Spinning threading action optimization method and system based on path deep learning

Abstract

The invention provides a spinning threading action optimization method and system based on path deep learning, and the method comprises the steps: firstly obtaining a continuous three-dimensional point cloud sequence of a yarn broken end region, yarn physical attributes and mechanical arm dynamic parameters, and then constructing a tension-moment constraint related to the tensile strength of yarn according to the physical attributes and the dynamic parameters, meanwhile, on the basis of the three-dimensional point cloud sequence, a motion acceleration constraint associated with the anti-jitter capacity and an obstacle avoidance safety distance constraint associated with the obstacle distribution density are constructed, and then a loss function of a pre-constructed mechanical arm path generation neural network is optimized according to the three kinds of dynamic constraints linked with a scene; according to the invention, the first success rate and the track stability of the threading action can be improved, the threading failure caused by snapping, throwing or collision is reduced, the threading time is shortened on the premise of ensuring the safety of yarns, and the reliability of spinning operation is improved.

Description

Technical Field

[0001] This application relates to the field of spinning equipment control technology, specifically to a method and system for optimizing spinning and threading actions based on path deep learning. Background Technology

[0002] Spinning is a fundamental and core link in the textile industry, and its continuity directly determines production efficiency and the quality of the final yarn. During the high-speed operation of key equipment such as spinning frames, yarn breakage is inevitable due to factors such as mechanical friction, tension fluctuations, weak fiber points, or external interference. After a breakage occurs, the broken yarn must be promptly and accurately re-threaded through a series of complex yarn-guiding components, including the yarn guide hook, ring, and traveler, to restore the production flow. The efficiency and success rate of this threading operation are one of the main bottlenecks restricting the level of automation in spinning and the overall improvement of production capacity.

[0003] Currently, the trajectory planning of robotic arms or actuators used in automated threading relies heavily on preset fixed path templates, simplified position servo control, or rule adjustments based on limited sensor feedback. This approach has limitations when facing the actual working conditions of the spinning field. It lacks the ability to effectively perceive and adaptively model the multi-source, time-varying, and uncertain physical states of the work object and the working environment. As a result, the generated motion trajectories are often not robust enough and are prone to execution failure and low reliability when dealing with actual disturbances on site, making it difficult to meet the requirements of high success rate and high stability fully automated threading. Summary of the Invention

[0004] This application provides a method and system for optimizing spinning and threading actions based on path deep learning. It can solve the technical problems of existing spinning and threading trajectory planning methods, which are difficult to adapt to the dynamic uncertainties of the yarn itself and the production environment, resulting in insufficient robustness of the generated actions and easy execution failure and low reliability when dealing with actual disturbances on site.

[0005] A first aspect of this application provides a method for optimizing spinning and threading actions based on path deep learning, the method comprising: Obtain the continuous three-dimensional point cloud sequence of the yarn breakage area, the maximum tensile stress that can be withstood, the yarn diameter, the dynamic parameters of the robotic arm, and the pre-constructed robotic arm path generation network; Based on the maximum withstandable tensile stress, yarn diameter, and robotic arm dynamics parameters, a tension-torque constraint related to yarn tensile strength is constructed. Based on the continuous three-dimensional point cloud sequence and the dynamic parameters of the robotic arm, motion acceleration constraints related to anti-shake capability and obstacle avoidance safety distance constraints related to obstacle distribution density are constructed respectively. The loss function of the pre-constructed robotic arm path generation network is optimized based on the tension-torque constraint, motion acceleration constraint, and obstacle avoidance safety distance constraint to obtain the optimized robotic arm path generation network. Based on the optimized robotic arm path generation network, an optimized robotic arm spinning and threading motion trajectory sequence is determined. The robotic arm executes the optimized robotic arm spinning and threading motion trajectory sequence to complete the spinning and threading motion.

[0006] In one possible implementation, constructing the tension-torque constraint related to yarn tensile strength based on the maximum withstandable tensile stress, yarn diameter, and robotic arm dynamics parameters includes: Calculate the maximum allowable tension of the yarn based on the maximum withstandable tensile stress and the yarn diameter; The actual tension and tension influence coefficient applied by the robotic arm to the yarn are obtained, and the maximum output torque of the robotic arm is extracted from the robotic arm dynamic parameters. The tension risk ratio is obtained by calculating the ratio of the actual tension to the maximum permissible tension. Calculate the tension dynamic scaling factor based on the tension influence coefficient and tension risk ratio; Based on the maximum output torque of the robotic arm, the tension risk ratio, and the tension dynamic scaling factor, the upper limit constraint value of the joint torque is calculated, and the tension-torque constraint is constructed.

[0007] In one possible implementation, constructing motion acceleration constraints related to anti-shake capability based on the continuous three-dimensional point cloud sequence and the robotic arm dynamic parameters includes: A yarn point cloud sequence is segmented from the continuous three-dimensional point cloud sequence; Based on the yarn point cloud sequence, the yarn anti-vibration capability parameters are determined by calculating the displacement variance. Obtain the jitter influence coefficient and extract the rated maximum joint angular acceleration of the robotic arm from the robotic arm dynamic parameters; Calculate the dynamic scaling factor of acceleration based on the vibration influence coefficient and the yarn vibration resistance parameter; Based on the acceleration dynamic scaling factor and the rated maximum joint angular acceleration of the robotic arm, the upper limit of the joint angular acceleration is calculated, and motion acceleration constraints are constructed.

[0008] In one possible implementation, determining the yarn anti-shake capability parameter based on the yarn point cloud sequence by calculating the displacement variance includes: For each frame of yarn point cloud sequence, the yarn centroid is extracted to obtain the three-dimensional centroid coordinates of the yarn. For the three-dimensional centroid coordinates of adjacent yarns in the yarn point cloud sequence, calculate the Euclidean distance between the three-dimensional centroid coordinates of adjacent yarns to obtain the inter-frame displacement sequence; Calculate the standard deviation of the inter-frame displacement sequence based on the inter-frame displacement sequence; Based on the standard deviation, the yarn vibration resistance parameter is calculated using a linear mapping method.

[0009] In one possible implementation, the step of constructing an obstacle avoidance safety distance constraint related to obstacle distribution density based on the continuous three-dimensional point cloud sequence and the robotic arm dynamic parameters includes: Extract a frame of 3D point cloud from the continuous 3D point cloud sequence, perform semantic segmentation on the frame of 3D point cloud, and obtain the 3D point cloud of the machine obstacle. Based on the three-dimensional point cloud of the machine obstacle, the surface area of ​​the obstacle and the total surface area of ​​the region are calculated using the voxelized exposed area method. Calculate the obstacle distribution density parameter based on the surface area of ​​the obstacle and the total surface area of ​​the region; Obtain the basic safety distance and obstacle density influence coefficient; Calculate the dynamic obstacle avoidance safety distance scaling factor based on the obstacle distribution density parameter and the obstacle density influence coefficient; Based on the dynamic obstacle avoidance safety distance scaling factor and the basic safety distance, the upper limit of the robot arm's safety distance is calculated, and the obstacle avoidance safety distance constraint is constructed.

[0010] In one possible implementation, the step of calculating the surface area of ​​the obstacle and the total surface area of ​​the region based on the three-dimensional point cloud of the obstacle using a voxelized exposed area method includes: Define the three-dimensional spatial computation range and divide the three-dimensional spatial computation range into a uniform voxel grid; If a voxel contains at least one point of the obstacle point cloud, it is marked as an obstacle voxel. If a voxel contains at least one point cloud point of any class, it is marked as an occupied voxel. Based on the grid occupancy information of the obstacle voxels and the occupier voxels, the surface area of ​​the obstacle and the total surface area of ​​the region are determined.

[0011] In one possible implementation, optimizing the loss function of the pre-constructed robotic arm path generation network based on the tension-torque constraint, motion acceleration constraint, and obstacle avoidance safety distance constraint to obtain an optimized robotic arm path generation network includes: The tension-torque constraint is converted into a first penalty term, which generates a positive penalty when the joint torque of the robotic arm exceeds the upper limit constraint value of the joint torque. The motion acceleration constraint is transformed into a second penalty term, which generates a positive penalty when the joint angular acceleration of the robotic arm exceeds the upper limit of the joint angular acceleration. The obstacle avoidance safety distance constraint is transformed into a third penalty term, which generates a positive penalty when the real-time distance between the end of the robotic arm and the obstacle is less than the upper limit of the robotic arm safety distance. Assign a first weight, a second weight, and a third weight to the first penalty item, the second penalty item, and the third penalty item, respectively; The weighted first penalty term, second penalty term, and third penalty term are added to the original loss function of the pre-built robotic arm path generation network to obtain the optimized loss function and the optimized robotic arm path generation network.

[0012] In one possible implementation, the formulas for the tension-torque constraint, the motion acceleration constraint, and the obstacle avoidance safety distance constraint are expressed as follows: , In the formula, Here is the predicted value of the joint torque at time t. This is the maximum output torque of the robotic arm. The tension influence coefficient is... The actual tension applied to the yarn by the robotic arm. This is the maximum permissible tension of the yarn; , In the formula, Here is the predicted value of the joint angular acceleration at time t. The rated maximum joint angular acceleration of the robotic arm, This is the jitter impact coefficient. This refers to the yarn's resistance to vibration. In the formula, Let t be the actual distance between the end effector of the robotic arm and the nearest obstacle. Based on the safe distance, The obstacle density influence coefficient. This is the obstacle distribution density parameter.

[0013] In one possible implementation, the loss function of the pre-constructed robotic arm path generation network is optimized based on the tension-torque constraint, motion acceleration constraint, and obstacle avoidance safety distance constraint to obtain the optimized robotic arm path generation network. The expression formula of the loss function of the optimized robotic arm path generation network is as follows: , In the formula, Generate the total loss function value for the path generation network training. The joint angle at time t is predicted by the network. The reference joint angle at time t, These are the weighting coefficients for trajectory matching error. These are the weighting coefficients for trajectory smoothness constraints. These are the weighting coefficients for tension-related constraints. Here is the predicted value of the joint torque at time t. This represents the original upper limit of the joint torque. The tension influence coefficient is... The actual yarn tension at time t. The maximum allowable tension of the yarn. These are the weighting coefficients for jitter suppression constraints. Here is the predicted value of the joint angular acceleration at time t. The rated maximum angular acceleration of the robotic arm, This is the jitter impact coefficient. This refers to the yarn's resistance to vibration. The weighting coefficients for obstacle avoidance constraints. Based on the safe distance, The obstacle density influence coefficient. The obstacle distribution density parameter, Let t be the real-time minimum distance between the end effector of the robotic arm and the obstacle.

[0014] This example provides a method for optimizing spinning threading actions based on deep learning. First, a continuous 3D point cloud sequence of the yarn breakage area, yarn physical properties, and robotic arm dynamic parameters are acquired. Then, tension-torque constraints related to yarn tensile strength are constructed based on the physical properties and dynamic parameters. Simultaneously, motion acceleration constraints related to anti-vibration capability and obstacle avoidance safety distance constraints related to obstacle distribution density are constructed based on the 3D point cloud sequence. Next, the loss function of the pre-constructed robotic arm path generation neural network is optimized based on these three types of scene-linked dynamic constraints to obtain the optimized network. Finally, this network generates and executes the robotic arm threading trajectory sequence. By adaptively integrating multi-source real-time perceived physical constraints into the deep learning trajectory generation process, the robotic arm can understand and respond online to the uncertainties brought about by yarn flexibility, susceptibility to vibration, and dense obstacle environments. This significantly improves the first-time success rate and trajectory stability of the threading action, effectively reducing threading failures caused by breakage, slippage, or collisions. While ensuring yarn safety, the threading time is shortened, meeting the reliable requirements of fully automated, high-rhythm continuous operation in spinning production.

[0015] A second aspect of this application provides a spinning and threading action optimization system based on path deep learning, the apparatus comprising: The acquisition unit is used to acquire the continuous three-dimensional point cloud sequence of the yarn breakage area, the maximum tensile stress that can be withstood, the yarn diameter, the robot arm dynamic parameters, and the pre-constructed robot arm path generation network; The first processing unit is used to construct a tension-torque constraint related to the tensile strength of the yarn based on the maximum withstandable tensile stress, yarn diameter and robotic arm dynamic parameters. The second processing unit is used to construct motion acceleration constraints related to anti-shake capability and obstacle avoidance safety distance constraints related to obstacle distribution density based on the continuous three-dimensional point cloud sequence and the dynamic parameters of the robotic arm, respectively. The third processing unit is used to optimize the loss function of the pre-constructed robotic arm path generation network based on the tension-torque constraint, motion acceleration constraint and obstacle avoidance safety distance constraint, so as to obtain the optimized robotic arm path generation network. The execution unit is used to determine an optimized robotic arm spinning and threading motion trajectory sequence based on the optimized robotic arm path generation network, and the robotic arm executes the optimized robotic arm spinning and threading motion trajectory sequence to complete the spinning and threading motion.

[0016] A third aspect of this application provides a terminal including a processor, an input device, an output device, and a memory, wherein the processor, input device, output device, and memory are interconnected, wherein the memory is used to store a computer program, the computer program including program instructions, and the processor is configured to invoke the program instructions to execute the step instructions as described in the path deep learning-based spinning and threading action optimization method in the first aspect of this application.

[0017] A fourth aspect of this application provides a computer-readable storage medium storing a computer program for electronic data interchange, wherein the computer program causes a computer to perform some or all of the steps described in the path deep learning-based spinning and threading action optimization method of the first aspect of this application.

[0018] A fifth aspect of this application provides a computer program product, comprising a non-transitory computer-readable storage medium storing a computer program operable to cause a computer to perform some or all of the steps described in the path deep learning-based spinning and threading action optimization method of the first aspect of this application. The computer program product may be a software installation package. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This application provides a schematic diagram of the overall process for optimizing spinning and threading actions based on path deep learning in an embodiment of the present application. Figure 2 This application provides a schematic diagram of the overall structure of a spinning and threading action optimization system based on path deep learning, as an embodiment of the present application. Figure 3 This application provides a schematic diagram of the structure of a terminal. Figure label: Acquisition unit-1, first processing unit-2, second processing unit-3, third processing unit-4, execution unit-5. Detailed Implementation

[0021] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] The terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0023] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application can be combined with other embodiments.

[0024] To better understand the path-based deep learning-based spinning and threading motion optimization method provided in this application embodiment, the following is a brief introduction to the scenarios in which this method is applied. Currently, the trajectory planning of robotic arms or actuators used for automated threading largely relies on preset fixed path templates, simplified position servo control, or rule adjustments based on limited sensor feedback. These methods have fundamental limitations when facing the actual working conditions of the spinning field: they are difficult to adapt to the dynamic uncertainties of the yarn itself and the production environment. For example, as a flexible material, the spatial posture, vibration state, and tolerable tension range of yarn after breakage change in real time; simultaneously, obstacles such as yarn guides and rollers are densely arranged and space is limited within the machine area. Existing planning strategies lack the ability to uniformly perceive, quantify, and integrate these multi-dimensional, time-varying physical constraints into the decision model, resulting in poor robustness of the generated threading trajectory. When dealing with random disturbances in real-world scenarios, problems such as grasping failure, yarn breakage, or collisions with machine parts easily occur, failing to meet the requirements for highly reliable and high-success-rate intelligent threading.

[0025] The path deep learning-based spinning and threading motion optimization method is applied to the path deep learning-based spinning and threading motion optimization system. Figure 1 A schematic diagram of the overall process for optimizing spinning and threading actions based on path deep learning is shown. Figure 1 As shown, it includes: S1. Obtain the continuous three-dimensional point cloud sequence of the yarn breakage area, the maximum tensile stress that can be withstood, the yarn diameter, the robot arm dynamic parameters, and the pre-constructed robot arm path generation network.

[0026] For the continuous three-dimensional point cloud sequence of the yarn breakage area, a depth camera can be installed at the end of the robotic arm to continuously capture real-time images of the breakage area of ​​the spinning machine. The acquired depth image sequence is used to generate a continuous three-dimensional point cloud sequence to represent the position of the yarn in space.

[0027] In this context, for any frame of the depth image sequence and any frame of the 3D point cloud in the continuous 3D point cloud sequence, the camera is mounted at the end of the robotic arm, with the camera and the end of the robotic arm fixed in place, and the shooting range covers the decapitation area. The formulas for generating a frame of depth image and point cloud are as follows: , In the formula, Let these be the coordinates of the camera's principal point. For camera focal length, These are the pixel coordinates. This represents the depth value.

[0028] Among them, the maximum withstandable tensile stress is the maximum withstandable tensile stress of the current yarn, which can be collected in real time by a miniature tension-stress detection module installed at the end gripper of the robotic arm.

[0029] The yarn diameter can be found in the product parameter manual provided by the yarn manufacturer. The robotic arm dynamic parameters, including the maximum output torque and the rated maximum joint angular acceleration, can be found in the robotic arm parameter manual provided by the robotic arm manufacturer.

[0030] The pre-built robotic arm path generation network is a DNN network, consisting of an encoder and a decoder. The encoder encodes the three-dimensional coordinates of the input yarn endpoints and the target channel position into a fused task context feature vector, and the decoder decodes the feature vector into a time-smoothed joint angle sequence.

[0031] Furthermore, the loss function for training the pre-built robotic arm path generation network is shown below: , In the formula, The original total loss value for the pre-built robotic arm path generation network. The total number of time steps for the trajectory. This is the robot arm joint angle vector predicted by the network at time t. Let be the true joint angle vector at time t. These are the weighting coefficients for the trajectory matching error term. These are the weighting coefficients for the trajectory smoothness constraint term.

[0032] S2. Based on the maximum withstandable tensile stress, yarn diameter, and robotic arm dynamics parameters, construct a tension-torque constraint related to yarn tensile strength.

[0033] In one possible implementation, step S2 includes the following sub-steps: S2-1. Calculate the maximum allowable tension of the yarn based on the maximum withstandable tensile stress and yarn diameter.

[0034] Firstly, based on the fundamental principles of materials mechanics, the yarn is approximated as a uniform cylinder with a cross-sectional area of... Through yarn diameter The calculations are as follows: , In the formula, The cross-sectional area of ​​the yarn. This refers to the yarn diameter.

[0035] After obtaining the cross-sectional area of ​​the yarn, and combining it with the maximum tensile stress the yarn can withstand, the maximum allowable tension that the yarn can withstand without breaking can be calculated, as shown below: , In the formula, The maximum allowable tension of the yarn. This represents the maximum tensile stress that the yarn can withstand. This represents the cross-sectional area of ​​the yarn. The maximum allowable tension of the yarn calculated in this step serves as a key input parameter for subsequent tension risk assessment and torque constraint construction, providing a dynamic safety upper limit for the torque output of the robotic arm. During the threading process, if the actual tension applied to the yarn by the robotic arm approaches or exceeds this threshold, the system will automatically tighten the joint torque constraint, forcibly reduce the movement speed, or adjust the gripping force, thereby fundamentally preventing yarn breakage due to overload stretching. This step transforms the flexible material properties of the yarn into quantifiable mechanical boundaries, providing a clear anti-breakage protection basis for the subsequent path generation network, thus improving the reliability of the threading action and the integrity of the yarn.

[0036] S2-2. Obtain the actual tension and tension influence coefficient applied to the yarn by the robotic arm, and extract the maximum output torque of the robotic arm from the robotic arm dynamic parameters.

[0037] The actual tension applied to the yarn by the robotic arm can be collected by a miniature tension-stress detection module installed at the end gripper of the robotic arm.

[0038] The tension influence coefficient is an empirical parameter used to quantify the impact of yarn tension on the upper limit of the joint torque of the robotic arm. It can be based on the analysis and learning of a large amount of historical threading operation data. Specifically, during the training phase before system deployment, operation records containing different yarn types, different breakage postures, and different threading results are collected. The actual tension at key moments t, the corresponding robotic arm joint torque, and whether yarn breakage occurred are recorded. Through data regression analysis or a lightweight machine learning model, the negative correlation between actual tension and the safe upper limit of joint torque is fitted under the premise of ensuring yarn safety, thereby determining the value of the tension influence coefficient. According to empirical data, the value range of the tension influence coefficient is usually between 0.3 and 0.5. The tension influence coefficient reflects the sensitivity of yarn tension risk to the limit of robotic arm output under specific production conditions and is solidified as prior knowledge in the system's parameter database or the input features of the path generation network, and is called during each threading task initialization.

[0039] The maximum output torque of the robotic arm can be obtained from the equipment manufacturer's technical specifications.

[0040] S2-3. Calculate the ratio of the actual tension to the maximum permissible tension to obtain the tension risk ratio.

[0041] The formula for calculating the tension risk ratio is as follows: , In the formula, The tension risk ratio at time t is a dimensionless scalar. The actual tension applied to the yarn by the robotic arm. This represents the maximum permissible tension of the yarn. Used to reflect the percentage of the tension currently applied to the yarn relative to its breaking limit, when This means that the actual tension has reached or exceeded the yarn's theoretical breaking limit, posing an extremely high risk of breakage. In practical control, a warning threshold can be set (e.g., ...). When the risk ratio exceeds this threshold, a high-level alarm is triggered and a more stringent constraint mechanism is activated.

[0042] S2-4. Calculate the tension dynamic scaling factor based on the tension influence coefficient and tension risk ratio.

[0043] The tension dynamic scaling factor is used to flexibly adjust the safety boundary of the joint torque. A dynamic scaling coefficient can be determined based on real-time risk levels and historical experience. Specifically, the calculation formula for the tension dynamic scaling factor is as follows: , In the formula, Let be the tension dynamic scaling factor at time t. The tension influence coefficient is... Let be the tension risk ratio at time t. The tension dynamic scaling factor reveals that the greater the tension on the yarn, the more proportionally the upper limit of the robotic arm joint torque should be reduced to avoid the risk of breakage. The calculated tension dynamic scaling factor is a value between... The real-time dynamic parameters will be used as core variables in the next step of the joint torque upper limit constraint formula, thereby enabling the robotic arm output limit to be adaptively and continuously adjusted according to the real-time stress of the yarn, which can reflect the accuracy and adaptability of dynamic protection of flexible targets.

[0044] S2-5. Based on the maximum output torque of the robotic arm, the tension risk ratio, and the tension dynamic scaling factor, calculate the upper limit constraint value of the joint torque and construct the tension-torque constraint.

[0045] The formula for expressing the upper limit constraint value of joint torque is as follows: , In the formula, This represents the upper limit constraint value for joint torque.

[0046] The expression formula for tension-torque constraint is as follows: , In the formula, Here is the predicted value of the joint torque at time t. This is the maximum output torque of the robotic arm. The tension influence coefficient is... The actual tension applied to the yarn by the robotic arm. This represents the maximum permissible tension of the yarn.

[0047] In this example, the tension-torque constraint limits the maximum output force of the robotic arm. Risk ratio of real-time tension to yarn Through the tension influence coefficient Coupled. When the actual tension When the value is increased, the upper limit on the right side decreases accordingly, which can force the network to generate or the controller to execute a trajectory with less force and gentler force, thus avoiding the yarn from being broken due to the robotic arm moving too violently.

[0048] In this example, the basic security boundary is... Definition, and The term, acting as a dynamic scaling factor varying within the interval (0, 1), enables real-time contraction and expansion of the safety boundary. For example, when... When the actual tension reaches more than 70% of the yarn's tensile strength, the scaling factor may drop to below 0.5, the upper limit of the joint torque is significantly tightened to less than 50% of the original maximum output, and the system automatically enters the low tension protection mode.

[0049] In this example, the calculated This can be used as a hard constraint or transformed into a penalty term in the loss function, and input into the pre-built robotic arm path generation network. During training or inference, the network must ensure that the predicted joint torques at each time step of the output trajectory sequence are within acceptable limits. This constraint ensures that the generated threading path is compatible with the yarn's tensile properties at the dynamic level.

[0050] In summary, this example demonstrates how to obtain information from yarn material properties ( ), real-time sensor data ( From the core control quantity of the robotic arm (joint torque) The closed-loop constraint modeling is used. The constructed tension-torque constraint is not a fixed threshold, but a dynamic value that adaptively adjusts according to the stress state of the yarn. This allows the torque output of the robotic arm to match the bearing capacity of the flexible yarn, thereby improving the success rate of threading and preventing yarn breakage.

[0051] S3. Based on the continuous three-dimensional point cloud sequence and the dynamic parameters of the robotic arm, construct motion acceleration constraints related to anti-shake capability and obstacle avoidance safety distance constraints related to obstacle distribution density.

[0052] Step S3 includes a sub-step of constructing motion acceleration constraints related to anti-shaking capability based on the continuous three-dimensional point cloud sequence and the robotic arm dynamic parameters, and a sub-step of constructing obstacle avoidance safety distance constraints related to obstacle distribution density based on the continuous three-dimensional point cloud sequence and the robotic arm dynamic parameters.

[0053] In one possible implementation, a motion acceleration constraint related to anti-jitter capability is constructed based on the continuous three-dimensional point cloud sequence and the robotic arm dynamic parameters, including: S3-1. Segment the yarn point cloud sequence from the continuous three-dimensional point cloud sequence.

[0054] This can be achieved by using a deep learning-based instance segmentation model to process continuous 3D point cloud sequences frame by frame. Specifically, the segmentation network is trained offline using a large-scale labeled spinning scene point cloud dataset, where each point is labeled with categories such as "yarn," "machine," "yarn guide hook," and "background." The trained network model is capable of performing point-by-point semantic prediction on the input single-frame 3D point cloud and outputting the category label for each 3D point.

[0055] Furthermore, in the online application phase, each frame in the continuous 3D point cloud sequence obtained in step S1 is... (It should be noted that,) The frame number is input into the trained segmentation network. The network processes the point cloud... Each point in Predict a probability distribution for a category and take the category with the highest probability as the label for that point. Subsequently, based on the point labels, all points labeled "yarn" are extracted to form the yarn point cloud corresponding to that frame. : , By repeating this operation on each frame of the sequence, a yarn point cloud sequence synchronized with the original sequence can be obtained. ,in This represents the total number of frames.

[0056] In this example, the segmentation model can use a point cloud processing architecture such as PointNet++ or KPConv. The loss function can combine classification cross-entropy loss and regularization terms to ensure accurate segmentation boundaries and robustness to the characteristics of long, thin, and curved yarns. Through this step, the system can eliminate interference from static obstacles such as machine tools and lock onto dynamically changing yarn targets, providing a reliable data foundation for subsequent point cloud sequence-based analysis of yarn motion states.

[0057] S3-2. Determine the yarn anti-vibration capability parameters by calculating the displacement variance based on the yarn point cloud sequence.

[0058] In one possible implementation, step S3-2 includes the following sub-steps: S3-2-1. Extract the yarn centroid from the yarn point cloud sequence of each frame to obtain the three-dimensional centroid coordinates of the yarn.

[0059] Among them, each frame of yarn point cloud can be processed. Calculate the arithmetic mean coordinates of all its points to use as the three-dimensional centroid of the yarn in that frame. The three-dimensional centroid of the yarn The calculation formula is as follows: , In the formula, The coordinates of the yarn's three-dimensional centroid are... For the first The number of points in the yarn point cloud. For the first The three-dimensional centroid coordinates of each frame of yarn are used to summarize the spatial shape of each frame of yarn into a representative feature point, which facilitates the subsequent calculation of inter-frame displacement and simplifies motion state analysis.

[0060] S3-2-2. For the three-dimensional centroid coordinates of adjacent yarns in the yarn point cloud sequence, calculate the Euclidean distance between the three-dimensional centroid coordinates of adjacent yarns to obtain the inter-frame displacement sequence.

[0061] Among them, let there be a total The point cloud of the yarn in frame, corresponding to the centroid sequence is: Calculate the Euclidean distance between the centroids of adjacent frames. Euclidean distance The calculation formula is as follows: In the formula, For the first Frame to the The displacement of the centroid of the frame yarn. Euclidean distance. Calculating Euclidean distance allows for the quantification of yarn positional changes over consecutive moments, reflecting the amplitude of yarn movement and thus assessing the intensity of vibration.

[0062] S3-2-3. Calculate the standard deviation of the inter-frame displacement sequence based on the inter-frame displacement sequence.

[0063] Specifically, the standard deviation of the inter-frame displacement sequence is calculated. It can reflect the degree of fluctuation in yarn displacement. The larger the value, the more violent the yarn shaking and the more unstable the movement.

[0064] S3-2-4. Based on the standard deviation, calculate the yarn vibration resistance parameter using a linear mapping method.

[0065] One approach is to first set a maximum permissible standard deviation of displacement. (This can be based on historical data or experimental calibration) Mapped to a linear function The range is used to obtain the yarn's vibration resistance parameter. Yarn vibration resistance parameters The formula for expressing this is as follows: , In the formula, This is a parameter representing the yarn's resistance to vibration; a higher value indicates stronger resistance. The standard deviation of the inter-frame displacement sequence. For the maximum permissible standard deviation of displacement, if ,but This indicates that the yarn has extremely weak anti-vibration ability, and can transform the displacement fluctuation statistics into a normalized anti-vibration ability score, providing a quantitative basis for subsequent motion acceleration constraints.

[0066] In this example, the yarn vibration resistance parameter This is a dynamic index between 0 and 1, used to quantify the motion stability of the current yarn in space. A higher value indicates less yarn jitter and greater shape stability; a lower value indicates more severe yarn jitter and higher pose uncertainty. This parameter is directly input into subsequent motion acceleration constraints to dynamically adjust the upper limit of the robotic arm's joint angular acceleration: when... When the speed is low, the system automatically limits the robotic arm's acceleration to prevent the yarn from swinging further or slipping out of its grip due to rapid movement; when... At higher speeds, the acceleration limit can be appropriately relaxed to improve threading efficiency. This can be achieved by introducing... The system enables online sensing and adaptive response to the dynamic state of the yarn, thereby improving the robustness and success rate of the threading action while ensuring operational stability.

[0067] S3-3, Obtain the jitter influence coefficient and extract the rated maximum joint angular acceleration of the robotic arm from the robotic arm dynamic parameters.

[0068] The jitter impact coefficient β is an empirical parameter used to quantify the degree to which yarn jitter affects the upper limit of the joint angular acceleration of the robotic arm. Its value typically ranges from 0.6 to 0.8, and the specific value is determined based on the analysis and learning of a large amount of historical threading operation data. Specifically, during the training phase before system deployment, offline or online operation data containing different yarn types, different jitter states, and different threading results can be collected. Through data regression analysis or a lightweight machine learning model, the negative correlation between the yarn jitter resistance parameter δ_{vib} and the upper limit of the safe joint angular acceleration is fitted, thereby determining the optimal value of β. This coefficient is then embedded as prior knowledge in the system's parameter database.

[0069] The rated maximum joint angular acceleration of the robotic arm can be obtained from the equipment technical specifications provided by the robotic arm manufacturer.

[0070] S3-4. Calculate the acceleration dynamic scaling factor based on the vibration influence coefficient and the yarn anti-vibration capability parameter.

[0071] Among them, the yarn's anti-vibration capability parameter can be calculated in real time based on the vibration influence coefficient β. Calculate a dynamic scaling factor The safety upper limit for joint angular acceleration is used to adjust the acceleration dynamic scaling factor in real time. The formula for calculating the acceleration dynamic scaling factor is as follows: , In the formula, The acceleration dynamic scaling factor at time t is a value between Real-time scalar, This is the jitter impact coefficient. This is a parameter representing the yarn's resistance to vibration.

[0072] In this example, the real-time vibration state of the yarn can be measured. This is converted into a dynamic scaling factor using an empirical coefficient β. When the yarn vibrates violently... Small, and When the value is large, the scaling factor approaches 1, having little impact on the upper limit of acceleration, especially when the yarn is very stable. Big, and When the value is small, the scaling factor decreases significantly, thereby tightening the upper limit of acceleration and thus achieving adaptive safety boundary adjustment based on yarn condition.

[0073] S3-5. Based on the acceleration dynamic scaling factor and the rated maximum joint angular acceleration of the robotic arm, calculate the upper limit of the joint angular acceleration and construct motion acceleration constraints.

[0074] The formula for expressing the upper limit of joint angular acceleration is as follows: , In the formula, This is the upper limit of the knot angular acceleration. This is the rated maximum joint angular acceleration of the robotic arm. This is the jitter impact coefficient. This is a parameter representing the yarn's resistance to vibration.

[0075] The formula for the motion acceleration constraint is shown below: , In the formula, Here is the predicted value of the joint angular acceleration at time t. The rated maximum joint angular acceleration of the robotic arm, This is the jitter impact coefficient. This is a parameter for the yarn's resistance to vibration. The purpose of this example is to concretize the dynamic scaling factor calculated in the preceding steps into an executable, time-varying upper limit constraint on joint angular acceleration. This constraint, acting as a hard rule or penalty term in the loss function, forces the predicted acceleration of the trajectory output by the path generation network at any given time. This dynamic limit must not be exceeded.

[0076] In this example, when the yarn's anti-vibration ability is weak, the upper limit value on the right side of the constraint is automatically reduced, forcing the robotic arm to move in a smoother manner with less acceleration, thereby avoiding the yarn from swinging, falling off, or even breaking due to the robotic arm's rapid start-stop or turning.

[0077] In this example, the motion acceleration constraint is not a fixed value, but changes dynamically with the real-time state of the yarn. When the yarn state is stable, the robotic arm is allowed to move efficiently with higher acceleration; when the yarn state is unstable, it automatically switches to robust mode to prioritize the reliability of the operation.

[0078] In this example, motion acceleration constraints can be transformed into a penalty term in the loss function and thus deeply integrated into the training and inference process of the path generation network. This enables the robotic arm path generation network to learn how to reach the target while also learning how to move in a yarn-friendly manner, thereby improving the robustness and first-time success rate of the threading action in complex dynamic environments.

[0079] In summary, this example demonstrates how constructing and applying motion acceleration constraints enables the system to achieve online sensing and closed-loop response to the dynamic physical characteristics of yarn, thereby improving the intelligence level and success rate of the fully automated threading system.

[0080] In one possible implementation, constructing obstacle avoidance safety distance constraints related to obstacle distribution density based on the continuous three-dimensional point cloud sequence and robotic arm dynamic parameters includes: S3-6. Extract a frame of three-dimensional point cloud from the continuous three-dimensional point cloud sequence, segment the frame of three-dimensional point cloud, and obtain the three-dimensional point cloud of the machine obstacle.

[0081] The segmentation step follows the same principle as the preceding steps.

[0082] S3-7. Based on the three-dimensional point cloud of the machine obstacle, calculate the surface area of ​​the obstacle and the total surface area of ​​the region using the voxelized exposed area method.

[0083] In one possible implementation, step S3-7 includes the following sub-steps: S3-7-1. Define the three-dimensional spatial calculation range and divide the three-dimensional spatial calculation range into a uniform voxel grid.

[0084] This can be centered on the yarn breakage area and extended to adjacent areas that the robotic arm's threading path might traverse. An axis-aligned cubic bounding box is defined as the 3D spatial computation range. Its minimum corner coordinates are The coordinates of the largest corner point are .

[0085] Then, exist Divide the three dimensions evenly into sides of length . A cubic grid, where each small cube is called a voxel. The value can be set based on the average density of the point cloud or prior experience, for example... Possible values ​​are: to between.

[0086] In this example, the continuous 3D space is discretized into a regular voxel grid, providing a basic data structure for subsequent point cloud spatial distribution statistics. Discretization allows for efficient determination of whether each voxel contains point cloud data and the occupancy status, enabling the calculation of obstacle surface area and total region surface area.

[0087] S3-7-2-A. If a voxel contains at least one point of the obstacle point cloud, it is marked as an obstacle voxel.

[0088] S3-7-2-B: If a voxel contains at least one point cloud point of any class, it is marked as an occupied voxel.

[0089] Among them, the 3D point cloud of machine obstacles obtained by traversing steps S3-6 and the original frame point cloud (Includes all category points). For each voxel : Furthermore, examine whether there exists anything belonging to this voxel space. If at least one point exists, then the voxel is marked as an obstacle voxel, denoted as . Otherwise, record as .

[0090] Furthermore, examine whether there exists any original point cloud within the voxel space. Any point (including all categories such as yarn, obstacles, and background). If at least one point exists, the voxel is marked as occupied voxel, denoted as . Otherwise, record as .

[0091] In this example, the voxel mesh is binarized by traversing the point cloud. The obstacle voxels reflect the occupancy of obstacles in space and are the basis for calculating the exposed area of ​​obstacles. The occupied voxels reflect the occupancy of all physical entities (including obstacles, yarn, etc.) in the entire working area and are the basis for calculating the total surface area of ​​the area.

[0092] S3-7-3. Based on the grid occupancy information of the obstacle voxels and the occupier voxels, determine the surface area of ​​the obstacle and the total surface area of ​​the region.

[0093] This involves traversing all elements marked as obstacles (i.e., The voxels of the obstacle. For each obstacle voxel, examine its position in... Adjacent voxels in six directions (i.e., indexed by) Is the voxel a barrier voxel?

[0094] Furthermore, if a voxel in a certain adjacent direction is not an obstacle voxel (i.e. If the voxel is outside the calculation range, then this face of the current obstacle voxel is considered exposed; obstacle surface area. It is approximately equal to the sum of the areas of all exposed surfaces of the obstacles.

[0095] Furthermore, using the same principle, but traversing the object becomes all occupying voxels (i.e. For each occupied voxel, check if its six adjacent voxels are occupied voxels.

[0096] Furthermore, if a voxel in a certain adjacent direction is not an occupying voxel, then the face of the currently occupying voxel is considered the exposed surface of the region boundary; the total surface area of ​​the region. It is approximately equal to the sum of the boundary exposure areas of all occupied voxels.

[0097] In this example, voxel neighborhood analysis allows estimation of the area of ​​a continuous surface and the surface area of ​​an obstacle from a discrete voxel mesh. This reflects the total area of ​​obstacles exposed within the calculation region, representing the total surface area of ​​the region. It reflects the total area of ​​the composite surface composed of all entities within the entire working area.

[0098] S3-8. Calculate the obstacle distribution density parameter based on the surface area of ​​the obstacle and the total surface area of ​​the region.

[0099] Among them, in obtaining the surface area of ​​the obstacle and total surface area of ​​the region Then, the obstacle distribution density parameter can be obtained by calculating the ratio of the two. This parameter is used to quantify the density of obstacles on the surface of the current workspace. The calculation formula is as follows: In the formula, The obstacle distribution density parameter is a dimensionless scalar between 0 and 1. The surface area of ​​the obstacles represents the total exposed area of ​​all obstacles. The total surface area represents the total exposed surface area of ​​all entities (including obstacles, yarns, etc.) within the calculated area.

[0100] In this example, the distribution of obstacles in three-dimensional space is quantified into a simple density index. The higher the value, the larger the exposed area of ​​the obstacle in the space, meaning the more crowded the environment and the more difficult it is to avoid obstacles; the lower the value, the more open the space. It can be used as a quantitative input for subsequent dynamic obstacle avoidance safety distance calculation, enabling the system to adaptively adjust the safety boundary according to environmental complexity.

[0101] S3-9, Obtain the basic safety distance and obstacle density influence coefficient.

[0102] The basic safety distance can be determined based on the physical dimensions of the robotic arm and the spinning machine, as well as safety regulations. For example, it can be set by measuring the gap between the end effector of the robotic arm (such as a gripper) and the narrowest passage on the machine (such as between two yarn guide hooks), and allowing a certain margin (such as 5-10 mm). In this example, [the following is a more detailed description of the safety distance]. The value range is set to 5 mm to 10 mm.

[0103] Among them, the obstacle density influence coefficient This is an empirical parameter used to quantify the impact of obstacle density on the adjustment range of the safety distance. A larger value indicates a more sensitive system response to environments with dense obstacles, and a greater range for safety distance expansion. Specifically, during the training phase before system deployment, this parameter can be obtained through regression analysis of historical obstacle-crossing operation data (which can include successful and collision cases under different obstacle densities) or through reinforcement learning tuning. In this example, The value ranges from 0.8 to 1.2.

[0104] S3-10. Calculate the dynamic obstacle avoidance safety distance scaling factor based on the obstacle distribution density parameter and the obstacle density influence coefficient.

[0105] Among them, it can be based on obstacle distribution density parameters and obstacle density influence coefficient Calculate a dynamic scaling factor This is used to adjust the base safety distance in real time based on the level of environmental congestion. The scaling factor reflects the proportion by which the safety distance needs to be increased relative to the base value under the current obstacle density.

[0106] The formula for calculating the dynamic obstacle avoidance safety distance scaling factor is as follows: , In the formula, The dynamic obstacle avoidance safety distance scaling factor is a real-time scalar greater than or equal to 1. The obstacle density influence coefficient. This is the obstacle distribution density parameter.

[0107] In this example, the degree of environmental crowding ( Through influence coefficient This is converted into a linear scaling factor. When there are no obstacles in the environment ( When obstacles are dense ( ), the scaling factor is 1, and the safety distance remains at the base value; when obstacles are dense ( Close to 1) and When the scaling factor is large, it is significantly greater than 1, indicating that the safety distance needs to be increased significantly to cope with the higher collision risk. The scaling factor provides the core adjustment coefficient for the next step of calculating the upper limit of the dynamic safety distance, realizing the environmental adaptive capability of obstacle avoidance constraints.

[0108] S3-11. Based on the dynamic obstacle avoidance safety distance scaling factor and the basic safety distance, calculate the upper limit of the robot arm's safety distance and construct the obstacle avoidance safety distance constraint.

[0109] Among them, the upper limit of dynamic safe distance The calculation formula is as follows: In the formula, This represents the dynamic upper limit of the safe distance for the robotic arm. Basic safe distance, The obstacle density influence coefficient. This is the obstacle distribution density parameter.

[0110] The formula for the obstacle avoidance safety distance constraint is shown below: In the formula, Let t be the actual distance between the end effector of the robotic arm and the nearest obstacle. Based on the safe distance, The obstacle density influence coefficient. This is the obstacle distribution density parameter.

[0111] In this example, the static base safety distance can be used. Scaling factor reflecting real-time environmental congestion Combined, an adaptive safe distance threshold is generated. This means that the safety boundary is no longer a fixed value, but can change with the density of obstacles in the environment. Stretch.

[0112] In this example, the computed upper limit is transformed into a clear inequality constraint that can be directly utilized by the path planning algorithm. This rule mandates that all movement trajectories of the robotic arm must meet this distance condition.

[0113] In this example, the obstacle avoidance safety distance constraint can be directly transformed into a penalty term in the path generation network loss function. When the trajectory predicted by the network leads to... When a safe distance is violated, a penalty will be activated, thereby guiding the network to optimize its trajectory during training and inference and actively avoid obstacles.

[0114] In summary, through the obstacle avoidance safety distance constraints established in this step, the system achieves a closed loop from environmental perception to decision boundary generation. Compared to traditional fixed safety distance methods, this significantly improves obstacle avoidance intelligence in environments with dense, unstructured obstacles, such as spinning machines. When the system detects dense obstacles, it automatically increases the safety distance requirement, forcing the planned path to be more conservative and circuitous, fundamentally reducing the risk of collision. In open areas, it allows for more direct and efficient paths, thus achieving dynamic optimization of threading efficiency while ensuring absolute safety.

[0115] S4. Optimize the loss function of the pre-constructed robotic arm path generation network based on the tension-torque constraint, motion acceleration constraint, and obstacle avoidance safety distance constraint to obtain the optimized robotic arm path generation network.

[0116] In one possible implementation, step S4 includes the following sub-steps: S4-1. The tension-torque constraint is converted into a first penalty term, which generates a positive penalty when the joint torque of the robotic arm exceeds the upper limit constraint value of the joint torque.

[0117] Among them, the tension-torque constraint inequality This is transformed into a computable penalty term. Specifically, at each time step of network training or inference... Calculate the predicted joint torque The difference between the value and the dynamic upper limit. If If the dynamic upper limit is exceeded (i.e., the constraint is violated), a positive penalty is generated, and the penalty amount is proportional to the amount exceeded; if it is not exceeded, the penalty is zero. This penalty term, as the "first penalty term," forces the network to learn and generate a trajectory with gentle force that will not cause overload stretching to the yarn, fundamentally preventing the yarn from breaking due to excessive movement of the robotic arm.

[0118] S4-2. The motion acceleration constraint is converted into a second penalty term, which generates a positive penalty when the joint angular acceleration of the robotic arm exceeds the upper limit of the joint angular acceleration.

[0119] Among them, the motion acceleration constraint Converted into a penalty term. At each time step. Check the predicted joint angular acceleration Does it exceed the upper limit of its dynamic adjustment based on the yarn's vibration state? If it does, a positive penalty is applied based on the amount of excess. The purpose of this "second penalty" is to guide the network to automatically generate a smooth, low-acceleration motion trajectory when the yarn vibrates violently, avoiding increased yarn swaying or gripping failure due to rapid movement, thereby improving the stability of the threading process in dynamic environments.

[0120] S4-3. The obstacle avoidance safety distance constraint is converted into a third penalty term, which generates a positive penalty when the real-time distance between the end of the robotic arm and the obstacle is less than the upper limit of the robotic arm safety distance.

[0121] Among them, the obstacle avoidance safety distance constraint This is converted into a penalty. The penalty is based on the real-time distance between the robotic arm's end effector and the nearest obstacle. Perform the calculation: If If the distance is less than the dynamic safety distance limit (i.e., there is a risk of collision), a penalty proportional to the distance shortage is applied. This "third penalty" forces the network to consider the distribution density of obstacles in the environment when planning paths, and automatically plans more conservative detour trajectories that maintain a greater safety distance in high-density areas, effectively preventing collisions between the robotic arm and machine components.

[0122] S4-4. Assign a first weight, a second weight, and a third weight to the first penalty term, the second penalty term, and the third penalty term, respectively.

[0123] Among these, an adjustable weight coefficient can be assigned to the first, second, and third penalty terms, denoted as the first weight. Second weight and third weight These weights are used to balance the relative importance of different types of constraints in the total loss function. For example, in production processes where yarn is particularly vulnerable, they can be used to improve... To enhance protection against breakage; in areas with extremely dense obstacles, it can improve... Prioritize obstacle avoidance safety. The weights can be set based on domain experience or determined through hyperparameter tuning.

[0124] S4-5. Add the weighted first penalty term, second penalty term, and third penalty term to the original loss function of the pre-built robotic arm path generation network to obtain the optimized loss function and the optimized robotic arm path generation network.

[0125] The loss function of the optimized robotic arm path generation network is expressed as follows: , In the formula, Generate the total loss function value for the path generation network training. The joint angle at time t is predicted by the network. The reference joint angle at time t, These are the weighting coefficients for trajectory matching error. These are the weighting coefficients for trajectory smoothness constraints. These are the weighting coefficients for tension-related constraints. Here is the predicted value of the joint torque at time t. This represents the original upper limit of the joint torque. The tension influence coefficient is... The actual yarn tension at time t. The maximum allowable tension of the yarn. These are the weighting coefficients for jitter suppression constraints. Here is the predicted value of the joint angular acceleration at time t. The rated maximum angular acceleration of the robotic arm, This is the jitter impact coefficient. This refers to the yarn's resistance to vibration. The weighting coefficients for obstacle avoidance constraints. Based on the safe distance, The obstacle density influence coefficient. The obstacle distribution density parameter, Let t be the real-time minimum distance between the end effector of the robotic arm and the obstacle.

[0126] In this example, the path generation network transforms from a model that only learns motion patterns into an intelligent decision-making entity capable of understanding and responding to complex physical scenarios online. It can directly convert multi-source, time-varying sensory information (yarn tension, jitter, obstacle density) into training signals, giving the generated threading trajectory inherent robustness, safety, and adaptability. This significantly improves the first-time success rate and trajectory stability of the threading action, effectively reducing threading failures caused by breakage, slippage, or collisions. Under the premise of ensuring the safety of yarn and equipment, it meets the reliable requirements of fully automated, high-rhythm continuous operation in spinning production.

[0127] S5. Based on the optimized robotic arm path generation network, determine the optimized robotic arm spinning and threading motion trajectory sequence, and the robotic arm executes the optimized robotic arm spinning and threading motion trajectory sequence to complete the spinning and threading action.

[0128] This example provides a method for optimizing spinning threading actions based on deep learning. First, a continuous 3D point cloud sequence of the yarn breakage area, yarn physical properties, and robotic arm dynamic parameters are acquired. Then, tension-torque constraints related to yarn tensile strength are constructed based on the physical properties and dynamic parameters. Simultaneously, motion acceleration constraints related to anti-vibration capability and obstacle avoidance safety distance constraints related to obstacle distribution density are constructed based on the 3D point cloud sequence. Next, the loss function of the pre-constructed robotic arm path generation neural network is optimized based on these three types of scene-linked dynamic constraints to obtain the optimized network. Finally, this network generates and executes the robotic arm threading trajectory sequence. By adaptively integrating multi-source real-time perceived physical constraints into the deep learning trajectory generation process, the robotic arm can understand and respond online to the uncertainties brought about by yarn flexibility, susceptibility to vibration, and dense obstacle environments. This significantly improves the first-time success rate and trajectory stability of the threading action, effectively reducing threading failures caused by breakage, slippage, or collisions. While ensuring yarn safety, the threading time is shortened, meeting the reliable requirements of fully automated, high-rhythm continuous operation in spinning production.

[0129] For those consistent with the above, please refer to Figure 2 , Figure 2This application provides a schematic diagram of a spinning and threading action optimization system based on path deep learning, as an embodiment of the present application. Figure 2 As shown, the system includes: Acquisition Unit 1 is used to acquire the continuous three-dimensional point cloud sequence of the yarn breakage area, the maximum tensile stress that can be withstood, the yarn diameter, the dynamic parameters of the robotic arm, and the pre-constructed robotic arm path generation network.

[0130] The first processing unit 2 is used to construct a tension-torque constraint related to the tensile strength of the yarn based on the maximum withstandable tensile stress, yarn diameter, and robotic arm dynamic parameters.

[0131] The second processing unit 3 is used to construct motion acceleration constraints related to anti-shake capability and obstacle avoidance safety distance constraints related to obstacle distribution density, respectively, based on the continuous three-dimensional point cloud sequence and the dynamic parameters of the robotic arm.

[0132] The third processing unit 4 is used to optimize the loss function of the pre-constructed robotic arm path generation network according to the tension-torque constraint, motion acceleration constraint and obstacle avoidance safety distance constraint, so as to obtain the optimized robotic arm path generation network.

[0133] Execution unit 5 is used to determine an optimized robotic arm spinning and threading motion trajectory sequence based on the optimized robotic arm path generation network, and the robotic arm executes the optimized robotic arm spinning and threading motion trajectory sequence to complete the spinning and threading motion.

[0134] For examples consistent with the above embodiments, please refer to... Figure 3 , Figure 3 A schematic diagram of a terminal structure provided in an embodiment of this application is shown in the figure. It includes a processor, an input device, an output device, and a memory. The processor, input device, output device, and memory are interconnected. The memory is used to store a computer program, which includes program instructions. The processor is configured to call the program instructions. The program includes instructions for performing the following steps. Obtain the continuous three-dimensional point cloud sequence of the yarn breakage area, the maximum tensile stress that can be withstood, the yarn diameter, the dynamic parameters of the robotic arm, and the pre-constructed robotic arm path generation network.

[0135] Based on the maximum withstandable tensile stress, yarn diameter, and robotic arm dynamics parameters, a tension-torque constraint related to yarn tensile strength is constructed.

[0136] Based on the continuous three-dimensional point cloud sequence and the dynamic parameters of the robotic arm, motion acceleration constraints related to anti-shake capability and obstacle avoidance safety distance constraints related to obstacle distribution density are constructed respectively.

[0137] The loss function of the pre-constructed robotic arm path generation network is optimized based on the tension-torque constraint, motion acceleration constraint, and obstacle avoidance safety distance constraint to obtain the optimized robotic arm path generation network.

[0138] Based on the optimized robotic arm path generation network, an optimized robotic arm spinning and threading motion trajectory sequence is determined. The robotic arm executes the optimized robotic arm spinning and threading motion trajectory sequence to complete the spinning and threading motion.

[0139] In this example, a continuous three-dimensional point cloud sequence of the yarn breakage area, the yarn's physical properties, and the robotic arm's dynamic parameters are first obtained. Then, tension-torque constraints related to the yarn's tensile strength are constructed based on the physical properties and dynamic parameters. Simultaneously, motion acceleration constraints related to anti-vibration capability and obstacle avoidance safety distance constraints related to obstacle distribution density are constructed based on the three-dimensional point cloud sequence. Then, the loss function of the pre-constructed robotic arm path generation neural network is optimized based on the above three types of scene-linked dynamic constraints to obtain the optimized network. Finally, the network generates and executes the robotic arm's threading motion trajectory sequence. By adaptively integrating multi-source real-time perceived physical constraints into the deep learning trajectory generation process, the robotic arm can understand and respond online to the uncertainties brought about by the yarn's flexibility, susceptibility to vibration, and dense obstacle environment. This significantly improves the first-time success rate and trajectory stability of the threading action, effectively reduces threading failures caused by breakage, slippage, or collision, and shortens threading time while ensuring yarn safety, meeting the reliable requirements of fully automated, high-rhythm continuous operation in spinning production.

[0140] The above mainly describes the solutions of the embodiments of this application from the perspective of the method execution process. It is understood that, in order to achieve the above functions, the terminal includes the corresponding hardware structure and / or software modules for executing each function. Those skilled in the art should readily recognize that, in conjunction with the units and algorithm steps of the various examples described in the embodiments provided herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0141] This application embodiment can divide the terminal into functional units according to the above method example. For example, each function can be divided into a separate functional unit, or two or more functions can be integrated into one processing unit. The integrated unit can be implemented in hardware or as a software functional unit. It should be noted that the unit division in this application embodiment is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods.

[0142] This application also provides a computer storage medium storing a computer program for electronic data interchange, which causes a computer to perform some or all of the steps of any of the path deep learning-based spinning and threading action optimization methods described in the above method embodiments.

[0143] This application also provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program that causes a computer to perform some or all of the steps of any of the path deep learning-based spinning and threading action optimization methods described in the above method embodiments.

[0144] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0145] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0146] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical or other forms.

[0147] 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.

[0148] Furthermore, the functional units in the various embodiments of the application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software program module.

[0149] If the integrated unit is implemented as a software program module and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0150] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage device, which may include: a flash drive, a read-only memory, a random access memory, a magnetic disk, or an optical disk, etc.

[0151] The embodiments of this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for optimizing spinning and threading actions based on path deep learning, characterized in that, include: Obtain the continuous three-dimensional point cloud sequence of the yarn breakage area, the maximum tensile stress that can be withstood, the yarn diameter, the dynamic parameters of the robotic arm, and the pre-constructed robotic arm path generation network; Based on the maximum withstandable tensile stress, yarn diameter, and robotic arm dynamics parameters, a tension-torque constraint related to yarn tensile strength is constructed. Based on the continuous three-dimensional point cloud sequence and the dynamic parameters of the robotic arm, motion acceleration constraints related to anti-shake capability and obstacle avoidance safety distance constraints related to obstacle distribution density are constructed respectively. The loss function of the pre-constructed robotic arm path generation network is optimized based on the tension-torque constraint, motion acceleration constraint, and obstacle avoidance safety distance constraint to obtain the optimized robotic arm path generation network. Based on the optimized robotic arm path generation network, an optimized robotic arm spinning and threading motion trajectory sequence is determined. The robotic arm executes the optimized robotic arm spinning and threading motion trajectory sequence to complete the spinning and threading motion.

2. The method for optimizing spinning and threading actions based on path deep learning according to claim 1, characterized in that, The step of constructing a tension-torque constraint related to yarn tensile strength based on the maximum withstandable tensile stress, yarn diameter, and robotic arm dynamic parameters includes: Calculate the maximum allowable tension of the yarn based on the maximum withstandable tensile stress and the yarn diameter; The actual tension and tension influence coefficient applied by the robotic arm to the yarn are obtained, and the maximum output torque of the robotic arm is extracted from the robotic arm dynamic parameters. The tension risk ratio is obtained by calculating the ratio of the actual tension to the maximum permissible tension. Calculate the tension dynamic scaling factor based on the tension influence coefficient and tension risk ratio; Based on the maximum output torque of the robotic arm, the tension risk ratio, and the tension dynamic scaling factor, the upper limit constraint value of the joint torque is calculated, and the tension-torque constraint is constructed.

3. The method for optimizing spinning and threading actions based on path deep learning according to claim 1, characterized in that, The step of constructing motion acceleration constraints related to anti-shake capability based on the continuous three-dimensional point cloud sequence and the robotic arm dynamic parameters includes: A yarn point cloud sequence is segmented from the continuous three-dimensional point cloud sequence; Based on the yarn point cloud sequence, the yarn anti-vibration capability parameters are determined by calculating the displacement variance. Obtain the jitter influence coefficient and extract the rated maximum joint angular acceleration of the robotic arm from the robotic arm dynamic parameters; Calculate the dynamic scaling factor of acceleration based on the vibration influence coefficient and the yarn vibration resistance parameter; Based on the acceleration dynamic scaling factor and the rated maximum joint angular acceleration of the robotic arm, the upper limit of the joint angular acceleration is calculated, and motion acceleration constraints are constructed.

4. The method for optimizing spinning and threading actions based on path deep learning according to claim 3, characterized in that, The step of determining the yarn anti-vibration capability parameters based on the yarn point cloud sequence by calculating the displacement variance includes: For each frame of yarn point cloud sequence, the yarn centroid is extracted to obtain the three-dimensional centroid coordinates of the yarn. For the three-dimensional centroid coordinates of adjacent yarns in the yarn point cloud sequence, calculate the Euclidean distance between the three-dimensional centroid coordinates of adjacent yarns to obtain the inter-frame displacement sequence; Calculate the standard deviation of the inter-frame displacement sequence based on the inter-frame displacement sequence; Based on the standard deviation, the yarn vibration resistance parameter is calculated using a linear mapping method.

5. The method for optimizing spinning and threading actions based on path deep learning according to claim 1, characterized in that, The step of constructing obstacle avoidance safety distance constraints based on the continuous three-dimensional point cloud sequence and the robotic arm's dynamic parameters includes: Extract a frame of 3D point cloud from the continuous 3D point cloud sequence, perform semantic segmentation on the frame of 3D point cloud, and obtain the 3D point cloud of the machine obstacle. Based on the three-dimensional point cloud of the machine obstacle, the surface area of ​​the obstacle and the total surface area of ​​the region are calculated using the voxelized exposed area method. Calculate the obstacle distribution density parameter based on the surface area of ​​the obstacle and the total surface area of ​​the region; Obtain the basic safety distance and obstacle density influence coefficient; Calculate the dynamic obstacle avoidance safety distance scaling factor based on the obstacle distribution density parameter and the obstacle density influence coefficient; Based on the dynamic obstacle avoidance safety distance scaling factor and the basic safety distance, the upper limit of the robot arm's safety distance is calculated, and the obstacle avoidance safety distance constraint is constructed.

6. The method for optimizing spinning and threading actions based on path deep learning according to claim 5, characterized in that, The step of calculating the surface area and total surface area of ​​the obstacle using the voxelized exposed area method based on the three-dimensional point cloud of the obstacle on the machine platform includes: Define the three-dimensional spatial computation range and divide the three-dimensional spatial computation range into a uniform voxel grid; If a voxel contains at least one point of the obstacle point cloud, it is marked as an obstacle voxel. If a voxel contains at least one point cloud point of any class, it is marked as an occupied voxel. Based on the grid occupancy information of the obstacle voxels and the occupier voxels, the surface area of ​​the obstacle and the total surface area of ​​the region are determined.

7. The method for optimizing spinning and threading actions based on path deep learning according to claim 1, characterized in that, The optimization of the loss function of the pre-constructed robotic arm path generation network based on the tension-torque constraint, motion acceleration constraint, and obstacle avoidance safety distance constraint to obtain the optimized robotic arm path generation network includes: The tension-torque constraint is converted into a first penalty term, which generates a positive penalty when the joint torque of the robotic arm exceeds the upper limit constraint value of the joint torque. The motion acceleration constraint is transformed into a second penalty term, which generates a positive penalty when the joint angular acceleration of the robotic arm exceeds the upper limit of the joint angular acceleration. The obstacle avoidance safety distance constraint is transformed into a third penalty term, which generates a positive penalty when the real-time distance between the end of the robotic arm and the obstacle is less than the upper limit of the robotic arm safety distance. Assign a first weight, a second weight, and a third weight to the first penalty item, the second penalty item, and the third penalty item, respectively; The weighted first penalty term, second penalty term, and third penalty term are added to the original loss function of the pre-built robotic arm path generation network to obtain the optimized loss function and the optimized robotic arm path generation network.

8. The method for optimizing spinning and threading actions based on path deep learning according to claim 1, characterized in that, The formulas for the tension-torque constraint, motion acceleration constraint, and obstacle avoidance safety distance constraint are shown below in sequence: ， In the formula, Here is the predicted value of the joint torque at time t. This is the maximum output torque of the robotic arm. The tension influence coefficient is... The actual tension applied to the yarn by the robotic arm This is the maximum permissible tension of the yarn; ， In the formula, Let be the predicted value of the joint angular acceleration at time t. The rated maximum joint angular acceleration of the robotic arm, This is the jitter impact coefficient. This refers to the yarn's resistance to vibration. In the formula, Let t be the actual distance between the end effector of the robotic arm and the nearest obstacle. Based on the safe distance, The obstacle density influence coefficient. This is the obstacle distribution density parameter.

9. The method for optimizing spinning and threading actions based on path deep learning according to claim 1, characterized in that, The loss function of the pre-constructed robotic arm path generation network is optimized based on the tension-torque constraint, motion acceleration constraint, and obstacle avoidance safety distance constraint to obtain the optimized robotic arm path generation network. The expression formula of the loss function of the optimized robotic arm path generation network is as follows: ， In the formula, Generate the total loss function value for the path generation network training. The joint angle at time t is predicted by the network. The reference joint angle at time t, These are the weighting coefficients for trajectory matching error. These are the weighting coefficients for trajectory smoothness constraints. These are the weighting coefficients for tension-related constraints. Here is the predicted value of the joint torque at time t. This represents the original upper limit of the joint torque. The tension influence coefficient is... The actual yarn tension at time t. The maximum allowable tension of the yarn. These are the weighting coefficients for jitter suppression constraints. Here is the predicted value of the joint angular acceleration at time t. The rated maximum angular acceleration of the robotic arm, This is the jitter impact coefficient. This refers to the yarn's resistance to vibration. The weighting coefficients for obstacle avoidance constraints. Based on the safe distance, The obstacle density influence coefficient. The obstacle distribution density parameter, Let t be the real-time minimum distance between the end effector of the robotic arm and the obstacle.

10. A spinning and threading action optimization system based on path deep learning, characterized in that, include: The acquisition unit is used to acquire the continuous three-dimensional point cloud sequence of the yarn breakage area, the maximum tensile stress that can be withstood, the yarn diameter, the robot arm dynamic parameters, and the pre-constructed robot arm path generation network; The first processing unit is used to construct a tension-torque constraint related to the tensile strength of the yarn based on the maximum withstandable tensile stress, yarn diameter and robotic arm dynamic parameters. The second processing unit is used to construct motion acceleration constraints related to anti-shake capability and obstacle avoidance safety distance constraints related to obstacle distribution density based on the continuous three-dimensional point cloud sequence and the dynamic parameters of the robotic arm, respectively. The third processing unit is used to optimize the loss function of the pre-constructed robotic arm path generation network based on the tension-torque constraint, motion acceleration constraint and obstacle avoidance safety distance constraint, so as to obtain the optimized robotic arm path generation network. The execution unit is used to determine an optimized robotic arm spinning and threading motion trajectory sequence based on the optimized robotic arm path generation network, and the robotic arm executes the optimized robotic arm spinning and threading motion trajectory sequence to complete the spinning and threading motion.

Images