A method for sensing deviation of a clamped posture of a tool of a tool grinder and correcting a grinding path

Abstract

The application discloses a kind of knife tool clamping posture deviation perception and sharpening path correction method, it is related to tool processing technical field, including the following specific steps: after filtering to original clamping state point set, form has clamping state measured point set;Point position structure is solved based on the internal transmission relationship matrix of reference surface flattening coordinate set;After obtaining the clamping posture deviation energy of clamping attitude deviation based on the difference of multi-scale hole life spectrum and profile multi-scale hole life spectrum, the six degrees of freedom clamping posture deviation of actual tool clamping relative to ideal clamping is obtained;After each control point is executed compensation correction by row extraction. The application constructs reference grid and executes surface measurement flat evolution, flattens tool complex surface to two-dimensional plane, and calibrates profile multi-scale hole life spectrum and solves internal transmission relationship matrix, so that subsequent deviation perception can fully reflect the overall morphological characteristics of tool profile.

Description

Technical Field

[0001] This invention relates to the field of cutting tool processing technology, specifically to a method for sensing the clamping posture deviation of a tool in a grinding machine and correcting the grinding path. Background Technology

[0002] In the sharpening process of a knife grinding machine, the accuracy of the tool clamping posture is the prerequisite for ensuring the precision of the cutting edge. However, during actual clamping, the tool inevitably experiences translational deviation along the X, Y, and Z directions and rotational deviation around the three coordinate axes.

[0003] Using a dial indicator to take points on the outer circumference of the tool to calculate radial runout can avoid translational and rotational misalignment, but it is difficult to fully describe the true pose of the entire tool profile in space. Current online measurement methods, such as using line-through lasers or contact probes to acquire point cloud data of the tool surface in the clamped state and comparing it with theoretical models to infer deviations, can describe the tool profile's pose in space. However, they clearly do not consider the problem that the spatial curvature of the tool profile directly interferes with the distance measurement between point clouds. During optimization, the geometric distortion of the surface itself is easily misjudged as clamping posture deviation, resulting in significant discrepancies between the calculated rotation and translation values ​​and the actual posture. Furthermore, when the clamping deviation is large, these algorithms are prone to getting trapped in local extrema, exhibiting insufficient convergence stability, thus compromising the accuracy of subsequent grinding path corrections. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method for sensing tool clamping posture deviation and correcting grinding path in a sharpening machine, comprising the following specific steps: S1. Generate a sampling grid based on the geometric information extracted from the computer design model of the tool. Generate a set of profile reference points by reading the spatial coordinates. Obtain the original clamping state point set based on the sensor acquisition of the tool in the clamping state. After filtering the original clamping state point set, form a set of measured points in the clamping state. S2. Construct a reference grid based on the profile reference point set, and perform surface metric flattening evolution on the reference grid. Take the set of planar coordinates corresponding to each node of the evolved flattening metric grid as the reference surface flattening coordinate set. Based on the reference surface flattening coordinate set, calibrate the profile multi-scale hole lifetime spectrum, and solve the internal transmission relationship matrix of the point structure based on the reference surface flattening coordinate set. S3. Based on the initial clamping posture deviation, it is converted into a rigid body transformation matrix through exponential mapping. Each point in the clamping state measured point set is combined with the rigid body transformation matrix to form the current attempted clamping point set. After completing the spatial transformation, the current attempted clamping point set is repeated after the S2 operation to obtain the current attempted flattened coordinate set. Based on the internal transmission relationship matrix, the structural transmission relationship between the current attempted point set and the reference point set is established, and the multi-scale hole life spectrum of the current tool state is constructed. Based on the difference between the multi-scale hole life spectrum and the profile multi-scale hole life spectrum, the clamping posture deviation energy is obtained, and the six-degree-of-freedom clamping posture deviation of the actual tool clamping relative to the ideal clamping is obtained. S4. A translation correction vector is formed based on the six-degree-of-freedom clamping attitude deviation, and compensation correction is performed on each control point after extracting it row by row.

[0005] Preferably, step S1 is based on the tool computer design model, and with the tool rotation axis as the reference direction, spiral sampling lines are generated on the surface of the tool computer design model along the circumferential and axial directions at equal arc length intervals. The three-dimensional coordinates of spatial points are extracted based on the spiral sampling lines to form an ordered array as a set of profile reference points. Thus, through the pre-calibrated rigid body transformation relationship from the tool design coordinate system to the positioning reference coordinate system of the grinding machine fixture, the coordinates of all points in the set are uniformly mapped to the clamping evaluation reference system. Based on the line laser profile sensor in the working area of ​​the grinding machine, and after the laser line projected by the line laser profile sensor forms a preset angle with the tool axis, the tool is rotated at a constant speed around the axis under the drive of the workpiece axis, and the line laser profile sensor is translated at a constant speed along the guide rail parallel to the axis. The position of the two-dimensional profile point formed by the reflection of the laser line on the tool surface is obtained by the synchronous triggering of each encoder. After continuously collecting the complete scanning stroke, the original clamping state point set is obtained, and the profile point is reconstructed into a three-dimensional point in the machine tool coordinate system. S1 is to eliminate the interference of sensor noise and point density differences on subsequent deviation perception. The original clamping state point set is filtered. Specifically, the space where the original clamping state point set is located is divided into three-dimensional grid cells. The side length of the grid cell is automatically set according to the density of the original points inside it. That is, in areas with high point density such as the cutting edge, the side length is automatically reduced to retain the sharp geometric features, and in smooth areas of the tool body, the side length is automatically increased to smooth random fluctuations. The arithmetic mean of the three coordinate components of all the original points inside each non-empty grid cell is calculated to form a representative point. All original points in the non-empty grid cells are cleared to form a clamping state measured point set with uniform point density and preserved cutting edge features.

[0006] Preferably, S2 is based on taking each sampling point in the contour reference point set as a node and establishing the connection relationship between the points through the k nearest neighbor method. Specifically, it calculates the straight-line distance from each node to the other nodes, connects the other nodes with the smallest distance to form an edge, traverses all three-node combinations, and when there are three nodes with edges between each pair, fills the area enclosed by the three edges as a triangular facet to obtain a triangular mesh that seamlessly covers all nodes and has no topological defects as a reference mesh. Based on each node located inside the baseline grid, find all triangular faces with the node as the vertex. Based on the current length of the three sides of each triangular face, and through the relationship between the side length and the included angle, calculate the interior angle value of the node's location. The interior angles of all the triangular faces of the node are summed, and the difference between 2π and the sum is taken as the current Gaussian curvature of the node. If the difference is positive, it means that the surface is convex in the local area of ​​the node; if it is negative, it is concave; and if it is zero, it means that the local area is flat. Set the target Gaussian curvature of each internal node in the reference grid to zero. Based on the topological connectivity and edge length of the reference grid, construct a system of linear equations with the scaling factor correction of each node as unknowns. It should be noted that the coefficients of the system of equations are composed of the cotangent values ​​of the angles of the triangles containing the nodes at both ends of each edge. The constant on the right side of the system of equations is the difference between the current Gaussian curvature and the target curvature of zero. The fine-tuning amount of the scaling factor of each node is obtained by solving the system of linear equations.

[0007] Preferably, S2 is based on each edge in the reference grid, finds the two nodes connected by each edge, adds the corresponding two current scaling factors, obtains the edge length scaling factor by calculating the sum of the natural constant e, and uses the product of the edge length scaling factor and the edge length as the new length of the corresponding edge. The lengths of all edges in the entire grid are updated synchronously. After updating the length of each edge of the reference grid, iteratively calculate the current absolute value of Gaussian curvature of all nodes in the current reference grid. If the maximum value of the current absolute value of Gaussian curvature is less than a preset threshold, stop iterating, use the current reference grid as the flatness metric grid, and use the maximum value of the current absolute value of Gaussian curvature as the reference flatness metric.

[0008] Preferably, in step S2, a node is randomly selected on the flatness measurement grid as the origin and assigned a two-dimensional plane coordinate origin value. An edge connected to the origin is selected and assigned coordinates along the horizontal axis. Each triangular facet is processed in breadth-first order. The plane coordinates of all nodes constitute a reference surface flattening coordinate set, and the node area weight of each node is assigned. Based on the reference surface flattening coordinate set, the edge length under the reference flatness measurement is used as the path length. The shortest path search algorithm is executed to calculate the shortest path length between any two nodes as the flattening geodesic distance. The flattening geodesic distance between all node pairs is stored in a matrix. Based on the observation scale parameter that is set to gradually increase from zero and compared with the flattening geodesic distance, a shape structure including nodes, edges, triangular facets, and tetrahedrons is generated to form a multi-scale topological growth sequence. The multi-scale porosity spectrum of the profile is obtained by executing a continuous cohomology algorithm.

[0009] Preferably, in the S2 step, each node is divided by the sum of the area weights of all nodes in the reference surface flattened coordinate set to obtain the normalized quality of the node, and the normalized quality of all nodes constitutes the reference source distribution, and the square of the Euclidean distance between any two nodes in the reference surface flattened coordinate is used as the movement cost. After introducing the reference source distribution and the movement cost, S2 solves the transmission relationship in the coordinate set of the flattened reference surface to form an internal transmission relationship matrix.

[0010] Preferably, step S3 is based on three rotational deviations and three translational deviations to form the clamping attitude deviation to be determined. The clamping attitude deviation to be determined is represented by a vector containing six values. After setting the initial values ​​of the three rotational deviations and three translational deviations to zero, the clamping attitude deviation to be determined is converted into a four-row, four-column rigid body transformation matrix through exponential mapping. By traversing each point in the set of measured clamping state points, the three-dimensional coordinates are expanded into homogeneous coordinates, which means adding a fourth component with a value of 1 after the original three coordinate values. The rigid body transformation matrix is ​​then multiplied by the column vector of the homogeneous coordinates on the left, and the first three components of the resulting vector are used as the transformed spatial positions. After repeating the operation on the set of measured clamping state points, the current attempted clamping point set is obtained. After repeating the operation on the current attempted clamping point set, the obtained flatness measurement grid is flattened onto a two-dimensional plane to obtain the current attempted flattened coordinate set. The area weights of all nodes in the current attempted flattened coordinate set are normalized to form the attempt source distribution, and based on... Based on the baseline target distribution in the flattened coordinate set of the baseline surface, the squared Euclidean distance between each node in the attempt source distribution and each node in the baseline target distribution in their respective flattened coordinate system is calculated to construct a cross-domain movement cost matrix. Because in actual cross-domain operations, not only spatial distance but also the inherent structural constraints of the tool shape must be considered, this application provides an internal transmission relationship matrix to correct the cross-domain movement cost matrix. This includes: using the sum of all elements in each row of the attempt source distribution node and the baseline target distribution node in the internal transmission relationship matrix as the ideal dispersion of the attempt source distribution node, and the sum of all elements in each column as the ideal aggregation of the baseline target distribution node; obtaining the structural compatibility coefficient based on the geometric mean of the ideal dispersion and the ideal aggregation; dividing the element value of any row and column in the cross-domain movement cost matrix by the structural compatibility coefficient to obtain the corrected cross-domain movement cost; and solving the corrected movement cost matrix through the transmission relationship to form the current cross-domain transmission relationship matrix.

[0011] Preferably, step S3 merges all nodes of the currently attempted flattened coordinate set and all nodes of the reference surface flattened coordinate set into a node set, and constructs a multi-scale topology growth sequence according to the connection weight between any two nodes in the obtained node set. The calibration operation of step S2 is repeated on the obtained multi-scale topology growth sequence to obtain the transmission sensing multi-scale hole lifetime spectrum. Based on each interval in the transmission sensing multi-scale hole lifetime spectrum and the profile multi-scale hole lifetime spectrum as a point on a two-dimensional plane, the projection points of all points to the diagonal are introduced as an augmented set. Under the condition that interval points are allowed to be paired with diagonal points, the sum of the squares of the Euclidean distances between all paired points is minimized, and the square root of the minimum sum is taken as the clamping attitude deviation energy.

[0012] Preferably, in step S3, based on the acquired clamping attitude deviation energy, the partial derivative of each component of the clamping attitude deviation energy is calculated using the chain rule, and then combined into a gradient vector. A one-dimensional search is performed based on the opposite direction of the gradient vector. Starting from the initial step size, the initial step size is repeatedly halved. The current step size is multiplied by the negative gradient vector as a correction value, and the clamping attitude deviation, which includes three rotational deviations and three translational deviations, is updated. The updated clamping attitude deviation is used as the current attempted attitude. The original clamping state measured point set is re-performed with spatial transformation. The final clamping attitude deviation is negative and converted into a rigid body transformation through exponential mapping. The output is a six-degree-of-freedom clamping attitude deviation that includes three rotational angle deviations and three translational deviations.

[0013] Preferably, step S4 constructs a three-row, three-column rotation correction matrix based on the three rotational deviation angles of the six-degree-of-freedom clamping posture deviation, and combines the translational deviations into a translational correction vector. Based on the standard grinding path program, the position coordinates of the tool path control points and the corresponding grinding wheel axis direction vector are extracted line by line, and compensation correction is performed on each control point. The original position coordinates and axis vectors in the standard grinding path program segment are all replaced with the corrected values. The grinding machine drives each motion axis of the machine tool based on the standard grinding path program segment, so that the grinding motion of the grinding wheel relative to the tool compensates for the clamping posture deviation.

[0014] Beneficial effects This invention provides a method for sensing tool clamping posture deviation and correcting grinding path in a knife grinding machine, which has the following beneficial effects: By constructing a reference mesh and performing surface metric flattening evolution, the complex curved surface of the tool is flattened onto a two-dimensional plane. This preserves the inherent geometric structure of the surface while eliminating distortion interference caused by bending. Furthermore, the multi-scale hole life spectrum of the profile is calibrated and the internal transmission relationship matrix is ​​solved, enabling subsequent deviation perception to fully reflect the overall morphological characteristics of the tool profile, rather than relying on the positional information of a few discrete points. Finally, the perceived six-degree-of-freedom clamping posture deviation is converted into a translation correction vector and a rotation correction matrix through rigid body transformation. After extracting the control point information in the standard grinding path program line by line, compensation correction is performed, so that the corrected grinding path can offset the influence of the clamping posture deviation and ensure that the grinding motion of the grinding wheel relative to the tool is consistent with the theoretical trajectory. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating a method for sensing tool clamping posture deviation and correcting grinding path in a sharpening machine according to the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Please see Figure 1 This invention provides a method for sensing tool clamping posture deviation and correcting grinding path in a knife grinding machine, comprising the following steps: S1. Generate a sampling grid based on the geometric information extracted from the computer design model of the tool. Generate a set of profile reference points by reading the spatial coordinates. Obtain the original clamping state point set based on the sensor acquisition of the tool in the clamping state. After filtering the original clamping state point set, form a set of measured points in the clamping state. S1 is based on the computer design model of the tool, and with the tool rotation axis as the reference direction, spiral sampling lines are generated on the surface of the computer design model along the circumferential and axial directions at equal arc length intervals. The three-dimensional coordinates of spatial points are extracted based on the spiral sampling lines to form an ordered array as a set of profile reference points. Thus, through the pre-calibrated rigid body transformation relationship from the tool design coordinate system to the positioning reference coordinate system of the grinding machine fixture, the coordinates of all points in the set of points are uniformly mapped to the clamping evaluation reference system. Based on the line laser profile sensor in the working area of ​​the grinding machine, and after the laser line projected by the line laser profile sensor forms a preset angle with the tool axis, the tool is rotated at a constant speed around the axis under the drive of the workpiece axis, and the line laser profile sensor is translated at a constant speed along the guide rail parallel to the axis. The position of the two-dimensional profile point formed by the reflection of the laser line on the tool surface is obtained by the synchronous triggering of each encoder. After continuously collecting the complete scanning stroke, the original clamping state point set is obtained, and the profile point is reconstructed into a three-dimensional point in the machine tool coordinate system. S1 is to eliminate the interference of sensor noise and point density differences on subsequent deviation perception. The original clamping state point set is filtered. Specifically, the space where the original clamping state point set is located is divided into three-dimensional grid cells. The side length of the grid cell is automatically set according to the density of the original points inside it. That is, in areas with high point density such as the cutting edge, the side length is automatically reduced to retain the sharp geometric features, and in smooth areas of the tool body, the side length is automatically increased to smooth random fluctuations. The arithmetic mean of the three coordinate components of all the original points inside each non-empty grid cell is calculated to form a representative point. All original points in the non-empty grid cells are cleared to form a clamping state measured point set with uniform point density and preserved cutting edge features.

[0018] S2. Construct a reference grid based on the profile reference point set, and perform surface metric flattening evolution on the reference grid. Take the set of planar coordinates corresponding to each node of the evolved flattening metric grid as the reference surface flattening coordinate set. Based on the reference surface flattening coordinate set, calibrate the profile multi-scale hole lifetime spectrum, and solve the internal transmission relationship matrix of the point structure based on the reference surface flattening coordinate set. The S2 method is based on taking each sampling point in the contour reference point set as a node and establishing the connection relationship between the points through the k-nearest neighbor method. Specifically, it calculates the straight-line distance from each node to the other nodes, connects the other nodes with the smallest distance to form an edge, traverses all three-node combinations, and when there are three nodes with edges between each pair, fills the area enclosed by the three edges as a triangular patch to obtain a triangular mesh that seamlessly covers all nodes and has no topological defects as a reference mesh. Because the tool surface has complex curved shapes such as helical grooves and flank faces, directly calculating the distance between points on it will introduce geometric distortion caused by the curvature of the surface. This may lead to errors in the subsequent perception of rigidity deviations caused by clamping posture. Therefore, we need to perform metric flattening on the ideal reference mesh to change the edge length distribution of the mesh, so that the curvature of each local part of the mesh approaches zero, that is, the Gaussian curvature approaches zero, while ensuring that the angular relationship on the mesh remains unchanged. This results in a flat mesh with the same angular structure as the original surface. Based on each node located inside the reference mesh, we find all triangular patches with the node as the vertex. Based on the current length of the three sides in each triangular patch, we calculate the interior angle value of the node's location by solving the relationship between the side length and the included angle. The interior angles of all the triangular faces of the node are summed, and the difference between 2π and the sum is taken as the current Gaussian curvature of the node. If the difference is positive, it means that the surface is convex in the local area of ​​the node; if it is negative, it is concave; and if it is zero, it means that the local area is flat. Set the target Gaussian curvature of each internal node in the reference grid to zero. Based on the topological connectivity and edge length of the reference grid, construct a system of linear equations with the scaling factor correction of each node as unknowns. It should be noted that the coefficients of the system of equations are composed of the cotangent values ​​of the angles of the triangles containing the nodes at both ends of each edge. The constant on the right side of the system of equations is the difference between the current Gaussian curvature and the target curvature of zero. The fine-tuning amount of the scaling factor of each node is obtained by solving the system of linear equations.

[0019] S2 is based on each edge in the reference grid. It finds the two nodes connected by each edge, adds the two current scaling factors, and obtains the edge length scaling factor by calculating the sum of the natural constant e. The product of the edge length scaling factor and the edge length is used as the new length of the corresponding edge. The length of all edges in the entire grid is updated synchronously. After updating the length of each edge of the reference grid, iteratively calculate the current absolute value of Gaussian curvature of all nodes in the current reference grid. If the maximum value of the current absolute value of Gaussian curvature is less than a preset threshold, stop iterating, use the current reference grid as the flatness measurement grid, and use the maximum value of the current absolute value of Gaussian curvature as the reference flatness measurement. The preset threshold in this application is 10⁻⁶.

[0020] S2 selects a node on the flatness metric grid as the origin and assigns it a two-dimensional plane coordinate origin value. It also selects an edge connected to the origin and assigns it coordinates along the horizontal axis. Each triangular facet is processed in breadth-first order. Specifically, for a triangle with known two node plane coordinates and the lengths of its three sides under the flatness metric, the plane coordinates of the third node are calculated by finding the intersection of the two circles. If a node is derived from multiple triangles with different candidate coordinates, the arithmetic mean is taken as the final coordinate. The plane coordinates of all nodes constitute a set of flattened coordinates of the reference surface, and each node is assigned a node area weight. The node area weight of each node is one-third of the sum of the areas of all triangular faces around the node under the flatness metric. The weight is proportional to the local influence range of the node on the surface. Based on the reference surface flattening coordinate set, using the edge length under the reference flatness metric as the path length, a shortest path search algorithm is executed to calculate the shortest path length between any two nodes as the flattened geodesic distance. The flattened geodesic distances between all node pairs are stored in a matrix. Based on the observation scale parameter, which is set to gradually increase from zero, and compared with the flattened geodesic distance, a shape structure including nodes, edges, triangular facets, and tetrahedrons is generated. This application further discloses the process: when the observation scale parameter exceeds the flattened geodesic distance of a node pair, an edge is added between that pair of nodes; when the parameter increases to the point where every pair of three nodes has an edge, a triangular facet is generated; when all triangular facets exist between four nodes, a tetrahedron is generated. As the observation scale parameter continuously increases, the generated shape... The structure grows from isolated nodes into complex connected bodies, forming a multi-scale topological growth sequence. By executing a continuous cohomology algorithm, a multi-scale hole lifetime spectrum of the profile is obtained. The lifetime spectrum generated above is a compact topological feature signature of the ideal shape of the tool that is insensitive to rigid body transformations. Each interval in the spectrum corresponds to the observation scale at which any hole feature on the tool becomes identifiable and at what scale it is filled. The execution process is roughly as follows: by recording the observation time of the first appearance of each closed loop composed of edges, and the final time when the closed loop disappears after being completely filled by triangular facets or higher-dimensional structures, all such observation time-final time pairs are recorded in the form of intervals to obtain a set of intervals, which is the final multi-scale hole lifetime spectrum of the profile.

[0021] S2 divides each node by the sum of the area weights of all nodes in the reference surface flattened coordinate set to obtain the normalized quality of the node, and constructs a reference source distribution with the normalized quality of all nodes. Since the reference source distribution here solves the transmission structure from itself to itself within the point set, the same normalized quality distribution is used as the target distribution for transportation, and the square of the Euclidean distance between any two nodes in the reference surface flattened coordinate is used as the movement cost. After introducing the reference source distribution and the movement cost, S2 solves the transmission relationship in the coordinate set of the flattened reference surface to form an internal transmission relationship matrix.

[0022] Therefore, in the specific implementation process, this application provides a specific solution process as follows: The kernel matrix is ​​constructed based on the movement cost matrix composed of multiple node pairs. That is, for any pair of nodes i and j, the movement cost is taken out, and the natural constant e is calculated to the power of the result. The resulting value is filled into the i-th row and j-th column of the kernel matrix. The significance of this step is that this application transforms the movement cost into the structural affinity between two points. The smaller the movement cost, the higher the affinity, and the matrix element is approximately close to 1. Two auxiliary adjustment arrays are then initialized: the source scaling factor array and the target scaling factor array. The length of both arrays is equal to the total number of nodes, and the initial value of all elements in the arrays is set to the value 1. These two arrays will be used to alternately scale the rows and columns of the kernel matrix. After completing the above construction work, this application enters the iterative solution process: Divide each element of the baseline source distribution array by the corresponding element in the result array obtained by left multiplying the affinity kernel matrix by the current target scaling factor array, and update the corresponding position of the source scaling factor array with the obtained quotient value, so that the sum of the rows of the current transmission scheme approximates the source distribution. Divide each element of the baseline target distribution array by the corresponding element in the result array after left multiplication of the transpose of the affinity kernel matrix and the updated source scaling factor array. Update the corresponding position of the target scaling factor array with the quotient value to make the column sum of the current transmission scheme approximate the target distribution. Repeat these two steps alternately until the maximum value of the change in all elements of the source scaling factor array and the target scaling factor array in the two iterations is less than the preset minimum tolerance. After the iteration stops, for each pair of nodes i and j, the i-th element of the final source scaling factor array, the j-th element of the target scaling factor array, and the i-th row and j-th column element of the affinity kernel matrix are multiplied together. The resulting product is the structural transmission quality from node i to node j. The calculation results of all node pairs are arranged in rows and columns to form the ideal internal transmission relationship matrix.

[0023] S3. Based on the initial clamping posture deviation, it is converted into a rigid body transformation matrix through exponential mapping. Each point in the clamping state measured point set is combined with the rigid body transformation matrix to form the current attempted clamping point set. After completing the spatial transformation, the current attempted clamping point set is repeated after the S2 operation to obtain the current attempted flattened coordinate set. Based on the internal transmission relationship matrix, the structural transmission relationship between the current attempted point set and the reference point set is established, and the multi-scale hole life spectrum of the current tool state is constructed. Based on the difference between the multi-scale hole life spectrum and the profile multi-scale hole life spectrum, the clamping posture deviation energy is obtained, and the six-degree-of-freedom clamping posture deviation of the actual tool clamping relative to the ideal clamping is obtained. The S3 method is based on three rotational deviations and three translational deviations to form the clamping attitude deviation to be determined. The clamping attitude deviation to be determined is represented by a vector containing six values. After setting the initial values ​​of the three rotational deviations and three translational deviations to zero, the clamping attitude deviation to be determined is converted into a four-row, four-column rigid body transformation matrix through exponential mapping. The conversion steps are to construct an antisymmetric matrix from the three rotational values, calculate the matrix exponent of the antisymmetric matrix using the Rodrigues formula to obtain the third-order rotation matrix, and calculate the coefficient matrix in conjunction with the translational values ​​using the antisymmetric matrix and its norm. Finally, the rigid body transformation matrix is ​​formed. By traversing each point in the measured point set of the clamping state, the three-dimensional coordinates are expanded into homogeneous coordinates. This means adding a fourth component with a value of 1 after the original three coordinate values. The homogeneous coordinate column vector is then multiplied by the rigid body transformation matrix, and the first three components of the resulting vector are used as the transformed spatial positions. After repeating this operation on the measured point set of the clamping state, the current attempted clamping point set is obtained. The point set obtained in this application simulates the position that the tool surface point should be in the ideal clamping reference system when the clamping posture deviation is the value represented by the current vector. After repeating the S2 operation on the current set of clamping points, the obtained flat metric grid is flattened onto a two-dimensional plane to obtain the current set of flattened coordinates. This step has been disclosed above and will not be repeated here. The area weights of all nodes in the current attempt flattening coordinate set are normalized to form the attempt source distribution. Based on the benchmark target distribution of the benchmark surface flattening coordinate set, the squared Euclidean distance between each node in the attempt source distribution and each node in the benchmark target distribution in their respective flattening coordinate systems is calculated, and a cross-domain movement cost matrix is ​​constructed. Because in actual cross-domain operations, not only spatial distance but also the inherent structural constraints of the tool shape must be considered, this application provides an internal transmission relationship matrix to correct the cross-domain movement cost matrix. This includes: using the sum of all elements in each row of the attempt source distribution node and the benchmark target distribution node in the internal transmission relationship matrix as the ideal dispersion of the attempt source distribution node, and the sum of all elements in each column as the ideal aggregation of the benchmark target distribution node; obtaining the structural coordination coefficient based on the geometric mean of the ideal dispersion and the ideal aggregation; and dividing the element value of any row and column in the cross-domain movement cost matrix by the structural coordination coefficient to obtain the corrected cross-domain movement cost. The modified movement cost matrix is ​​solved using the transmission relationship to form the current cross-domain transmission relationship matrix.

[0024] S3 merges all nodes of the current attempt flattening coordinate set and all nodes of the reference surface flattening coordinate set into a node set, and constructs a multi-scale topology growth sequence based on the connection weight between any two nodes in the obtained node set. The connection weight is determined by the output of the above steps. Specifically, if both nodes are from the current attempt set, their connection weight is the flattened geodesic distance calculated on the flattened metric grid of the current attempt set; if both nodes are from the reference set, their connection weight is the flattened geodesic distance calculated on the reference flattened metric grid; another case is if one node is from the current attempt set and the other node is from the reference set, the Euclidean distance between the two nodes is calculated, and the connection weight is the ratio of the median of all flattened geodesic distances in the current attempt set to the Euclidean distance. The definition of connection weight in this application makes the connection weight of node pairs with higher transmission quality smaller, and they will be connected earlier in multi-scale topology growth. Repeat the calibration operation of S2 on the obtained multi-scale topological growth sequence to obtain the transmission sensing multi-scale hole lifetime spectrum. Based on the transmission sensing multi-scale hole lifetime spectrum and the profile multi-scale hole lifetime spectrum, each interval is treated as a point on a two-dimensional plane. The projection points of all points onto the diagonal are introduced as an augmented set, that is, the observation time is equal to the final time. Under the condition that the interval point is allowed to be paired with the diagonal point, the sum of the squares of the Euclidean distances between all paired points is minimized, and the square root of the minimum sum is taken as the clamping attitude deviation energy. The above scheme of this application is based on Wasserstein distance.

[0025] S3 calculates the partial derivative of each component of the clamping attitude deviation energy based on the obtained clamping attitude deviation energy using the chain rule, and combines them into a gradient vector. Then, a one-dimensional search is performed in the opposite direction of the gradient vector. Starting from the initial step size, the initial step size is repeatedly halved until two conditions are met simultaneously: First, the deviation energy recalculated after updating the attitude vector with the new step size is lower than the original energy, and the decrease is not less than the square of the gradient norm multiplied by a proportion of the step size. Second, the gradient vector norm at the new parameter is decayed compared with the original gradient norm. The current step size is multiplied by the negative gradient vector as a correction amount, and the clamping attitude deviation value containing three rotational deviations and three translational deviations is updated. The updated clamping attitude deviation is used as the current attempt attitude. The spatial transformation is re-executed on the original clamping state measured point set. This refers to the spatial transformation in step S3. The final clamping attitude deviation is negative and then converted into a rigid body transformation through exponential mapping. The output has a six-degree-of-freedom clamping attitude deviation containing three rotation angle deviations and three translational deviations.

[0026] S4. A translation correction vector is formed based on the six-degree-of-freedom clamping attitude deviation, and compensation correction is performed on each control point after extracting it row by row.

[0027] The S4 constructs a three-row, three-column rotation correction matrix based on the three rotational deviation angles of the six-degree-of-freedom clamping posture deviation, and combines the translational deviations into a translational correction vector. Based on the standard grinding path program, the position coordinates of the tool path control points and the corresponding grinding wheel axis direction vector are extracted line by line, and compensation correction is performed on each control point. The original position coordinates and axis vectors in the standard grinding path program segment are all replaced with the corrected values. The grinding machine drives each motion axis of the machine tool based on the standard grinding path program segment, so that the grinding motion of the grinding wheel relative to the tool compensates for the clamping posture deviation.

[0028] Furthermore, the compensation correction includes: for the position coordinates, first add the corresponding component of the translation correction vector to each of the three coordinate components to obtain the translation-corrected intermediate position coordinates; then multiply the column vector formed by the intermediate position coordinates by the rotation correction matrix on the left, and the three components of the resulting column vector are the final corrected tool position; for the grinding wheel axis direction vector, directly multiply the column vector formed by the rotation correction matrix on the left to obtain the corrected direction vector, and scale it to a unit length.

[0029] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0030] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0031] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for sensing tool clamping posture deviation and correcting grinding path in a knife grinding machine, characterized in that: The specific steps include the following: S1. Generate a sampling grid based on the geometric information extracted from the computer design model of the tool. Generate a set of profile reference points by reading the spatial coordinates. Obtain the original clamping state point set based on the sensor acquisition of the tool in the clamping state. After filtering the original clamping state point set, form a set of measured points in the clamping state. S2. Construct a reference grid based on the profile reference point set, and perform surface metric flattening evolution on the reference grid. Take the set of planar coordinates corresponding to each node of the evolved flattening metric grid as the reference surface flattening coordinate set. Based on the reference surface flattening coordinate set, calibrate the profile multi-scale hole lifetime spectrum, and solve the internal transmission relationship matrix of the point structure based on the reference surface flattening coordinate set. S3. Based on the initial clamping posture deviation, it is converted into a rigid body transformation matrix through exponential mapping. Each point in the clamping state measured point set is combined with the rigid body transformation matrix to form the current attempted clamping point set. After completing the spatial transformation, the current attempted clamping point set is repeated after the S2 operation to obtain the current attempted flattened coordinate set. Based on the internal transmission relationship matrix, the structural transmission relationship between the current attempted point set and the reference point set is established, and the multi-scale hole life spectrum of the current tool state is constructed. Based on the difference between the multi-scale hole life spectrum and the profile multi-scale hole life spectrum, the clamping posture deviation energy is obtained, and the six-degree-of-freedom clamping posture deviation of the actual tool clamping relative to the ideal clamping is obtained. S4. A translation correction vector is formed based on the six-degree-of-freedom clamping attitude deviation, and compensation correction is performed on each control point after extracting it row by row.

2. The method for sensing tool clamping posture deviation and correcting grinding path in a sharpening machine according to claim 1, characterized in that: S1 is based on the tool computer design model, and with the tool rotation axis as the reference direction, spiral sampling lines are generated on the surface of the tool computer design model along the circumferential and axial directions at equal arc length intervals. The three-dimensional coordinates of spatial points are extracted based on the spiral sampling lines to form an ordered array as a set of profile reference points. Based on the line laser profile sensor in the working area of ​​the grinding machine, and after the laser line projected by the line laser profile sensor forms a preset angle with the tool axis, the tool is rotated at a constant speed around the axis under the drive of the workpiece axis, and the line laser profile sensor is translated at a constant speed along the guide rail parallel to the axis. The position of the two-dimensional profile point formed by the reflection of the laser line on the tool surface is obtained by the synchronous triggering of each encoder. After continuously collecting the complete scanning stroke, the original clamping state point set is obtained. The S1 filter is performed on the original clamping state point set. Specifically, the space where the original clamping state point set is located is divided into three-dimensional grid cells. The arithmetic mean of the three coordinate components of all original points located in each non-empty grid cell is calculated to form a representative point. All original points in the non-empty grid cells are then cleared to form the clamping state measured point set.

3. A method for sensing tool clamping posture deviation and correcting grinding path in a sharpening machine according to claim 2, characterized in that: The S2 is based on taking each sampling point in the contour reference point set as a node and establishing the connection relationship between the points through the k nearest neighbor method. Specifically, it calculates the straight-line distance from each node to the other nodes, connects the other nodes with the smallest distance to form an edge, traverses all three-node combinations, and when there are three nodes with edges between each pair, fills the area enclosed by the three edges as a triangular facet to obtain the triangular mesh as the reference mesh. Based on each node located inside the baseline grid, find all triangular faces with the node as the vertex. Based on the current length of the three sides of each triangular face, and through the relationship between the side length and the included angle, calculate the interior angle value of the node's location. The interior angles of all the triangular faces of the node are summed, and the difference between 2π and the sum is used as the current Gaussian curvature of the node. Set the target Gaussian curvature of each node located inside the reference grid to zero. Based on the topological connectivity and edge length of the reference grid, construct a system of linear equations with the scaling factor correction of each node as unknowns. Solve the system of linear equations to obtain the fine-tuning amount of the scaling factor of each node.

4. A method for sensing tool clamping posture deviation and correcting grinding path in a sharpening machine according to claim 3, characterized in that: S2 is based on each edge in the reference grid, finds the two nodes connected by each edge, adds the two current scaling factors, obtains the edge length scaling factor by calculating the sum of the natural constant e, and uses the product of the edge length scaling factor and the edge length as the new length of the corresponding edge. After updating the length of each edge of the reference grid, iteratively calculate the current absolute value of Gaussian curvature of all nodes in the current reference grid. If the maximum value of the current absolute value of Gaussian curvature is less than a preset threshold, stop iterating, use the current reference grid as the flatness metric grid, and use the maximum value of the current absolute value of Gaussian curvature as the reference flatness metric.

5. A method for sensing tool clamping posture deviation and correcting grinding path in a sharpening machine according to claim 1, characterized in that: S2 is based on selecting any node on the flat metric grid as the origin and assigning it the origin value of the two-dimensional plane coordinates. It also selects an edge connected to the origin and assigns it coordinates along the horizontal axis. Each triangular facet is processed one by one in breadth-first order. The plane coordinates of all nodes constitute a reference surface flattening coordinate set, and the node area weight of each node is assigned. Based on the reference surface flattening coordinate set, the edge length under the reference flatness metric is used as the path length. The shortest path search algorithm is executed to calculate the shortest path length between any two nodes as the flattened geodesic distance. The flattened geodesic distance between all node pairs is stored in a matrix. Based on the observation scale parameter that is set to gradually increase from zero and compared with the flattened geodesic distance, a shape structure including nodes, edges, triangular facets and tetrahedrons is generated to form a multi-scale topological growth sequence. The multi-scale porosity spectrum of the profile is obtained by executing the continuous homology algorithm.

6. The method for sensing tool clamping posture deviation and correcting grinding path in a sharpening machine according to claim 1, characterized in that: The S2 divides each node by the sum of the area weights of all nodes in the reference surface flattened coordinate set to obtain the normalized quality of the node, and constructs a reference source distribution based on the normalized quality of all nodes, and uses the square of the Euclidean distance between any two nodes in the reference surface flattened coordinate as the movement cost. After introducing the reference source distribution and the movement cost, S2 solves the transmission relationship in the coordinate set of the flattened reference surface to form an internal transmission relationship matrix.

7. The method for sensing tool clamping posture deviation and correcting grinding path in a sharpening machine according to claim 6, characterized in that: The S3 is based on three rotational deviations and three translational deviations to form the clamping attitude deviation to be determined, and the initial values ​​of the three rotational deviations and three translational deviations are all set to zero. The clamping attitude deviation to be determined is converted into a rigid body transformation matrix through exponential mapping. By traversing each point in the set of measured points in the clamping state, the three-dimensional coordinates are expanded into homogeneous coordinates. The rigid body transformation matrix is ​​then multiplied by the column vector of the homogeneous coordinates. The first three components of the resulting vector are used as the transformed spatial positions. This process is repeated on the set of measured points in the clamping state to obtain the current set of clamping points. After repeating the S2 operation on the current set of clamping points, the obtained flat metric grid is flattened onto a two-dimensional plane to obtain the current set of flattened coordinates. The area weights of all nodes in the current attempt flattened coordinate set are normalized to form the attempt source distribution. Based on the benchmark target distribution of the benchmark surface flattened coordinate set, the squared Euclidean distance between each node in the attempt source distribution and each node in the benchmark target distribution in their respective flattened coordinate systems is calculated to construct a cross-domain movement cost matrix. The sum of all elements in each row of the internal transmission relationship matrix between the attempt source distribution nodes and the benchmark target distribution nodes is used as the ideal dispersion of the attempt source distribution nodes, and the sum of all elements in each column is used as the ideal aggregation of the benchmark target distribution nodes. The structural coordination coefficient is obtained based on the geometric mean of the ideal dispersion and the ideal aggregation. The element value of any row and column in the cross-domain movement cost matrix is ​​divided by the structural coordination coefficient to obtain the corrected cross-domain movement cost. The modified movement cost matrix is ​​solved using the transmission relationship to form the current cross-domain transmission relationship matrix.

8. The method for sensing tool clamping posture deviation and correcting grinding path in a sharpening machine according to claim 7, characterized in that: S3 merges all nodes of the current attempted flattening coordinate set and all nodes of the reference surface flattening coordinate set into a node set, and constructs a multi-scale topology growth sequence based on the connection weight between any two nodes in the obtained node set. Repeat the calibration operation of S2 on the obtained multi-scale topological growth sequence to obtain the transmission sensing multi-scale hole lifetime spectrum. Based on the transmission sensing multi-scale hole lifetime spectrum and the profile multi-scale hole lifetime spectrum, each interval is treated as a point on a two-dimensional plane. The projection points of all points to the diagonal are introduced as an augmented set. Under the condition that the interval points are allowed to be paired with the diagonal points, the sum of the squares of the Euclidean distances between all paired points is minimized, and the square root of the minimum sum is taken as the clamping attitude deviation energy.

9. The method for sensing tool clamping posture deviation and correcting grinding path in a sharpening machine according to claim 1, characterized in that: S3 calculates the partial derivative of each component of the clamping attitude deviation energy based on the obtained clamping attitude deviation energy using the chain rule, and combines them into a gradient vector. And perform a one-dimensional search based on the opposite direction of the gradient vector. Starting from the initial step size, repeatedly halve the initial step size, multiply the current step size by the negative gradient vector as the correction amount, and update the clamping attitude deviation value containing three rotational deviations and three translational deviations. The updated clamping attitude deviation is used as the current attempted attitude. The spatial transformation is re-executed on the original clamping state measured point set. The final clamping attitude deviation is negative and converted into a rigid body transformation through exponential mapping. The output has a six-degree-of-freedom clamping attitude deviation including three rotation angle deviations and three translational deviations.

10. The method for sensing tool clamping posture deviation and correcting grinding path in a sharpening machine according to claim 1, characterized in that: The S4 constructs a rotation correction matrix based on the three rotational deviation angles of the six-degree-of-freedom clamping posture deviation, and combines the translational deviation into a translation correction vector. Based on the standard grinding path program, it extracts the position coordinates of the tool path control points and the corresponding grinding wheel axis direction vector line by line, and performs compensation correction for each control point. It also replaces all the original position coordinates and axis vectors in the standard grinding path program segment with the corrected values. The grinding machine drives each motion axis of the machine tool based on the standard grinding path program segment, so that the grinding motion of the grinding wheel relative to the tool compensates for the clamping posture deviation.

Images