A joint positioning method and system for a fine yarn joint robot

Abstract

The present application relates to the technical fields of intelligent manufacturing and textile automation, and especially relates to a splicing positioning method and system for a spun yarn splicing robot, which comprises the following steps: adjusting the parallelism of the spun yarn splicing robot and the spun yarn machine table to obtain a parallel state of the pose of the spun yarn splicing robot; obtaining yarn breakage position information; under the parallel state of the pose, collecting point cloud data and image information by the spun yarn splicing robot; obtaining an axis direction, an axis position point and a cylinder radius according to the point cloud data to determine the axis position of the spool rod; determining a bottom reference position according to the image information, and combining the axis position of the spool rod to realize the splicing positioning of the spun yarn splicing robot. The present application overcomes the parking error and high reflection interference of the spun yarn splicing robot, stably estimates the axis of the spool rod under incomplete point cloud, and accurately positions the bottom reference by fusing image information, thereby improving the reliability and accuracy of splicing.

Description

Technical Field

[0001] This invention relates to the fields of intelligent manufacturing and textile automation technology, and in particular to a splice positioning method and system for yarn splicing robots. Background Technology

[0002] Currently, the automatic splicing robots in spinning workshops mainly include two types: track-mounted splicing robots and traveling splicing robots.

[0003] Track-mounted splicing robots typically run along fixed tracks on spinning machines. The robot body reaches the target spindle position through track guidance, mechanical limits, or fixed station calibration. This type of solution offers high positioning accuracy, and the splicing actuator can complete the splicing operation according to a pre-taught fixed trajectory. However, track-mounted solutions require the addition of dedicated tracks, support structures, and power and communication mechanisms around the spinning machine, resulting in significant equipment modifications and high installation costs. Furthermore, they usually require one or more rows of machines to be equipped with dedicated track-mounted robots, making them difficult to adapt to the application requirements of traditional workshops with multiple machines, multiple channels, and low modification costs.

[0004] To reduce modification costs, existing technologies have gradually developed into itinerant automated splicing robots based on Automated Guided Vehicle (AGV) chassis. These robots typically reach the vicinity of the yarn breakage station via QR code, magnetic navigation, laser navigation, or visual navigation, and then a robotic arm or splicing actuator approaches the target component along a preset trajectory. Some solutions use cameras to identify the yarn breakage area, roller area, or spindle bar area, and perform simple corrections to the robot's end effector based on the target position in the image. This approach can reduce track modification costs to some extent and improve the robot's reusability across multiple spinning machines.

[0005] However, existing roving splicing robots still have significant positioning errors. After the AGV reaches the target spindle, it can usually only achieve macroscopic stopping, and its lateral, longitudinal, and posture errors are difficult to completely eliminate. Due to the confined working space of the spinning machine splicing operation, and the small size and close proximity of components such as the spindle bar, rollers, traveler, and yarn guide hook, relying solely on the AGV's stopping accuracy and the robotic arm's fixed teaching trajectory can easily lead to deviations between the end effector and the target components, thus affecting the reliability of subsequent yarn suction, threading, and splicing actions.

[0006] Existing visual positioning solutions also utilize the spindle rod as a spatial reference for localization. The common approach involves extracting the spindle rod's edges using a 2D camera, or acquiring a 3D point cloud of the spindle rod using a Red Green Blue-Depth (RGB-D) camera, a binocular camera, or a structured light camera. Then, methods such as cylindrical fitting, least squares fitting, Random Sample Consensus (RANSAC) fitting, or principal component analysis are used to estimate the spindle rod's axis, and a robot operation reference coordinate system is established based on the axis position. However, spinning machine spindle rods are mostly made of metal, which, under the influence of workshop lighting, machine reflections, and oil stains, easily creates highly reflective areas. This results in false edges in the 2D image, and bright reflective areas are easily misidentified as the true contour. Simultaneously, the presence of fly waste, yarn occlusion, and background structural interference within the workshop leads to unstable visual features of the target area.

[0007] Furthermore, since cameras typically only observe the spindle bar from one side, the resulting 3D point cloud is often not a complete cylindrical surface, but rather a localized curved or incomplete surface. For such non-closed point clouds, traditional direct least-squares cylindrical fitting is easily affected by reflective noise, missing edge points, and outliers, leading to unstable estimations of the axis direction, axis position, and radius. Some existing solutions require adding a black background or masking structure behind the spindle bar to enhance contrast, but this approach requires additional modifications to the spinning machine, resulting in high implementation costs and hindering large-scale adoption in traditional workshops.

[0008] Therefore, the existing technology lacks a positioning method suitable for AGV-type splicing robots in spinning workshops that can stably extract the spindle axis and bottom reference point under conditions such as AGV parking error, high metal reflectivity, fly waste interference, incomplete point cloud and narrow working space, and convert them into the positioning reference required for robot splicing operations. Summary of the Invention

[0009] To address the shortcomings of existing technologies, this invention provides a splice positioning method and system for spinning splice robots, which is particularly suitable for automated splicing operations in spinning machine workshops and other environments where component positioning is required.

[0010] To achieve the above objectives, in a first aspect, the present invention provides a splice positioning method for a yarn splicing robot, the method comprising the following steps: adjusting the parallelism between the yarn splicing robot and the yarn spinning machine to obtain a parallel pose state of the yarn splicing robot; acquiring yarn breakage location information; in the parallel pose state, collecting point cloud data and image information through the yarn splicing robot; obtaining the axial direction, axial position point, and cylinder radius based on the point cloud data to determine the spindle rod axial position; determining the bottom reference position based on the image information, and combining the spindle rod axial position to achieve splice positioning of the yarn splicing robot.

[0011] Optionally, adjusting the parallelism between the yarn splicing robot and the yarn spinning machine to obtain a parallel pose state of the yarn splicing robot includes: adjusting the parallelism using a laser displacement sensor so that the yarn splicing robot and the yarn spinning machine remain parallel to obtain the parallel pose state.

[0012] Optionally, the step of acquiring the yarn breakage location information, in the parallel pose state, involves collecting point cloud data and image information through the yarn splicing robot, including: determining the yarn breakage location information based on a preset teaching position and the yarn breakage dataset; and acquiring the point cloud data and image information based on the yarn breakage location information using a data acquisition device carried by the robotic arm of the yarn splicing robot.

[0013] Optionally, the step of obtaining the axis direction, axis position point, and cylinder radius based on the point cloud data to determine the spindle rod axis position includes: defining a spindle rod point cloud set based on the point cloud data; centering the spindle rod point cloud set to obtain an initial axis; constructing a set of parameters to be optimized based on the initial axis using the current axis direction, current axis position point, and current cylinder radius; calculating the radial distance and calculating the radial residual based on the radial distance; constructing a robust objective function based on the radial residual; iteratively updating the set of parameters to be optimized using the robust objective function to obtain parameter increments; constructing a change criterion for the objective function, and determining the axis direction, axis position point, and cylinder radius in combination with the parameter increments to obtain the spindle rod axis position.

[0014] Optionally, the step of centering the point cloud set of the spindle rods to obtain the initial axis includes: obtaining the center of the point cloud as a reference position based on the point cloud set of the spindle rods, and translating the point cloud data to obtain a center point; constructing a spatial distribution matrix using the center point to determine the direction of the initial axis; projecting the center of the point cloud onto a plane perpendicular to the direction of the initial axis to obtain the position point of the initial axis; obtaining the initial axis distance of each point in the point cloud data, and using the median of the initial axis distance as the initial value of the cylinder radius; and determining the initial axis by combining the initial axis direction, the initial axis position point, and the initial value of the cylinder radius.

[0015] Optionally, calculating the radial distance and calculating the radial residual based on the radial distance includes: calculating the relative vector from each point in the point cloud data to the current axis position point, and calculating the projection length of the relative vector in the direction of the current axis; removing the axial component in the relative vector based on the projection length to obtain a radial vector perpendicular to the direction of the current axis; obtaining the magnitude of the radial vector as the radial distance; and using the difference between the radial distance and the current cylinder radius as the radial residual.

[0016] Optionally, the step of iteratively updating the set of parameters to be optimized using the robust objective function to obtain the parameter increment includes: merging the increments of the current axis direction, the current axis position point, and the current cylinder radius to obtain an increment vector; forming a residual vector from the radial residuals of each point in the point cloud data; calculating the partial derivative matrix of the residual vector with respect to the local parameters; constructing a diagonal matrix based on the optimization weights of each point in the point cloud data; and solving the increment vector using the least squares equation by combining the residual vector, the partial derivative matrix, and the diagonal matrix to obtain the parameter increment.

[0017] Optionally, determining the bottom reference position based on the image information includes: preprocessing the image information to obtain a feature tensor, and mapping the feature tensor to obtain a feature matrix; defining a semantic subspace for the feature space and setting an orthogonal projection basis; calculating a soft-assignment probability vector for the visual feature blocks in the feature matrix belonging to the semantic subspace in conjunction with the orthogonal projection basis; aggregating the soft-assignment probability vectors to obtain a global probability weight vector for the semantic subspace; constructing a diagonal feature scaling matrix, and updating the visual feature blocks in conjunction with the orthogonal projection basis and the global probability weight vector to determine the bottom reference position.

[0018] Optionally, the step of combining the spindle rod axis position to realize the splice positioning of the yarn splicing robot includes: converting the spindle rod axis position and the bottom reference position into the working positioning reference of the yarn splicing robot to realize the splice positioning.

[0019] Secondly, this invention provides a splice positioning system for a yarn splicing robot. The system executes the splice positioning method for the yarn splicing robot provided by this invention. The system includes an input device, an output device, a processor, and a memory, which are interconnected. The memory stores a computer program, which includes program instructions, and the processor is configured to invoke the program instructions. This invention, through high-performance hardware collaboration, ensures stable execution of the method and facilitates business deployment and expansion.

[0020] Compared with the prior art, the present invention has at least the following beneficial effects: First, compared with traditional two-dimensional visual edge localization methods, this application introduces a reflection suppression visual feature processing mechanism to address the high reflectivity problem of metal spindle rods in spinning workshops. This mechanism distinguishes and models the real spindle rod edge, metal reflection interference, and environmental fly hair interference, and suppresses abnormally bright features through statistical feature weights, thereby reducing the probability of reflective areas being misjudged as real edges and improving the stability of spindle rod edge extraction under complex lighting conditions.

[0021] Second, compared with traditional direct cylindrical fitting or least squares fitting methods, this application addresses the problems of non-closed point clouds, local missing points, and excessive noise in the spindle rod point cloud under a single-sided view by employing constrained axis parameterization and robust radial residual optimization methods to estimate the spindle rod axis. This method can reduce the influence of outliers and incomplete arc surfaces on the axis estimation results, making the estimation of the spindle rod axis direction, axis position, and radius more stable.

[0022] Third, compared with methods that rely solely on 3D point clouds to determine the bottom point, this application uses 3D point clouds to estimate the spindle axis and combines this with relatively complete edge and bottom features in a 2D image to determine the bottom reference point. This 2D-3D fusion method can compensate for the shortcomings of RGB-D point clouds, which are prone to missing or having high noise at the bottom edge, thus obtaining a positioning reference that is more suitable for use as a reference for splicing operations.

[0023] Fourth, compared with traditional Transformer-based global self-attention visual processing methods, the reflection suppression module in this application does not need to calculate the pairwise correlations between all visual feature blocks. Instead, it calculates feature weights through subspace projection and second-order statistics. As the number of visual feature blocks increases, this method can reduce computational overhead and is more suitable for deployment in industrial control computers or edge computing devices used in splicing robots in spinning workshops. Attached Figure Description

[0024] Figure 1 This is a flowchart of a splice positioning method for a yarn splicing robot according to an embodiment of the present invention; Figure 2 This is a schematic diagram of parallelism adjustment according to an embodiment of the present invention; Figure 3 This is a framework diagram of a splice positioning system for a yarn splicing robot according to an embodiment of the present invention. Detailed Implementation

[0025] Specific embodiments of the present invention will now be described in detail. It should be noted that the embodiments described herein are for illustrative purposes only and are not intended to limit the invention. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that these specific details are not necessary to practice the invention. In other instances, well-known circuits, software, or methods have not been specifically described to avoid obscuring the invention.

[0026] Throughout this specification, references to "an embodiment," "an embodiment," "an example," or "an example" mean that a particular feature, structure, or characteristic described in connection with that embodiment or example is included in at least one embodiment of the invention. Therefore, the phrases "in an embodiment," "in an embodiment," "an example," or "an example" appearing in various places throughout the specification do not necessarily refer to the same embodiment or example. Furthermore, specific features, structures, or characteristics can be combined in one or more embodiments or examples in any suitable combination and / or sub-combination. Moreover, those skilled in the art will understand that the illustrations provided herein are for illustrative purposes and are not necessarily drawn to scale.

[0027] Please see Figure 1 One embodiment of the present invention provides a splice positioning method for a yarn splicing robot, the method comprising the following steps: S1. Adjust the parallelism between the yarn splicing robot and the yarn spinning machine to obtain the parallel position of the yarn splicing robot.

[0028] In this embodiment, when the traveling yarn splicing robot stops at the yarn breakage position, based on the laser displacement sensor, the parallelism between the yarn splicing robot (AGV) and the spinning machine is corrected by point laser in conjunction with the yarn splicing robot, so that the entire robot is parallel to the spinning machine, thereby obtaining the parallel position state of the yarn splicing robot.

[0029] In an optional embodiment, the parallelism between the spinning splicing robot and the spinning machine can be corrected by a dedicated leveling mechanism.

[0030] Please see Figure 2 The diagram illustrates parallelism adjustment. The upper rectangle represents the spinning machine, while the lower, tilted rectangle represents the spinning splicing robot. Points 1 and 2 are laser displacement sensors, and the dashed line indicates the parallelism between the spinning splicing robot and the spinning machine. When the traveling spinning splicing robot stops at the yarn breakage position on the spinning machine, it uses laser displacement sensors 1 and 2 to collect position data. This data, combined with an AGV or a dedicated leveling mechanism, corrects the parallelism of the robot relative to the spinning machine, adjusting the tilted robot to a parallel state with the spinning machine.

[0031] S2. Obtain the yarn breakage location information. In the parallel posture state, collect point cloud data and image information through the yarn splicing robot.

[0032] In this embodiment, the robotic arm or end effector of the yarn splicing robot carries a data acquisition device (RGB-D camera) to a preset area to determine the specific yarn breakage location information and acquire point cloud data and image information.

[0033] First, the robotic arm or end effector is adjusted to a preset teaching position. This teaching position is determined by the AGV stopping at the working spindle position during testing, ensuring parallelism with the machine. Using the standard spindle position as the teaching object, the robotic arm is controlled to move, carrying the RGB-D camera, so that the target yarn breakage area is within the camera's effective field of view, and the distance between the camera and the target is within the camera's effective measurement range. The pose of the robotic arm's end effector relative to the robot's base coordinate system in this state is recorded as the preset teaching pose. In actual deployment, the robotic arm or end effector acquires information from this preset teaching position.

[0034] Secondly, the location of the yarn breakage is located by using a pre-deployed yarn breakage dataset: by comparing the yarn breakage location with the actual dataset, the left and right deviation of the yarn splicing robot relative to the spinning machine is determined. The robotic arm carries the camera and moves left and right. By recognizing the position of the frame, the yarn breakage location is placed in the center of the field of view, correcting the left and right parking error of the AGV, and then it moves vertically down to the bottom of the spindle rod.

[0035] Finally, point cloud data and image information were acquired: point cloud data of the bottom of the spindle rod was acquired using an RGB-D camera and transmitted via USB. Since the acquired spindle rod point cloud contained irrelevant background points, the background points of the workpiece were removed using the Open3D function to extract target features. To facilitate subsequent rapid calculations, yarn or other noise points in the 3D point cloud were downsampled and filtered. RGB images were then acquired using the RGB-D camera as image information.

[0036] S3. Based on the point cloud data, obtain the axis direction, axis position point and cylinder radius to determine the axis position of the spindle rod.

[0037] In this embodiment, since the point cloud data acquired at the bottom of the spindle is a non-closed cylindrical point cloud, the point cloud data at the bottom of the spindle is processed to determine the corresponding spindle axis position. Addressing the issue that RGB-D cameras can only acquire spindle point clouds from one side, resulting in point clouds exhibiting non-closed arc surfaces, partial gaps, and mixed reflective noise, an axis reconstruction method based on radial consistency constraints is proposed. This method does not rely on a complete cylindrical cross-section or end face contour, but utilizes the geometric characteristic that the radial distance from the incomplete arc surface points to the true central axis should remain consistent. The spindle axis is represented as a constrained unit direction, orthogonal axis points, and radius parameters. Robust weighted iterative optimization is used to jointly solve for the axis position, axis direction, and radius. This method can stably recover the spindle central axis under conditions of single-sided observation, incomplete point clouds, and strong noise interference, providing a reliable spatial reference for subsequent bottom reference point determination and robot joint positioning. The specific process of the axis reconstruction method is as follows: S31. Define a spindle rod point cloud set based on the point cloud data.

[0038] In this embodiment, the definition of the spindle rod point cloud set satisfies the following relationship:

[0039] in, For the collection of points on the spindle rod, For the first A three-dimensional point, For point index, For three-dimensional space, The number of points.

[0040] Specifically, three-dimensional points satisfy the following relationship:

[0041] in, For the first A three-dimensional point, For the first points Coordinate components, For the first points Coordinate components, For the first points Coordinate components, This is a transpose.

[0042] S32. The initial axis is obtained by centering the set of points on the yarn spindle rods.

[0043] First, average all points in the point cloud set of the spindle rods to obtain the spatial center of the point cloud, which satisfies the following relationship:

[0044] in, For the center of the point cloud, For the number of points, For point index, For the first A three-dimensional point.

[0045] Secondly, using the center of the point cloud as a reference position, the entire point cloud data is translated to obtain the centralized coordinates of the centralized point, satisfying the following relationship:

[0046] in, For the first Centralized coordinates of each point For the first A three-dimensional point, The center of the point cloud.

[0047] Next, a spatial distribution matrix is ​​constructed using the centralized points to describe the degree of dispersion of the acquired spindle point cloud data in different directions, satisfying the following relationship:

[0048] in, It is a spatial distribution matrix. For the number of points, For point index, For the first Centralized coordinates of each point It is the outer product of vectors.

[0049] Subsequently, based on the spatial distribution matrix, the direction that maximizes the projection variance is selected from the set of unit vectors as the initial axis direction of the spindle cylinder axis, satisfying the following relationship:

[0050] in, The initial axis direction, For candidate unit directions, It is a 2-norm. This is the spatial distribution matrix.

[0051] It should be noted that the candidate unit direction is any direction in three-dimensional space that satisfies... The direction vector.

[0052] Then, the center of the point cloud is projected onto a plane perpendicular to the initial axis direction to obtain the initial axis position point, satisfying the following relationship:

[0053] in, The initial axis position point, It is a third-order identity matrix. The initial axis direction, The center of the point cloud.

[0054] Finally, the initial axis distance from each point in the point cloud data to the initial axis direction is calculated, and the median is taken as the initial value of the cylindrical radius of the spindle rod, satisfying the following relationship:

[0055] in, Let the initial value be the radius of the cylinder. This represents the median. For point index, It is a third-order identity matrix. The initial axis direction, For the first A three-dimensional point, This is the initial axis position point.

[0056] In this embodiment, the initial axis is determined by combining the initial axis direction, the initial axis position point, and the initial value of the cylinder radius.

[0057] S33. Based on the initial axis, construct a set of parameters to be optimized using the current axis direction, the current axis position point, and the current cylinder radius.

[0058] In this embodiment, the initial axis of the spindle is constrained, including the following steps: First, restrict the current axis direction to a unit vector to avoid the length of the direction vector affecting distance calculation, satisfying the following relationship:

[0059] in, This indicates the current axis direction.

[0060] Second, define the current axis position point as the point on the current axis that is closest to the origin. At this point, the point is orthogonal to the direction of the current axis and satisfies the following relationship:

[0061] in, This is the current position point on the axis. This indicates the current axis direction.

[0062] Third, construct a three-dimensional orthogonal basis consisting of two horizontal basis vectors and the current axis direction as the orthogonal basis matrix, satisfying the following relationship:

[0063] in, It is an orthogonal basis matrix. This represents two unit basis vectors perpendicular to the current axis direction. This indicates the current axis direction.

[0064] Fourth, the three directions in the orthogonal basis matrix are mutually perpendicular and have a length of 1, satisfying the following relationship:

[0065] in, It is an orthogonal basis matrix. It is a third-order identity matrix.

[0066] Fifth, represent the current axis position point with two unit basis vectors perpendicular to the current axis direction, so that the current axis position point automatically satisfies the constraint of being orthogonal to the current axis direction, satisfying the following relationship:

[0067] in, This is the current position point on the axis. These are the two coordinate parameters of the current axis position point in the horizontal plane. Represents two unit basis vectors perpendicular to the direction of the current axis.

[0068] It should be noted that the coordinate parameters are first obtained from the projection of the point cloud center onto the axis normal plane, and then iteratively updated as variables to be optimized in the subsequent radial residual minimization process.

[0069] Sixth, unify the current axis direction, current axis position point, and current cylinder radius into a set of parameters to be optimized, satisfying the following relationship:

[0070] in, Let be the set of parameters to be optimized for the cylinder. It is an orthogonal basis matrix. These are the two coordinate parameters of the current axis position point in the horizontal plane. This represents the current cylinder radius.

[0071] S34. Calculate the radial distance and calculate the radial residual based on the radial distance.

[0072] First, calculate the relative vectors of each point in the point cloud data to the current axis position point, satisfying the following relationship:

[0073] in, For the first The relative vector from each point to the current axis position point For the first A three-dimensional point, This is the current position point on the axis.

[0074] Secondly, calculate the projection length of the relative vector along the current axis direction, satisfying the following relationship:

[0075] in, For the first The projection length of each point along the current axis direction. The current axis direction, For the first The relative vector from each point to the current axis position.

[0076] Next, the axial component of the relative vector is removed to obtain the radial vector perpendicular to the current axis direction, satisfying the following relationship:

[0077] in, For the first radial vector of points For the first The relative vector from each point to the current axis position point For the first The projection length of each point along the current axis direction. This indicates the current axis direction.

[0078] Subsequently, the shortest distance from each point in the point cloud data to the current axis is represented by the magnitude of the radial vector. The shortest distance is then used as the radial distance, satisfying the following relationship:

[0079] in, For the first The radial distance between points For the first The radial vector of each point.

[0080] Then, the radial distance is expressed in equivalent square root form for easier algorithm implementation, satisfying the following relationship:

[0081] in, For the first The radial distance between points For the first The relative vector from each point to the current axis position point For the first The projection length of each point along the current axis direction.

[0082] Finally, the difference between the radial distance and the current cylinder radius is used to represent the deviation of the point relative to the cylinder surface to obtain the radial residual, which satisfies the following relationship:

[0083] in, For the first The radial residual at each point, For the first The radial distance between points This represents the current cylinder radius.

[0084] S35. Construct a robust objective function based on the radial residual.

[0085] In this embodiment, when the radial residual does not exceed the threshold, the squared error is used to maintain the fitting accuracy, satisfying the following relationship:

[0086] in, For loss function, For the first The radial residual at each point, The threshold value is used.

[0087] It should be noted that the threshold is determined by the median absolute deviation of the radial residuals under the current model, and is obtained by multiplying the median by a set coefficient of 1.3.

[0088] In this embodiment, when the radial residual exceeds the threshold, a linear growth method is used to suppress the influence of outliers, satisfying the following relationship:

[0089] in, For loss function, For the first The radial residual at each point, The threshold value is used.

[0090] In this embodiment, for trusted points, their full weights are retained, satisfying the following relationship:

[0091] in, For the first Optimization weights for each point For the first The radial residual at each point, The threshold value is used.

[0092] In this embodiment, for large residual points, their weights are reduced according to the magnitude of the radial residuals, satisfying the following relationship:

[0093] in, For the first Optimization weights for each point For the threshold, For the first The radial residuals at each point.

[0094] It should be noted that, The smaller the value, the weaker the influence of that point on the fitting result.

[0095] Furthermore, the weighted robust residuals of all points are summed to form the robust objective function for fitting the spindle bar, which satisfies the following relationship:

[0096] in, For the robust objective function, For the number of points, For point index, For the first Optimization weights for each point For loss function, For the first The radial residuals at each point.

[0097] In this embodiment, the cylinder parameters that minimize the robust objective function are determined, satisfying the following relationship:

[0098] in, For the optimized cylinder parameters, Let be the set of parameters to be optimized for the cylinder. The objective function is robust.

[0099] S36. The parameter increment is obtained by iteratively updating the set of parameters to be optimized using the robust objective function.

[0100] First, the increments of the current axis direction, current axis position, and current cylinder radius in one iteration are combined to obtain the increment vector, which satisfies the following relationship:

[0101] in, For incremental vectors, The increment is in the axial direction. The increment of the axis position point, This is the increment of the cylinder's radius.

[0102] Second, the radial residuals of each point in the point cloud data are used to form a column vector as the residual vector, satisfying the following relationship:

[0103] in, For the residual vector, For the first The radial residual at each point, For the first The radial residual at each point, For the first The radial residuals at each point.

[0104] Third, the Jacobian matrix of the residual vector with respect to the local parameter vector is calculated as the partial derivative matrix, satisfying the following relationship:

[0105] in, It is a partial derivative matrix. For the residual vector, This is a local parameter vector.

[0106] Fourth, the optimization weights for each point are written in diagonal matrix form for easier matrix solving, satisfying the following relationship:

[0107] in, It is a diagonal matrix. This indicates the construction of a diagonal matrix. For the first Optimization weights for each point For the first Optimization weights for each point For the first Optimize the weights of each point.

[0108] Fifth, the parameter increment is obtained by solving the least squares equation with damping terms for the increment vector, thereby improving the iterative stability and satisfying the following relationship:

[0109] in, It is a partial derivative matrix. It is a diagonal matrix. The damping coefficient is... It is the identity matrix. For incremental vectors, This is the residual vector.

[0110] It should be noted that when the current model and point cloud data have good consistency, the damping coefficient is reduced to make the parameter update closer to the Gauss-Newton method, thereby improving the local convergence speed. When the current parameter update cannot reduce the objective function, the damping coefficient is increased to reduce the parameter update amplitude and improve iterative stability. The preset value range is 2-10, and the actual value should be adjusted according to the camera and light source used.

[0111] Sixth, update the orthogonal basis matrix through rotation index mapping, ensuring that the updated basis vectors remain orthogonal, satisfying the following relationship:

[0112] in, This is the orthogonal basis matrix for the next iteration. Represents the rotation index mapping, For the first Rotation increment of the next iteration Let be the orthogonal basis matrix of the current iteration.

[0113] Seventh, update the two coordinate parameters of the axis position based on the obtained increment, satisfying the following relationship:

[0114] in, The coordinate parameters for the next iteration. The coordinate parameters for the current iteration. This represents the increment of the coordinate parameters.

[0115] Furthermore, the cylinder radius is updated based on the obtained radius increment, satisfying the following relationship:

[0116] in, Let the radius of the cylinder be the radius of the next iteration. Let the radius of the cylinder be the radius of the current iteration. This is the radius increment.

[0117] In this embodiment, convergence is determined as follows: when the parameter increment is sufficiently small, the iteration can be considered to have reached a stable state, satisfying the following relationship:

[0118] in, For the first The increment vector of the next iteration. The parameter is the convergence threshold for incremental convergence.

[0119] It should be noted that the parameter increment convergence thresholds are set with corresponding allowable changes, and the parameter increments are normalized before joint judgment; the thresholds for axis position and radius increments are both 0.005mm, and the threshold for axis direction increments is... The normalization convergence threshold is set to 1.

[0120] S37. Construct a criterion for the change of the objective function, and determine the axis direction, the axis position point and the cylinder radius by combining the parameter increment, so as to obtain the axis position of the spindle rod.

[0121] In this embodiment, a criterion for the change of the objective function is constructed, satisfying the following relationship:

[0122] in, For the first Robust objective function for the next iteration For the first The parameter set for the next iteration. For the first Robust objective function for the next iteration For the first The parameter set for the next iteration. The threshold for the change of the objective function.

[0123] It should be noted that the optimization process can be considered to have converged when the changes in the robust objective function in two consecutive iterations are less than the set threshold for the change in the objective function. The dimensionless relative rate of change is used for setting, and is taken as... The cylindrical parameter iterative optimization is terminated when the relative change of the objective function does not exceed the threshold twice consecutively, and the parameter increment simultaneously meets the convergence condition.

[0124] In this embodiment, the final position of the spindle rod axis is determined, and the final cylindrical axis is represented by the optimized axis position point and axis direction, satisfying the following relationship:

[0125] in, For the final cylinder axis, For axis parameters, This is the final axis position point. This represents the final axis direction.

[0126] It should be noted that the axis parameter is used to represent the distance moved along the axis direction. It is determined by the orthogonal projection of the point onto the current axis direction. After optimization, the axis position and axis direction are determined, so the value of this parameter can be obtained from the camera coordinate system.

[0127] Furthermore, the final cylinder diameter satisfies the following relationship:

[0128] in, For the final cylinder diameter, This is the final cylinder radius.

[0129] In this embodiment, the spindle axis position is determined based on the final axis direction, axis position point, and cylinder radius.

[0130] Compared with traditional direct cylindrical fitting or least squares fitting methods, this invention addresses the problems of non-closed point clouds, local missing points, and excessive noise in the spindle rod point cloud under a single-sided view by employing constrained axis parameterization and robust radial residual optimization methods to estimate the spindle rod axis. This method can reduce the influence of outliers and incomplete arc surfaces on the axis estimation results, making the estimation of the spindle rod axis direction, axis position, and radius more stable.

[0131] S4. Determine the bottom reference position based on the image information, and combine it with the position of the spindle rod axis to realize the joint positioning of the yarn splicing robot.

[0132] In this embodiment, due to edge missingness of point cloud data or interference from the spinning machine scene, there is noise and missingness in the point cloud of the spindle rod edge and its end. The RGB image obtained by the RGB-D camera has high completeness, but the spinning machine workshop environment also has reflective parts that interfere with the collected spindle rod dataset. In order to locate the reference coordinates, a visual module that enhances reflection is introduced for training so that the spindle rod target can be accurately obtained and the reference point can be located using the edge line.

[0133] Compared with traditional two-dimensional visual edge localization methods, this invention introduces a reflection suppression visual feature processing mechanism to address the high reflectivity problem of metal spindle rods in spinning workshops. This mechanism distinguishes and models the real spindle rod edge, metal reflection interference, and environmental fly hair interference, and suppresses abnormally bright features through statistical feature weights, thereby reducing the probability of reflective areas being misjudged as real edges and improving the stability of spindle rod edge extraction under complex lighting conditions.

[0134] Specifically, in order to effectively suppress reflective noise and determine the edge features of the spindle rod, and considering the high reflective environment of the workshop, the traditional computationally expensive global self-attention mechanism was abandoned, and a feature compression and noise reduction algorithm relying on statistical characteristics was introduced; the specific calculation mechanism is as follows: First, a visual feature module for learning 2D spindle data is introduced to preprocess the image information to obtain feature tensors, and then the feature tensors are mapped to feature matrices.

[0135] Specifically, after the image pixel values ​​are normalized, they are input into the visual feature encoder with an input size of 352×288 pixels. The convolutional encoder outputs a feature tensor and expands its spatial dimension into a feature matrix.

[0136] The above feature tensors satisfy the following relationship:

[0137] in, For the characteristic tensor, For feature dimension, For parameters Visual feature encoder, For preprocessing images, The size of the feature map.

[0138] Expanding the two-dimensional spatial positions of the feature tensor yields the feature matrix, which satisfies the following relationship:

[0139] in, The characteristic matrix, For the characteristic tensor, For real matrix space, For feature dimension, The total number of visual feature blocks. The size of the feature map.

[0140] It should be noted that the feature matrix satisfies the following relationship:

[0141] in, The characteristic matrix, It is a visual feature block.

[0142] Secondly, based on the semantic annotation of actual spinning workshop images, a semantic subspace is defined, including the subspace of the bottom edge of the real spindle bar, the subspace of metal reflection interference, the subspace of yarn or fly interference, and the subspace of background structure. Image annotation software such as LabelStudio is used for polyline or region masks. After mapping the labeled polylines and masks to the feature map positions, feature samples of each category are extracted, and singular value decomposition is performed on the centered feature matrix of each category to initialize the orthogonal projection basis.

[0143] The above singular value decomposition of the centered eigenma matrix satisfies the following relationship:

[0144] in, For the centered feature matrix, Let be the left vector matrix of the space corresponding to the bottom of the spindle rod obtained from singular value decomposition. Let be the singular value matrix of the space corresponding to the bottom of the spindle rod obtained by singular value decomposition. This is the right vector matrix of the space corresponding to the bottom of the spindle rod obtained from singular value decomposition.

[0145] Pick Middle and front The column vectors corresponding to the larger singular values ​​constitute the i-th... The initial orthogonal projection basis of each semantic subspace satisfies the following relationship:

[0146] in, As the initial orthogonal projection basis, Let be the left vector matrix of the space corresponding to the bottom of the spindle rod obtained from singular value decomposition. For the first The dimension of each semantic subspace (take the dimension that can retain 95% of the cumulative singular values).

[0147] Furthermore, the orthogonal projection basis satisfies the following relationship:

[0148] in, As the orthogonal projection basis, for An identity matrix of order 1.

[0149] Next, for any number of... A visual feature block The soft-assigned probability vector belonging to the m-th semantic subspace The calculation satisfies the following relationship:

[0150] in, For soft-assignment probability vectors, Learnable parameters for controlling the smoothness of assignment, As the orthogonal projection basis, For the first A visual feature block.

[0151] It should be noted that the learnable parameter controlling the smoothness of the assignment is initially set to 0.1 and adjusted during backpropagation in subsequent model training.

[0152] Then, after calculating all visual feature blocks, the global probability weight vector of the m-th semantic subspace can be obtained by aggregation. The probability of all visual feature blocks to the m-th semantic subspace is obtained by arranging and normalizing the probability according to their spatial position.

[0153] Subsequently, in order to extract the structure of the features assigned to each semantic subspace, this module does not calculate the pairwise dot product between visual feature blocks, but instead extracts the features on the projection basis. Using the second-order moment statistics, a diagonal feature scaling matrix is ​​constructed specifically to suppress reflectivity anomalies. It satisfies the following relationship:

[0154] in, The diagonal eigenscale matrix, This indicates the construction of a diagonal matrix. It is a monotonically decreasing nonlinear activation function. As the orthogonal projection basis, The characteristic matrix, This is the global probability weight vector.

[0155] It should be noted that the nonlinear activation function simplifies to Therefore, when the characteristic response in certain directions appears abrupt for high-frequency reflective noise on the spindle due to a lack of group statistical commonality, the diagonal feature scaling matrix... It will dynamically output minimal weights that approach 0, thereby automatically filtering out these interference directions during low-rank projection.

[0156] Compared with traditional Transformer-based global self-attention visual processing methods, the reflection suppression module of this invention does not need to calculate the pairwise correlations between all visual feature blocks. Instead, it calculates feature weights through subspace projection and second-order statistics. As the number of visual feature blocks increases, this method can reduce computational overhead and is more suitable for deployment in industrial control computers or edge computing devices used in splicing robots in spinning workshops.

[0157] Finally, by combining the residual structure with reflection suppression weights, feature enhancement and update for a single visual feature block are completed, satisfying the following relationship:

[0158] in, For the updated feature matrix, The characteristic matrix, For the iteration step size coefficient of network optimization, For visual feature blocks, The total number of semantic subspaces. For the index of the semantic subspace, As the orthogonal projection basis, The diagonal eigenscale matrix, This is the global probability weight vector.

[0159] In this embodiment, through the dimensionality reduction and reconstruction calculation driven by the statistical features of the aforementioned visual feature blocks, the module can accurately remove erroneous features caused by ambient high light reflection without increasing quadratic computing power consumption. This results in the output of a set of spindle rod edge features with high robustness, providing reliable data support for subsequent high-precision reference point positioning using the side edge lines. A recognition and positioning network specifically designed for this type of reflective yarn spindle rod in the spinning workshop is trained in this way. By identifying and determining the spindle rod components and edge lines, the bottom reference position is determined, facilitating subsequent splice positioning.

[0160] Because 3D point clouds are prone to edge loss and depth noise at the bottom of the spindle, this invention does not rely solely on the end points of the point cloud to determine the bottom reference position. Instead, it fuses the robustly estimated spindle axis position with the edge and bottom features of the spindle extracted from the RGB image to determine the bottom reference position of the spindle. This bottom reference position serves as a key reference point for the robot's splicing coordinate system, used by the robotic arm to perform subsequent yarn suction, threading, or splicing actions.

[0161] Compared to methods that rely solely on 3D point clouds to determine the bottom point, this invention uses 3D point clouds to estimate the spindle axis and combines this with relatively complete edge and bottom features from a 2D image to determine the bottom reference point. This 2D-3D fusion method can compensate for the shortcomings of RGB-D point clouds, which are prone to missing or having high noise at the bottom edges, thus obtaining a more suitable positioning reference for splicing operations.

[0162] In this embodiment, the spindle axis position and bottom reference position are converted into the working positioning reference of the yarn splicing robot to achieve splice positioning.

[0163] Please see Figure 3 In an optional embodiment, the present invention provides a splice positioning system for a yarn splicing robot. The system includes an input device, an output device, a processor, and a memory, all interconnected. The memory stores a computer program comprising program instructions. The processor is configured to invoke the program instructions to execute specific steps as described in the relevant embodiments of the splice positioning method for a yarn splicing robot provided by the present invention. The splice positioning system for a yarn splicing robot provided by the present invention has a complete and stable structure, enhancing the overall applicability and practical application capability of the present invention.

[0164] In summary, the present invention provides a splice positioning method and system for a spinning splice robot. First, the parallelism between the spinning splice robot and the spinning machine is adjusted to achieve a parallel pose for the robot. Second, the yarn breakage location information is acquired, and point cloud data and image information are collected by the spinning splice robot while maintaining a parallel pose. Next, the axis direction, axis position point, and cylinder radius are obtained based on the point cloud data to determine the spindle axis position. Finally, the bottom reference position is determined based on the image information, and the splice positioning of the spinning spindle axis is achieved in conjunction with the spindle axis position. The method of this invention is easy to understand, computationally simple, requires minimal workload, and is convenient for engineering applications, providing a theoretical foundation and technical support for the further development of intelligent manufacturing and textile automation technologies.

[0165] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A splice positioning method for a yarn splicing robot, characterized in that, Includes the following steps: The parallelism between the spinning splicing robot and the spinning machine is adjusted to obtain the parallel position of the spinning splicing robot. Obtain the yarn breakage location information, and in the parallel posture state, collect point cloud data and image information through the yarn splicing robot; Based on the point cloud data, the axis direction, axis position point, and cylinder radius are obtained to determine the axis position of the spindle rod; The bottom reference position is determined based on the image information, and the joint positioning of the yarn splicing robot is achieved by combining the position of the spindle rod axis.

2. The splice positioning method for a yarn splicing robot according to claim 1, characterized in that, The adjustment of the parallelism between the spinning splicing robot and the spinning machine to obtain a parallel pose state for the spinning splicing robot includes: The parallelism is adjusted by a laser displacement sensor so that the yarn splicing robot is parallel to the yarn spinning machine to obtain the parallel posture state.

3. The splice positioning method for a yarn splicing robot according to claim 1, characterized in that, The step of obtaining the yarn breakage location information, in the parallel pose state, involves collecting point cloud data and image information through the yarn splicing robot, including: Based on the preset teaching position, the yarn breakage location information is determined through the yarn breakage dataset; Based on the yarn breakage location information, the point cloud data and image information are acquired using the robotic arm of the yarn splicing robot, which carries a data acquisition device.

4. The splice positioning method for a yarn splicing robot according to claim 1, characterized in that, The step of obtaining the axis direction, axis position point, and cylinder radius based on the point cloud data to determine the axis position of the spindle rod includes: Define a spindle rod point cloud set based on the aforementioned point cloud data; The initial axis is obtained by centering the set of points on the yarn spindle rods; Based on the initial axis, a set of parameters to be optimized is constructed using the current axis direction, the current axis position point, and the current cylinder radius; Calculate the radial distance, and calculate the radial residual based on the radial distance; A robust objective function is constructed based on the radial residual; The parameter increment is obtained by iteratively updating the set of parameters to be optimized using the robust objective function. A criterion for the change of the objective function is constructed, and the axis direction, the axis position point, and the cylinder radius are determined in combination with the parameter increment to obtain the axis position of the spindle rod.

5. The splice positioning method for a yarn splicing robot according to claim 4, characterized in that, The process of centering the point cloud set of the yarn spindles to obtain the initial axis includes: The point cloud center is obtained as a reference position based on the point cloud set of the yarn spindle rod, and the point cloud data is translated to obtain the center point; A spatial distribution matrix is ​​constructed using the centralized points to determine the initial axis direction; The center of the point cloud is projected onto a plane perpendicular to the initial axis direction to obtain the initial axis position point; Obtain the initial axial distance of each point in the point cloud data, and use the median of the initial axial distance as the initial value of the cylinder radius; The initial axis is determined by combining the initial axis direction, the initial axis position point, and the initial value of the cylinder radius.

6. The splice positioning method for a yarn splicing robot according to claim 4, characterized in that, The calculation of the radial distance, and the calculation of the radial residual based on the radial distance, includes: Calculate the relative vector from each point in the point cloud data to the current axis position point, and calculate the projection length of the relative vector in the direction of the current axis. The axial component in the relative vector is removed based on the projection length to obtain a radial vector perpendicular to the current axis direction; The magnitude of the radial vector is obtained as the radial distance; The difference between the radial distance and the current cylinder radius is used as the radial residual.

7. The splice positioning method for a yarn splicing robot according to claim 4, characterized in that, The step of iteratively updating the set of parameters to be optimized using the robust objective function to obtain parameter increments includes: The increment vector is obtained by merging the increments of the current axis direction, the current axis position point, and the current cylinder radius; The radial residuals of each point in the point cloud data are combined to form a residual vector; Calculate the partial derivative matrix of the residual vector with respect to the local parameters; Construct a diagonal matrix based on the optimized weights of each point in the point cloud data; The parameter increment is obtained by solving the increment vector using the least squares equation, combining the residual vector, the partial derivative matrix, and the diagonal matrix.

8. The splice positioning method for a yarn splicing robot according to claim 1, characterized in that, Determining the bottom reference position based on the image information includes: The image information is preprocessed to obtain a feature tensor, and the feature tensor is mapped to obtain a feature matrix; Define a semantic subspace for the feature space and set an orthogonal projection basis; Based on the orthogonal projection basis, calculate the soft assignment probability vector of the visual feature blocks in the feature matrix belonging to the semantic subspace; The soft-assigned probability vectors are aggregated to obtain the global probability weight vector of the semantic subspace; Construct a diagonal feature scaling matrix, and update the visual feature block by combining the orthogonal projection basis and the global probability weight vector to determine the bottom reference position.

9. The splice positioning method for a yarn splicing robot according to claim 1, characterized in that, The method of positioning the yarn splicing robot by combining the position of the spindle axis includes: The spindle axis position and the bottom reference position are converted into the working positioning reference of the yarn splicing robot to achieve the splice positioning.

10. A splice positioning system for a yarn splicing robot, characterized in that, The system includes an input device, an output device, a processor, and a memory, which are interconnected. The memory stores a computer program, which includes program instructions. The processor is configured to invoke the program instructions to execute the splice positioning method for a yarn splice robot as described in any one of claims 1-9.

Images