A robotic milling path error prediction method, system, and storage medium

Abstract

The application provides a robot milling path error prediction method, system and storage medium, and the method comprises the following steps: acquiring robot sensing information and workpiece sensing information, confirming robot feature information and workpiece feature information respectively according to the robot sensing information and the workpiece sensing information; performing dimension conversion on the robot feature information and the workpiece feature information, determining a robot feature matrix and a workpiece feature matrix, combining the robot feature matrix and the workpiece feature matrix to determine tool position feature at each tool position; arranging the tool position feature in sequence and constructing a sliding window of a preset size, selecting all tool position features in the sliding window through the sliding window to fuse and determine fused features; inputting the fused features into a preset hybrid neural network model to predict the contour error of the robot at different tool positions. The application solves the problem that there is no method for accurately predicting the robot milling path error with high precision in the prior art.

Description

Technical Field

[0001] This invention relates to the field of error prediction and compensation, and in particular to a method, system and storage medium for predicting errors in robot milling paths. Background Technology

[0002] In the field of milling, robotic milling offers advantages over multi-axis CNC machine tools for machining large, complex, and irregularly shaped parts, including lower cost, greater flexibility, increased intelligence, and a larger operating space. However, industrial robots suffer from low structural rigidity and trajectory accuracy, particularly the deformation of their end effectors under milling forces. Furthermore, insufficient dynamic performance and uneven workpiece allowance distribution lead to significant contour errors during machining, severely impacting machining quality and efficiency, and preventing the required precision for finishing. To ensure that robotic milling achieves the necessary machining quality and efficiency, error prediction and compensation for the robotic milling path are essential.

[0003] Currently, traditional error compensation methods are mostly based on static models or empirical rules, which are insufficient to handle complex nonlinear dynamic errors in machining. In international research, some advanced manufacturing companies and research institutions have attempted to introduce error prediction methods based on dynamic models or simple machine learning algorithms, but their ability to capture complex dynamic characteristics is insufficient, and the fusion processing of multi-feature information is still in the exploratory stage. Domestic research started later and mostly focuses on the analysis and local optimization of single features, lacking systematic and intelligent solutions. This results in the inability to achieve high-precision prediction and real-time compensation of contour errors during machining, making robotic milling unable to meet the demands of high-precision machining. Summary of the Invention

[0004] Based on this, the purpose of this invention is to provide a method, system, and storage medium for predicting robot milling path errors, aiming to solve the problem that there is no method in the prior art that can accurately predict robot milling path errors with high precision.

[0005] A method for predicting robot milling path errors according to an embodiment of the present invention includes:

[0006] Robot sensing information and workpiece sensing information are acquired through preset sensing devices, and robot feature information and workpiece feature information are confirmed respectively based on the robot sensing information and workpiece sensing information.

[0007] The robot feature information and the workpiece feature information are dimensionally transformed to determine the robot feature matrix and the workpiece feature matrix, and the robot feature matrix and the workpiece feature matrix are combined to determine the tool position feature at each tool position point;

[0008] The knife point features are arranged in order, and a sliding window of a preset size is constructed. All knife point features within the sliding window are selected and fused to determine the fused features.

[0009] The fused features are input into a preset hybrid neural network model to predict the contour error of the robot at different tool positions, thereby adjusting the milling path of the robot.

[0010] In addition, the robot milling path error prediction method according to the above embodiments of the present invention may also have the following additional technical features:

[0011] Furthermore, the step of confirming workpiece feature information based on the workpiece sensing information includes:

[0012] The point cloud information of the workpiece is determined based on the workpiece sensing information, and the sphere is divided with each tool point as the center and the diameter of the tool as the diameter, to determine the sphere region and the point cloud coordinates of each tool point.

[0013] The coordinates of the regional point cloud are converted to workpiece point cloud data in the tool coordinate system. The tool coordinate system corresponding to the tool position point is a coordinate system with the tool position point as the origin, the tool feed direction as the X-axis, and the tool axis direction as the Z-axis.

[0014] The workpiece feature information is determined by unifying the dimensions of the workpiece point cloud data.

[0015] Furthermore, the step of determining the workpiece feature information by unifying the dimensions of the workpiece point cloud data includes:

[0016] In the tool coordinate system, a grid with a preset grid length is divided on the three planes xy, yz and xz. The preset grid length is less than the minimum spacing of the scanned point cloud so that the number of projected points in each grid is no greater than one.

[0017] The grids in the xy, yz, and xz planes are treated as different matrices, each with a size of . The workpiece point cloud data is then projected onto the xy, yz, and xz planes respectively to determine the features of the three planes. , That is, the distribution matrix of the workpiece scan point cloud at the tool position point in the tool coordinate system. The radius of the cutting tool The preset grid length;

[0018] The characteristic expressions of the three planes are:

[0019]

[0020] Where is the minimum spacing of the scanned point cloud. and Represents the first in the projection plane and Each grid.

[0021] Furthermore, the robot feature information includes robot end-effector stiffness information, robot spatial pose information, and relative position information between the tool and the workpiece. The step of confirming the robot feature information based on the robot sensing information includes:

[0022] The robot's joint angle information is determined based on the robot's sensor information, and the joint angle information is the robot's spatial pose information;

[0023] The machining tool position information is determined based on the robot sensing information, and the machining tool position information is the relative position information between the tool and the workpiece;

[0024] The robot end effector stiffness matrix, i.e., the robot end effector stiffness information, is determined by a set of preset equations based on the robot's sensor information.

[0025] The preset set of equations is as follows:

[0026]

[0027] in, Indicates that the robot is under external force Stiffness deformation under action, The joint angle is The corresponding Jacobian matrix at time, For robot joint deformation, This is the corresponding robot joint stiffness matrix. Represents the transpose of the robot's Jacobian matrix. This is the stiffness matrix of the robot's end effector.

[0028] Further, the step of performing dimensional transformation on the robot feature information to determine the robot feature matrix includes:

[0029] Based on the robot's feature information, the robot's original feature expression is determined as follows: , For robot joint angles, For machining tool position information;

[0030] The robot feature kernel matrix dictionary is determined based on the robot's original feature expression and feature kernel function;

[0031] The expression for the feature kernel function is:

[0032]

[0033] in, and express The i and the j List, The mean is 0, and the covariance is ;

[0034] The expression for the robot feature kernel matrix dictionary is:

[0035] .

[0036] Further, the step of performing dimensional transformation on the workpiece feature information to determine the workpiece feature matrix includes:

[0037] Construct a Gaussian convolution kernel and normalize the Gaussian convolution kernel;

[0038] The workpiece convolution feature matrix is ​​obtained by convolving the normalized Gaussian convolution with the workpiece feature information, and the convolution feature matrix is ​​combined to form the workpiece feature matrix.

[0039] The expression for the Gaussian convolution kernel is:

[0040]

[0041] in, and Represent matrices respectively row and column indexes, and These are the mean and standard deviation in two directions of a two-dimensional Gaussian distribution, respectively.

[0042] The normalized Gaussian convolution kernel expression is as follows:

[0043]

[0044] in, ;

[0045] The expression for the workpiece convolution feature matrix is:

[0046]

[0047] in, row and column indexes, and These are the convolution kernels The row and column indices, and the corresponding fill length and move step are respectively and ;

[0048] The expression for the workpiece feature matrix is ​​as follows: .

[0049] Further, the step of inputting the fused features into a preset hybrid neural network model to predict the contour error of the robot at different tool positions includes:

[0050] Spatial and temporal features are extracted from the fused features, and the extracted spatial and temporal features are encoded to capture the long-term dependencies between the extracted features.

[0051] Regression prediction is performed in the input fully connected layer to obtain the machining error at each tool position point.

[0052] Another objective of this invention is to provide a robot milling path error prediction system, the system comprising:

[0053] The feature information determination module is used to acquire robot sensing information and workpiece sensing information through preset sensing devices, and to confirm robot feature information and workpiece feature information respectively based on the robot sensing information and workpiece sensing information.

[0054] The feature matrix determination module is used to perform dimensional transformation on the robot feature information and the workpiece feature information, determine the robot feature matrix and the workpiece feature matrix, and combine the robot feature matrix and the workpiece feature matrix to determine the tool position feature at each tool position point;

[0055] The fusion feature determination module is used to arrange the knife point features in sequence, construct a sliding window of a preset size, and select all the knife point features within the sliding window for fusion to determine the fusion feature;

[0056] The error prediction module is used to input the fused features into a preset hybrid neural network model to predict the contour error of the robot at different tool positions, and then adjust the milling path of the robot.

[0057] Another objective of this invention is to provide a storage medium storing a computer program that, when executed by a processor, implements the steps of the robot milling path error prediction method described above.

[0058] Another object of the present invention is to provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the above-described robot milling path error prediction method.

[0059] This invention extracts feature information from both the workpiece and the robot by collecting information from sensing devices. Then, through dimensional transformation, it determines robot and workpiece feature matrices with the same dimensions, allowing these two types of features to be combined to determine the tool position features at each tool position point. The robot's feature matrix and the workpiece's feature information each contain information about multiple angles. Real-time acquisition and fusion of this information enables accurate judgment of robot milling path errors in dynamic and complex scenarios. Furthermore, after determining the tool position features, a sliding window is constructed, fusing all features within the window. The fused features are then used as input to a preset model for error prediction. This is because in actual machining, the point cloud at a machining point is not only related to the features at the current point, but also to the machining features before and after the current tool position point, which in turn affect the machining error at that point. This method fuses multiple tool position features before and after the current tool position point with the current tool position feature, transforming the original three-dimensional feature data into four-dimensional structured data. Each four-dimensional structured data includes data features of the current tool position point and multiple tool position points before and after it, further improving the accuracy and precision of error prediction. Therefore, this invention solves the problem in the prior art of lacking a method that can accurately predict robot milling path errors with high precision. Attached Figure Description

[0060] Figure 1 This is a flowchart of the robot milling path error prediction method in the first embodiment of the present invention;

[0061] Figure 2 This is a structural block diagram of the robot milling path error prediction system in the second embodiment of the present invention;

[0062] Figure 3 This is a schematic diagram of the structure of the electronic device in the third embodiment of the present invention;

[0063] Figure 4 This is a schematic diagram illustrating the construction of the training set for the hybrid neural network model provided in this embodiment of the invention;

[0064] Figure 5 This is a schematic diagram of a hybrid neural network provided in an embodiment of the present invention;

[0065] The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation

[0066] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0067] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0068] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0069] Example 1

[0070] Please see Figure 1 The diagram shows a flowchart of the robot milling path error prediction method in the first embodiment of the present invention, which specifically includes steps S01-S04.

[0071] S01, acquire robot sensing information and workpiece sensing information through preset sensing devices, and confirm robot feature information and workpiece feature information respectively based on the robot sensing information and workpiece sensing information.

[0072] Specifically, the step of confirming workpiece feature information based on the workpiece sensing information includes: determining the workpiece point cloud information based on the workpiece sensing information, and dividing the sphere with each tool point as the center and the tool diameter as the diameter to determine the spherical region and the region point cloud coordinates corresponding to each tool point; converting the region point cloud coordinates to workpiece point cloud data in the tool coordinate system, wherein the tool coordinate system corresponding to the tool point is a coordinate system with the tool point as the origin, the tool feed direction as the X-axis, and the tool axis direction as the Z-axis.

[0073] Furthermore, the workpiece point cloud data is dimensionally unified to determine workpiece feature information. The step of dimensionally unifying the workpiece point cloud data to determine workpiece feature information includes: dividing the workpiece point cloud data into grids of a preset grid length on the xy, yz, and xz planes in the tool coordinate system. The preset grid length is less than the minimum spacing of the scanned point cloud, so that the number of projected points in each grid is no greater than one; the grids in the xy, yz, and xz planes are respectively regarded as different matrices, with a matrix size of... The workpiece point cloud data is then projected onto the xy, yz, and xz planes respectively to determine the features of the three planes. , That is, the distribution matrix of the workpiece scan point cloud at the tool position point in the tool coordinate system. The radius of the cutting tool The preset grid length;

[0074] The characteristic expressions of the three planes are:

[0075]

[0076] Where is the minimum spacing of the scanned point cloud. and Represents the first in the projection plane and Each grid.

[0077] In practical implementation, the point cloud of the workpiece is first calculated, at each tool position point. At the tool position point, draw a sphere based on the tool diameter and calculate the coordinates of all point clouds within the sphere. The point cloud coordinates within the sphere are considered to represent the meshing characteristics of the workpiece and tool at that tool position point. Transform all point clouds within the corresponding tool range at the tool position point to a tool coordinate system with the tool position point as the origin, the feed direction as the x-axis, and the tool axis direction as the z-axis. The corresponding tool radius is... Since the number of point clouds extracted at different tool positions may vary, further data processing is needed to transform the workpiece point cloud at each tool position into workpiece features of uniform dimension. First, a mesh is created in the xy, yz, and xz planes of the tool coordinate system, with a mesh length of [missing information]. The length of the grid division satisfies ,in The minimum spacing for scanning the point cloud, i.e., the resolution of the blue light scanner, is used to ensure that at most one projection point can appear in each divided grid. The grids in the xy, yz, and xz planes are treated as different matrices, with a matrix size of... The point cloud data transformed into the tool coordinate system is projected onto the xy, yz, and xz planes respectively. If the corresponding mesh in the xy, yz, and xz planes... If a projection point exists, the value of the corresponding xy, yz, xz plane matrix at that grid point is set to 1; otherwise, it is set to 0. and Represents the first in the projection plane and A grid. Features of three planes are obtained. , .

[0078] Specifically, the robot feature information includes robot end effector stiffness information, robot spatial pose information, and relative position information between the tool and the workpiece. The step of confirming the robot feature information based on the robot sensing information includes: determining the robot's joint angle information based on the robot sensing information, wherein the joint angle information is the robot spatial pose information; determining the machining tool position information based on the robot sensing information, wherein the machining tool position information is the relative position information between the tool and the workpiece; and determining the robot end effector stiffness matrix, i.e., the robot end effector stiffness information, based on the robot sensing information through a preset set of equations.

[0079] The preset set of equations is as follows:

[0080]

[0081] in, Indicates that the robot is under external force Stiffness deformation under action, The joint angle is The corresponding Jacobian matrix at time, For robot joint deformation, This is the corresponding robot joint stiffness matrix. Represents the transpose of the robot's Jacobian matrix. This is the stiffness matrix of the robot's end effector.

[0082] In practical implementation, the main causes of machining errors in robot-machined curved surfaces include poor robot motion accuracy and stiffness deformation during machining. When batch machining complex curved surface parts, the machining path and parameters, including tool speed and feed rate, are usually the same for each part. Therefore, theoretically, machining errors can be compensated for using trial cutting. However, for some cast housing parts, the remaining cutting allowance for each part may differ, resulting in different surface contour errors after machining, which may even exceed the tolerance requirements of the drawing. When the machining parameters are the same, the surface contour error of the machined part is only related to the robot's posture at the tool position and the undeformed chip thickness of the part at that tool position. Therefore, the robot's posture at each tool position can be used as the robot's machining feature, and the undeformed chip thickness of the part with uneven allowance at that tool position can be used as the part's machining feature. Finally, by fusing the robot's and part's machining features, a neural network model can be used to predict the machining contour error of the curved surface part with uneven allowance.

[0083] Robots have low stiffness, making them prone to stiffness deformation when subjected to external forces at their end effector, leading to machining errors. The end effector stiffness represents the robot's ability to resist deformation under external forces, and it is only related to the robot's posture and joint stiffness. Typically, a robot joint is considered as an elastic torsional spring, and the stiffness deformation of the end effector is the cumulative result of the stiffness deformation of each joint. Under external forces, the robot undergoes stiffness deformation. According to the principle of virtual work, the total virtual work generated by the end effector due to external forces is equal to the total virtual work of the equivalent force applied to the joints. Therefore, this equality can be used to determine the stiffness matrix of the robot's end effector.

[0084] S02, perform dimensional transformation on the robot feature information and the workpiece feature information to determine the robot feature matrix and the workpiece feature matrix, and combine the robot feature matrix and the workpiece feature matrix to determine the tool position feature at each tool position point.

[0085] Specifically, to combine the robot's features with the workpiece's features, the robot's features need to be structured. The robot's stiffness, pose, and the relative position of the tool and workpiece exhibit non-linear relationships. Constructing a Gaussian kernel function maps the robot's sample features from the original space to a higher-dimensional space. This kernel function mapping allows the robot's sample features to be adjusted into image features of arbitrary dimensions, mitigating the linear inseparability of feature information in low-dimensional space and preserving the structured advantages of current feature fusion. For a given tool position point, the robot's sample features are represented in the original space as follows: Introducing Gaussian eigenkernel function Expand sample features to A square matrix of dimensions, using express.

[0086] Feature kernel function It can be represented as:

[0087]

[0088] in and Represents the original space The i and the j List, The mean is 0, and the covariance is ,feature The new vector constructed by combining it with other features can be represented as:

[0089]

[0090] The different features of the robot follow a Gaussian distribution, and these features are arranged in an ordered manner to obtain a Gaussian feature kernel matrix dictionary. The upgraded feature matrix dictionary effectively preserves the information of robot stiffness, joint angles, and tool position points, and avoids overlap between feature signals. It is expressed as follows:

[0091]

[0092] Errors in robotic milling are related not only to the robot's posture during machining but also to the workpiece's machining characteristics, such as feed rate, spindle speed, and machining allowance. Even with consistent machining parameters, inconsistencies in machining allowances due to casting variations affect the consistency of machining accuracy across different parts. In more severe cases, excessive machining allowances can lead to out-of-tolerance machining errors, resulting in part scrap. Therefore, it is necessary to construct the allowance feature information of casting parts with uneven allowances from the point cloud data and combine this information with the robot's pose information to achieve accurate error prediction in robotic milling of workpieces with uneven allowances.

[0093] Specifically, since the planar feature matrix of the workpiece obtained in step S01 is very sparse, directly inputting it into the neural network model may cause excessive computational overhead and low prediction accuracy. At the same time, the size of the obtained workpiece feature matrix is ​​much larger than the robot's feature matrix. Therefore, it is necessary to perform dimensionality reduction on the obtained workpiece feature matrix so that the robot's features and the workpiece feature matrix can match each other and increase the model's prediction accuracy.

[0094] To ensure consistency between workpiece features and robot features, a two-dimensional Gaussian convolution kernel of size k*k is constructed. Represented as:

[0095]

[0096] in and Represent matrices respectively row and column indexes, and These represent the mean and standard deviation in two directions of a two-dimensional Gaussian distribution, respectively. To ensure that the workpiece features are not introduced with additional gain or attenuation after convolution, the Gaussian convolution kernel needs to be normalized.

[0097]

[0098] in Suppose we use Gaussian convolution kernels to process the features of the three planes respectively. , After convolution, the workpiece features can be represented as:

[0099]

[0100] in row and column indexes, and These are the convolution kernels The row and column indices, and the corresponding fill length and move step are respectively and The size N of the matrix after convolution can be expressed as:

[0101]

[0102] The three matrices resulting from the convolution are then combined to represent the machining features of the workpiece. It contains three dimensionality-reduced planar feature matrices, with corresponding dimensions of . The robot's features can be combined with these features to form the predictive features of the entire system at that tool position. , as input features of the neural network model.

[0103] S03, arrange the knife point features in order, construct a sliding window of a preset size, and select all the knife point features within the sliding window for fusion to determine the fused features.

[0104] Specifically, such as Figure 4As shown, the constructed robot and workpiece point cloud machining features can effectively reflect the error-related features at the machining point. However, in actual machining, the point cloud at the machining point is not only related to the features at the current point, but the machining features before and after the current machining tool position also affect the machining error at the current point. To solve this problem, the constructed features at the tool position are first... Arrange the points according to their order, then construct a sliding window with a size of [size missing]. This represents the 2 points before and after the current tool position, respectively. L The features of each tool position are fused with the features of the current tool position, where L represents the machining error of the current tool position affected by the machining toolpath before and after. L The influence of individual blade positions, through this method, will reduce the original... The three-dimensional feature data is transformed into The four-dimensional structured data includes the data features of the current tool position point and L tool positions before and after it. When the number of feature points before the initial tool position point and after the last tool position point is insufficient, zero padding of the same matrix size is added. In this case, the tool and workpiece are not in contact, and it will not affect the error at the tool position point.

[0105] S04, the fused features are input into a preset hybrid neural network model to predict the contour error of the robot at different tool positions, and then the milling path of the robot is adjusted.

[0106] Specifically, such as Figure 5 As shown, the proposed hybrid neural network model consists of a convolutional neural network (CNN layer) and a bidirectional long short-term memory network (BILSTM layer). This network first uses the CNN to extract spatial and temporal features from the input, then uses the BILSTM layer to further encode the extracted features and capture the long-term dependencies between them. These features are then input into a fully connected layer for regression prediction, yielding the machining error at each tool position caused by changes in robot posture and allowance. The proposed model not only extracts locally relevant features from the input data but also simultaneously captures forward and backward temporal dependencies, demonstrating excellent performance in processing multidimensional complex data.

[0107] Convolutional neural networks extract information from images using convolutional kernels of different sizes and multiple filters to extract various information from images. 3D convolutional kernels can slide in three directions, simultaneously extracting information features in the height, width, and depth of the image, and rearranging the extracted features in three-dimensional space to capture the spatial and temporal features of the input data. Pooling layers are used to downsample the output of the convolutional layers to reduce the data dimensionality and effectively prevent overfitting.

[0108] In robotic milling, the error at each tool position is influenced not only by the previous machining path but also by the uncut portion. Considering only the influence of a single time direction may not accurately predict machining errors. The BILSTM layer captures bidirectional dependencies in the data by combining forward and backward time series, utilizing both past and future information to make more accurate predictions. The BILSTM layer takes features extracted by the CNN layer as input and fuses contextual local features to generate hybrid features representing global information.

[0109] Ultimately, the constructed neural network model predicts the contour error of the robot at different tool positions, thereby improving the machining accuracy of the robot when milling irregular large and complex curved parts.

[0110] In summary, the robot milling path error prediction method in the above embodiments of the present invention extracts feature information of the workpiece and the robot respectively through information collected by sensing devices, and then determines the robot feature matrix and workpiece feature matrix with the same dimensions through dimensional transformation, so that the two types of features can be combined to determine the tool position features at each tool position point. The robot feature matrix and the workpiece feature information each contain information on multiple angle conditions. By collecting and fusing this information in real time, the robot milling path error status in dynamic and complex scenarios can be accurately determined. Furthermore, after determining the tool position features, a sliding window is constructed, and the sliding window... All features are fused, and the fused features are then used as input to a preset model for error prediction. This is because in actual machining, the point cloud at a machining point is not only related to the features at the current point, but also to the machining features before and after the current tool position. This method fuses the features of multiple tool positions before and after the current tool position with the current tool position feature, transforming the original three-dimensional feature data into four-dimensional structured data. Each four-dimensional structured data includes the data features of the current tool position and multiple tool positions before and after it, further improving the accuracy and precision of error prediction. Therefore, this invention solves the problem of the lack of a method in the prior art for high-precision and accurate prediction of robot milling path errors.

[0111] Example 2

[0112] Please see Figure 2 The diagram shows a structural block diagram of the robot milling path error prediction system proposed in the second embodiment of the present invention. The robot milling path error prediction system 200 includes: a feature information determination module 21, a feature matrix determination module 22, a fusion feature determination module 23, and an error prediction module 24, wherein:

[0113] The feature information determination module 21 is used to acquire robot sensing information and workpiece sensing information through a preset sensing device, and to confirm robot feature information and workpiece feature information respectively based on the robot sensing information and workpiece sensing information.

[0114] The feature matrix determination module 22 is used to perform dimensional transformation on the robot feature information and the workpiece feature information, determine the robot feature matrix and the workpiece feature matrix, and combine the robot feature matrix and the workpiece feature matrix to determine the tool position feature at each tool position point;

[0115] The fusion feature determination module 23 is used to arrange the knife point features in sequence, construct a sliding window of a preset size, and select all the knife point features within the sliding window for fusion to determine the fusion feature;

[0116] Error prediction module 24 is used to input the fused features into a preset hybrid neural network model to predict the contour error of the robot at different tool positions, and then adjust the milling path of the robot.

[0117] Furthermore, the robot feature information includes robot end-effector stiffness information, robot spatial pose information, and relative position information between the tool and the workpiece. The feature information determination module 21 includes:

[0118] The regional point cloud coordinate determination unit is used to determine the point cloud information of the workpiece based on the workpiece sensing information, and to divide the sphere with each of the tool points as the center and the diameter of the tool as the diameter, and to determine the spherical region and regional point cloud coordinates corresponding to each of the tool points.

[0119] The coordinate system transformation unit is used to transform the coordinates of the regional point cloud to the workpiece point cloud data under the tool coordinate system. The tool coordinate system corresponding to the tool position point is a coordinate system with the tool position point as the origin, the tool feed direction as the X-axis, and the tool axis direction as the Z-axis.

[0120] The dimension-unified unit is used to determine the workpiece feature information by unifying the dimensions of the workpiece point cloud data.

[0121] The joint angle information determination unit is used to determine the joint angle information of the robot based on the robot sensing information, wherein the joint angle information is the robot's spatial pose information.

[0122] The machining tool position information determination unit is used to determine the machining tool position information based on the robot sensing information, wherein the machining tool position information is the relative position information between the tool and the workpiece;

[0123] The robot end effector stiffness matrix determination unit is used to determine the robot end effector stiffness matrix, i.e., the robot end effector stiffness information, based on the robot sensing information through a preset set of equations.

[0124] The preset set of equations is as follows:

[0125]

[0126] in, Indicates that the robot is under external force Stiffness deformation under action, The joint angle is The corresponding Jacobian matrix at time, For robot joint deformation, This is the corresponding robot joint stiffness matrix. Represents the transpose of the robot's Jacobian matrix. This is the stiffness matrix of the robot's end effector.

[0127] Furthermore, the same unit in the same dimension includes:

[0128] The mesh division sub-unit is used to divide the tool coordinate system into meshes of a preset mesh length on the three planes xy, yz and xz. The preset mesh length is less than the minimum spacing of the scanned point cloud so that the number of projection points in each mesh after division is no greater than one.

[0129] Planar features define sub-elements, used to treat the meshes in the xy, yz, and xz planes as distinct matrices, with a matrix size of... The workpiece point cloud data is then projected onto the xy, yz, and xz planes respectively to determine the features of the three planes. , That is, the distribution matrix of the workpiece scan point cloud at the tool position point in the tool coordinate system. The radius of the cutting tool The preset grid length;

[0130] The characteristic expressions of the three planes are:

[0131]

[0132] Where is the minimum spacing of the scanned point cloud. and Represents the first in the projection plane and Each grid.

[0133] Furthermore, the feature matrix determination module 22 includes:

[0134] The robot original feature expression determination unit is used to determine the robot's original feature expression based on the robot feature information. , For robot joint angles, For machining tool position information;

[0135] The robot feature kernel matrix dictionary determination unit is used to determine the robot feature kernel matrix dictionary based on the robot's original feature expression and feature kernel function.

[0136] The expression for the feature kernel function is:

[0137]

[0138] in, and express The i and the j List, The mean is 0, and the covariance is ;

[0139] The expression for the robot feature kernel matrix dictionary is:

[0140] .

[0141] A normalization processing unit is used to construct Gaussian convolution kernels and normalize the Gaussian convolution kernels.

[0142] The workpiece feature matrix determination unit is used to obtain the workpiece convolution feature matrix by convolving the normalized Gaussian convolution with the workpiece feature information, and to combine the convolution feature matrices to form the workpiece feature matrix.

[0143] The expression for the Gaussian convolution kernel is:

[0144]

[0145] in, and Represent matrices respectively row and column indexes, and These are the mean and standard deviation in two directions of a two-dimensional Gaussian distribution, respectively.

[0146] The normalized Gaussian convolution kernel expression is as follows:

[0147]

[0148] in, ;

[0149] The expression for the workpiece convolution feature matrix is:

[0150]

[0151] in, row and column indexes, and These are the convolution kernels The row and column indices, and the corresponding fill length and move step are respectively and ;

[0152] The expression for the workpiece feature matrix is ​​as follows: .

[0153] Furthermore, the error prediction module 24 includes:

[0154] The feature dependency determination unit is used to extract spatial and temporal features from the fused features, and encode the extracted spatial and temporal features to capture the long-term dependencies between the extracted features.

[0155] An error prediction unit is used to perform regression prediction on the fully connected layer to obtain the machining error at each tool position point.

[0156] The functions or operation steps implemented by the above modules are largely the same as those in the above method embodiments, and will not be repeated here.

[0157] Example 3

[0158] In another aspect, the present invention also proposes an electronic device, please refer to [link to relevant documentation]. Figure 3 The diagram shows an electronic device according to the third embodiment of the present invention, including a memory 20, a processor 10, and a computer program 30 stored in the memory and executable on the processor. When the processor 10 executes the computer program 30, it implements the robot milling path error prediction method as described above.

[0159] In some embodiments, the processor 10 may be a central processing unit (CPU), controller, microcontroller, microprocessor or other data processing chip, used to run program code stored in memory 20 or process data, such as executing access restriction programs.

[0160] The memory 20 includes at least one type of readable storage medium, such as flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 20 can be an internal storage unit of an electronic device, such as the hard disk of the electronic device. In other embodiments, the memory 20 can also be an external storage device of the electronic device, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. Furthermore, the memory 20 can include both internal and external storage units of the electronic device. The memory 20 can be used not only to store application software and various types of data of the electronic device, but also to temporarily store data that has been output or will be output.

[0161] It should be pointed out that, Figure 3 The structure shown does not constitute a limitation on the electronic device. In other embodiments, the electronic device may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0162] This invention also proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the robot milling path error prediction method described above.

[0163] Those skilled in the art will understand that the logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can mean any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0164] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0165] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0166] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0167] The above embodiments merely illustrate several implementation methods of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this patent should be determined by the appended claims.

Claims

1. A robot milling path error prediction method, characterized by, The method comprises: obtaining robot sensing information and workpiece sensing information through a preset sensing device, and confirming robot feature information and workpiece feature information according to the robot sensing information and the workpiece sensing information respectively; dimensionally converting the robot feature information and the workpiece feature information, determining a robot feature matrix and a workpiece feature matrix, and combining the robot feature matrix and the workpiece feature matrix to determine tool position feature at each tool position; arranging the tool position feature in sequence, constructing a preset size sliding window, and selecting all the tool position features in the sliding window through the sliding window to fuse and determine a fusion feature; inputting the fusion feature into a preset hybrid neural network model to predict the contour error of the robot at different tool positions, and then adjusting the milling path of the robot; wherein the step of confirming the workpiece feature information according to the workpiece sensing information comprises: determining the point cloud information of the workpiece according to the workpiece sensing information, and dividing the sphere with each tool position as the center and the diameter of the tool as the diameter to determine the spherical region corresponding to each tool position and the region point cloud coordinates; transforming the region point cloud coordinates into workpiece point cloud data in the tool coordinate system, wherein the tool coordinate system corresponding to the tool position is a coordinate system with the tool position as the origin, the feed direction of the tool as the X axis, and the tool axis direction as the Z axis; dimensionally unifying the workpiece point cloud data to determine the workpiece feature information; wherein the robot feature information comprises robot end stiffness information, robot spatial pose information, and relative position information of the tool and the workpiece, and the step of confirming the robot feature information according to the robot sensing information comprises: determining the joint angle information of the robot according to the robot sensing information, which is the robot spatial pose information; determining the machining tool position information according to the robot sensing information, which is the relative position information of the tool and the workpiece; determining the robot end stiffness matrix, i.e. the robot end stiffness information, according to the robot sensing information through a preset equation set; the preset equation set is: wherein, represents the stiffness deformation of the robot under an external force , is the Jacobian matrix corresponding to the joint angles , is the joint deformation of the robot, is the corresponding Jacobian stiffness matrix of the robot, represents the transpose of the Jacobian matrix of the robot, is the end-effector stiffness matrix of the robot; wherein the step of dimensionally converting the robot feature information to determine the robot feature matrix comprises: According to the robot feature information, a robot original feature expression is determined as , is a robot joint angle, is machining tool position information; determining a robot feature kernel matrix dictionary according to the robot original feature expression and the feature kernel function; the expression of the feature kernel function is: wherein, and denote the first i and the second j column, the mean of the first ; and the expression of the robot feature kernel matrix dictionary is: wherein the step of inputting the fusion feature into a preset hybrid neural network model to predict the contour error of the robot at different tool positions comprises: extracting the spatial dimension feature and the time dimension feature in the fusion feature, and encoding the extracted spatial dimension feature and time dimension feature to capture the long-term dependency between the extracted features; inputting into a full connection layer for regression prediction to obtain the machining error at each tool position.

2. The robotic milling path error prediction method of claim 1, wherein, the step of dimensionally unifying the workpiece point cloud data to determine the workpiece feature information comprises: Grid division is performed on three planes of the tool coordinate system xy, yz and xz with a preset grid length, the preset grid length being smaller than the minimum interval of the point cloud, so that the number of projection points in each grid after division is not greater than one; The grids in the xy, yz, and xz planes are treated as different matrices, each with a size of . The workpiece point cloud data is then projected onto the xy, yz, and xz planes respectively to determine the features of the three planes. , That is, the distribution matrix of the workpiece scan point cloud at the tool position point in the tool coordinate system. The radius of the cutting tool The preset grid length; The characteristic expression of the three planes is: wherein dminis the minimum distance of the scanned point cloud, and represents the projection plane in the and grid.

3. The robotic milling path error prediction method of claim 2, wherein, The step of performing dimension conversion on the workpiece feature information includes: A Gaussian convolution kernel is constructed, and the Gaussian convolution kernel is normalized; The workpiece convolution feature matrix is obtained by convolution of the normalized Gaussian convolution and the workpiece feature information, and the convolution feature matrix is combined to form the workpiece feature matrix; The Gaussian convolution kernel expression is: where, and denote the row and column indices of the matrix and are the mean and standard deviation in the two directions of the two-dimensional Gaussian distribution, respectively.​ The normalized Gaussian convolution kernel expression is: wherein ; The workpiece convolution feature matrix expression is: wherein, the row and column indexes of the convolution kernel, and are the row and column indexes of the convolution kernel respectively, and the corresponding padding length and moving step are and respectively; The workpiece feature matrix expression is .

4. A robotic milling path error prediction system, characterized by, The system is used to implement the robot milling path error prediction method according to any one of claims 1 to 3. A feature information determination module is configured to acquire robot sensing information and workpiece sensing information through a preset sensing device, and to determine robot feature information and workpiece feature information according to the robot sensing information and the workpiece sensing information, respectively; A feature matrix determination module is configured to perform dimension conversion on the robot feature information and the workpiece feature information, to determine a robot feature matrix and a workpiece feature matrix, and to combine the robot feature matrix and the workpiece feature matrix to determine tool position features at each tool position. A fusion feature determination module is configured to arrange the tool position features in sequence, to construct a sliding window of a preset size, and to select all the tool position features in the sliding window through the sliding window to fuse the tool position features, thereby determining fusion features. An error prediction module is configured to input the fusion features into a preset hybrid neural network model to predict contour errors of the robot at different tool positions, and to adjust the milling path of the robot.

5. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the steps of the robot milling path error prediction method according to any one of claims 1 to 3.

6. An electronic device, comprising: The computer program is stored in the memory and executable on the processor, and the processor implements the robot milling path error prediction method according to any one of claims 1 to 3 when executing the program.

Images