured value of surface roughness, or categories such as "qualified," "over-polished," and "with vibration marks" determined by visual / tactile monitoring (which can be quantified into numerical values, such as qualified = 0, with vibration marks = 1, over-polished = 0.5). The quantified value is used as the surface quality evaluation index Q corresponding to the segment. Furthermore, the final label is:

[0053] The penalty coefficient is obtained by mapping the quality evaluation index Q (e.g., when Q is acceptable). =0, indicating the presence of vibrational ripples. =0.3), The preset weights for the impact on quality are set between 0 and 1. This formula means that even if a segment has little measured vibration, if it leads to poor quality results, its "risk label" will be increased; conversely, a segment with slightly larger vibration but good results may have a relatively lower label value. This makes the model learn not only the vibration itself, but also the "effective risk" of vibration causing quality deterioration.

[0054] A2. Using the dynamic fingerprint features of all segments after labeling as the training dataset, train the pre-selected neural network structure, continuously optimize the model parameters, and obtain the final process vibration sensitivity prediction model.

[0055] For example, the pre-selected neural network structure can be set as a multilayer perceptron (MLP). The model takes fixed-dimensional fragment dynamic fingerprint features as input, passes through several hidden layers containing non-linear activation functions, and finally outputs the predicted value through a linear activation function in the output layer. Using mean squared error (MSE) as the loss function, a backpropagation algorithm (such as the Adam optimizer) is used to optimize the model parameters on the training set until the model converges and achieves satisfactory prediction accuracy on independent validation sets. The trained model encapsulates a complex nonlinear mapping relationship from path segment features to their vibration excitation risk.

[0056] To ensure the model's generalization ability, a large amount of experimental data can be used to drive the model. Its internal parameters are obtained through training on a large amount of grinding experimental data covering different working conditions. During the experiment, the actual vibration signals during the segment's execution need to be recorded synchronously and processed as training labels. This can be a machine learning model (such as a neural network or support vector machine) trained on a large amount of historical experimental data. Its core function is to establish a mapping relationship from the static features of the segment to its dynamic behavior (vibration excitation), providing a quantitative cost basis for subsequent time-series optimization.

[0057] The scalar prediction value output by the model is the vibration process sensitivity index predicted for that segment. Its numerical range is related to the training labels. Consistency indicates the level of system resonance risk that is expected to be triggered when executing this segment. The higher the value, the more "dangerous" the segment is in terms of vibration.

[0058] S302, Optimization Problem: Define the execution sequence (i.e., execution order) of a path segment as follows: ,in It is the index of the m-th executed segment in the sequence within the entire set, aiming to find an optimal path segment execution sequence. To minimize a cost function used to evaluate the quality of sequences:

[0059] in, For the corresponding execution sequence Cost value, Let be the vibration process sensitivity index of the m-th executed segment in the sequence. To minimize the cumulative vibration cost item, this item reduces the total vibration excitation throughout the entire processing process, thus avoiding the continuous accumulation of vibration energy. For peak vibration cost, minimizing this term aims to prevent the occurrence of a single extremely high-risk segment in the sequence, thereby avoiding extreme conditions that could trigger severe vibration or resonance. The total idle distance traveled is the term representing the distance traveled in the sequence. The following is the total distance the tool needs to move between different segments. Optimizing this is aimed at reducing the invalid movement distance when the tool jumps between different segments, thus ensuring processing efficiency. , , The non-negative weight coefficients for each cost item, summing to 1, are used to balance the trade-off between vibration suppression and processing efficiency. They can be set according to specific process requirements (the specific application scenario determines the relative importance of cumulative vibration, peak vibration, and backlash distance in the overall optimization objective, set by process experts or determined through Pareto front analysis of multi-objective optimization, and their specific values ​​are set according to the different emphasis requirements of the specific process on efficiency, surface quality, and accuracy).

[0060] For example, the total empty travel distance item It can be set to: ,in, For fragments The endpoint coordinates, For fragments The starting coordinates.

[0061] It should be noted that, , , All results are normalized and used to unify dimensions; this invention will not elaborate on this.

[0062] S303, during the optimization process, process constraints are defined to minimize the cost function value. The optimization cost function is solved, and the optimal path segment execution sequence is output. .

[0063] The process constraints are set as follows: Full coverage constraint: the entire set Each path segment must be in the sequence It appears only once; Intra-layer continuity constraint: For layers belonging to the same attitude-stable layer Multiple path segments should be prioritized for sequential execution in a sequence to minimize unnecessary tool pose switching. For example, this constraint can be implemented through the cost function. This can be achieved by adding a penalty term to sequences that violate this continuity, or by directly imposing restrictions in the encoding / mutation operations of the genetic algorithm to ensure that sequences meet basic process constraints. This invention will not elaborate on or limit these restrictions. For example, segments within the same pose-stable layer should be executed as continuously as possible to avoid frequent, long-distance pose jumps. When constraints are violated, the value of this term increases (this option is only a supplement; it is used within the function). The term also implicitly implies the goal of minimizing movement distance, indicating fewer posture jumps.

[0064] For example, solving the optimization cost function can be configured as follows: a metaheuristic algorithm (such as a genetic algorithm or simulated annealing algorithm) is used to solve the optimization cost function. The algorithm uses multiple randomly generated initial sequences that satisfy basic process constraints (such as all segments must be visited once) as a population. During the iteration process, new candidate sequences are generated through operations such as selection, crossover, and mutation, and the cost function is always used. The fitness of each sequence is evaluated. The solution process eventually converges to an optimal or near-optimal path segment execution sequence. The sequence Mathematically, this makes the total cost Minimization, physically speaking, means planning a processing schedule with smoother vibration excitation, lower peak risk, and improved efficiency. (Sequence) Significant suppression of predicted vibration excitation was achieved while satisfying process constraints (to ensure that the sequence meets basic process constraints, such as segments within the same attitude-stabilized layer should be executed as continuously as possible to avoid frequent, long-distance attitude jumps).

[0065] Therefore, the pre-defined global timing optimization mechanism uses the predicted vibration process sensitivity index... Based on this core principle, the execution order of all path segments is rearranged to generate an optimal processing sequence with smoother vibration excitation. The aim is to manage the dynamic stability of the processing process from a time perspective, proactively suppressing harmful vibrations by optimizing the execution order of path segments.

[0066] S105 generates the final grinding operation instruction based on the optimal path segment execution sequence and controls the grinding equipment to execute it.

[0067] Specifically, the grinding operation instruction is generated as follows: Execute sequence according to optimal path segment Process each path segment in a predetermined order. For each segment, generate one or more specific numerical control (NC) instructions, which may include: Motion command: Drive the tool to move precisely from the current point to the starting point of the segment, and then move along the segment path at a constant or variable speed to the end point. The coordinate values ​​are the actual execution path point coordinates after wear compensation. Posture command: Controls the tool to adjust to the representative posture corresponding to this segment. ; Process parameter command: Sets the feed rate for this segment during execution. Spindle speed wait; Auxiliary function commands: such as turning the coolant on / off.

[0068] In other examples, step S105 further includes: sending the generated complete NC program to the controller of the grinding equipment (industrial robot or CNC machine tool), the controller parses and executes the program, and drives the equipment to complete the grinding operation of the entire workpiece; during the execution process, the system can continue to monitor the tool wear W(t) in real time, and can recalculate the compensation amount δ(t) for subsequent unexecuted path segments according to a preset strategy (such as after machining a sub-region), so as to achieve true online adaptation; at the same time, the vibration sensor can monitor the actual vibration level for subsequent optimization model updates and learning.

[0069] Therefore, by simultaneously collecting and fusing the three-dimensional geometric information and surface material information of the workpiece, digital perception of the workpiece's physical properties is achieved. Based on adaptive region division according to material distribution, the system can identify "hidden" functional zones on the workpiece surface caused by materials or processes, thus laying the foundation for subsequent differentiated process parameter settings. The effect of this level is that it overcomes the limitation of traditional pure geometric planning in "ignoring" material characteristics, providing data support for "material-specific grinding" from the source, and effectively alleviating the problem of over-grinding or under-grinding caused by uneven material distribution. A real-time monitoring and dynamic compensation mechanism for tool wear is introduced. By establishing a correlation model between wear amount and path correction amount, the execution path can be automatically fine-tuned according to the current state of the tool, thereby offsetting the processing error caused by tool size wear. The effect of this level is that it gives the grinding path the ability to self-adjust over time, so that the processing effect can remain relatively stable within the tool life cycle, improving the quality consistency of batch processing or long-term continuous operation. For the first time, vibration timing optimization is deeply integrated into the path planning process. By quantitatively evaluating the excitation potential of different path segments on system vibration, and based on this, the execution sequence of all path segments is globally optimized. This allows for the proactive planning of a processing time sequence with smoother vibration excitation and lower energy accumulation. The effect of this level is that, from the perspective of dynamic process management, it significantly reduces the negative impact of harmful vibration on trajectory tracking accuracy and surface processing quality. It not only solves the problem that reasonable local paths may lead to vibration deterioration due to the global execution sequence, but also achieves forward-looking management and optimization of the dynamic characteristics of the processing process through data-driven methods, thereby obtaining better processing stability and surface quality overall.

[0070] In summary, by simultaneously collecting geometric and hardness data, a "physical digital twin" of the workpiece is constructed, giving it not only shape but also the "intrinsic properties" of the material. Based on this, it can intelligently identify natural machining zones formed on the workpiece surface due to material differences and plan a preliminary scheme for each zone that balances efficiency (stable posture layers) and geometric accuracy (theoretical path). Furthermore, a dynamic dual-loop compensation mechanism is introduced: the inner loop is for real-time geometric compensation for tool wear, ensuring that the cutting depth does not decrease over time; the outer loop is based on data-driven vibration sensitivity prediction and global timing optimization, which actively "smooths out" dynamic load fluctuations during the machining process by intelligently scheduling the execution order of each path segment, suppressing vibration accumulation from the time dimension.

[0071] Ultimately, the output of this embodiment is no longer a static tool path, but an executable set of operation instructions that can self-adjust according to the actual state of the workpiece (material, geometry) and the real-time status of the machining system (tool wear), and has been pre-optimized for vibration risk in the time series.

[0072] Example 2 Figure 2 This is a schematic diagram of a grinding path planning system according to an embodiment of the present invention.

[0073] like Figure 2 As shown, a path refining planning system includes: an acquisition module, a partitioning module, and an optimization module; The acquisition module is used to acquire the three-dimensional geometric shape data and surface physical property data of the target workpiece, and generate a physical feature fusion dataset through spatiotemporal registration; The partitioning module is used to divide the surface of the target workpiece to be polished into multiple processing sub-regions based on the physical feature fusion dataset, and to partition at least one attitude stabilizing layer for each processing sub-region; using a preset path planning mechanism, an initial polishing path is generated for each attitude stabilizing layer, and the initial polishing path is composed of multiple path segments to form an initial path segment set. The optimization module is used to generate an optimal path segment execution sequence based on each path segment in the initial path segment set using a preset global timing optimization mechanism; generate the final grinding operation instruction based on the optimal path segment execution sequence; and control the grinding equipment to execute it.

[0074] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0075] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0076] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0077] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0078] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0079] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0080] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for planning a polishing path, characterized in that, include: S101, acquire the three-dimensional geometric shape data and surface physical property data of the target workpiece, and generate a physical feature fusion dataset through spatiotemporal registration; S102, based on the physical feature fusion dataset, divide the surface of the target workpiece to be polished into multiple processing sub-regions, and divide at least one attitude-stabilizing sheet into each processing sub-region; S103, using a preset path planning mechanism, an initial polishing path is generated for each attitude-stabilized layer. The initial polishing path consists of multiple path segments, forming an initial path segment set. S104: Based on each path segment in the initial path segment set, the optimal path segment execution sequence is generated using a preset global timing optimization mechanism; S105 generates the final grinding operation instruction based on the optimal path segment execution sequence and controls the grinding equipment to execute it.

2. The method for planning the polishing path according to claim 1, characterized in that, In S101, the physical feature fusion dataset is represented as follows: , Let i be the coordinates of the i-th point. Let N be the hardness value at the i-th point, and N be the number of point clouds.

3. The method for planning the polishing path according to claim 2, characterized in that, The process of dividing the workpiece surface to be polished into multiple processing sub-regions specifically includes: Based on the hardness and coordinate information in the physical feature fusion dataset D, the surface to be polished is divided into M spatially connected processing sub-regions with uniform material, and the average hardness value of each processing sub-region is calculated. .

4. The method for planning the polishing path according to claim 3, characterized in that, The step of dividing each processing sub-region into at least one attitude-stabilizing layer specifically includes: For each processing sub-region Extract the set of unit normal vectors corresponding to its surface point set, and divide it into groups using a preset clustering algorithm. There are 3 normal vector clusters, each cluster defining an attitude-stabilized sheet and its representative grinding attitude.

5. The method for planning the polishing path according to claim 4, characterized in that, The preset path planning mechanism specifically includes: S201, based on each attitude stabilization layer, plans a theoretical path of tool contact points that continuously covers the layer. The theoretical path is defined by a series of ordered theoretical path points. S202, based on the current tool wear and the average hardness value of the current machining sub-region, the current radial compensation amount is obtained using a preset dynamic geometric compensation model; S203, For each theoretical path point in the theoretical path, apply a radial compensation amount along its corresponding surface normal direction to obtain the actual execution path point; S204, define a straight-line motion segment between adjacent actual execution path points as a path segment, and combine all path segments of all layers to form the initial path segment set. J represents the total number of segments.

6. The method for planning the polishing path according to claim 5, characterized in that, Each path segment, in addition to being bound to start and end point coordinates, is also associated with its corresponding processing sub-region, attitude stabilization layer, representative attitude, and based on... Recommended process parameters.

7. The method for planning the polishing path according to claim 1, characterized in that, The preset global timing optimization mechanism specifically includes: S301: Based on each path segment, extract the dynamic fingerprint features of the segment, input them into the pre-trained process vibration sensitivity prediction model, and output the vibration process sensitivity index, a scalar that characterizes the risk of vibration excitation. S302, Optimization Problem: Define the execution sequence of the path segment as follows ,in It is the index of the m-th executed segment in the sequence within the entire set, aiming to find an optimal path segment execution sequence. To minimize a cost function used to evaluate the quality of sequences. ; S303, during the optimization process, process constraints are defined to minimize the cost function value. The optimization cost function is solved, and the optimal path segment execution sequence is output. .

8. The method for planning the polishing path according to claim 7, characterized in that, The fragment dynamic fingerprint features include: geometric motion features, process load features, system interaction features, and context features.

9. The method for planning a polishing path according to claim 8, characterized in that, The pre-trained process vibration sensitivity prediction model is obtained as follows: A1. Collect a large number of historical polishing data records in the same current scenario. Based on each path segment of each historical polishing data record, generate the corresponding segment dynamic fingerprint feature and label it. The label content is set to the vibration process sensitivity index that the segment actually corresponds to. A2. Using the dynamic fingerprint features of all segments after labeling as the training dataset, train the pre-selected neural network structure, continuously optimize the model parameters, and obtain the final process vibration sensitivity prediction model.

10. A system for planning a polishing path, characterized in that, The system, which uses the polishing path planning method as described in any one of claims 1-9, comprises: an acquisition module, a partitioning module, and an optimization module; The acquisition module is used to acquire the three-dimensional geometric shape data and surface physical property data of the target workpiece, and generate a physical feature fusion dataset through spatiotemporal registration; The partitioning module is used to divide the surface of the target workpiece to be polished into multiple processing sub-regions based on the physical feature fusion dataset, and to partition at least one attitude stabilizing layer for each processing sub-region; using a preset path planning mechanism, an initial polishing path is generated for each attitude stabilizing layer, and the initial polishing path is composed of multiple path segments to form an initial path segment set. The optimization module is used to generate an optimal path segment execution sequence based on each path segment in the initial path segment set using a preset global timing optimization mechanism; generate the final grinding operation instruction based on the optimal path segment execution sequence; and control the grinding equipment to execute it.

Images