Key error identification method for tool curved surface of five-axis numerical control grinding machine

Abstract

The application relates to the technical field of three-dimensional error modeling, in particular to a key error identification method for a tool curved surface of a five-axis numerical control grinding machine, and more particularly to a key error identification method for a tool curved surface of a five-axis numerical control grinding machine, which comprises the following steps: step 1: obtaining a three-dimensional structure model of the five-axis grinding machine, and constructing a three-dimensional pose error model of each working component in the five-axis grinding machine, wherein the three-dimensional pose error model is the pose error between the ideal position and the actual position of each working component; wherein the pose error comprises a plurality of position-independent geometric errors and a plurality of position-dependent geometric errors; the application breaks through the limitation of a single analysis method, significantly improves the robustness and reliability of key error identification, and provides a core theoretical basis for subsequent accurate error compensation and efficient improvement of the machining precision of the five-axis grinding machine.

Description

Technical Field

[0001] This application relates to the field of three-dimensional error modeling technology, and more specifically, to a method for identifying key errors in the curved surface of a five-axis CNC grinding machine tool. Background Technology

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

[0003] When machining complex tool surfaces (such as helical end mills and ball end mills) on a five-axis CNC grinding machine, the machining accuracy is significantly affected by the inherent geometric errors of the machine tool. These geometric errors mainly include positioning errors, straightness errors, and angular deviations (pitch, yaw, roll) on linear axes (X / Y / Z), as well as rotational errors, axis drift, and inter-axis perpendicularity errors on rotary axes (A / C axes). These errors originate from machine tool manufacturing and assembly defects, guideway wear, thermal deformation, and dynamic loads, causing the actual tool trajectory to deviate from the theoretical path. In five-axis simultaneous machining, multi-axis geometric errors are transmitted to the tool end via nonlinear coupling of the kinematic chain, ultimately resulting in significant contour deviations and shape distortions on the tool's working surfaces (such as the rake face, flank face, and cutting edge).

[0004] To improve machining accuracy, current mainstream methods compensate machine tools for errors by establishing a mapping model between geometric errors and machining errors (such as homogeneous coordinate transformation models based on multibody system theory, error sensitivity matrices, etc.). This model identifies key geometric error terms, predicts their impact on the tool tip (TCP) pose, and generates compensation commands in the CNC system. On structurally simple three-axis machine tools or rotationally symmetric workpieces (such as shaft parts), this method can effectively reduce machining errors.

[0005] The aforementioned compensation strategies based on relative geometric errors have significant limitations in machining complex curved surfaces. For tools with variable curvature and non-developable surface features (such as chip grooves in helical end mills, spherical surfaces in ball end mills, and blade facet / back surfaces in turbine blade end mills), the contact point between the tool and the workpiece dynamically changes with the spatial pose of the cutting edge. The transmission characteristics of geometric errors to different regions of the surface exhibit strong nonlinearity and spatial heterogeneity. Relying solely on machine tool geometric error parameters cannot accurately characterize the error distribution across the entire working surface of the tool, and it is particularly difficult to capture local distortions in key functional areas such as the microstructure of the cutting edge and the transition zone between the rake and flank faces. Summary of the Invention

[0006] The summary section of this application is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description section below. This summary section is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0007] Some embodiments of this application propose a method for identifying key errors in the curved surface of a five-axis CNC grinding machine tool to solve the technical problems mentioned in the background section above.

[0008] As a first aspect of this application, some embodiments of this application provide a method for identifying key errors in the curved surface of a five-axis CNC grinding machine tool, comprising the following steps:

[0009] Step 1: Obtain the 3D structural model of the five-axis grinding machine, and construct the 3D pose error model of each working part in the five-axis grinding machine. The 3D pose error model represents the pose error between the ideal position and the actual position of each working part.

[0010] The pose error includes multiple position-independent geometric errors and multiple position-dependent geometric errors;

[0011] Step 2: Based on the machine tool kinematic chain, calculate the ideal pose transformation relationship and the actual pose transformation relationship between the tool coordinate system and the grinding wheel coordinate system, and construct a pose error model between the tool coordinate system and the grinding wheel coordinate system.

[0012] Step 3: Establish a tool surface pose error model based on the grinding wheel's machining trajectory;

[0013] Step 4: Calculate the sensitivity of each pose error in the tool surface pose error model to sort the pose errors.

[0014] This invention constructs a machine tool component pose error model incorporating both position-independent and position-dependent geometric errors. Based on the kinematic chain, it accurately derives the actual pose error between the tool and grinding wheel coordinate systems. Combined with the grinding wheel machining trajectory, it establishes an error model directly mapped to the tool surface position points. Furthermore, by calculating and ranking the sensitivity of each geometric error term to the surface pose error, this method can accurately identify the key geometric error terms that have the most significant impact on the global contour deviation of complex tool surfaces (such as the chip groove of a helical end mill and the spherical surface of a ball end mill) and the local distortion of key functional areas (such as the cutting edge and the rake / flank transition zone). This effectively overcomes the limitations of traditional relative geometric error compensation strategies in machining variable curvature and non-developable surfaces due to the strong nonlinearity of error propagation and spatial heterogeneity.

[0015] Furthermore, in step 2: the motion chain of the five-axis grinding machine is: W-A-C-Y-F-X-Z-T. Where W represents the tool coordinate system, A represents the A rotary axis, C represents the C rotary axis, Y represents the Y linear axis, X represents the X linear axis, F represents the spindle flange, Z represents the Z linear axis, and T represents the grinding wheel coordinate system.

[0016] Furthermore, step 2 includes the following steps:

[0017] Step 21: Based on the CNC commands for the three linear axes and two rotary axes, construct the motion transformation matrices for the three linear axes and two rotary axes;

[0018] Step 22: Construct the grinding wheel coordinate system and the tool coordinate system, and establish the ideal pose transformation relationship between the tool coordinate system and the grinding wheel coordinate system;

[0019] Step 23: Construct the motion matrix of related errors caused by position-related geometric errors based on the motion transformation matrix and the 3D pose error model;

[0020] Step 24: Construct an independent error motion matrix caused by position-independent geometric errors based on the 3D pose error model;

[0021] Step 25: Construct the actual pose transformation relationship between the tool coordinate system and the grinding wheel coordinate system based on the relevant error motion matrix and the irrelevant error motion matrix;

[0022] Step 26: Based on the ideal pose transformation relationship between the tool coordinate system and the grinding wheel coordinate system and the actual pose transformation relationship between the grinding wheel coordinate system and the tool coordinate system, construct a pose error model between the tool coordinate system and the grinding wheel coordinate system.

[0023] This application independently constructs a "related error motion matrix" (step 23) caused by position-related geometric errors and a "non-related error motion matrix" (step 24) caused by position-independent geometric errors using a pose error model. These two types of error motion matrices are combined with an ideal kinematic chain (step 25) to accurately reconstruct the "actual pose transformation relationship" between the grinding wheel and tool coordinate systems, encompassing the influence of all key geometric errors. By comparing the ideal and actual pose transformation relationships, a high-fidelity "tool-grinding wheel pose error model" is established (step 26). This model not only fully characterizes the nonlinear coupling and transmission process of multi-axis (linear axes X / Y / Z and rotary axes A / C) geometric errors in complex kinematic chains, but also accurately reflects the combined effect of error sources (such as angular error, axis drift, and perpendicularity error) on the tool end-effector pose (position and attitude).

[0024] Furthermore, step 3 includes the following steps:

[0025] Step 31: Establish the expression for any point on the rotating surface of the grinding wheel in the grinding wheel coordinate system;

[0026] Step 32: Construct the expression for any point on the grinding wheel's rotating surface in the tool coordinate system;

[0027] Step 33: Based on the expression of any point on the grinding wheel's rotating surface in the tool coordinate system, construct the envelope surface equation of the envelope region during the grinding process;

[0028] Step 34: Substitute the solution of the envelope surface equation and the axis vector into the pose error model to obtain the tool surface pose error model.

[0029] This invention establishes a precise geometric expression for any point on the grinding wheel's rotating surface in the grinding wheel coordinate system (step 31) and transforms it to the tool coordinate system (step 32). Based strictly on the relative motion relationship between the grinding wheel and the tool during the grinding process, it constructs an "envelope surface equation" describing the tool surface formed by the grinding wheel envelope (step 33). The "tool-grinding wheel pose error model" established in step 2, which includes the influence of multi-axis geometric errors, is dynamically introduced into the envelope calculation process (step 34). By substituting the solution results of the envelope surface equation (i.e., theoretical contact points / lines) and the "axis vectors" characterizing the tool's posture into the pose error model, a "tool surface position error model" directly related to the machine tool's geometric error source and the specific spatial position of the tool's working surface is established. This model can accurately predict the contour deviation introduced by the grinding wheel envelope motion at any specified position point (not just the tip point TCP) on the tool's complex curved surface (such as the rake face, flank face, variable curvature chip groove, and spherical surface) under actual (including error) machine tool motion conditions. It solves the core problem that traditional methods cannot accurately characterize the spatial heterogeneity distribution of geometric errors across the entire surface of non-developable, variable curvature cutting tools (especially fine structures such as the cutting edge and the transition zone between the rake and flank faces).

[0030] Furthermore, step 32 includes the following steps:

[0031] Step 321: Construct the grinding motion pose matrix set of the grinding wheel during the helical motion process. The grinding motion pose matrix set includes rotation matrix and translation matrix.

[0032] Step 322: Based on the pose transformation of the grinding wheel motion, decompose the rotation matrix and translation matrix into higher-order rotation matrix and higher-order translation matrix;

[0033] Step 323: Align the grinding wheel coordinate system with the tool coordinate system. Based on the origin coordinates and the grinding wheel axis vector, construct the expressions for the origin coordinates and the grinding wheel axis vector in the tool coordinate system.

[0034] Step 324: Combine the expressions for the origin coordinates of the grinding wheel in the tool coordinate system, the expression for the grinding wheel axis vector in the tool coordinate system, the higher-order rotation matrix, and the higher-order translation matrix to obtain the expression for any point on the grinding wheel's rotating surface in the tool coordinate system during the grinding motion.

[0035] This scheme dynamically incorporates the "tool-grinding wheel pose error model" established in step 2, which includes the influence of multi-axis geometric errors, into the envelope calculation process (step 34). By substituting the solution results of the envelope surface equation (i.e., theoretical contact points / lines) and the "axis vectors" characterizing the tool posture into the pose error model, a "tool surface pose error model" directly related to the machine tool geometric error source and the specific spatial position of the tool's working surface is established. This model can accurately predict the contour deviation introduced by the grinding wheel envelope motion at any specified position point (not just the tip point TCP) on the complex curved surface of the tool (such as the rake face, flank face, variable curvature chip groove, spherical surface) under the actual (including error) machine tool motion state. This fundamentally solves the core problem that traditional methods cannot accurately characterize the spatial heterogeneity distribution of geometric errors across the entire non-developable, variable curvature tool surface (especially the micro-structures such as the cutting edge and the rake / flank face transition zone).

[0036] Furthermore, step 33 includes the following steps:

[0037] Step 33 includes the following steps:

[0038] Step 331: Based on the principle of envelope surface, determine the envelope region during the grinding process and establish the spatial envelope surface;

[0039] Step 332: Define the spatial envelope. Defined as the envelope surface of the spiral groove, the parametric equations of the spiral groove envelope surface are obtained by solving the envelope profile.

[0040] This invention, through the principle of envelope formation (step 331), precisely constructs a "spatial envelope surface S" in three-dimensional space, describing the envelope region formed by the movement of the grinding wheel relative to the tool. Targeting the core feature of the tool—the chip groove surface—the spatial envelope surface S is concretized as a "spiral groove envelope surface" (step 332), and its precise "parametric equation" is successfully solved. This parametric equation rigorously characterizes the complete geometric process of the grinding wheel's envelope motion generating the spiral groove surface under ideal conditions (without machine tool errors), providing a precise geometric and kinematic basis for the subsequent step (step 34) to dynamically introduce machine tool geometric errors (through a pose error model) and calculate the positional deviation of any point on the tool's spiral groove surface during actual machining.

[0041] Furthermore, step 4 includes the following steps:

[0042] Step 41: Using the tool surface pose error model as a mapping function, with m pose errors as input and surface pose error as output, construct the surface pose error vector;

[0043] Step 42: Calculate the first-order sensitivity index and global sensitivity index of m pose errors to tool surface errors using the Sobol method;

[0044] Step 43: Use the random forest algorithm to identify pose errors and obtain the quantitative FIS for each pose error;

[0045] Step 44: Assign weights to the first-order sensitivity index, the global sensitivity index, and the quantitative FIS to obtain the comprehensive sensitivity index for each pose error;

[0046] Step 45: Sort the pose errors according to the comprehensive sensitivity index and identify the pose errors that have a significant impact on the tool surface error.

[0047] This invention parameterizes the tool surface pose error model, explicitly taking m pose errors (including position-independent and position-dependent geometric errors) as inputs and surface pose errors as outputs to construct a surface pose error vector (step 41), laying the foundation for systematic error impact analysis. The core innovation lies in employing a multi-dimensional, complementary sensitivity analysis method, effectively overcoming the limitations of single sensitivity analysis methods in highly nonlinear, highly coupled error propagation systems. It comprehensively considers the main effects, interaction effects, and nonlinear contributions of error terms, significantly improving the comprehensiveness, robustness, and engineering applicability of key error term identification.

[0048] Furthermore, in step 41:

[0049] The expression for the surface pose error vector is:

[0050] ;

[0051] in: , Represents the surface pose error vector; Represents the pose error vector; Let x be the mapping function between the pose error vector and the surface pose error vector. m This represents the pose error of the m-th term, where m = 41.

[0052] Furthermore, step 42 includes the following steps:

[0053] Step 421: Obtain the probability distribution and Sobol sequence of m pose errors, and generate two independent first parameter sampling matrices. And the second parameter sampling matrix B.

[0054] Step 422: The Column B, excluding the first Constructing the combination of the remaining columns of a column ;

[0055] Step 423: Calculate the output value based on the mapping function. , and ;

[0056] Step 424: Use , and The sample mean is used to approximate the total variance and conditional variance, and the first-order sensitivity index and global sensitivity index of each pose error to the tool surface error are calculated using the Sobol method.

[0057] ;

[0058] ;

[0059] in: For the first The first-order sensitivity index for each pose error; For the first Global sensitivity index for individual pose errors; The sampling matrix for the first parameter The The second parameter sampling matrix B Represents a constant; Indicates the number of samples; Let represent the total variance, n' represent the sample index, f represent the mapping function, and i represent the pose error index.

[0060] Furthermore, step 43 includes the following steps:

[0061] Step 431: Collect several training samples, and then... For each decision tree, Bootstrap sampling is performed from several training samples, and the samples not sampled from the training samples are used as the Out-of-Body (OOB) samples of decision tree j. j Sample, using OOB j The OOB error of decision tree j is calculated for the sample, denoted as . ;

[0062] Step 432: For all OOBs j Sample pose error x i Randomly add noise interference, recalculate the OOB error of decision tree j, and denot it as . .

[0063] Step 433: For all For each decision tree, the feature importance score of each pose error is calculated using the random forest algorithm:

[0064] ;

[0065] in: The pose error is xi Feature importance score; For the first The OOB error is calculated based on the corresponding OOB data selected by each decision tree. For features of all OOB data samples OOB error recalculated after randomly adding noise interference.

[0066] Step 44 includes the following steps:

[0067] Step 441: Take the absolute values ​​of the first-order sensitivity index and the global sensitivity index and map them to the interval [0,1] to obtain the fusion sensitivity index:

[0068] ;

[0069] ;

[0070] in: and They are the first-order sensitivity indices. and global sensitivity index Take the absolute value and map it to the value after the interval [0,1]. and These are all first-order sensitivity indices. The maximum and minimum values ​​in; and These are all global sensitivity indices. The maximum and minimum values ​​in;

[0071] Step 442: Take the absolute value of the feature importance score and map it to the interval [0,1] to obtain the feature importance score:

[0072] ;

[0073] in: To assign feature importance scores Take the absolute value and map it to the value in the interval [0,1]. and The importance scores for all features are respectively The maximum and minimum values ​​in the range.

[0074] Step 443: Combine the first-order sensitivity index, the global sensitivity index, and the feature importance score to obtain the comprehensive sensitivity index. ;

[0075] ;

[0076] ;

[0077] The overall sensitivity index; , as well as These are the first weighting coefficient, the second weighting coefficient, and the third weighting coefficient, respectively. For the first Individual pose error The proportion, For the first Individual pose error The proportion; For the first Individual pose error The percentage.

[0078] The technical solution of this application embodiment has at least the following advantages and beneficial effects:

[0079] This invention presents a method for identifying key errors in the surface of five-axis grinding tools, integrating the Sobol method and random forest. By constructing a tool surface error model, it overcomes the shortcomings of existing technologies that only focus on the relative pose error between the tool and the workpiece, thus providing a more comprehensive description of actual machining errors. Furthermore, by incorporating the sensitivity analysis of the Sobol method and random forest, it achieves efficient and robust identification of key geometric errors in five-axis grinding machines. Its significant technical effects are mainly reflected in the following three aspects:

[0080] (1) A breakthrough has been achieved in modeling accuracy: By comprehensively considering the machine tool kinematic chain, all 41 geometric errors and the grinding wheel envelope grinding principle, the established tool surface error model can accurately reflect the quantitative mapping relationship between geometric errors and actual machined surface errors, fundamentally overcoming the limitation of traditional methods that rely solely on tool pose errors and cannot accurately describe complex surface machining errors;

[0081] (3) Innovations were achieved in the reliability of identification: The proposed dual-channel collaborative analysis framework creatively integrates the global variance analysis capability of the Sobol method with the data-driven feature importance assessment capability of random forest; by defining the comprehensive sensitivity index (CSI), the independent contribution and interaction of various errors were effectively quantified, which has both theoretical rigor and adaptability to complex nonlinear relationships. Compared with the traditional single sensitivity analysis method, it significantly improves the robustness and anti-interference capability of key error identification, and avoids the perspective limitations and misjudgment risks of single methods;

[0082] In summary, this invention provides a scientific basis for accurately locating the key error sources that have the greatest impact on machining accuracy, provides a theoretical basis for subsequent precise error compensation, and concentrates resources to prioritize the handling of core issues, thereby effectively improving the grinding accuracy of five-axis grinding machines with higher efficiency and lower cost, which has important engineering practical significance. Attached Figure Description

[0083] Figure 1 A flowchart illustrating the method for identifying key errors in the curved surface of a five-axis CNC grinding machine tool;

[0084] Figure 2 This is a schematic diagram of the rotating surface of a grinding wheel;

[0085] Figure 3 This is a schematic diagram illustrating the principle of envelope grinding.

[0086] Figure 4 This is a flowchart of Example 2. Detailed Implementation

[0087] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments. The same reference numerals in the accompanying drawings represent the same components. It should be noted that the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the described embodiments of this application without creative effort are within the scope of protection of this application.

[0088] Compared to the embodiments shown in the accompanying drawings, feasible embodiments within the scope of this application may have fewer components, other components not shown in the drawings, different components, components arranged differently, or components with different connections, etc. Furthermore, two or more components in the drawings may be implemented in a single component, or a single component shown in the drawings may be implemented as multiple separate components.

[0089] refer to Figure 1 Example 1: A method for identifying key errors in the curved surface of a five-axis CNC grinding machine tool, comprising the following steps:

[0090] Step 1: Obtain the 3D structural model of the five-axis grinding machine, and construct the 3D pose error model of each working part in the five-axis grinding machine. The 3D pose error model represents the pose error between the ideal position and the actual position of each working part.

[0091] Five-axis grinding machines are common machining machines in the field of tool processing. The working parts of a five-axis grinding machine include three linear axes (X, Y, and Z) and two rotary axes (A and C). These five axes have a theoretical position in the design. However, during installation, due to factors such as installation process and part clearance, there is an installation deviation between the actual position of the five axes and the ideal position, resulting in positional error.

[0092] Pose error is divided into multiple position-independent geometric errors and multiple position-dependent geometric errors.

[0093] Position-independent geometric errors are related to the installation process and consist of 11 items, including three straightness errors between the X, Y, and Z linear axes. , , ) and eight installation errors of the two rotating axes A and C ;

[0094] The specific details are shown in Table 1:

[0095] Table 1 shows the position-independent geometric errors;

[0096] .

[0097] in, The offset error of the rotation center of axis A in the Y direction.

[0098] The offset error of the rotation center of axis A in the Z direction;

[0099] The offset error of the rotation center of axis C in the X direction;

[0100] The offset error of the rotation center of axis C in the Y direction;

[0101] The tilt error of the rotation axis A about the Z-axis;

[0102] The tilt error of the rotation axis C about the X-axis;

[0103] The tilt error of the rotation axis C about the Y-axis;

[0104] The tilt error of the rotation axis A about the Y-axis;

[0105] Position-related geometric errors are caused by manufacturing defects and wear of parts, totaling 30 items.

[0106] The specific details are shown in Table 2:

[0107] Table 2 shows the location-related geometric errors;

[0108] .

[0109] δx(y), δy(y), and δz(y) represent the displacement errors produced by the Y-axis in the X, Y, and Z directions, respectively.

[0110] δx(c), δy(c), and δz(c) represent the displacement errors produced by the C rotation axis in the X, Y, and Z directions, respectively.

[0111] δx(a), δy(a), and δz(a) represent the displacement errors produced by rotation axis A in the X, Y, and Z directions, respectively.

[0112] δx(x), δy(x), and δz(x) represent the displacement errors produced by the X-axis in the X, Y, and Z directions, respectively.

[0113] δx(z), δy(z), and δz(z) represent the displacement errors produced by the Z-axis in the X, Y, and Z directions, respectively.

[0114] εx(y), εy(y), and εz(y) represent the angular errors produced by the Y-axis in the X, Y, and Z directions, respectively.

[0115] εx(x), εy(x), and εz(x) represent the angular errors produced by the X-axis in the X, Y, and Z directions, respectively.

[0116] εx(z), εy(z), and εz(z) represent the angular errors produced by the Z-axis in the X, Y, and Z directions, respectively.

[0117] εx(a), εy(a), and εz(a) represent the angular errors produced by rotation axis A in the X, Y, and Z directions, respectively.

[0118] εx(c), εy(c), and εz(c) represent the angular errors produced by the C rotation axis in the X, Y, and Z directions, respectively.

[0119] Thus, the pose error model includes position-independent geometric errors related to the installation process, as well as position-dependent geometric errors caused by part manufacturing defects and wear.

[0120] The key to this solution lies in summarizing or screening the geometric errors that have the greatest impact on the machining from the pose error model, so as to compensate and correct them in a targeted manner (to be proposed separately).

[0121] Step 2: Based on the machine tool kinematic chain, calculate the ideal pose transformation relationship and the actual pose transformation relationship between the tool coordinate system and the grinding wheel coordinate system, and construct a pose error model between the tool coordinate system and the grinding wheel coordinate system;

[0122] This solution requires the use of a 5-axis machine tool to machine curved surface tools, with a grinding wheel as the machining head, and the position and angle controlled by the 5-axis machine tool.

[0123] Furthermore, the motion chain of the five-axis grinding machine is: W-A-C-Y-F-X-Z-T. Where W represents the tool coordinate system, A represents the A rotary axis, C represents the C rotary axis, Y represents the Y linear axis, X represents the X linear axis, F represents the spindle flange, Z represents the Z linear axis, and T represents the grinding wheel coordinate system. The interface at the end of the spindle where the grinding wheel is mounted is positioned and its orientation is determined by the subsequent motion axes.

[0124] Furthermore, step 2 includes the following steps:

[0125] Step 21: Based on the CNC commands for the three linear axes and two rotary axes, construct the motion transformation matrices for the three linear axes and two rotary axes;

[0126] ;

[0127] ;

[0128] ;

[0129] ;

[0130] ;

[0131] Where x, y, z, a, and c are the numerical control commands for the X-axis, Y-axis, Z-axis, A-axis, and C-axis, respectively. , , , , , These are the motion transformation matrices for the X-axis, Y-axis, Z-axis, A-axis, and C-axis under the CNC commands x, y, z, a, and c, respectively.

[0132] Step 22: Construct the grinding wheel coordinate system and the tool coordinate system, and establish the ideal pose transformation relationship between the tool coordinate system and the grinding wheel coordinate system. ;

[0133] ;

[0134] This represents the ideal pose transformation relationship between the tool coordinate system and the grinding wheel coordinate system. This is the motion transformation matrix between the grinding wheel coordinate system and the Z-axis coordinate system; This is the motion transformation matrix between the X-axis and Z-axis coordinate systems under the numerical control command z; This is the motion transformation matrix between the bed coordinate system and the X-axis coordinate system under the CNC command x; This is the motion transformation matrix between the bed coordinate system and the Y-axis coordinate system under the CNC command y; This is the motion transformation matrix between the Y-axis and C-axis coordinate systems under CNC command c; This is the motion transformation matrix between the C-axis coordinate system and the A-axis coordinate system under CNC command a; This is the motion transformation matrix between the A-axis coordinate system and the tool coordinate system. -This represents the inverse operation of a matrix.

[0135] The bed coordinate system, X-axis coordinate system, and Z-axis coordinate system are not three completely independent coordinate systems. The X-axis coordinate system is constructed to describe the single-axis motion along the X-axis. It uses the machine origin of the X-axis as its origin, and its X-axis direction is completely consistent with the X-axis direction of the bed coordinate system. It is a local reference frame for single-axis motion. The Z-axis coordinate system is constructed to describe the single-axis motion along the Z-axis and is a local reference frame for single-axis motion along the Z-axis.

[0136] Step 23: Construct the motion matrix of related errors caused by position-related geometric errors based on the motion transformation matrix and the 3D pose error model;

[0137] ;

[0138] in, Let N represent the motion matrix with relevant error, and let N represent the index of the axis, N∈{X linear axis, Y linear axis, Z linear axis, A rotation axis, C rotation axis};

[0139] For example, when N equals the rotation axis C, then = ;

[0140] ;

[0141] The motion matrix representing the C-axis of rotation is related to the error. The interpretation of the relevant parameters can be found in the position-related geometric errors.

[0142] Step 24: Construct an independent error motion matrix caused by position-independent geometric errors based on the 3D pose error model;

[0143] ;

[0144] ;

[0145] ;

[0146] ;

[0147] ;

[0148] The linear motion matrix representing the independent error along the Z-axis. The independent error motion matrix representing the X-axis. The independent error motion matrix representing the Y-axis. This represents the motion matrix of independent error along the C rotation axis. This represents the motion matrix of the independent error of the rotation axis A.

[0149] Step 25: Construct the actual pose transformation relationship between the tool coordinate system and the grinding wheel coordinate system based on the relevant error motion matrix and the irrelevant error motion matrix;

[0150] ;

[0151] in: This represents the actual pose transformation relationship between the tool coordinate system and the grinding wheel coordinate system. This is the motion transformation matrix between the grinding wheel coordinate system and the Z-axis coordinate system. This is the motion transformation matrix between the A-axis coordinate system and the tool coordinate system;

[0152] The linear motion matrix representing the independent error along the Z-axis. This represents the motion matrix representing the correlation error along the Z-axis.

[0153] The independent error motion matrix representing the X-axis. This represents the motion matrix representing the correlation error along the X-axis.

[0154] The independent error motion matrix representing the Y-axis. The motion matrix representing the correlation error along the Y-axis;

[0155] This represents the motion matrix of independent error along the C rotation axis. This represents the motion matrix related to the C-axis rotation error.

[0156] This represents the motion matrix with independent error along the rotation axis A. This represents the motion matrix representing the correlation error along the rotation axis A. express The inverse matrix, express The inverse matrix.

[0157] Step 26: Based on the ideal pose transformation relationship between the tool coordinate system and the grinding wheel coordinate system, and the actual pose transformation relationship between the grinding wheel coordinate system and the tool coordinate system, construct a pose error model between the tool coordinate system and the grinding wheel coordinate system;

[0158] ;

[0159] in: A model of the pose error between the tool coordinate system and the grinding wheel coordinate system. This represents the actual pose transformation relationship between the tool coordinate system and the grinding wheel coordinate system. This represents the ideal pose transformation relationship between the tool coordinate system and the grinding wheel coordinate system.

[0160] Step 3: Establish a tool surface pose error model based on the grinding wheel's machining trajectory;

[0161] Step 3 includes the following steps:

[0162] Step 31: Establish the expression for any point on the rotating surface of the grinding wheel in the grinding wheel coordinate system;

[0163] Specifically: such as Figure 2 As shown, the origin of the grinding wheel coordinate system Located at the center of the large end face of the grinding wheel, The plane coincides with the large end face of the grinding wheel. Align with the axis of the grinding wheel. , , These represent the horizontal axis, vertical axis, and height axis of the grinding wheel coordinate system, respectively. , , It is parallel to the Z-axis, Y-axis and X-axis respectively.

[0164] Any point on the rotating surface of the grinding wheel The expression in the grinding wheel coordinate system is:

[0165] ;

[0166] in: The equation for the profile of the grinding wheel generatrix is ​​given. This is the distance from a point on the surface of revolution to the large end face; This refers to the rotation angle of the grinding wheel.

[0167] Step 32: Construct the expression for any point on the grinding wheel's rotating surface in the tool coordinate system;

[0168] Step 32 includes the following steps:

[0169] Step 321: Construct the grinding motion pose matrix set of the grinding wheel during the helical motion process. The grinding motion pose matrix set includes rotation matrix and translation matrix.

[0170] ;

[0171] Represents the rotation matrix. Represents the translation matrix. This represents the motion transformation matrix of rotary axis A under CNC command a. This represents the motion transformation matrix of the C-axis rotation under the CNC command c. This is the motion transformation matrix between the X-axis coordinate system and the Z-axis coordinate system. This is the motion transformation matrix between the bed coordinate system and the X-axis coordinate system;

[0172] like Figure 3 During the grinding process, the cutting tool can be considered stationary, and the grinding wheel, guided by the helical cutting edge, rotates around the coordinate axis. While rotating, it also performs translational motion, so the rotation matrix and translation matrix can be decomposed, as in step 322.

[0173] Step 322: Based on the pose transformation of the grinding wheel motion, decompose the rotation matrix and translation matrix into a higher-order rotation matrix M and a higher-order translation matrix r;

[0174] ;

[0175] ;

[0176] ;

[0177] in, For intermediate parameters, For higher-order rotation matrices, The X-axis is the direction vector of the line, which is also the tool axis vector; The direction vector of the Y-axis; The direction vector of the Z-axis; For the guide, The helical rotation angle of the grinding wheel. Install angle for grinding wheel, The installation height of the grinding wheel;

[0178] Step 323: Align the grinding wheel coordinate system with the tool coordinate system. Construct the origin coordinate system of the grinding wheel based on the origin coordinates of the grinding wheel and the grinding wheel axis vector. The expression for the grinding wheel axis vector in the tool coordinate system:

[0179] ;

[0180] ;

[0181] This represents the expression for the coordinates of the grinding wheel origin in the tool coordinate system. This represents the expression for the grinding wheel axis vector in the tool coordinate system.

[0182] When constructing the expression for step 323, it is necessary to assume that the grinding wheel is in its initial position. When the grinding wheel coordinate system coincides with the tool coordinate system, the origin coordinates and the grinding wheel axis vector are respectively... and ,T represents the transpose symbol.

[0183] Step 324: Establish the origin coordinates of the grinding wheel. The expressions in the tool coordinate system, the expression for the grinding wheel axis vector in the tool coordinate system, the higher-order rotation matrix M, and the higher-order translation matrix r are used to obtain any point on the grinding wheel's rotating surface during the grinding motion. Expression in the tool coordinate system :

[0184] ;

[0185] ;

[0186] in: For rotation matrix, The X-axis direction vector is the tool axis vector; The Y-axis direction vector; The Z-axis direction vector; Here is the equation for the rotating surface of the grinding wheel; It is a high-order translation matrix; The helical rotation angle of the grinding wheel; Set the angle for the grinding wheel; The installation height of the grinding wheel; For intermediate parameters, For the guide, The equation for the profile of the grinding wheel generatrix is ​​given. This is the distance from a point on the surface of revolution to the large end face; The rotation angle of the grinding wheel;

[0187] Step 33: Based on the expression of any point on the grinding wheel's rotating surface in the tool coordinate system, construct the envelope surface equation of the envelope region during the grinding process;

[0188] Step 33 includes the following steps:

[0189] Step 331: Based on the principle of envelope surface, determine the envelope region during the grinding process and establish the spatial envelope surface S;

[0190] Specifically, by using the envelope principle to calculate the envelope region during the grinding process, the position of the grinding wheel in the tool coordinate system can be obtained. A family of curved surfaces, denoted as ,in Indicates parameters in the family. The magnitude is the helical rotation angle of the grinding wheel; thus, the parameter values ​​in the family represent the helical rotation angle of the grinding wheel. The lower rotating surface.

[0191] When surface Satisfy: any point on it Belongs to the family of curved surfaces A surface in the middle, and at that point If a surface shares a common tangent plane with the surface of that family, then the surface... Called The envelope of .

[0192] If point Envelope Any point on, that point Definitely in And satisfy the system of equations:

[0193]

[0194] in: and Let represent the equations of the family of surfaces and the derivative equations of the family of surfaces, respectively.

[0195] Remove parameters The spatial envelope can then be obtained. ,Right now:

[0196] , Indicates the envelope symbol; Point Coordinates in the tool coordinate system.

[0197] Step 332: Define the spatial envelope. Defined as the envelope surface of the spiral groove, and the envelope contour is solved to obtain the spiral groove envelope surface. Parametric equations;

[0198] Specifically:

[0199] During the grinding process of a grinding wheel, the spatial envelope surface Defined as the envelope surface of a spiral groove, we solve for it and its envelope profile, and establish the identity:

[0200] ;

[0201] Right now:

[0202] ;

[0203] This section is about points The coordinates in the tool coordinate system and the coordinates in the grinding wheel coordinate system correspond to each other. and points These are descriptions of the same point in different coordinate systems. Here is the equation for the rotating surface of the grinding wheel;

[0204] Right now:

[0205]

[0206] ;

[0207] right Taking the partial derivative, we get:

[0208] ;

[0209] in, Point In the tool coordinate system, T represents the matrix transpose symbol, and F represents the equation of the family of surfaces. This is the distance from a point on the surface of revolution to the large end face; For the rotation angle of the grinding wheel, The X-axis direction vector is the tool axis vector. The Y-axis direction vector, Let M be the Z-axis direction vector, M be the higher-order rotation matrix, r be the higher-order translation matrix, and R(h) be the equation of the grinding wheel generatrix profile. The derivative of the equations for the family of surfaces, Equivalent to , Indicates the derivative symbol. Indicates parameters in the family. The size is the helical rotation angle of the grinding wheel. It means that it is always equal to, Represents the higher-order translation matrix r pair Find the matrix after partial derivative and transpose. -1 This represents the matrix inversion operation.

[0210] Based on the expression of any point on the grinding wheel's rotating surface in the tool coordinate system, we obtain:

[0211] ;

[0212] Substituting into the above equation, we get:

[0213] ;

[0214] Solving the above equation yields the envelope equation:

[0215] ;

[0216] The above formula Substituting the expression for any point on the grinding wheel's rotating surface in the tool coordinate system during grinding motion, we can obtain the envelope surface of the helical groove. Parametric equation form:

[0217] ;

[0218] in: Point The expression in the tool coordinate system, Let be the distance from a point on the surface of revolution to the large end face. For the rotation angle of the grinding wheel, The X-axis direction vector is the tool axis vector. The Y-axis direction vector, Let M be the Z-axis direction vector, M be the higher-order rotation matrix, r be the higher-order translation matrix, and R(h) be the equation of the grinding wheel generatrix profile. Here is the equation for the rotating surface of the grinding wheel; The derivative of the equations for the family of surfaces, The derivative of the equations for the family of surfaces, Equivalent to , , For the guide, Here, R(h) = ah + b is the equation for the grinding wheel generatrix, and 'a' represents the taper coefficient of the grinding wheel. Install angle for grinding wheel, Indicates the rotation angle of the grinding wheel. The installation height of the grinding wheel, The higher-order translation matrix r with respect to parameters The matrix obtained by taking the partial derivative and then transposing it; The matrix obtained by taking the partial derivative of the higher-order translation matrix r with respect to the parameter α. The resulting matrix, Envelope The parametric equations, i.e., the envelope surface equations;

[0219] Step 34: Substitute the solution of the envelope surface equation and the axis vector into the pose error model. The tool surface pose error model is obtained;

[0220] The tool surface pose error model is represented as:

[0221] ;

[0222] ;

[0223] in: Indicates the positional error of the tool surface; Indicates the tool surface pose error; , and These represent the position errors in the X, Y, and Z directions, respectively. , and These represent the angular errors in the X, Y, and Z directions, respectively. This represents the ideal pose transformation relationship between the tool coordinate system and the grinding wheel coordinate system. This represents the ideal pose transformation relationship between the tool coordinate system and the grinding wheel coordinate system. This represents the position coordinates of point j on the envelope plane; The attitude of point j on the envelope surface is represented by T, which represents the matrix device symbol.

[0224] Step 4: Calculate the sensitivity of each pose error in the tool surface pose error model to sort the pose errors.

[0225] In this scheme, step 4 mainly involves sorting the pose errors introduced in step 1. For ease of understanding, position-independent geometric errors and position-dependent geometric errors are collectively referred to as pose errors. In fact, pose errors include the 3+8+30 errors recorded in step 1.

[0226] The tool surface pose error model is the sole basis for simplifying the parameter system, and its core mapping function is the mathematical correlation carrier between the error source and the pose error.

[0227] Step 41: Using the tool surface pose error model as a mapping function, with m pose errors as input and surface pose error as output, construct the surface pose error vector, expressed as:

[0228] ;

[0229] in: , Represents the surface pose error vector; Represents the pose error vector; Let x be the mapping function between the pose error vector and the surface pose error vector. m Let m represent the pose error of the m-th term, and T represent the matrix transpose.

[0230] The pose error vector consists of the 41 errors corresponding to the pose error model provided in step 1, where m=41;

[0231] The surface pose error vector is a set of parameters used to describe the machining accuracy of the tool surface.

[0232] Step 42: Calculate the first-order sensitivity index and global sensitivity index of m pose errors to tool surface errors using the Sobol method;

[0233] Step 42 includes the following steps:

[0234] Step 421: Obtain the probability distribution and Sobol sequence of m pose errors, and generate two independent first parameter sampling matrices. The second parameter sampling matrix B;

[0235] Step 422: The Column B, excluding the first Constructing the combination of the remaining columns of a column ;

[0236] Step 423: Calculate the output value based on the mapping function. ;

[0237] Step 424: Use The sample mean is used to approximate the total variance and conditional variance, and the first-order sensitivity index and global sensitivity index of each pose error to the tool surface error are calculated using the Sobol method.

[0238] ;

[0239] ;

[0240] in: For the first The first-order sensitivity index for each pose error; For the first Global sensitivity index for individual pose errors; The sampling matrix for the first parameter The The second parameter sampling matrix B Represents a constant; Indicates the number of samples; Let represent the total variance, n' represent the sample index, f represent the mapping function, and i represent the pose error index.

[0241] The principle of step 42 is to use Sobol global sensitivity analysis, through variance decomposition and a specific sampling strategy, to quantify the independent influence of each of the 41 tool pose errors on the final surface machining accuracy (first-order sensitivity index) and the total influence including the interaction (global sensitivity index).

[0242] Step 43: Use the random forest algorithm to identify pose errors and obtain the quantitative FIS for each pose error;

[0243] Step 43 includes the following steps:

[0244] Step 431: Collect several training samples, and then... For each decision tree, Bootstrap sampling is performed from several training samples, and the samples not sampled from the training samples are used as the Out-of-Body (OOB) samples of decision tree j. j Sample, using OOB j The OOB error of decision tree j is calculated for the sample, denoted as . ;

[0245] Step 432: For all OOBs j Sample pose error xi Randomly add noise interference, recalculate the OOB error of decision tree j, and denot it as . .

[0246] Step 433: For all For each decision tree, the feature importance score of each pose error is calculated using the random forest algorithm:

[0247] ;

[0248] in: The pose error is x i Feature importance score; For the first The OOB error is calculated based on the corresponding OOB data selected by each decision tree. For features of all OOB data samples OOB error recalculated after randomly adding noise interference.

[0249] Step 43 works by employing the out-of-bag (OOB) error ranking importance method of random forests to evaluate the importance of pose error features. Specifically, after training the random forest using the samples generated in step 41, for each decision tree, the baseline prediction error is calculated using the unselected samples in its bootstrap sampling; subsequently, only the features x in the samples are randomly shuffled. i The value of x is used to recalculate the prediction error; by comparing the change in error of the same tree before and after the features are shuffled, a significant increase in error indicates that x... i Predicting this tree is crucial; ultimately, the error variation across all trees is averaged to obtain a global importance score for the feature. A higher score indicates a greater contribution of that feature to the predicted surface pose error. The core of this method lies in using OOB samples as a natural validation set and quantifying feature importance by observing the degree of model performance degradation through the destruction of single feature information.

[0250] Step 44: Assign weights to the first-order sensitivity index, the global sensitivity index, and the quantitative FIS to obtain the comprehensive sensitivity index for each pose error.

[0251] Step 44 includes the following steps:

[0252] Step 441: Take the absolute values ​​of the first-order sensitivity index and the global sensitivity index and map them to the interval [0,1] to obtain the fusion sensitivity index:

[0253] ;

[0254] in: and They are the first-order sensitivity indices. and global sensitivity index Take the absolute value and map it to the value after the interval [0,1]. and These are all first-order sensitivity indices. The maximum and minimum values ​​in; and These are all global sensitivity indices. The maximum and minimum values ​​in;

[0255] Step 442: Take the absolute value of the feature importance score and map it to the interval [0,1] to obtain the feature importance score:

[0256] ;

[0257] in: To assign feature importance scores Take the absolute value and map it to the value in the interval [0,1]. and The importance scores for all features are respectively The maximum and minimum values ​​in;

[0258] Step 443: Combine the first-order sensitivity index, the global sensitivity index, and the feature importance score to obtain the comprehensive sensitivity index. ;

[0259] This application defines a novel Comprehensive Sensitivity Index (CSI) to quantify the independent contributions and interactions of pose errors, building upon the Sobol-based sensitivity index and the random forest-based FIS. Specifically, the first... Individual pose error , , The normalized (proportion) can be:

[0260] ;

[0261] ;

[0262] in: The overall sensitivity index; , as well as These are the first weighting coefficient, the second weighting coefficient, and the third weighting coefficient, respectively. For the first Individual pose error The proportion, For the first Individual pose error The proportion; For the first Individual pose error The percentage.

[0263] Step 45: Sort the pose errors according to the comprehensive sensitivity index to identify the pose errors that significantly affect the tool surface error. Pose error refers to position and orientation error, that is, the collective term for position and angle error.

[0264] refer to Figure 4 Example 2: Example 1 can use the comprehensive sensitivity index to sort the pose errors of each position, and the sorting results are mainly used for tool processing compensation.

[0265] Generally, neural network models are used in tool machining compensation. The input to this model is the pose error (i.e., the influencing factor), and the output after processing by the neural network model is compensation information. The compensation information is usually the compensation amount (6 degrees of freedom) for the relative pose of the tool and grinding wheel.

[0266] Different compensation schemes will choose neural network models with different structures. Generally, the more complex the neural network structure, the higher the corresponding compensation accuracy. However, in practice, neural network models cannot take in all pose error information and need to selectively select some parameters.

[0267] Therefore, in the technical solution provided in Embodiment 1, the pose error ranking can help determine the parameters input to the neural network model. For example, after obtaining the pose error ranking using the solution in Embodiment 1, if the neural network model needs to input 6 parameters, then the top 6 pose errors in the ranking can be selected.

[0268] Existing solutions typically first sort the pose errors, then select the top-ranked pose errors (e.g., 6) based on the number of input channels of the neural network model (e.g., 5 or 6 parameters), and then proceed with the subsequent training of the neural network model.

[0269] This training method ignores the influence of unselected parameters on the compensation result, which can easily lead to overfitting of the model. Therefore, Example 2 provides the following technical solution based on Example 1:

[0270] The method for identifying key errors in the surface of a five-axis CNC grinding machine tool further includes methods for generating a first weighting coefficient, a second weighting coefficient, and a third weighting coefficient. These methods include the following steps:

[0271] S1: Construct a selection neural network model. The input to the selection neural network model is the measured pose error parameter set, and the output is the first weight coefficient. Second weighting coefficient Third weighting coefficient The pose error parameter set includes the 41 pose error parameters listed in Example 1.

[0272] S2: Update the comprehensive sensitivity index of each pose error according to the first weight coefficient, second weight coefficient, and third weight coefficient output by the selected neural network model, and sort the top X pose errors according to the updated comprehensive sensitivity index sorting results, and input them into the multi-channel feature extraction module;

[0273] S3: Multi-channel feature extraction module, used to extract each pose error. The number of multi-channel feature extraction layers is equal to the number of pose error inputs X. Each pose error is input to the multi-channel feature extraction module to extract the hidden features.

[0274] The multi-channel feature extraction module includes multiple independent fully connected layer networks, which can extract the hidden features of pose error;

[0275] In this scheme, multiple independent fully connected layers are used to extract the latent features of pose error, which essentially involves identifying which information within the pose error is more relevant. The multi-channel feature extraction module is essentially a preprocessing module; after preprocessing the pose error, it is input into the compensation neural network model for subsequent feature fusion and output.

[0276] S4: Input each hidden feature into the compensation neural network model to obtain the compensation scheme. Use the difference between the machining error and the machining accuracy after the actual operation of the compensation scheme as a label to back-optimize the compensation neural network model and filter the neural network. The compensation scheme is the compensation amount of the relative pose of the tool and the grinding wheel.

[0277] This application does not impose requirements on the structure or style of the compensation neural network model or the screening neural network model. Only a positive feedback neural network is required. Based on the deviation between the actual processing error and the processing accuracy after execution, the model parameters within the compensation neural network model are corrected, thereby gradually reducing the deviation between the predicted results and the actual results.

[0278] In practice, both the selection neural network model and the compensation neural network model use multilayer perceptron (MLP) networks. Generally speaking, the network structure of the compensation neural network model is more complex, that is, it has more fully connected layers.

[0279] The following is a detailed description of the training process:

[0280] (1) Prepare multiple training samples. Each training sample includes a set of measured pose error parameters and machining errors. The set of pose error parameters includes the 41 pose error parameters listed in Example 1, and the machining accuracy is the machining accuracy required by the tool system.

[0281] (2) Input the pose error parameter set from the training samples into the screening neural network model to obtain the first weight coefficient. Second weighting coefficient Third weighting coefficient According to the first weighting coefficient Second weighting coefficient Third weighting coefficient Calculate the comprehensive sensitivity index of each pose error, sort them according to the comprehensive sensitivity index, and select X pose errors that are equal to the number of inputs to the compensation neural network model.

[0282] (3) The selected X pose errors are input into the corresponding X independent fully connected layer networks (channels) in the multi-channel feature extraction module. Each independent fully connected layer network is dedicated to processing a key error parameter, transforming (extracting) its original value into a more abstract latent feature vector containing richer information.

[0283] (4) Input the latent feature vectors into the compensation neural network model. The compensation neural network model calculates based on these latent feature vectors and ultimately outputs a specific compensation scheme (e.g., predicted correction values ​​for each machine tool motion command requiring compensation). Apply the "compensation scheme" predicted by the compensation neural network model to the machine tool control system (or simulate its application in a simulation model) and measure the new actual machining error. Compare this new actual machining error with the original labels (expected machining accuracy / target error) provided in the training samples and calculate the difference (loss). This difference represents how effective the compensation scheme predicted by the current network is (the smaller the difference, the better).

[0284] (5) Backward optimization (training network): Utilizing the calculated loss (difference in processing error) to simultaneously optimize two neural networks:

[0285] Compensation neural network model optimization: This is the most direct approach. The loss signal is propagated back from the output layer of the compensation neural network model to its input layer through a backpropagation algorithm, adjusting all connection weights and bias parameters within the model. The goal is to enable the compensation neural network model to predict a better compensation scheme (i.e., less actual processing error after application) given the same key error latent features in the future.

[0286] The loss signal also propagates back to the selection neural network model. Although the loss is calculated at the output of the compensation neural network model, the input to the compensation neural network model (i.e., the latent features of the X key errors) is selected by the selection neural network model. Therefore, the loss signal travels in the reverse direction of the data flow, through the multi-channel feature extraction module (whose parameters are usually fine-tuned), and finally back to the compensation neural network model. This prompts the selection neural network model to learn and adjust its internal weight parameters so that its next output λ1, λ2, λ3 can select a more critical set of X pose error parameters that are more helpful for the compensation neural network model to make accurate compensation predictions.

[0287] Iterative Loop: Repeat steps 3 through 8 for all samples (or batches) in the training dataset. In each round (or batch) of training, the network parameters are fine-tuned based on the calculated loss. This process is repeated until the predictive performance of the two networks stabilizes (i.e., the loss no longer decreases significantly), or the preset number of training rounds is reached.

[0288] The loss function used during training can be the cross-entropy loss function.

[0289] The internal parameters of the multi-channel feature extraction module are also fine-tuned during the iteration process. In practice, the weights of the multi-channel feature extraction module for different pose parameters can be frozen.

[0290] Specifically, the multi-channel feature extraction module contains multiple independent fully connected networks, each with its own weight parameters.

[0291] During the first training, three pose parameters, A, B, and C, were input to the first, second, and third fully connected layers of the multi-channel feature extraction module, resulting in the first, second, and third network parameters, respectively.

[0292] During the second training iteration, A and B are still input into the multi-channel feature extraction module, while C is replaced by D. Therefore, the first and second fully connected layers still use the parameters of the first and second networks for this training, while the third fully connected layer requires initialization of its network parameters.

[0293] If the pose parameters C need to be re-input to the multi-channel feature extraction module during the third training, the third fully connected layer network will reload the third network parameters left over from the previous training of pose parameters C.

[0294] Thus, this approach avoids the problem of decreased accuracy in extracting latent features from pose parameters due to frequent changes in pose parameters.

[0295] The above are merely preferred embodiments of this application and are not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for identifying key errors in the curved surface of a five-axis CNC grinding machine tool, characterized in that, Includes the following steps: Step 1: Obtain the three-dimensional structural model of the five-axis grinding machine, and construct the three-dimensional pose error model of each working part in the five-axis grinding machine. The three-dimensional pose error model is the pose error of each working part between the ideal position and the actual position. The pose error includes multiple position-independent geometric errors and multiple position-dependent geometric errors; Step 2: Based on the machine tool kinematic chain, calculate the ideal pose transformation relationship and the actual pose transformation relationship between the tool coordinate system and the grinding wheel coordinate system, and construct a pose error model between the tool coordinate system and the grinding wheel coordinate system; Step 3: Establish a tool surface pose error model based on the machining trajectory of the grinding wheel; Step 4: Calculate the sensitivity of each pose error in the tool surface pose error model to sort the pose errors; Step 4 includes the following steps: Step 41: Use the tool surface pose error model as a mapping function, take the m pose errors as input and the surface pose error as output, and construct the surface pose error vector. Step 42: Calculate the first-order sensitivity index and global sensitivity index of m pose errors to tool surface error using the Sobol method; Step 43: Use the random forest algorithm to identify pose errors and obtain the quantitative FIS for each pose error; Step 44: Assign weights to the first-order sensitivity index, the global sensitivity index, and the quantitative FIS to obtain the comprehensive sensitivity index for each pose error; Step 45: Sort the pose errors according to the comprehensive sensitivity index and identify the pose errors that have a significant impact on the tool surface error; Step 42 includes the following steps: Step 421: Obtain the probability distribution and Sobol sequence of the m pose errors, and generate two independent first parameter sampling matrices. The second parameter sampling matrix B; Step 422: [The text appears to be incomplete and contains several grammatical errors. A more accurate translation would require The Column B, excluding the first Constructing by combining the remaining columns of a column ; Step 423: Calculate the output value based on the mapping function. and ; Step 424: Use and The sample mean is used to approximate the total variance and conditional variance, and the first-order sensitivity index and global sensitivity index of each pose error to the tool surface error are obtained by using the Sobol method. ； ； in: For the first The first-order sensitivity index for each pose error; For the first Global sensitivity index for individual pose errors; The sampling matrix for the first parameter The Column and second parameter sampling matrix B divided by the first The matrix constructed by combining the remaining columns outside the main columns; Represents a constant; Indicates the number of samples; Let represent the total variance, n' represent the sample index, f represent the mapping function, and i represent the pose error index. Step 43 includes the following steps: Step 431: Collect several training samples and configure N' decision trees. For each decision tree, Bootstrap sampling is performed from several training samples, and the samples not sampled from the training samples are used as the Out-of-Body (OOB) samples of decision tree j. j Sample, using OOB j The OOB error of decision tree j for sample calculation is denoted as . ; Step 432: For all OOBs j Sample pose error x i Randomly introduce noise interference, recalculate the OOB error of decision tree j, and denot it as . ; Step 433: For all For each decision tree, the feature importance score of each pose error is calculated using the random forest algorithm: ； in: The pose error x i Feature importance score; For the first The OOB error is calculated for each decision tree that selects the corresponding OOB data. Let x be the pose error for all OOB data samples. i The OOB error is recalculated after randomly adding noise interference, where i represents the index of the pose error; Step 44 includes the following steps: Step 441: Take the absolute values ​​of the first-order sensitivity index and the global sensitivity index and map them to the interval [0,1] to obtain the fusion sensitivity index: ; ; in: and They are the first-order sensitivity indices. and global sensitivity index Take the absolute value and map it to the value after the interval [0,1]; and These are all first-order sensitivity indices. The maximum and minimum values ​​in; and These are all global sensitivity indices. The maximum and minimum values ​​in; Step 442: Take the absolute value of the feature importance score and map it to the interval [0,1] to obtain the feature importance score: ； in: To assign feature importance scores Take the absolute value and map it to the value in the interval [0,1]. and The importance scores for all features are respectively The maximum and minimum values ​​in; Step 443: Combine the first-order sensitivity index, the global sensitivity index, and the feature importance score to obtain the comprehensive sensitivity index. ; ; ; in: The overall sensitivity index; , as well as These are the first weighting coefficient, the second weighting coefficient, and the third weighting coefficient, respectively. For the first Individual pose error The proportion, For the first Individual pose error The proportion; For the first Individual pose error The percentage.

2. The method for identifying key errors in the curved surface of a five-axis CNC grinding machine tool according to claim 1, characterized in that, In step 2, the motion chain of the five-axis grinding machine is: W-A-C-Y-F-X-Z-T; where W represents the coordinate system, A represents the A rotary axis, C represents the C rotary axis, Y represents the Y linear axis, X represents the X linear axis, F represents the spindle flange, Z represents the Z linear axis, and T represents the grinding wheel coordinate system.

3. The method for identifying key errors in the curved surface of a five-axis CNC grinding machine tool according to claim 2, characterized in that, Step 2 includes the following steps: Step 21: Construct the motion transformation matrix of the three linear axes and two rotary axes based on the CNC commands of the three linear axes and two rotary axes; Step 22: Construct the grinding wheel coordinate system and the tool coordinate system, and establish the ideal pose transformation relationship between the tool coordinate system and the grinding wheel coordinate system; Step 23: Construct the motion matrix of related errors caused by position-related geometric errors based on the motion transformation matrix and the 3D pose error model; Step 24: Construct an independent error motion matrix caused by position-independent geometric errors based on the 3D pose error model; Step 25: Construct the actual pose transformation relationship between the tool coordinate system and the grinding wheel coordinate system based on the relevant error motion matrix and the irrelevant error motion matrix; Step 26: Based on the ideal pose transformation relationship between the tool coordinate system and the grinding wheel coordinate system and the actual pose transformation relationship between the grinding wheel coordinate system and the tool coordinate system, construct a pose error model between the tool coordinate system and the grinding wheel coordinate system.

4. The method for identifying key errors in the curved surface of a five-axis CNC grinding machine tool according to claim 1, characterized in that, Step 3 includes the following steps: Step 31: Establish the expression for any point on the rotating surface of the grinding wheel in the grinding wheel coordinate system; Step 32: Construct the expression for any point on the grinding wheel's rotating surface in the tool coordinate system; Step 33: Based on the expression of any point on the grinding wheel's rotating surface in the tool coordinate system, construct the envelope surface equation of the envelope region during the grinding process; Step 34: Substitute the solution of the envelope surface equation and the axis vector into the pose error model to obtain the tool surface pose error model.

5. The method for identifying key errors in the curved surface of a five-axis CNC grinding machine tool according to claim 4, characterized in that, Step 32 includes the following steps: Step 321: Construct the grinding motion pose matrix set of the grinding wheel during the helical motion process. The grinding motion pose matrix set includes the rotation matrix and the translation matrix. Step 322: Based on the pose transformation of the grinding wheel motion, decompose the rotation matrix and translation matrix into higher-order rotation matrix and higher-order translation matrix; Step 323: Align the grinding wheel coordinate system with the tool coordinate system. Based on the origin coordinates and the grinding wheel axis vector, construct the expressions for the origin coordinates and the grinding wheel axis vector in the tool coordinate system. Step 324: Combine the expressions for the origin coordinates of the grinding wheel in the tool coordinate system, the expression for the grinding wheel axis vector in the tool coordinate system, the higher-order rotation matrix, and the higher-order translation matrix to obtain the expression for any point on the grinding wheel's rotating surface in the tool coordinate system during the grinding motion.

6. The method for identifying key errors in the curved surface of a five-axis CNC grinding machine tool according to claim 4, characterized in that, Step 33 includes the following steps: Step 33 includes the following steps: Step 331: Based on the principle of envelope surface, determine the envelope region during the grinding process and establish the spatial envelope surface S; Step 332: Define the spatial envelope. Defined as the envelope surface of the spiral groove, and the envelope contour is solved to obtain the spiral groove envelope surface. The equation of the envelope surface.

Images