Adaptive Machining Method and System for Industrial Robots Based on Digital Twin

Abstract

This invention relates to the field of adaptive machining technology for industrial robots, and provides an adaptive machining method and system for industrial robots based on digital twins. It includes constructing a digital twin model synchronized with the robotic arm, comprising a reference point cloud, a virtual tool, a material removal influence field model, and a pre-planned trajectory, solving the problem of disconnect between virtual and actual machining. A depth camera is used to acquire workpiece point clouds online, which are then registered with the reference point cloud to create a real-time point cloud, enabling real-time capture of machining dynamics and avoiding shape deviations caused by clamping errors and material fluctuations. The virtual tool is driven to move along the trajectory, and the influence field model is invoked to calculate and accumulate displacement vectors to generate a predicted point cloud. This predicted point cloud is compared with the reference point cloud to obtain a deviation field point cloud. Region clustering is used to identify deviation regions and associate them with trajectory segments. Compensation vectors and speed adjustment coefficients are calculated and sent to the controller to achieve adaptive compensation, dynamically adjusting machining parameters, improving the machining accuracy of complex workpieces, reducing defects, and lowering rework costs.

Description

Technical Field

[0001] This invention relates to the field of adaptive machining technology for industrial robots, and more specifically, to an adaptive machining method and system for industrial robots based on digital twins. Background Technology

[0002] The content in this section only provides background information related to this invention and may not constitute prior art.

[0003] In modern manufacturing, industrial robots, with their flexible operating capabilities, have been widely used in the processing of various workpieces, becoming one of the core equipment for improving processing efficiency and promoting the upgrading of production automation. Especially in the processing of complex-shaped workpieces, the multi-degree-of-freedom motion characteristics of industrial robots give them advantages that traditional processing equipment cannot match, enabling them to complete the processing tasks of various complex curved surfaces and irregular structures. Currently, industrial robot processing mainly relies on pre-set fixed processing parameters and motion trajectories. Through pre-planned tool paths and feed rates, the physical actuators are driven to complete the processing operations. Some technical solutions introduce virtual simulation methods to assist in the pre-planning of processing trajectories, thereby reducing actual trial and error costs.

[0004] However, existing industrial robot machining technologies still suffer from numerous insurmountable shortcomings, hindering their application in high-precision and highly adaptable machining scenarios. In current technologies, virtual simulation models often exhibit significant disconnect from the actual machining process. The virtual environment cannot capture and reflect various dynamic changes during actual machining in real time, leading to deviations between the virtually planned machining trajectory and actual machining requirements. During actual machining, factors such as minute deviations during workpiece clamping, fluctuations in material properties, tool wear during machining, and workpiece deformation caused by machining heat can all cause deviations between the actual workpiece surface morphology and the preset design model. Existing machining methods struggle to detect these real-time deviations online and cannot dynamically adjust the machining trajectory and parameters in a timely manner. Furthermore, even if some technologies can collect relevant information about the workpiece surface, they mostly make simple corrections to deviations at single locations. They cannot accurately identify areas where deviations are concentrated, and it is even more difficult to correlate deviation areas with corresponding machining trajectories. They cannot formulate personalized compensation strategies for different deviation areas, which leads to machining accuracy that cannot meet the machining requirements of complex workpieces. This easily results in machining defects, affecting product quality and machining efficiency. It also increases the cost of subsequent rework, limiting the further promotion and application of industrial robots in high-precision adaptive machining scenarios. Summary of the Invention

[0005] The purpose of this invention is to provide an adaptive machining method and system for industrial robots based on digital twins, in order to improve the aforementioned problems. To achieve the above objective, the technical solution adopted by this invention is as follows:

[0006] In a first aspect, this application provides an adaptive machining method for industrial robots based on digital twins, comprising:

[0007] Construct a digital twin model synchronized with the physical manipulator on the industrial robot; the digital twin model includes a reference point cloud generated by the target design model, a virtual tool model, a material removal influence field model, and a pre-planned machining trajectory; the material removal influence field model defines the spatial displacement influence function of the virtual tool model on the surrounding material points under different motion postures;

[0008] During the processing, point cloud data of the workpiece surface is acquired online using a depth camera; the point cloud data is registered and aligned with the reference point cloud in the digital twin model, and the registered point cloud data is loaded into the digital twin model as the real-time workpiece point cloud;

[0009] The virtual tool model is driven to move along a pre-planned machining trajectory. For each discrete position on the machining trajectory, based on the geometric parameters and motion posture of the virtual tool model, the material removal influence field model is called to calculate the expected displacement vector of the virtual tool model to each point in the real-time workpiece point cloud at that position. The expected displacement vectors of all points in the real-time workpiece point cloud along the entire machining trajectory are accumulated, and the spatial position of each point in the real-time workpiece point cloud is superimposed and corrected to generate a predicted machining point cloud.

[0010] The predicted processing point cloud is compared point by point with the corresponding reference point cloud to calculate and generate a deviation field point cloud containing spatial position deviation.

[0011] Perform regional clustering analysis on the deviation field point cloud to identify spatial clustering regions where the deviation exceeds the preset range; for each spatial clustering region, associate it with the corresponding machining segment on the pre-planned machining trajectory; calculate the tool path compensation vector and feed rate adjustment coefficient of the machining segment based on the average direction and magnitude of the deviation within the spatial clustering region.

[0012] The toolpath compensation vector and feed rate adjustment coefficient are sent to the controller of the physical robot to adjust the actual movement of the physical robot in the machining segment.

[0013] Furthermore, the construction of the influence field model for material removal specifically includes:

[0014] Based on the geometry and dimensions of the virtual tool model, a spatial influence domain is defined around the tool's motion envelope. Within the influence domain, a displacement weight function related to the tool's center distance and relative angle is preset for each spatial point, so that the closer the point is to the tool's cutting edge and the more direct the angle, the larger the magnitude of the expected displacement vector is assigned, and the direction is consistent with the tool's motion trend at that point.

[0015] Furthermore, the range of the spatial influence domain is set based on the maximum cutting radius of the virtual tool model.

[0016] Furthermore, the point cloud data is registered and aligned with the reference point cloud in the digital twin model, specifically including:

[0017] From the point cloud data and the reference point cloud, extract the point set of the same fixed structural surface outside the processing target area as the registration reference;

[0018] Based on the registration datum, the point cloud data is coarsely registered to the global coordinate system of the reference point cloud by calculating the rotation and translation transformation matrices.

[0019] Based on coarse registration, the iterative nearest point algorithm is used to perform fine registration between the point cloud data and the reference point cloud. Through iterative optimization, the average distance between the points in the point cloud data and the corresponding points in the reference point cloud is minimized until the preset registration accuracy threshold is reached.

[0020] Furthermore, the expected displacement vector of the virtual tool model at this location relative to each point in the real-time workpiece point cloud is calculated, specifically including:

[0021] The real-time workpiece point cloud is traversed using preset methods to obtain the expected displacement vectors of all points; the preset methods include:

[0022] For each point in the real-time workpiece point cloud, based on its spatial coordinates in the digital twin model, determine whether it is located within the spatial influence domain of the material removal influence field model corresponding to the current pose of the virtual tool model; if not, determine that the expected displacement vector of the point is zero; if so, based on the distance and direction of the point relative to the cutting edge of the virtual tool model, call the displacement influence function corresponding to the point in the material removal influence field model to calculate and obtain the expected displacement vector of the point.

[0023] Furthermore, traversing the real-time workpiece point cloud also includes:

[0024] Prioritize traversing the point cloud data that is closest to the current position of the virtual tool model and has not yet been traversed.

[0025] Furthermore, the spatial position of each point in the real-time workpiece point cloud is superimposed and corrected, specifically including:

[0026] Based on the accumulated expected displacement vector of each point, the original spatial coordinates of each point in the real-time workpiece point cloud are superimposed with the total expected displacement vector corresponding to that point. The superposition result is used as the updated spatial position of that point to generate the final coordinates of that point in the predicted processing point cloud.

[0027] Furthermore, before sending the toolpath compensation vector and feed rate adjustment coefficient to the controller of the physical robot, the following steps are also included:

[0028] The adjusted machining trajectory segment is virtually simulated in the digital twin model. After verifying that there is no risk of collision, the final compensation command is sent to the controller of the physical robot.

[0029] Secondly, this application also provides an adaptive machining system for industrial robots based on digital twins, comprising:

[0030] The twin model construction module is used to build a digital twin model that is synchronized with the physical manipulator on the industrial robot. The digital twin model includes a reference point cloud generated by the target design model, a virtual tool model, a material removal influence field model, and a pre-planned machining trajectory. The material removal influence field model defines the spatial displacement influence function of the virtual tool model on the surrounding material points under different motion postures.

[0031] The point cloud registration and loading module is used to collect point cloud data of the workpiece surface online through a depth camera during the processing; register and align the point cloud data with the reference point cloud in the digital twin model; and load the registered point cloud data into the digital twin model as the real-time workpiece point cloud.

[0032] The predictive point cloud generation module is used to drive the virtual tool model to move along the pre-planned machining trajectory. For each discrete position on the machining trajectory, based on the geometric parameters and motion posture of the virtual tool model, the material removal influence field model is called to calculate the expected displacement vector of the virtual tool model to each point in the real-time workpiece point cloud at that position. The expected displacement vectors of all points in the real-time workpiece point cloud along the entire machining trajectory are accumulated, and the spatial position of each point in the real-time workpiece point cloud is superimposed and corrected to generate the predictive machining point cloud.

[0033] The deviation field point cloud generation module is used to compare the predicted processing point cloud with the corresponding reference point cloud point by point, and calculate and generate a deviation field point cloud containing spatial position deviation.

[0034] The compensation parameter calculation module is used to perform regional clustering analysis on the deviation field point cloud and identify spatial clustering regions where the deviation exceeds the preset range; for each spatial clustering region, it is associated with the corresponding machining segment on the pre-planned machining trajectory; based on the average direction and magnitude of the deviation within the spatial clustering region, the tool path compensation vector and feed rate adjustment coefficient of the machining segment are calculated.

[0035] The motion parameter adjustment module is used to send the toolpath compensation vector and feed rate adjustment coefficient to the controller of the physical robot, so as to adjust the actual motion of the physical robot in the machining segment.

[0036] Thirdly, this application also provides an electronic device, including:

[0037] Memory, used to store computer programs;

[0038] A processor is used to implement the method steps as described in the first aspect when executing a computer program.

[0039] The beneficial effects of this invention are as follows:

[0040] This invention overcomes the disconnect between virtual simulation and actual machining in existing technologies by constructing a digital twin model synchronized with the physical manipulator of an industrial robot. This model includes a reference point cloud generated from a target design model, a virtual tool model, a material removal influence field model, and a pre-planned machining trajectory. The virtual environment synchronously reflects the actual machining state, preventing deviations between virtual planning and actual requirements. During machining, a depth camera is used to collect point cloud data of the workpiece surface online. This data is then registered and aligned with the reference point cloud in the digital twin model and loaded into the model as the real-time workpiece point cloud. This enables real-time capture of dynamic changes during actual machining, effectively avoiding the problem of workpiece surface morphology deviating from the preset model due to factors such as workpiece clamping deviations and material property fluctuations, which are then imperceptible. Subsequently, the virtual tool model is driven to move along the pre-planned machining trajectory. At each discrete position on the trajectory, the virtual tool's position is combined with several... The parameters and motion posture are used to call the material removal influence field model to calculate the expected displacement vector of the virtual tool to each point in the real-time workpiece point cloud and accumulate and correct it to generate a predicted machining point cloud. This ensures that it can accurately reflect the expected shape of the workpiece after machining and provides a reliable basis for deviation identification. Then, the predicted machining point cloud is compared with the reference point cloud point by point to generate a deviation field point cloud. Through regional clustering analysis, spatial clustering areas with deviations exceeding the preset range are identified and associated with corresponding machining trajectory segments. This breaks the limitation of existing technologies that can only make simple corrections to a single position. Then, based on the average direction and magnitude of the deviation in the region, the tool path compensation vector and feed rate adjustment coefficient of the machining segment are calculated and sent to the physical robot controller to adjust the actual motion. This realizes personalized compensation for different deviation areas, timely and dynamically adjusts the machining trajectory and parameters, effectively improves the machining accuracy of complex workpieces, reduces machining defects, and ensures product quality and machining efficiency. Attached Figure Description

[0041] Figure 1 A flowchart illustrating an adaptive machining method for industrial robots based on digital twins, provided by this invention;

[0042] Figure 2 A schematic diagram of an adaptive machining system for industrial robots based on digital twins provided by the present invention;

[0043] Figure 3 This is a schematic diagram of the structure of an electronic device provided by the present invention.

[0044] In the diagram: 201, Twin Model Construction Module; 202, Point Cloud Registration Loading Module; 203, Predicted Point Cloud Generation Module; 204, Deviation Field Point Cloud Generation Module; 205, Compensation Parameter Calculation Module; 206, Motion Parameter Adjustment Module. Detailed Implementation

[0045] 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. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0046] like Figure 1 As shown in the embodiment of the present invention, an adaptive machining method for industrial robots based on digital twins includes:

[0047] S101, construct a digital twin model synchronized with the physical manipulator on the industrial robot; the digital twin model includes a reference point cloud generated by the target design model, a virtual tool model, a material removal influence field model, and a pre-planned machining trajectory; the material removal influence field model defines the spatial displacement influence function of the virtual tool model on the surrounding material points under different motion postures.

[0048] Specifically, the digital twin model of this invention is not a single virtual mapping, but integrates four core components: a reference point cloud, a virtual tool model, a material removal influence field model, and a pre-planned machining trajectory. These components work together to form a complete virtual machining simulation system. The reference point cloud is generated by the target design model and is essentially a point cloud representation of the ideal shape of the target workpiece. It serves as a benchmark for subsequent machining deviation comparison. Its principle is to accurately quantify the design dimensions and geometric shape of the workpiece through point cloud data, ensuring that the virtual simulation is consistent with the actual machining target. This provides a clear reference for the identification of subsequent real-time machining deviations and avoids the lack of a basis for deviation judgment. A virtual tool model is a proportional virtual replica of a physical machining tool, accurately matching the geometry, size, and cutting characteristics of the physical tool. The principle is to construct a virtual entity based on the actual parameters of the physical tool, achieving a virtual restoration of tool movement and cutting behavior. It can simulate the machining process in a virtual environment, completing trajectory verification and cutting prediction without occupying physical tools and workpieces. Pre-planned machining trajectories are tool movement paths preset based on the target workpiece's machining requirements, providing an initial basis for virtual tool movement. The principle is to combine workpiece machining technology and tool movement characteristics to plan a reasonable path, providing a basic trajectory for subsequent machining adjustments and reducing the cost of trajectory trial and error in physical machining.

[0049] The material removal influence field model is the core of this digital twin model. Its core principle is based on the mechanical properties and spatial interaction laws of material removal during tool cutting. It defines the spatial displacement influence function of the virtual tool on surrounding material points under different motion postures, achieving accurate virtual simulation of the material removal process. This model predicts the morphological changes of the workpiece surface after tool movement, providing a core calculation basis for subsequent machining deviation prediction, and allowing for prediction of machining effects without actual cutting. Its specific construction process is as follows:

[0050] First, based on the geometry and dimensions of the virtual tool model, a spatial influence domain is defined around the tool's motion envelope. The principle is that during the cutting process, the tool only affects the material within a certain range around its trajectory. This spatial influence domain must completely cover the material area that the tool might touch, ensuring that all potentially removable material points are included in the calculation, avoiding omissions of material points within the tool's effective range, and guaranteeing the completeness of the material removal prediction. Then, within this influence domain, a displacement weight function is preset for each spatial point, correlated with the distance to the tool center and the relative angle. Its design principle stems from the cutting action of the tool on the material in actual machining—the closer to the tool's cutting edge, the stronger the cutting force and the greater the material displacement; conversely, the more direct the relative angle, the higher the transmission efficiency of the cutting force, and the greater the corresponding material displacement. Therefore, the displacement weight function must be set so that points under the above two conditions obtain a larger expected displacement vector, and the displacement direction is consistent with the tool's movement trend at that point. This ensures that the predicted material displacement in the virtual environment highly matches the actual material removal situation in physical machining, improving prediction accuracy. The specific formula for the displacement weight function is as follows:

[0051] Defined at discrete locations The coordinates of the tool center point are The unit vector in the direction of the tool spindle is ; For any point in the real-time workpiece point cloud; point To the tool center axis (through) , direction is The vertical distance is The position vector of a point relative to the center of the tool. Perpendicular to The projection on the plane is From this we can know relative to the cutting direction of the tool The included angle Satisfy the following formula:

[0052] (1)

[0053] That is when and Consistent direction ( The impact should be greatest when (=0).

[0054] At the same time, it is necessary to satisfy the condition that the closer the distance and the more direct the angle, the greater the weight, thus yielding the following formula:

[0055] (2)

[0056] In the formula, The maximum expected displacement coefficient; indicating the optimal position ( =0, When =0), the maximum possible displacement amplitude of a material point under a single action. , These are the corresponding Gaussian attenuation coefficients, controlling the attenuation rate affected by distance and angle.

[0057] The spatial influence domain is set based on the maximum cutting radius of the virtual tool model. The principle is that the maximum cutting radius of the virtual tool determines the maximum spatial range that the tool can reach in a single cut. Setting the influence domain range based on this ensures that the influence domain can fully cover the effective area of ​​the tool cutting, while avoiding the inclusion of too many irrelevant spatial points, thus ensuring the completeness of the prediction and reducing the amount of invalid calculations.

[0058] S102, during the processing, point cloud data of the surface of the workpiece to be tested is collected online by a depth camera; the point cloud data is registered and aligned with the reference point cloud in the digital twin model, and the registered point cloud data is loaded into the digital twin model as the real-time workpiece point cloud;

[0059] Specifically, by utilizing the non-contact, high-precision, and real-time acquisition characteristics of depth cameras, the surface of the workpiece under test is continuously scanned during the processing, capturing the spatial coordinate information of the workpiece surface and generating point cloud data. This point cloud data is essentially a digital representation of the actual surface morphology of the workpiece under test, which can accurately reflect the real-time geometric state of the workpiece during processing. Non-contact acquisition can avoid secondary damage to the workpiece surface, ensuring the quality of workpiece processing. At the same time, the real-time acquisition mode can dynamically capture the morphological changes of the workpiece surface during processing, avoiding the information lag problem caused by offline acquisition, and providing timely data input for subsequent real-time deviation analysis.

[0060] The core principle of registering and aligning point cloud data with a reference point cloud in a digital twin model is to eliminate spatial positional deviations (including translation and rotation deviations) between the actual acquired point cloud and the reference point cloud using a specific algorithm. This ensures that the two types of point clouds are in the same global coordinate system, guaranteeing accurate comparison between the actual workpiece point cloud and the reference point cloud, which serves as an ideal benchmark. This registration and alignment process must strictly follow a preset procedure, which includes three key steps:

[0061] First, point sets from the same fixed structural surface outside the processing target area are extracted from both the point cloud data and the reference point cloud as registration references. The principle behind this step is that the processing target area undergoes morphological changes during processing, while the fixed structural surface outside it remains stable and unaffected by processing. Selecting this point set as the reference ensures the stability and consistency of the registration reference, avoiding registration deviations caused by fluctuations in the reference point set. Second, based on the selected registration reference, rotation and translation transformation matrices are calculated to coarsely register the point cloud data to the global coordinate system of the reference point cloud. This is achieved by using rotation and translation transformation matrices to eliminate the overall spatial offset between the point cloud data and the reference point cloud, realizing initial alignment of the two types of point clouds, quickly reducing the spatial deviation between them, decreasing the computational load of subsequent fine registration, improving registration efficiency, and preventing convergence difficulties in fine registration due to excessive initial deviation. Finally, based on the coarse registration, the iterative nearest point algorithm is used to perform fine registration between the point cloud data and the reference point cloud. Through continuous iterative optimization, the average distance between each point in the point cloud data and the corresponding point in the reference point cloud is minimized until the preset registration accuracy threshold is reached, ensuring that the registration accuracy meets the needs of subsequent processing deviation analysis and providing a guarantee for the accurate identification of processing deviations.

[0062] The registered point cloud data is loaded into the digital twin model as a real-time workpiece point cloud. The principle is to fuse the accurately registered actual workpiece point cloud data with the core components of the digital twin model, such as the reference point cloud and the virtual tool model. This allows the digital twin model to reflect the actual state of the workpiece during the physical machining process in real time, achieving real-time synchronization between virtual simulation and physical machining. The loading of the real-time workpiece point cloud makes the digital twin model no longer a virtual mapping detached from reality, but a real-time simulation carrier that can dynamically match the physical machining state. This provides real workpiece state data for subsequent virtual tool motion simulation, material removal effect prediction, and machining path compensation, ensuring the pertinence and accuracy of adaptive machining adjustments.

[0063] S103 drives the virtual tool model to move along the pre-planned machining trajectory. For each discrete position on the machining trajectory, based on the geometric parameters and motion posture of the virtual tool model, the material removal influence field model is called to calculate the expected displacement vector of the virtual tool model to each point in the real-time workpiece point cloud at that position. The expected displacement vectors of all points in the real-time workpiece point cloud along the entire machining trajectory are accumulated, and the spatial position of each point in the real-time workpiece point cloud is superimposed and corrected to generate the predicted machining point cloud.

[0064] Specifically, the virtual tool model moves along the pre-planned machining trajectory. This is based on the real-time synchronization between the digital twin model and the physical machining process. The pre-planned machining trajectory is used as the initial reference path for the virtual tool model's movement. Through motion simulation in the virtual environment, the complete motion process of the physical tool in actual machining is simulated. Without activating the physical robot and physical tool, the execution process of the tool machining trajectory can be reproduced in the virtual space, effectively avoiding losses such as tool wear and workpiece scrap that may be caused by trajectory trial and error in physical machining.

[0065] For each discrete position on the machining trajectory, based on the geometric parameters and motion posture of the virtual tool model, the material removal influence field model is invoked to calculate the expected displacement vector of the virtual tool model relative to each point in the real-time workpiece point cloud at that position. The principle is that the pre-planned machining trajectory is essentially a continuous tool motion path. However, in actual calculations, it is impossible to calculate each of the infinite number of positions on the continuous trajectory individually. Therefore, the continuous trajectory needs to be discretized into several evenly spaced discrete positions. Each discrete position can be considered as an instantaneous dwell position during the tool's motion. The virtual tool model will exhibit different motion postures (such as cutting angle, tool axis direction, etc.) at different discrete positions, and its geometric parameters (such as tool diameter, cutting edge length, etc.) are fixed. These parameters directly determine the cutting range and intensity of the tool's action on the surrounding material. The material removal influence field model, as the core of the digital twin model, predefines the spatial displacement influence function of the virtual tool on the surrounding material points under different motion postures. Therefore, at each discrete position, only the geometric parameters of the virtual tool and the current motion posture need to be input to call the model. Through the preset influence function, the expected displacement vector of the virtual tool on each material point in the real-time workpiece point cloud at that instantaneous position is calculated. This expected displacement vector is essentially a prediction of the cutting action of the tool on the material point at that position, resulting in a spatial positional shift of the material point. Its direction is consistent with the cutting direction of the tool, and its magnitude is related to the distance from the material point to the cutting edge of the tool.

[0066] The specific calculation process is as follows: The real-time workpiece point cloud is traversed using a preset method to obtain the expected displacement vectors of all points. This preset method includes: for each point in the real-time workpiece point cloud, based on its spatial coordinates in the digital twin model, determining whether it is located within the spatial influence domain of the material removal influence field model corresponding to the current pose of the virtual tool model; if not, the expected displacement vector of that point is determined to be zero; if so, based on the distance and direction of that point relative to the cutting edge of the virtual tool model, the displacement influence function corresponding to that point in the material removal influence field model is called to calculate and obtain the expected displacement vector of that point. The principle is that by judging and filtering material points affected by tool cutting through spatial coordinates, indiscriminate calculations are avoided for all point cloud data, ensuring calculation accuracy while reducing invalid calculations and improving the calculation efficiency of the expected displacement vector. Traversing the real-time workpiece point cloud also includes: prioritizing the traversal of the point cloud data closest to the current position of the virtual tool model that has not yet been traversed. The principle behind this priority traversal method is that point cloud data near the current position of the virtual tool is more likely to be within the spatial influence domain. Priority traversal can quickly identify material points affected by cutting, reduce invalid point cloud traversal far from the tool position, improve the calculation efficiency of the expected displacement vector, and shorten the time consumed by virtual simulation. For example, the continuous straight line trajectory from (0,0,5) to (100,0,5) is discretized at 1mm intervals to obtain 101 discrete positions. At one of the discrete positions (50,0,5), the cutting angle of the virtual tool model is 90° (i.e., the tool axis is perpendicular to the workpiece surface). At this time, the material removal influence field model is called, and the real-time workpiece point cloud is traversed according to a preset method: the untraversed point cloud data near (50,0,5) is traversed first. For each traversed point, it is determined whether it is located in the spatial influence domain (a cylindrical area with a radius of 2mm centered on the tool center) corresponding to the discrete position based on its spatial coordinates. If the coordinates of a point are (50.5,0,5.5), which is located in the spatial influence domain, the expected displacement vector is calculated by calling the displacement influence function based on its distance (0.5mm) from the tool cutting edge and its direction (slanted upward). If the coordinates of a point are (55,0,6), which is far from the spatial influence domain, its expected displacement vector is determined to be zero. In this way, the expected displacement vector calculation of all points is completed efficiently.

[0067] The formula for calculating the expected displacement vector is:

[0068] (3)

[0069] Finally, the expected displacement vectors of all points in the real-time workpiece point cloud along the entire machining trajectory are accumulated, and the spatial position of each point in the real-time workpiece point cloud is superimposed and corrected to generate the predicted machining point cloud. The principle is that during the movement of the virtual tool model along the entire machining trajectory, each discrete position will have an instantaneous cutting effect on the material points in the real-time workpiece point cloud. That is, each material point will be affected by the expected displacement vectors corresponding to multiple discrete positions, rather than the effect of a single discrete position. Therefore, it is necessary to accumulate the expected displacement vectors of each material point at all discrete positions to obtain the total expected displacement vector of that material point along the entire machining trajectory. The calculation formula is as follows:

[0070] (4)

[0071] In the formula, The total number of bits after discretizing the pre-planned processing trajectory; For discrete pose indexes =1, 2, ... .

[0072] Real-time workpiece point cloud is a digital representation of the actual surface morphology of the workpiece during machining. Each point has a clear original spatial coordinate. By superimposing the original spatial coordinates of each point with the corresponding total expected displacement vector, the spatial position of that point can be corrected. The corrected spatial coordinates are the final position of the material point after the entire virtual cutting process. All the corrected point cloud data are collected together to generate the predicted machining point cloud. The calculation formula is as follows:

[0073] (5)

[0074] In the formula, To predict the first point cloud in the processing point cloud Three-dimensional coordinate vectors of points; For the first point cloud of the real-time workpiece The original three-dimensional coordinate vectors of each point; all The resulting set is the predictive processing point cloud: .

[0075] The predicted machining point cloud is essentially a point cloud that predicts the final surface morphology of the workpiece after machining along the entire pre-planned trajectory. It provides a core basis for subsequent comparison with the reference point cloud and calculation of machining deviations. Continuing the above example, the original spatial coordinates of a certain material point in the real-time workpiece point cloud are (50,0,6). During the cutting process at all 101 discrete positions, this material point will be affected by the expected displacement vector at 80 of these discrete positions. After accumulating these 80 expected displacement vectors, the total expected displacement vector is (0,0,-1.2). The original coordinates (50,0,6) are superimposed and corrected with this total expected displacement vector to obtain the corrected coordinates (50,0,4.8). This coordinate is the predicted position of the material point after machining along the entire straight cutting trajectory. The point cloud formed after correcting all material points is the predicted machining point cloud of the flat workpiece after this straight cutting.

[0076] S104, compare the predicted processing point cloud with the corresponding reference point cloud point by point, and calculate and generate a deviation field point cloud containing spatial position deviation.

[0077] Specifically, based on the spatial coordinate correspondence after point cloud registration, the spatial position difference is calculated point-by-point between corresponding points (i.e., corresponding points) in the predicted processing point cloud and the reference point cloud. The coordinate difference of each corresponding point is the processing deviation at that location. All deviation points are integrated according to their spatial positions to form a deviation field point cloud containing the spatial position deviation information of each point. Essentially, this visualizes and quantifies the processing deviation in the form of a point cloud. The specific calculation formula is as follows:

[0078] Assuming that registration has been established With reference point cloud The corresponding point relationships between them (i.e.) and (These are the coordinates of the same surface location in the predicted and ideal states). The machining deviation vector is:

[0079] (6)

[0080] In the formula, For the first The processing deviation vector at each comparison point visually represents the offset (direction and magnitude) of the predicted surface relative to the ideal surface at that point. Reference point cloud and The corresponding number The three-dimensional coordinate vectors of each point (ideal position); all The resulting set is the bias field point cloud. .

[0081] S105, Perform regional clustering analysis on the deviation field point cloud to identify spatial clustering regions where the deviation exceeds the preset range; For each spatial clustering region, associate it with the corresponding machining segment on the pre-planned machining trajectory; Calculate the tool path compensation vector and feed rate adjustment coefficient of the machining segment based on the average direction and magnitude of the deviation within the spatial clustering region.

[0082] Specifically, first, a region clustering analysis is performed on the bias field point cloud (using Euclidean distance clustering pairs). Points in Clustering based on their spatial coordinates yields Spatially continuous clustered regions This method identifies spatial clusters where deviations exceed a preset range. Its core purpose is to avoid trajectory fluctuations and processing instability caused by direct adjustments based on individual deviation points. The deviation field point cloud contains spatial position deviation information for each corresponding point, essentially representing a digital and visual representation of processing deviations. The core principle of region clustering analysis is based on the correlation between point cloud spatial coordinates and the consistency of deviations. Discrete deviation points are aggregated into several spatial clusters according to the principle of spatial proximity and consistent deviation trends, achieving regional classification of deviations. This facilitates accurate location of concentrated deviation areas and provides a basis for subsequent targeted adjustments.

[0083] The identification of spatial clusters with deviations exceeding a preset range is based on a preset deviation threshold. This preset range is a deviation threshold pre-set based on the workpiece machining accuracy requirements. It combines workpiece design tolerances and machining process standards to determine a reasonable threshold and clearly define the acceptable machining deviation boundary. If the average or maximum deviation within a spatial cluster exceeds this threshold, it indicates that the actual machining of that area will not meet the accuracy requirements, necessitating trajectory and speed adjustments. The beneficial effect is the accurate selection of deviation areas requiring adjustment, avoiding resource waste and trajectory disorder caused by indiscriminate adjustments. The calculation formula is as follows:

[0084] (7)

[0085] In the formula, For each cluster region The average deviation; For the region The number of deviation points included; This is a preset deviation threshold. If... Greater than Then the region Adjustments are needed.

[0086] After identifying the spatial clustering regions that need adjustment, for each spatial clustering region, the corresponding processing segment on the pre-planned processing trajectory is associated. The principle is that there is a one-to-one correspondence between the spatial clustering regions of the deviation field point cloud and the pre-planned processing trajectory. The deviation of the clustering region is essentially caused by the virtual tool movement deviation within the corresponding processing segment. Through spatial coordinate mapping, it can be accurately associated with the specific processing segment on the pre-planned trajectory, clarifying the trajectory range corresponding to the source of the deviation, realizing the precise binding between the deviation and the processing trajectory, and ensuring that subsequent adjustments are only for the processing segment corresponding to the deviation.

[0087] After associating with the corresponding machining segment, the toolpath compensation vector and feed rate adjustment coefficient for the machining segment are calculated based on the average direction and magnitude of the deviation within the spatial clustering region. The principle is that the average direction of the deviation reflects the overall trend of toolpath offset, and the average magnitude reflects the degree of offset. The toolpath compensation vector must be opposite in direction and match the magnitude of the average deviation to correct the tool trajectory and offset the influence of the deviation. The feed rate adjustment coefficient is set according to the magnitude of the deviation. When the deviation is large, the feed rate is appropriately reduced to ensure smooth cutting and eliminate the deviation. When the deviation is small, the feed rate is maintained or increased, balancing accuracy and efficiency. This achieves adaptive adjustment of the machining trajectory and feed rate, accurately offsetting deviations and improving machining accuracy and stability. The specific calculation formula is as follows:

[0088] First calculate the regional average deviation vector :

[0089] (8)

[0090] In the formula, It characterizes the overall deviation trend and magnitude of the region.

[0091] The formula for calculating the toolpath compensation vector is:

[0092] (9)

[0093] In the formula, This is the toolpath compensation vector to be sent to the robot controller. Its direction is opposite to the average deviation, used to compensate for the deviation; This is the path compensation factor (usually set to 1 to indicate full compensation; it can also be set to a value between 0 and 1 to indicate partial compensation).

[0094] The formula for calculating the feed rate adjustment coefficient is:

[0095] (10)

[0096] In the formula, This is the feed rate adjustment coefficient; Adjust the sensitivity coefficient for speed.

[0097] In addition, the spatial position of each point in the real-time workpiece point cloud is superimposed and corrected, specifically including:

[0098] Based on the accumulated expected displacement vector of each point, the original spatial coordinates of each point in the real-time workpiece point cloud are superimposed with the total expected displacement vector corresponding to that point. The superposition result is used as the updated spatial position of that point to generate the final coordinates of that point in the predicted machining point cloud. The principle is to accurately restore the actual spatial position of the material point after the virtual tool cuts by vector superposition, providing accurate predicted machining point cloud data support for the subsequent generation of deviation field point cloud, and ensuring the accuracy of deviation calculation.

[0099] S106 sends the toolpath compensation vector and feed rate adjustment coefficient to the controller of the physical robot, adjusting the actual motion of the physical robot in the machining segment.

[0100] Specifically, the toolpath compensation vector and feed rate adjustment coefficient are adaptive adjustment parameters precisely calculated based on deviation clustering regions. As the core of motion control for the robot's controller, these parameters are received and converted into motion control signals for the robot, driving it to adjust its motion trajectory and feed rate in the corresponding machining segment. This ensures the workpiece's machining state matches the optimal adjustment scheme in the virtual simulation. Before sending the toolpath compensation vector and feed rate adjustment coefficient to the physical robot's controller, the adjusted machining trajectory segment is virtually simulated in a digital twin model. After verifying the absence of collision risks, the final compensation command is sent to the physical robot's controller. Using the virtual simulation to adjust the trajectory segment, the spatial relationship between the tool, workpiece, and robot itself can be simulated, identifying potential collision hazards and preventing unverified adjustment commands from being directly issued, which could damage the equipment. This proactively avoids collision risks, ensures the safety of the physical equipment and workpiece, and reduces equipment maintenance and workpiece scrap costs.

[0101] like Figure 2 As shown, based on the same inventive concept, this embodiment provides an industrial robot adaptive processing system based on digital twins, including:

[0102] The twin model construction module 201 is used to construct a digital twin model that is synchronized with the physical manipulator on the industrial robot. The digital twin model includes a reference point cloud generated by the target design model, a virtual tool model, a material removal influence field model, and a pre-planned machining trajectory. The material removal influence field model defines the spatial displacement influence function of the virtual tool model on the surrounding material points under different motion postures.

[0103] The point cloud registration and loading module 202 is used to collect point cloud data of the surface of the workpiece to be tested online through a depth camera during the processing; register and align the point cloud data with the reference point cloud in the digital twin model; and load the registered point cloud data into the digital twin model as the real-time workpiece point cloud.

[0104] The predictive point cloud generation module 203 is used to drive the virtual tool model to move along the pre-planned machining trajectory. For each discrete position on the machining trajectory, based on the geometric parameters and motion posture of the virtual tool model, the material removal influence field model is called to calculate the expected displacement vector of the virtual tool model to each point in the real-time workpiece point cloud at that position. The expected displacement vectors of all points in the real-time workpiece point cloud along the entire machining trajectory are accumulated, and the spatial position of each point in the real-time workpiece point cloud is superimposed and corrected to generate the predictive machining point cloud.

[0105] The deviation field point cloud generation module 204 is used to compare the predicted processing point cloud with the corresponding reference point cloud point by point, and calculate and generate a deviation field point cloud containing spatial position deviation.

[0106] The compensation parameter calculation module 205 is used to perform regional clustering analysis on the deviation field point cloud and identify spatial clustering regions where the deviation exceeds the preset range; for each spatial clustering region, it is associated with the corresponding machining segment on the pre-planned machining trajectory; based on the average direction and magnitude of the deviation within the spatial clustering region, the tool path compensation vector and feed rate adjustment coefficient of the machining segment are calculated.

[0107] The motion parameter adjustment module 206 is used to send the tool path compensation vector and feed rate adjustment coefficient to the controller of the physical robot to adjust the actual motion of the physical robot in the machining segment.

[0108] like Figure 3 As shown, based on the same inventive concept, this embodiment provides an electronic device, including:

[0109] Memory 302 is used to store computer programs;

[0110] Processor 301 is used to implement the method steps as described in the first aspect when executing a computer program.

[0111] It should be noted that the specific methods by which each module performs operations in the system described in the above embodiments have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0112] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An adaptive machining method for industrial robots based on digital twins, characterized in that, include: A digital twin model synchronized with the physical manipulator on an industrial robot is constructed. This digital twin model includes a reference point cloud generated from a target design model, a virtual tool model, a material removal influence field model, and a pre-planned machining trajectory. The material removal influence field model defines the spatial displacement influence function of the virtual tool model on surrounding material points under different motion postures. The construction of the material removal influence field model includes: defining a spatial influence domain around the tool's motion envelope based on the geometry and dimensions of the virtual tool model; within this influence domain, a displacement weight function associated with the tool's center distance and relative angle is preset for each spatial point, such that points closer to the tool's cutting edge and with a more direct angle are assigned a larger expected displacement vector magnitude, with the direction consistent with the tool's motion trend at that point. The displacement weight function is: In the formula, is defined at discrete locations. The coordinates of the tool center point are The unit vector in the direction of the tool spindle is ; For any point in the real-time workpiece point cloud; For point The perpendicular distance to the center axis of the tool; for The angle with the cutting direction of the tool, Let be the position vector of the point relative to the center of the tool. Perpendicular to The projection on the plane; For displacement weights; The maximum expected displacement coefficient; , These are the corresponding Gaussian decay coefficients; During the processing, point cloud data of the surface of the workpiece to be tested is collected online using a depth camera; the point cloud data is registered and aligned with the reference point cloud in the digital twin model, and the registered point cloud data is loaded into the digital twin model as the real-time workpiece point cloud; The virtual tool model is driven to move along the pre-planned machining trajectory. For each discrete position on the machining trajectory, based on the geometric parameters and motion posture of the virtual tool model, the material removal influence field model is called to calculate the expected displacement vector of the virtual tool model at that position relative to each point in the real-time workpiece point cloud. The expected displacement vectors of all points in the real-time workpiece point cloud along the entire machining trajectory are accumulated, and the spatial position of each point in the real-time workpiece point cloud is superimposed and corrected to generate a predicted machining point cloud. The predicted processing point cloud is compared point by point with the corresponding reference point cloud to calculate and generate a deviation field point cloud containing spatial position deviation. Regional clustering analysis is performed on the deviation field point cloud to identify spatial clustering regions where the deviation exceeds a preset range; for each spatial clustering region, the corresponding machining segment on the pre-planned machining trajectory is associated; based on the average direction and magnitude of the deviation within the spatial clustering region, the tool path compensation vector and feed rate adjustment coefficient of the machining segment are calculated; The toolpath compensation vector and feed rate adjustment coefficient are sent to the controller of the physical robot to adjust the actual movement of the physical robot in the machining segment.

2. The adaptive machining method for industrial robots based on digital twins according to claim 1, characterized in that, The range of the spatial influence domain is set according to the maximum cutting radius of the virtual tool model.

3. The adaptive machining method for industrial robots based on digital twins according to claim 1, characterized in that, The process of registering and aligning the point cloud data with the reference point cloud in the digital twin model specifically includes: From the point cloud data and the reference point cloud, extract the point set of the same fixed structural surface outside the processing target area as the registration reference; Based on the registration datum, the point cloud data is coarsely registered to the global coordinate system where the reference point cloud is located by calculating the rotation and translation transformation matrices. Based on coarse registration, the iterative nearest point algorithm is used to perform fine registration between the point cloud data and the reference point cloud. Through iterative optimization, the average distance between the points in the point cloud data and the corresponding points in the reference point cloud is minimized until a preset registration accuracy threshold is reached.

4. The adaptive machining method for industrial robots based on digital twins according to claim 1, characterized in that, The calculation of the expected displacement vector of the virtual tool model at this position relative to each point in the real-time workpiece point cloud specifically includes: The real-time workpiece point cloud is traversed using a preset method to obtain the expected displacement vectors of all points; the preset method includes: For each point in the real-time workpiece point cloud, based on its spatial coordinates in the digital twin model, it is determined whether it is located within the spatial influence domain of the material removal influence field model corresponding to the current pose of the virtual tool model; if not, the expected displacement vector of the point is determined to be zero; if so, based on the distance and direction of the point relative to the cutting edge of the virtual tool model, the displacement influence function corresponding to the point in the material removal influence field model is called to calculate and obtain the expected displacement vector of the point.

5. The adaptive machining method for industrial robots based on digital twins according to claim 4, characterized in that, The traversal of the real-time workpiece point cloud also includes: Prioritize traversing the point cloud data that is closest to the current position of the virtual tool model and has not yet been traversed.

6. The adaptive machining method for industrial robots based on digital twins according to claim 1, characterized in that, The process of superimposing and correcting the spatial position of each point in the real-time workpiece point cloud specifically includes: Based on the accumulated expected displacement vector of each point, the original spatial coordinates of each point in the real-time workpiece point cloud are superimposed with the total expected displacement vector corresponding to that point. The superposition result is used as the updated spatial position of that point to generate the final coordinates of that point in the predicted processing point cloud.

7. The adaptive machining method for industrial robots based on digital twins according to claim 1, characterized in that, Before sending the toolpath compensation vector and feed rate adjustment coefficient to the controller of the physical robot, the method further includes: The adjusted machining trajectory segment is virtually simulated in the digital twin model. After verifying that there is no risk of collision, the final compensation command is sent to the controller of the physical robot.

8. An adaptive machining system for industrial robots based on digital twins, used to implement the adaptive machining method for industrial robots based on digital twins as described in claim 1, characterized in that, include: The twin model construction module is used to construct a digital twin model synchronized with the physical manipulator on the industrial robot. The digital twin model includes a reference point cloud generated by the target design model, a virtual tool model, a material removal influence field model, and a pre-planned machining trajectory. The material removal influence field model defines the spatial displacement influence function of the virtual tool model on the surrounding material points under different motion postures. The point cloud registration and loading module is used to collect point cloud data of the surface of the workpiece to be tested online through a depth camera during the processing; register and align the point cloud data with the reference point cloud in the digital twin model; and load the registered point cloud data into the digital twin model as the real-time workpiece point cloud. The predictive point cloud generation module is used to drive the virtual tool model to move along the pre-planned machining trajectory. For each discrete position on the machining trajectory, based on the geometric parameters and motion posture of the virtual tool model, the material removal influence field model is called to calculate the expected displacement vector of the virtual tool model at that position relative to each point in the real-time workpiece point cloud. The expected displacement vectors of all points in the real-time workpiece point cloud along the entire machining trajectory are accumulated, and the spatial position of each point in the real-time workpiece point cloud is superimposed and corrected to generate the predictive machining point cloud. The deviation field point cloud generation module is used to compare the predicted processing point cloud with the corresponding reference point cloud point by point, and calculate and generate a deviation field point cloud containing spatial position deviation. The compensation parameter calculation module is used to perform regional clustering analysis on the deviation field point cloud and identify spatial clustering regions where the deviation exceeds a preset range; for each spatial clustering region, it is associated with the corresponding machining segment on the pre-planned machining trajectory; based on the average direction and magnitude of the deviation within the spatial clustering region, the tool path compensation vector and feed rate adjustment coefficient of the machining segment are calculated. The motion parameter adjustment module is used to send the tool path compensation vector and feed rate adjustment coefficient to the controller of the physical robot, so as to adjust the actual motion of the physical robot in executing the machining segment.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement, when executing the computer program, an adaptive machining method for an industrial robot based on digital twins as described in any one of claims 1 to 7.

Images