Double-mechanical-arm control system applied to different dyeing scenes

Abstract

The invention provides a double-mechanical-arm control system applied to different dyeing scenes, relates to the field of fabric dyeing control, and solves the technical problem that a fine cooperation strategy does not aim at different dip dyeing requirements. The method comprises the steps that a scene sensing module collects image information and physical information of a to-be-dyed fabric and judges the dyeing type; the central cooperative control and decision module is connected with the scene sensing module and used for calling a corresponding cooperative strategy according to the dyeing type to analyze and generate a double-mechanical-arm layered cooperative operation instruction set including the target movement track, the posture, the speed and the dip dyeing depth. And the double-mechanical-arm driving module receives the layered collaborative operation instruction set, calculates final motion parameters of all joints of the double mechanical arms and drives the double mechanical arms to perform control according to scenes. The self-adaptive updating module stores dyeing process parameter records to form a dyeing database. And optimizing and updating the collaborative strategy algorithm. The method is used in the fabric dyeing control process.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of fabric dyeing control, and in particular to a double-robot-arm control system applied to different dyeing scenes. BACKGROUND

[0002] In the field of automatic production of fabric dyeing, the double-robot-arm system has gradually been applied to the fabric dip dyeing process due to its cooperative work ability, and undertakes core tasks such as fabric clamping, dip dyeing posture adjustment and dye adsorption control. The control precision and cooperative strategy of the double-robot-arm system directly determine the dyeing efficiency and product quality. At present, the mainstream double-robot-arm dyeing control system in the industry generally adopts a simple synchronous motion mode, that is, the two arms perform immersion, lifting and other actions at a fixed interval and a unified speed, so that it cannot design a fine cooperative strategy for the dip dyeing needs of different fabrics and scenes. In the pure color fabric dyeing scene, it only relies on fixed clamping force and unified motion parameters for work. For fabrics with poor elasticity (such as some chemical fibers and blended fabrics), if the clamping force is too large, the fabric is easy to produce irreversible wrinkles in the dip dyeing process, which directly affects the flatness and appearance of the dyed fabric. If the clamping force is too loose, the fabric is easy to shift and sway in the dye, resulting in uneven contact of the fabric with the dye in different areas, and problems such as uneven dyeing depth, which reduces the dyeing uniformity of pure color fabrics. In the scene of patterned fabrics (fabrics containing multiple color areas and depth transition zones), there is also a lack of fine analysis ability of the pattern, and it is impossible to accurately identify the color depth, color ratio and range boundary of each color area of the fabric. The clamping point of the robot arm often falls into the color depth transition zone or the boundary area of different colors, so that the force in this area is uneven during dyeing, which causes the dye to penetrate and mix in the boundary area, destroys the edge definition of the pattern, and finally causes the pattern of the patterned fabric to be distorted and the boundary to be blurred after dyeing, which cannot meet the high-quality dyeing demand. SUMMARY

[0003] The present application provides a double-robot-arm control system applied to different dyeing scenes, which solves the technical problems that the prior art does not have a fine cooperative strategy for different dip dyeing needs, resulting in wrinkles or poor dyeing uniformity in pure color scenes, and uneven force in the boundary area of different colors in patterned scenes.

[0004] To achieve the above-mentioned purpose, the present application adopts the following technical solutions: In a first aspect, a double-robot control system applied to different dyeing scenarios includes: a scene perception module that collects image information and physical information of the fabric to be dyed, and determines the dyeing type, which is solid color or patterned color. A central collaborative control and decision module is connected to the scene perception module, and is used to call corresponding collaborative strategy analysis to generate a double-robot layered collaborative operation instruction set including target motion trajectory, posture, speed, and immersion depth. The collaborative strategy is a double-layer framework including an immersion posture optimization model for solid color type and a patterned color region adaptive adjustment model. The immersion posture optimization model is an algorithm that calculates the optimal double-robot clamping point, clamping angle, and immersion posture by analyzing fabric material, dye viscosity, drying speed, and real-time temperature, with the optimization objectives of maximizing the dye absorption per unit area of the fabric and minimizing the wrinkle deformation. The patterned color region adaptive adjustment model is an algorithm that dynamically calculates and adjusts the clamping position, angle, and timing and amplitude of the single-sided adjustment action of the double-robot by analyzing the color distribution, region boundary, and color depth information of the patterned color pattern, with the optimization objective of edge definition between different color regions. A double-robot driving module is connected to the central collaborative control and decision module, and is used to receive the layered collaborative operation instruction set and calculate the final motion parameters of each joint of the double-robot, and drive the double-robot to control in different scenes. In synchronization, the double-robot operates as a whole unit, and in local optimization, the double-robot independently performs fine-tuning actions. An adaptive updating module is connected to the central collaborative control and decision module, and is used to store dyeing process parameter records to form a dyeing database. Based on the comparison and analysis of historical data and real-time monitoring data, the collaborative strategy algorithm is optimized and updated, and the updated strategy is fed back to the central collaborative control and decision module.

[0005] Based on the above technical solution, in the double-robot control system applied to different dyeing scenarios provided in the present application, through the double-layer strategy framework of the solid color region immersion posture optimization model and the patterned color region adaptive adjustment model, autonomous coordination in different dyeing scenarios is achieved. In the solid color scenario, the optimization is performed with the multi-objective of "maximizing the absorption amount" and "minimizing the wrinkles", and the optimal clamping and motion parameters are solved. In the patterned color scenario, the edge definition is taken as the target, the clamping point is dynamically calculated, and the micro-motion to prevent color penetration is planned. Thus, the control strategy is changed from "one-size-fits-all" to "scenario-based and fine-tuned", and both the production efficiency of the absorption amount and the product quality of no wrinkles and clear edges are pursued.

[0006] In conjunction with the first aspect mentioned above, in one possible implementation, the scene perception module includes a perception layer and a decision layer: the perception layer is used to collect image information of the fabric to be dyed and physical information including material composition information, viscosity information of the dye in the vat, and real-time temperature information. The decision layer preprocesses and extracts features from the image information and physical information collected by the perception layer, calculates the variance of the image color feature distribution, and if the variance is less than or equal to a preset variance threshold, the dyeing type is determined to be solid color, and the material composition information, viscosity information, and real-time temperature information are sent to the central collaborative control and decision module. If the variance is greater than the preset variance threshold, the dyeing type is determined to be patterned, and the image segmentation submodule is activated to segment the surface image, identify the color and shade of different regions, and synchronously send the color, shade, image coordinate information of each region, along with the material composition information, viscosity information, and real-time temperature information, to the central collaborative control and decision module.

[0007] In conjunction with the first aspect mentioned above, in one possible implementation, the image segmentation submodule includes: converting the preprocessed RGB image of the fabric surface to be dyed from the RGB color space to the LAB color space, and extracting the color vector of each pixel. And calculate the average color difference ΔC between it and its 8 neighboring pixels. {avg} Here, an 8-neighbor pixel refers to eight pixels that are directly adjacent in the horizontal, vertical, and diagonal directions. The preprocessed RGB image of the fabric surface to be dyed is converted to grayscale, and the Sobel operator is used to calculate the gradient magnitude G of each pixel in the x and y directions. (x,y） And the average color difference ΔC {avg｝ and gradient magnitude G (x,y） Weighted fusion is performed to obtain the edge probability value P for each pixel. edge A mapping table between the sensitivity coefficient k, fabric material vector M, and target dyeing effect level E is established using historical data and stored in the database of the scene perception module. The basic segmentation threshold T is then obtained. base Based on the material vector M of the fabric to be dyed and the target dyeing effect level E, the mapping table is queried to obtain the current sensitivity coefficient k. Calculate the segmentation threshold T seg Where γ is the adjustment factor. The marginal probability value P... edge Greater than the dynamic segmentation threshold T seg Using the pixels as initial edge points, a region growing algorithm is used for image segmentation to form n candidate regions R. n , where n is the total number of candidate regions. Calculate the average brightness C of each pixel within all candidate regions. i Average saturation D i and contour coordinates point P i and area S i, and the area of all candidate regions is recorded as a region area set, where i is the index number of the region. The region area S i of the candidate region smaller than the preset area threshold is filtered out, and the remaining region information is output to the central collaborative control and decision module, where i' is the index of the filtered region, and S , Y

[0008] , and C are the region area, average brightness, average saturation, and contour coordinate point of the filtered region i, respectively. In combination with the first aspect, in a possible implementation, the dip-dye posture optimization model includes a data input layer, a relationship modeling layer, and an optimization calculation layer: the data input layer is configured to receive the fabric material vector M, the dye viscosity value V isc , the real-time temperature value T emp , and historical adsorption rate data from the scene perception module. The fabric material vector M is a multi-dimensional vector M = [m1, m 2, m n ] of the percentage content of cotton, hemp, silk, and chemical fiber material components measured by the near-infrared spectrum sensor, and satisfies ∑m n = 100%. The data input layer is also configured to receive the dye concentration value [Dye] from the dye vat concentration sensor and the elastic modulus distribution parameter E i corresponding to the current fabric material vector M. The relationship modeling layer obtains a material-dye adsorption relationship model trained from historical data, and runs the data input layer data through the material-dye adsorption relationship model to obtain the adsorption rate coefficient K ads and output, where f is a nonlinear mapping function trained by a neural network, used to represent the adsorption rate characteristics of the dye on the fabric under different material, viscosity, and temperature conditions. The effective dip-dye area A eff of the fabric and the dynamic weight factor ω are obtained, and after normalization, the adsorption rate coefficient K ads is combined to construct an optimization objective function with the maximum adsorption dye amount per unit area of the fabric and the minimum wrinkle deformation as the target, where Q ads is the total adsorption dye amount calculated by , S wrinkle is the wrinkle deformation degree calculated by , and Δx i is the displacement of the i-th grid element in the finite element model. The optimization calculation layer solves the objective function Z by an optimization algorithm to obtain the optimal combination of the clamping point coordinates, the clamping angle, and the dip posture that minimizes the value of Z, generates an executable collaborative operation instruction set, and outputs the set to the high-precision double-robot driving module.

[0009] In a possible implementation manner of the first aspect, the adaptive adjustment model comprises: obtaining the residual area information output by the scene perception module , extracting contour coordinate points P i ′ and solving the minimum circumscribed rectangle of the convex hull of all contour coordinate points P i ′ to obtain the minimum circumscribed rectangle R of the fabric. K points are uniformly sampled on the four sides of the minimum circumscribed rectangle R to form candidate clamping points, and all candidate clamping points on the opposite sides of the minimum circumscribed rectangle R are combined to form all possible opposite side point pairs to obtain a candidate clamping point pair set, each candidate clamping point (P A , P B ) in the candidate clamping point pair set represents a candidate clamping point P A of one side mechanical arm and a candidate clamping point P B of the other side mechanical arm. The center of gravity of the region i′ is calculated through the contour coordinate points P i ′, and the center of gravity of the region i′ is connected with the clamping point to obtain the length L of the calculation line, the vertical distance d （PA，PB） ′ of the center of gravity of the region i′ to the clamping point L i is obtained, and the support score S support is calculated through , wherein exp is an exponential function. The color boundary point E of the region i′ is extracted to obtain the distance d A of the clamping point P AE to the nearest color boundary point E and the distance d B of the clamping point P BE， to the nearest color boundary point E. The boundary avoidance score S boundary is calculated through . The boundary avoidance score S boundary and the support score S support are weighted and fused to obtain the clamping point comprehensive score S total . The maximum value of the clamping point comprehensive score S total is recorded as the candidate clamping point pair, and the clamping point coordinates (P Aopt , P Bopt） ) are output. Based on the fabric boundary tangent directions at the optimal clamping points P Aopt and P Bopt , the clamping angles θ A and θ B are calculated, and the tangent directions are obtained by calculating the differential of the contour points near the clamping points. Based on the residual area information and the dye viscosity value V isc , the independent adjustment action parameters required to prevent color penetration are calculated, and the independent adjustment action parameters and the clamping angles θ A and θ B are summarized to form an executable cooperative operation instruction set and output to the high-precision double mechanical arm driving module.

[0010] In conjunction with the first aspect above, in one possible implementation, calculating the independent adjustment parameters required to prevent color penetration includes: based on the dye viscosity value V isc Calculate the motion frequency F and motion amplitude A, and preset the start time T of the dyeing motion based on the dyeing process parameters. s and the end time T of the immersion action e Each color boundary is identified based on the intersection of the contours of two adjacent regions i′ and j′, and then... Calculate the contrast C of each color boundary. contrast In contrast C contrast When the color boundary exceeds the contrast threshold, retrieve the robotic arm identifier ID. Arm And summarize the amplitude of the movement A, the frequency of the movement F, and the start time of the immersion movement T. s and the end time T of the immersion action e， The independent adjustment action parameters are obtained, where the average saturation D in region i is... i′ The average saturation D of region j is less than j′ At that time, the robotic arm located on one side of region i′ is denoted as ID. Arm Otherwise, the robotic arm on the j′ side is denoted as ID. Arm .

[0011] In conjunction with the first aspect mentioned above, in one possible implementation, the dual-manipulator drive module includes: parsing an executable set of cooperative operation instructions, filtering the synchronous motion parameters of the two manipulators and recording them as reference cooperative motion layer instructions, wherein the reference cooperative motion layer instructions include the overall immersion speed V. t Overall lifting speed V I Overall immersion depth D. The independent motion parameters of a single robotic arm are recorded as independent adjustment commands, which include adjustment motion type T. action Amplitude of movement (A), frequency of movement (F), execution start time (T) s Execution end time T e and the function of robotic arm identification ID Arm Based on the baseline coordinated motion layer instructions, the ideal synchronous motion trajectory of the dual robotic arm end effectors in three-dimensional space is generated. The ideal synchronous motion trajectory is calculated using an inverse kinematics algorithm, and the output is the target angle sequence θ of each joint of the dual robotic arms. target(t） Where t is the time variable. Obtain the adjustment action type T. action The periodic function of amplitude A and frequency F is used to calculate the trajectory offset Δθ using a micro-motion model. (t） and through The trajectory offset Δθ (t） By superimposing this onto the target joint angle sequence of the corresponding robotic arm, the adjusted joint angle sequence θ is obtained. adjusted（t）The joint angle sequence θ adjusted（t） The angular velocity and angular acceleration of each joint are calculated, and a pulse width modulation signal is generated through a servo driving algorithm.

[0012] In combination with the first aspect, in a possible implementation, the micro-motion model includes: extracting effective micro-motion segments that prevent color penetration from end trajectory data of the mechanical arm in historical successful dyeing cases, and performing principal component analysis on the effective micro-motion segments to establish an association between the action primitives and the dyeing effect evaluation indicators, to obtain an action primitive library. According to the adjusted action type T action The corresponding target action primitive is selected from the action primitive library, the action amplitude A is mapped to the amplitude scaling coefficient of the action primitive, the action frequency F is mapped to the angular frequency of the action primitive, and the corresponding trajectory offset Δθ (t） .

[0013] In combination with the first aspect, in a possible implementation, the adaptive updating module includes: obtaining residual area information The area area S i and the average lightness C i ′ of the region i are extracted. The total area S total is calculated, and the average lightness variance V arc is calculated by calculating the variance of the average lightness C i ′ of all regions. Wherein C i ′ is a three-dimensional vector representing the color value in the LAB color space, and the variance calculation is respectively taken after averaging each component L, a, and b. The total area S total and the average lightness variance V arc of the historical successful dyeing cases are obtained, and the optimal speed ratio and amplitude ratio corresponding to the maximum dyeing uniformity f v and the edge sharpness f a are obtained through the trained neural network. The overall immersion speed V t and the overall pulling speed V I in the reference collaborative motion layer instruction are corrected by and to obtain the corrected overall immersion speed V t ′ and the overall pulling speed V I ′. The action amplitude A in the independent adjustment instruction layer instruction is corrected by to obtain the corrected amplitude A′. The corrected overall immersion speed V t ′, the overall pulling speed V I ′, and the amplitude A′ are summarized to obtain a layered collaborative operation instruction set, and are fed back to the double-robot driving module to recalculate the final motion parameters of each joint of the double robot, and drive the double robot to execute the motion.

[0014] In a second aspect, a dual-robot control device applied to different dyeing scenarios is provided, comprising: a communication unit and a processing unit; the communication unit is configured to collect image information and physical characteristic information of the fabric to be dyed by a scene perception module, and collect surface image information of the fabric to be dyed by a high-resolution industrial camera. The material composition information of the fabric is collected by a near-infrared spectrum sensor. The viscosity information of the dye in the dye vat is collected by a viscosity sensor. The real-time temperature information of the dye in the dye vat is collected by a temperature sensor; the processing unit is configured to determine the dyeing type, which is solid color or pattern color. The central collaborative control and decision module is connected to the scene perception module, and is configured to call corresponding collaborative strategy analysis to generate a dual-robot layered collaborative operation instruction set including target motion trajectory, posture, speed, and dip depth according to the dyeing type. The collaborative strategy is a double-layer framework including an immersion posture optimization algorithm for solid color type and a pattern region adaptive adjustment model for pattern color type. The immersion posture optimization algorithm is an algorithm for calculating the optimal clamping point, clamping angle, and immersion posture of the dual-robot by analyzing the fabric material, dye viscosity, drying speed, and real-time temperature, with the optimization objectives of maximizing the dye absorption per unit area of the fabric and minimizing the wrinkle deformation. The pattern region adaptive adjustment model is an algorithm for dynamically calculating and adjusting the clamping position, angle, and timing and amplitude of the single-side adjustment action of the dual-robot by analyzing the color distribution, region boundary, and color depth information of the pattern, with the optimization objective of edge definition between different color regions. A dual-robot driving module is connected to the central collaborative control and decision module, configured to receive the layered collaborative operation instruction set and calculate the final motion parameters of each joint of the dual-robot, drive the dual-robot to control in different scenes, and operate as a whole unit in synchronization, and independently fine-tune the action of the dual-robot in local optimization. An adaptive updating module is connected to the central collaborative control and decision module, configured to store dyeing process parameter records to form a dyeing database. Based on the comparison and analysis of historical data and real-time monitoring data, the collaborative strategy algorithm is optimized and updated, and the updated strategy is fed back to the central collaborative control and decision module.

[0015] In a third aspect, the application provides a dual-robot control device applied to different dyeing scenarios, comprising: a processor and a storage medium; the storage medium comprises instructions, and the processor is configured to run the instructions to implement the method described in the first aspect and any possible implementation manner of the first aspect. The dual-robot control device applied to different dyeing scenarios can be an electronic device or a chip in an electronic device.

[0016] In a fourth aspect, the present application provides a computer readable storage medium, which stores instructions. When the instructions are run on the dual-robot arm control device applied to different dyeing scenarios, the dual-robot arm control device applied to different dyeing scenarios executes the method described in the first aspect and any possible implementation manner of the first aspect.

[0017] In a fifth aspect, the present application provides a computer program product containing instructions. When the computer program product is run on the dual-robot arm control device applied to different dyeing scenarios, the dual-robot arm control device applied to different dyeing scenarios executes the method described in the first aspect and any possible implementation manner of the first aspect.

[0018] The present application provides a dual-robot arm control system applied to different dyeing scenarios. The dual-robot arm control system can realize autonomous coordination in different dyeing scenarios through a double-layer strategy framework of a solid color area immersion dyeing posture optimization model and a pattern color area self-adaptive adjustment model. In a solid color scenario, the dual-robot arm control system optimizes multiple targets of “maximum adsorption amount” and “minimum wrinkle” to solve optimal clamping and motion parameters. In a pattern color scenario, the dual-robot arm control system dynamically calculates clamping points and plans micro-motions to prevent color penetration with the target of “edge definition”. Thus, the control strategy of the dual-robot arm control system changes from “one-size-fits-all” to “scenario-based and refined”, and the production efficiency of adsorption amount and the product quality of no wrinkle and clear edge can be pursued at the same time.

[0019] It should be understood that the description of technical features, technical solutions, advantages or similar language in the present application does not imply that all features and advantages can be achieved in any single embodiment. On the contrary, it can be understood that the description of a feature or advantage means that the specific technical feature, technical solution or advantage is included in at least one embodiment. Therefore, the description of technical features, technical solutions or advantages in the specification does not necessarily refer to the same embodiment. Further, the technical features, technical solutions and advantages described in the embodiments can be combined in any appropriate manner. Those skilled in the art will understand that the embodiments can be implemented without one or more specific technical features, technical solutions or advantages of a particular embodiment. In other embodiments, additional technical features and advantages can be identified in specific embodiments that do not embody all embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 A system architecture diagram of a dual-robot arm control system applied to different dyeing scenarios is provided for the embodiments of the present application; Figure 2 A flowchart of a dual-robot arm control system applied to different dyeing scenarios is provided for the embodiments of the present application; Figure 3 Another flowchart of a double-robot arm control system applied to different dyeing scenarios provided by an embodiment of the present application is shown in FIG. 6; Figure 4 Another flowchart of a double-robot arm control system applied to different dyeing scenarios provided by an embodiment of the present application is shown in FIG. 6; Figure 5 Another flowchart of a double-robot arm control system applied to different dyeing scenarios provided by an embodiment of the present application is shown in FIG. 6; Figure 6 Another flowchart of a double-robot arm control system applied to different dyeing scenarios provided by an embodiment of the present application is shown in FIG. 6; Figure 7 Another flowchart of a double-robot arm control system applied to different dyeing scenarios provided by an embodiment of the present application is shown in FIG. 6. DETAILED DESCRIPTION

[0021] In the description of the present application, unless otherwise specified, “ / ” means “or”, for example, A / B can mean A or B. “And / or” in this document is only a description of the relationship between the associated objects, which means that there can be three relationships, for example, A and / or B, which can mean that A exists alone, A and B exist together, and B exists alone. In addition, “at least one” means one or more, and “multiple” means two or more. “First”, “second”, and the like do not limit the quantity and execution order, and “first”, “second”, and the like do not necessarily mean different.

[0022] It should be noted that in the present application, the words “exemplary” or “for example” are used to mean serving as an example, instance, or illustration. Any embodiment or design scheme described as “exemplary” or “for example” in the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. Rather, the use of “exemplary” or “for example” is intended to present the relevant concept in a specific manner.

[0023] To solve the problem that the prior art cannot automatically and accurately distinguish the dyeing type (such as solid color or pattern color), often relies on manual experience or simple threshold judgment, resulting in high misjudgment rate of dyeing type, affecting the accuracy of mechanical arm control decision. Secondly, the mechanical arm control adopts a "one-size-fits-all" fixed mode, which fails to dynamically adjust the collaborative strategy according to different dyeing scenes; in solid color dyeing, it is difficult to optimize the dye adsorption amount and fabric wrinkle deformation at the same time, often leading to low adsorption efficiency or fabric damage, while in pattern color dyeing, the color boundary processing is rough, which easily causes dye penetration problem, making the region edge unclear, reducing the dyeing quality. In addition, the existing control system lacks effective adaptive updating mechanism, and cannot optimize the control parameters based on historical process data and real-time monitoring information, resulting in rigid strategy and difficulty in coping with dynamic changes of fabric material, dye viscosity, etc. At the same time, the collaborative control of double mechanical arms also has shortcomings, the conflict between synchronous motion and local fine tuning has not been solved, and the double arms cannot independently execute fine actions to prevent color penetration while maintaining the overall unit operation. The embodiments of the present application provide a double mechanical arm control system applied to different dyeing scenes, which can realize autonomous coordination in different dyeing scenes through a double-layer strategy framework of solid color area immersion dyeing posture optimization model and pattern color area adaptive adjustment model, optimize the "maximum adsorption amount" and "minimum wrinkle" as multi-objectives in solid color scene, solve the optimal clamping and motion parameters; in pattern color scene, the edge definition is taken as the target, the clamping point is dynamically calculated, and the micro-motion to prevent color penetration is planned. Thus, the control strategy changes from "one-size-fits-all" to "scene-based and fine-tuned", which can pursue both production efficiency of adsorption amount and product quality of wrinkle-free and clear edge.

[0024] As shown in Figure 1 The embodiments of the present application provide a double mechanical arm control system applied to different dyeing scenes, which comprises: Step 101, the scene perception module collects image information and physical information of the fabric to be dyed, judges the dyeing type, and the dyeing type is solid color or pattern color.

[0025] The scene perception module refers to a system component responsible for collecting image information and physical information of the fabric to be dyed and judging the dyeing type. The image information refers to the surface image data of the fabric to be dyed collected by a high-resolution industrial camera. The physical information refers to physical characteristic data including fabric material composition information, viscosity information of dye in the vat, and real-time temperature information, etc. The dyeing type refers to the fabric dyeing category judged according to the image color feature distribution, which is divided into solid color or pattern color. Solid color refers to the dyeing type with uniform image color feature distribution and small variance. Pattern color refers to the dyeing type with uneven image color feature distribution and large variance, involving multiple colors or light and dark regions.

[0026] In some implementations, the scene perception module first collects the fabric surface image through a high-resolution industrial camera, collects the material composition information through a near-infrared spectrum sensor, collects the dye viscosity information through a viscosity sensor, and collects the real-time temperature information through a temperature sensor. That is, the collected image information and physical information are preprocessed and feature extracted, the image color feature distribution variance is calculated, and compared with the preset variance threshold, so as to judge the type of dyeing, so that the double robot can be controlled according to the actual needs of different types of scenes, thereby improving the flexibility and convenience of the overall use.

[0027] In step 102, the central collaborative control and decision module is connected to the scene perception module, and is used to call corresponding collaborative strategies according to the type of dyeing to generate a double-robot layered collaborative operation instruction set including target motion trajectory, posture, speed, and dip depth. The collaborative strategy is a double-layer framework including a dip posture optimization model for solid color type and a pattern area adaptive adjustment model for pattern type. The dip posture optimization model is an algorithm for calculating the optimal clamping point, clamping angle, and dip posture of the double robot by analyzing the fabric material, dye viscosity, drying speed, and real-time temperature, with the maximum absorption of dye per unit area of fabric and the minimum deformation of wrinkles as the optimization objective. The pattern area adaptive adjustment model is an algorithm for dynamically calculating and adjusting the clamping position, angle, and timing and amplitude of single-sided adjustment action of the double robot by analyzing the color distribution, area boundary, and color depth information of the pattern, with the edge definition between different color areas as the optimization objective.

[0028] The central collaborative control and decision module refers to the core processing unit connected to the scene perception module and calling corresponding collaborative strategies according to the type of dyeing. The collaborative strategy refers to a double-layer framework of a dip posture optimization model for solid color type and a pattern area adaptive adjustment model for pattern type, which can generate an algorithm framework for double-robot control instructions. The layered collaborative operation instruction set refers to a structured control instruction set containing parameters such as target motion trajectory, posture, speed, and dip depth.

[0029] In some implementations, the dyeing type judgment result from the scene perception module is received, and a corresponding cooperative strategy is called according to the dyeing type. If it is a solid color type, a dip dyeing posture optimization model is called, parameters such as fabric material vector, dye viscosity value, and real-time temperature value are received through a data input layer, the material-dye adsorption relationship model is run to obtain an adsorption rate coefficient, an optimization objective function is constructed with the maximum adsorption of dyes and the minimum deformation of wrinkles as the goal based on the effective dip dyeing area of the fabric and a dynamic weight factor, and finally the optimal clamping point coordinates, clamping angle, and dip posture combination are obtained by solving the optimization calculation layer. If it is a pattern type, a pattern area adaptive adjustment model is called, the remaining area information output by the scene perception module is obtained, the optimal clamping point pair is obtained by calculating the support score and the boundary avoidance score, the clamping angle is calculated based on the boundary tangent direction, and the independent adjustment action parameters required to prevent color penetration are calculated according to the area information and the dye viscosity value. Finally, the calculation results are parsed to generate a layered cooperative operation instruction set containing target motion trajectories, postures, speeds, and dip depths.

[0030] It should be noted that the double-layer frame design is adopted, and the dip dyeing posture optimization model is optimized for solid color and pattern types, so that the dye adsorption and fabric deformation are balanced by the multi-objective optimization function; the pattern area adaptive adjustment model dynamically selects the clamping point through the scoring mechanism to ensure the color boundary definition; and the generated layered cooperative operation instruction set contains detailed motion parameters, providing accurate control basis for double-robot driving.

[0031] Step 103, a double-robot driving module connected to the central cooperative control and decision module, for receiving the layered cooperative operation instruction set and calculating the final motion parameters of each joint of the double robot, driving the double robot to control in different scenes, operating as a whole unit in synchronization, and independently fine-tuning the action in local optimization.

[0032] Among them, the double-robot driving module refers to the functional unit connected to the central cooperative control and decision module and used for receiving the layered cooperative operation instruction set. The final motion parameters refer to the driving parameters such as angular velocity and angular acceleration of each joint of the double robot. The scene control refers to the process of differentiating the driving of the double robot according to the dyeing type (solid color or pattern). As a whole unit in synchronization refers to the double robot maintaining cooperative motion in a macroscopic sense, such as dipping or pulling at the same speed. Independent fine-tuning action in local optimization refers to the double robot performing independent adjustment action when needed, such as swinging or shaking a single arm.

[0033] In some implementations, the central collaborative control and decision module is connected first, and receives a hierarchical collaborative operation instruction set. The instruction set is parsed to filter out the benchmark collaborative motion layer instructions (including overall immersion speed, overall pulling speed, overall immersion depth, etc.) and the independent adjustment instruction layer instructions (including adjustment action type, action amplitude, action frequency, execution start time, execution end time, action mechanical arm identifier, etc.). Based on the benchmark collaborative motion layer instructions, the ideal synchronous motion trajectory of the double-robot end effector in three-dimensional space is generated through a kinematics inverse solution algorithm, and the output is the target angle sequence of each joint of the double robot. For the independent adjustment instruction layer instructions, the trajectory offset is calculated through the micro-action model, and the trajectory offset is superimposed on the joint target angle sequence of the corresponding robot to obtain the adjusted joint angle sequence.

[0034] Step 104, the adaptive updating module is connected to the central collaborative control and decision module, and is used to store the dyeing process parameter records to form a dyeing database. Based on the comparative analysis of historical data and real-time monitoring data, the collaborative strategy algorithm is optimized and updated, and the updated strategy is fed back to the central collaborative control and decision module.

[0035] The adaptive updating module refers to a functional unit connected to the central collaborative control and decision module and used to store dyeing process parameter records. The dyeing process parameter records refer to a collection of data such as environmental parameters and robot motion parameters monitored during each immersion process. The dyeing database refers to a relational database that stores dyeing process parameter records and can be queried and analyzed. The historical data refers to the past successful dyeing process records. The real-time monitoring data refers to the parameter data collected in real time during the current dyeing process. The comparative analysis refers to the process of comparing historical data and real-time monitoring data to identify differences.

[0036] In some implementations, the central collaborative control and decision module is connected to receive dyeing process parameter records from the dyeing process multi-parameter monitoring module. A unique identifier is generated for each record, which is associated with scene information, fabric information, and target color information for indexed storage in a relational database to form a dyeing database. The same scene in the dyeing database can be periodically analyzed to calculate the stable interval of key process parameters such as immersion speed and immersion time. When the central collaborative control and decision module issues a new instruction set, the preset parameters in the new instruction set are compared with the parameter stable interval of the historical successful records. If the parameters of the new instruction set deviate from the stable interval, a parameter correction suggestion is generated. If the actual parameters monitored in real time continuously deviate from the preset values of the instruction set, a system maintenance alarm is generated. Finally, the parameter correction suggestion is fed back to the central collaborative control and decision module for optimizing its internal collaborative strategy algorithm, such as adjusting the adsorption rate calculation in the immersion posture optimization model or the clamping point selection rule in the pattern area adaptive adjustment model.

[0037] Based on the above technical scheme, through the double-layer strategy framework of the solid color area immersion posture optimization model and the pattern color area self-adaptive adjustment model, autonomous coordination can be realized under different dyeing scenes, so that in the solid color scene, the multi-objective optimization of "maximum adsorption" and "minimum wrinkle" is carried out, and the optimal clamping and motion parameters are solved; in the pattern color scene, the edge definition is taken as the target, the clamping points are dynamically calculated, and the micro-motion for preventing color penetration is planned. Thus, the control strategy is changed from "one-size-fits-all" to "scene-based and refined", and the production efficiency of adsorption amount and the product quality of wrinkle-free and clear edge can be pursued at the same time. At the same time, by cooperating with the hierarchical collaborative operation instruction set containing the reference collaborative motion layer instruction and the independent adjustment instruction layer instruction, the dual-robot arms can complete high-precision synchronous motion as a whole unit to ensure consistent basic dyeing action, and can also independently perform controlled fine adjustment action to realize local optimization, such as preventing pattern color penetration, thereby solving the conflict problem between collaborative and independent tasks of the multi-arm system and improving the overall flexibility.

[0038] In a possible implementation manner of the embodiment of the present application, the scene perception module includes a perception layer and a decision layer, which can be implemented through the following steps 201 to 204, which are described in detail as follows. Figure 1 As shown in FIG. 2, the scene perception module includes a perception layer and a decision layer. Figure 2 The scene perception module can be implemented through the following steps 201 to 204, which are described in detail as follows. Step 201, the perception layer is configured to collect image information of the fabric to be dyed and physical information including material composition information, viscosity information of the dye in the vat and real-time temperature information.

[0039] The perception layer is responsible for directly collecting various input data of the fabric to be dyed in the scene perception module.

[0040] In some implementation manners, the perception layer photographs the fabric to be dyed through a high-resolution industrial camera, collects the surface image information, and obtains high-resolution digital image data; detects and outputs the percentage content of cotton, hemp, silk and chemical fiber components in the form of a multi-dimensional vector through a near-infrared spectrum sensor to scan the surface of the fabric; at the same time, the perception layer also collects the viscosity information of the dye in the vat through a viscosity sensor installed in the vat to monitor the viscosity of the dye in real time; collects the real-time temperature information through a temperature sensor to continuously read the temperature of the dye in the vat; all the collected image information and physical information (including material composition, viscosity and temperature) are summarized to provide the system with original and multi-modal input data basis.

[0041] In step 202, the decision layer pre-processes and extracts features from the image information and physical information collected by the perception layer, calculates the variance of the image color feature distribution, and if the variance is less than or equal to a preset variance threshold, determines that the dyeing type is solid color, and sends the material composition information, viscosity information, and real-time temperature information to the central collaborative control and decision module. If the variance is greater than the preset variance threshold, it is determined that the dyeing type is patterned, and the image segmentation submodule is started to segment the surface image, identify the color and depth of different regions, and send the color, depth, image coordinate information of each region, material composition information, viscosity information, and real-time temperature information to the central collaborative control and decision module.

[0042] The decision layer is the level responsible for data processing and type judgment in the scene perception module. Pre-processing is the operation of cleaning, standardizing and format converting the original image and physical data. Feature extraction is to calculate key indicators such as color distribution from the pre-processed image. The variance of the image color feature distribution is a statistical quantity that quantifies the dispersion degree of the pixel color value in the image. The preset variance threshold is a pre-set value used to compare the size of the variance. The image segmentation submodule is an algorithm component used to segment the patterned image into different color regions. The color and depth of different regions refer to the average color value and brightness value of each region after segmentation. Image coordinate information is the outline coordinate point position of each region in the image. The central collaborative control and decision module is the core processing unit that receives these information and generates control instructions.

[0043] In some implementations, the decision layer first receives the image information and physical information collected by the perception layer, then pre-processes the image information (such as removing noise and color space conversion), extracts features, and calculates the variance of the image color feature distribution; the calculated variance value is compared with the preset variance threshold, if the variance is less than or equal to the threshold, it is determined that the dyeing type is solid color, and the material composition information, viscosity information, and real-time temperature information are sent to the central collaborative control and decision module; if the variance is greater than the threshold, it is determined that the dyeing type is patterned, and the image segmentation submodule is started to segment the surface image, identify the color, depth, and image coordinate information of different regions; finally, these regional information and physical information are sent to the central collaborative control and decision module.

[0044] It should be noted that the preset variance threshold can be dynamically adjusted according to historical data to adapt to different fabric types.

[0045] The image segmentation submodule includes: The pre-processed surface RGB image of the fabric to be dyed is converted from the RGB color space to the LAB color space, and the color vector of each pixel point is extracted , and the average color difference AC of 8-neighborhood pixel points is calculated {avg}wherein the 8-neighborhood pixel points refer to 8 pixel points directly adjacent in horizontal, vertical, and diagonal directions.

[0046] wherein the image segmentation sub-module is a segmentation algorithm component in the scene perception module for processing the patterned fabric image. The RGB image is a color image composed of red, green, and blue channels. The LAB color space is a color model in which L represents lightness, and a and b represent color-opponent dimensions. The color vector [L, a, b] is the color numerical representation of each pixel point in the LAB space.

[0047] The pre-processed surface RGB image of the fabric to be dyed is subjected to grayscale processing, and the Sobel operator is used to calculate the gradient amplitude G (x,y） of each pixel point in the x and y directions. {avg｝ The average color difference ΔC (x,y） and the gradient amplitude G edge are weighted and fused to obtain the edge probability value P (x,y） of each pixel point.

[0048] wherein the grayscale processing is the process of converting the RGB image into a single-channel grayscale image. The Sobel operator is an edge detection operator used to calculate the gradient of the image. The gradient amplitude G edge is the amplitude of the Sobel operator in the x and y direction derivatives, used to represent the edge strength. The edge probability value P base is the probability value of each pixel point belonging to the edge obtained by weighted fusion of the average color difference and the gradient amplitude.

[0049] The mapping relationship table of the sensitivity coefficient k and the fabric material vector M, target dyeing effect level E is established through historical data and stored in the database of the scene perception module.

[0050] wherein the sensitivity coefficient k is a parameter for adjusting the segmentation sensitivity according to the fabric material and the target dyeing effect. The fabric material vector M is a multi-dimensional vector of the percentage content of the material composition measured by the near-infrared spectrum sensor. The target dyeing effect level E is the expected dyeing quality level. The mapping relationship table is a data table storing the corresponding relationship of the sensitivity coefficient k, material vector M, and effect level E.

[0051] The basic segmentation threshold T seg is obtained, and the current sensitivity coefficient k is obtained by querying the mapping relationship table according to the material vector M and the target dyeing effect level E of the current fabric to be dyed, and the segmentation threshold T base is calculated, wherein γ is the adjustment factor.

[0052] wherein the basic segmentation threshold T edge is the average segmentation threshold reference value obtained through historical data. The adjustment factor γ is a coefficient for adjusting the threshold calculation sensitivity.

[0053] edge probability value P edge greater than a dynamic segmentation threshold T seg as initial edge points, and using a region growing algorithm to perform image segmentation to form n candidate regions R n , wherein n is the total number of candidate regions.

[0054] calculating the average brightness C i , average saturation D i , and contour coordinate point P i , and region area S i of each pixel point in all candidate regions, obtaining the region area of all candidate regions, denoted as a region area set, wherein i is the index number of the region.

[0055] filtering out candidate regions with a region area S i less than a preset area threshold in the region area set, and outputting the remaining region information to a central collaborative control and decision module, wherein i' is the index of the filtered region, respectively, the region area, average brightness, average saturation, and contour coordinate point of the filtered region i.

[0056] , wherein the region growing algorithm is an image segmentation algorithm based on the similarity of seed points for region expansion. The candidate region R n is the image region formed by preliminary segmentation. The average brightness C i is the average value of the pixel brightness L in the candidate region. The average saturation D i is the average value of the pixel color saturation in the candidate region. The contour coordinate point P i is the set of coordinate points of the boundary of the candidate region. The region area S i is the number of pixels contained in the candidate region. The region area set is the set of all candidate region areas. The preset area threshold is the minimum limit of the area for filtering small regions.

[0057] In some implementations, the image segmentation submodule converts the preprocessed surface RGB image of the fabric to be dyed from the RGB color space to the LAB color space, extracts the color vector [L, a, b] of each pixel point, and calculates the average color difference AC arg of each pixel point and its 8-neighborhood pixel points; so that the color distribution is more uniform, which is conducive to accurate segmentation; then the same RGB image is subjected to grayscale processing, and the Sobel operator is used to calculate the gradient amplitude G (x,y） of each pixel point in the x direction and the y direction; and the average color difference AC arg and the gradient amplitude G (x,y） are weighted and fused to obtain the edge probability value P edge; meanwhile, a mapping relationship table of the sensitivity coefficient k and the fabric material vector M and the target dyeing effect grade E is established through historical data and stored in a database of the scene perception module; then, a basic segmentation threshold T base is obtained, and the current sensitivity coefficient k is obtained by querying the mapping relationship table according to the material vector M and the target dyeing effect grade E of the current fabric to be dyed, and the segmentation threshold T seg is calculated, thereby improving the flexibility of segmentation. That is, a pixel point with an edge probability value P edge greater than the dynamic segmentation threshold T seg is taken as an initial edge point, a region growing algorithm is used for image segmentation, color continuous regions are effectively processed, and the boundary integrity is ensured, thereby forming n candidate regions R n ; the average brightness C i , the average saturation D i , the contour coordinate point P i and the area S i of each pixel point in all candidate regions are calculated, and an area set of all candidate regions is obtained; finally, the candidate regions with the area S i less than a preset area threshold are filtered out, and the remaining region information is output to the central collaborative control and decision module.

[0058] Based on the above technical scheme, a multi-modal data acquisition system is constructed through hardware integration of an industrial camera, a near-infrared spectrometer, a viscosity sensor and a temperature sensor, so that the visual features and physical properties of the fabric can be synchronously obtained, comprehensive data support is provided for subsequent decision-making, the problem of misjudgment caused by complex material and dye property changes is avoided, and the traditional system relies on single visual data. At the same time, based on the software algorithm architecture of image color feature distribution variance calculation and threshold comparison, automatic dyeing type classification can be completed, the pure color type is directly transmitted to the physical parameters, the pattern type triggers the advanced segmentation process, and the problems of strong subjectivity, low efficiency and poor consistency in manual experience judgment of dyeing type are avoided. Secondly, the processing pipeline composed of LAB color space conversion, 8-neighborhood color difference calculation, Sobel gradient detection and weighted fusion algorithm can improve the detection accuracy of color boundaries and texture edges, especially the recognition ability in color gradient areas, and reduce the insufficient response to color gradient edges, so that the segmentation discontinuity occurs. Finally, through the mapping relationship table constructed through historical data, the material vector and the effect grade E are combined with the sensitivity coefficient, the segmentation threshold is dynamically optimized according to the fabric material and the dyeing requirement, and the segmentation robustness under different working conditions is improved. At the same time, the region growing algorithm is used to generate candidate regions and the area threshold filtering mechanism, which effectively filters out noise and irrelevant small regions while maintaining the integrity of the color region.

[0059] In a possible implementation of the embodiment of the present application, the dip dyeing posture optimization model including the data input layer, the relationship modeling layer and the optimization calculation layer can be implemented through the following steps 301 to step 204, which are described in detail as follows. Step 301, the data input layer is used to receive the fabric material vector M, the dye viscosity value V isc , the real-time temperature value T emp and the historical adsorption rate data from the scene perception module. The fabric material vector M is a multi-dimensional vector M = [m1, m2, …, m n ] of the percentage content of cotton, hemp, silk and chemical fiber material components measured by the near-infrared spectrum sensor, and satisfies ∑m n = 100%. The data input layer is also used to receive the dye concentration value [Dye] from the dye vat concentration sensor and the elastic modulus distribution parameter E i corresponding to the current fabric material vector M.

[0060] The dye viscosity value V isc is the viscosity measurement value of the dye in the dye vat, reflecting the flow characteristics of the dye. The real-time temperature value T emp is the real-time temperature data of the dye in the dye vat, affecting the dye adsorption process. The historical adsorption rate data is the recorded adsorption rate information of the dye on the fabric in the past dyeing process, which is used for model reference. The dye concentration value [Dye] is the concentration measurement value of the dye, which directly affects the calculation of the adsorption amount. The elastic modulus distribution parameter E i is the parameter corresponding to the fabric material vector M, which represents the elastic properties of the fabric and is used to evaluate the wrinkle deformation.

[0061] In some implementations, the data input layer obtains the fabric material vector M, the dye viscosity value V isc , the real-time temperature value T emp and the historical adsorption rate data from the scene perception module, and receives the dye concentration value [Dye] obtained by real-time sampling and calibration from the dye vat concentration sensor, reflecting the actual concentration of the dye in the dye vat; that is, the corresponding elastic modulus distribution parameter E i of the fabric can be queried from the preset database according to the current fabric material vector M, to describe the elastic properties of the fabric. Thus, after the above data is uniformly formatted and verified, it is transmitted to the relationship modeling layer to provide basic input for subsequent adsorption rate calculation and optimization objective function.

[0062] It should be noted that the percentage of the fabric material vector M is automatically verified by the algorithm, avoiding the problem of inconsistent data.

[0063] Step 302, the relationship modeling layer obtains the material-dye adsorption relationship model trained by the historical data, and runs the data input layer data through the material-dye adsorption relationship model The adsorption rate coefficient K was obtained. ads The output is f, which is a nonlinear mapping function obtained by training a neural network, used to characterize the adsorption rate characteristics of dyes on fabrics under different material, viscosity and temperature conditions.

[0064] The material-dye adsorption relationship model is a mathematical model trained through machine learning to describe the adsorption behavior of dyes on fabrics. The adsorption rate coefficient Kads is a key output parameter quantifying the adsorption rate of dyes under specific conditions. The neural network is a computational model that simulates the structure of the human brain by learning complex patterns through training. The nonlinear mapping function f is a mathematical function implemented by a neural network that can capture the complex nonlinear relationship between input and output. Material refers to the material composition of the fabric, such as cotton or linen. Viscosity is a measure of the internal friction of a fluid, affecting the flow and penetration of dyes. Temperature is a thermodynamic state parameter of the dye's environment, affecting the molecular motion speed and the adsorption process. Adsorption rate characteristics refer to the variation of the amount of dye adsorbed on the fabric per unit time.

[0065] In some implementations, a large amount of historical data is first retrieved from the system storage unit. This historical data can then be used for supervised training of the artificial neural network, using the fabric material vector M and the dye viscosity value V. isc and real-time temperature value T emp Using the actual measured adsorption rate as the training target and the backpropagation algorithm to continuously adjust the network weights, a well-trained material-dye adsorption relationship model is obtained. Then, the current operating condition data (fabric material vector M, measured dye viscosity value V) transmitted in real time from the data input layer is used. isc and real-time temperature value T emp The adsorption rate coefficient Kads is then input into the pre-trained material-dye adsorption relationship model and run. The model can then be calculated using the learned nonlinear mapping function f, and the adsorption rate coefficient Kads can be output and passed to the subsequent optimization calculation layer.

[0066] Step 303: Obtain the effective dyeing area A of the fabric. eff After normalization with the dynamic weighting factor ω, it is combined with the adsorption rate coefficient K. ads An optimization objective function is constructed to maximize the amount of dye adsorbed per unit area of ​​the fabric and minimize the wrinkle deformation. Q ads The total amount of adsorbed dye is obtained by passing through Calculations show that S wrinkle The degree of wrinkle deformation is determined by The calculation yields Δx i Let be the displacement of the i-th mesh element in the finite element model.

[0067] where A is the effective dyeing area of the fabric eff is the effective area of the fabric actually contacted with the dye during the dyeing process, which is calculated by measuring the fabric size and the dyeing conditions. The dynamic weight factor ω is an adjustable weight parameter between 0 and 1, which is used to balance the importance of different indicators in the optimization objective. The optimization objective function Z is a mathematical function used to minimize or maximize in the optimization process to achieve the goal, which combines the adsorbed dye amount and the wrinkle deformation at this time.

[0068] In some implementations, the effective dyeing area A of the fabric is obtained from the sensor or database eff and the dynamic weight factor ω, that is, the effective dyeing area A of the fabric eff and the dynamic weight factor ω are normalized (for example, using the minimum-maximum scaling method to map the value to the [0, 1] interval, eliminating the dimension effect), so that the normalized parameters are combined with the adsorption rate coefficient K ads to construct the optimization objective function Z .

[0069] It should be noted that the optimization objective function Z flexibly balances the two possibly conflicting goals of maximizing the adsorption amount and minimizing the wrinkle through the dynamic weight factor ω. The normalization process during its application ensures the comparability of different parameters, and the integral form of Q ads considers the time accumulation effect, and the finite element calculation of S wrinkle introduces the physical basis of the mechanical properties of the fabric, making the optimization more practical.

[0070] Step 304, the optimization calculation layer solves the objective function Z through an optimization algorithm to obtain the optimal clamping point coordinates, clamping angle and immersion posture combination that minimizes the value of Z, generates an executable cooperative operation instruction set, and outputs to the high-precision double-robot driving module.

[0071] where the optimal clamping point coordinates are the best position coordinates of the robot clamping the fabric. The clamping angle is the orientation angle parameter of the robot clamping device. The immersion posture combination is a set of overall posture parameters of the robot when it is immersed in the dye vat. The executable cooperative operation instruction set is a set of structured control instructions for driving the robot to work cooperatively. The high-precision double-robot driving module is a hardware and software unit for controlling the movement of the double robot.

[0072] In some implementations, the objective function Z and its related parameters are obtained from the relational modeling layer, and a suitable optimization algorithm is selected to iteratively solve the objective function Z. The partial derivatives of the Z value with respect to the variables such as the clamping point coordinates, the gripper angle, and the immersion posture combination are adjusted to gradually reduce the Z value until it converges to a minimum value. Then, during the solving process, the optimization algorithm evaluates the Z value under different variable combinations to ultimately determine the optimal clamping point coordinates, gripper angle, and immersion posture combination that minimize the Z value. Subsequently, the optimization calculation layer converts these optimal parameters into an executable collaborative work instruction set, including instructions such as target motion trajectories, postures, and speeds. Finally, the generated instruction set is output to the high-precision dual-robot driving module for actual control of the robot motion.

[0073] Based on the above technical solution, a material-dye adsorption relationship model is trained using historical data, which can adaptively learn and represent the complex, nonlinear coupling relationship between factors such as material, viscosity, temperature, and adsorption rate, improving the accuracy of key process parameter prediction and enabling the system to have reasoning ability similar to "expert experience". Thus, the system can effectively explore the complex internal rules of multiple interwoven factors, especially when facing new mixed materials or non-standard dyes, reducing prediction errors and improving color quality. At the same time, by constructing an explicit optimization objective function Z, the conflicting goals of "maximizing dye adsorption" (represented by Q ads ) and "minimizing wrinkle deformation" (represented by S wrinkle ) are unified in the same mathematical framework, and a dynamic weight factor ω is used for flexible trade-off. This can convert the vague and sensory "high-quality dyeing" standard into a quantifiable and calculable mathematical problem, thereby automatically finding the process parameters that achieve the best comprehensive effect under a specific weight, enabling precise control of quality targets. Finally, an optimization algorithm is called to automatically solve the objective function Z, and the optimal mathematical solution (such as a set of numerical values) is converted into an executable collaborative work instruction set (including the optimal clamping point coordinates, gripper angle, and immersion posture combination) in real time, and directly output to the high-precision dual-robot driving module, eliminating errors and delays that may be caused by human intervention and ensuring that the optimal process scheme can be accurately and quickly reproduced.

[0074] In a possible implementation of the embodiments of the present application, the adaptive adjustment model can be implemented through the following steps 401 to 407, which are described in detail as follows: Step 401: Obtain the remaining area information output by the scene perception module , extract the contour coordinate points P i ' and solve all contour coordinate points P iThe minimum circumscribed rectangle of the convex hull is the minimum circumscribed rectangle R of the fabric.

[0075] The convex hull refers to the smallest convex polygon that contains all the given point set. The minimum circumscribed rectangle refers to the smallest rectangle that can completely enclose the convex hull, which is often used to simplify the shape representation of the region.

[0076] In some implementations, the remaining region information retained after image segmentation and area filtering is obtained , and the contour coordinate points P of each region are extracted i to represent the precise position of the region boundary. The convex hull of all contour coordinate points P i is calculated to find the smallest convex polygon that contains all the contour points, thereby simplifying the shape of the region and eliminating the effect of concave parts; finally, the minimum rectangle that can completely enclose the convex hull is determined through geometric algorithms (such as the rotating calipers method or the axis-aligned bounding box calculation), thereby obtaining the minimum circumscribed rectangle R of the fabric as the reference for subsequent mechanical arm gripping point planning and motion control.

[0077] It should be noted that through the calculation of the convex hull and the minimum circumscribed rectangle, the complexity of the region shape of the patterned fabric is reduced, avoiding the computational burden caused by directly processing irregular contours, while ensuring the stability and reliability of the gripping point selection.

[0078] Step 402, uniformly sample k points on the four sides of the minimum circumscribed rectangle R to form candidate gripping points, and form all possible opposite side point pairs by grouping all candidate gripping points on opposite sides of the minimum circumscribed rectangle R to obtain a candidate gripping point pair set, each candidate gripping point (P A , P B ) in the candidate gripping point pair set represents a candidate gripping point P A of one side mechanical arm and a candidate gripping point P B of the other side mechanical arm.

[0079] The uniform sampling of k points refers to selecting k points on the four sides of the rectangle R in an equal interval manner to ensure uniform distribution of the sampling. The candidate gripping point refers to the point obtained by sampling, which can be used as a candidate point for possible gripping positions of the mechanical arm. The opposite side refers to the two opposite sides in the rectangle R, such as the upper and lower sides or the left and right sides. The opposite side point pair refers to the point pair composed of candidate gripping points on the opposite sides, representing a pair of possible gripping positions. The candidate gripping point pair set refers to the set of all possible opposite side point pairs, which is used for subsequent optimization selection.

[0080] In some implementations, based on the already obtained minimum bounding rectangle R of the fabric, each side is divided into k-1 equal segments to obtain k sampling points on each side, forming a candidate clamping point set; then, the opposite sides of rectangle R (such as top and bottom sides, left and right sides) are identified, and all candidate clamping points on the opposite sides are combined to generate all possible point pairs (such as each point on the top side is paired with each point on the bottom side, and each point on the left side is paired with each point on the right side). These point pairs can then be summarized into a candidate clamping point pair set, where each point pair (P) A P B ) Clearly indicate the candidate gripping point P of one side of the robotic arm. A And the candidate gripping point P of the other robotic arm B This provides basic data for subsequent clamping point scoring and selection.

[0081] It should be noted that by using uniform sampling and pairing opposite edge points, the candidate clamping points are ensured to cover the entire outline of the fabric, thus avoiding local deviations.

[0082] Step 403: Through the contour coordinate point P i Calculate the centroid of region i', and connect the centroid of region i' to the clamping point to obtain the length L of the calculation line. Obtain the line length L connecting the centroid of region i' to the clamping point. （PA，PB） vertical distance d i ',pass Calculate the support score S support , where exp is an exponential function.

[0083] Here, the centroid of region i′ is calculated from the contour coordinates and represents the center position of the region.

[0084] In some implementations, the arithmetic mean or geometric center of the contour coordinates Pi′ of each region i′ is calculated to determine the centroid coordinates of region i′, which can be determined based on a given pair of candidate gripping points (P). A ,P B ), connect P A and P B Form a line connecting the clamping points and calculate the length L of this line; then, for each region i′, calculate the line L connecting its centroid to the clamping point. (PA,PB） vertical distance d i ′, then you can pass Calculate the support score.

[0085] Step 404: Extract the color boundary point E of region i′ to obtain the clamping point P. A The distance d to the nearest color boundary point E AE and clamping point P B The distance d to the nearest color boundary point E BE， pass Compute the boundary avoidance score S boundary .

[0086] wherein the color boundary point E is a point in the region i' identified by image processing algorithm to indicate the color change or boundary between regions. The distance d AE is the Euclidean distance from the clamping point P A to the nearest color boundary point E, representing the spatial interval. The distance d BE is the Euclidean distance from the clamping point P B to the nearest color boundary point E, also representing the spatial interval.

[0087] In some implementations, based on the remaining region information , analyze the color gradient or edge detection algorithm of the region contour coordinate point P A i', extract the color boundary point E of each region i', so as to identify the point with obvious color change; that is, for a given candidate clamping point pair (P B , P A ), the distance from P AE to all color boundary points E is calculated, and the minimum value is taken as d B , and the distance from P BE to all color boundary points E is also calculated, and the minimum value is taken as d boundary , and the Euclidean distance formula is used to ensure accuracy; so as to calculate the boundary avoidance score S by the formula

[0088] Step 405, weight and fuse the boundary avoidance score S support and the support score S total to obtain the clamping point comprehensive score S total , and take the maximum value of the clamping point comprehensive score S Aopt as the candidate clamping point pair as the optimal clamping point pair, and output the clamping point coordinates (P Bopt） , P boundary .

[0089] In some implementations, based on the boundary avoidance score S support and the support score S boundary , the weight α of the boundary avoidance score S support and the weight β of the support score S total (α+β=1) are preset, so as to calculate the comprehensive score S boundary of each candidate clamping point pair by the formula S support =α×S total +β×S total ; then traverse all candidate clamping point pair sets, compare the S totalFind the maximum value; mark the candidate clamping point pair corresponding to the maximum value as the optimal clamping point pair, ensuring that it can simultaneously optimize boundary avoidance and area support; finally, output the coordinates (P) of the optimal clamping point pair. Aopt ,P Bopt） , where P Aopt P represents the coordinates of the optimal gripping point on one side of the robotic arm. Bopt The coordinates of the optimal gripping point on the other side of the robotic arm are used to provide precise input for subsequent robotic arm control.

[0090] Step 406: Based on the optimal clamping point P Aopt and P Bopt Calculate the clamping angle θ along the tangent direction of the fabric boundary. A and θ B The tangent direction is obtained by calculating the differential of the contour points near the clamping point.

[0091] Here, the fabric boundary refers to the fabric edge contour defined by the contour coordinate points. The tangent direction is the local tangent direction at a point on the fabric boundary, representing the instantaneous direction vector at that point. Clamping angle θ A and θ B This is the angular parameter of the robotic arm gripper relative to the tangential direction, used to adjust the gripping posture. The contour points are the set of discrete point coordinates on the fabric boundary, used to describe the boundary shape.

[0092] In some implementations, the optimal gripping point P is determined. Aopt and P Bopt Locate the corresponding position on the fabric boundary; then in P Aopt and P Bopt A set of contour points is selected nearby, and the tangent directions of these contour points are calculated through differential operations (such as the central difference method). That is, for each sequence of contour points near the clamping point, the coordinate differences between adjacent points are calculated and normalized to obtain the tangent vector; thus, the clamping angle θ is determined based on the tangent direction. A and θ B To optimize gripping stability, the robotic arm gripper is aligned with the tangential direction; finally, the calculated θ is... A and θ B It is integrated into the collaborative operation instruction set to drive the movement of the robotic arm.

[0093] It should be noted that the density of the selected contour points should be moderate. Too dense a density will lead to complex calculations, while too sparse a density will affect the accuracy of the direction. This will ensure the accuracy of the clamping angle by calculating the tangent direction through differentiation.

[0094] Step 407: Based on the remaining area information and dye viscosity value V isc Calculate the independent adjustment action parameters required to prevent color penetration, and combine the independent adjustment action parameters with the clamping angle θ.A and θ B The commands are compiled into an executable set of collaborative operation instructions and output to the high-precision dual robotic arm drive module.

[0095] Preventing color penetration refers to the process of avoiding the mixing of dyes between different color areas through the micro-motions of the robotic arm. Independently adjustable motion parameters are used to control the micro-motions performed by a single robotic arm, including the motion amplitude A, motion frequency F, and execution start time T. s and execution end time T c wait.

[0096] The independent adjustment parameters required to prevent color penetration include: Based on dye viscosity value V isc Calculate the motion frequency F and motion amplitude A, and preset the start time T of the dyeing motion based on the dyeing process parameters. s and the end time T of the immersion action e .

[0097] Each color boundary is identified based on the contour intersection of two adjacent regions i′ and j′, and then... Calculate the contrast C of each color boundary. contrast .

[0098] At contrast C contrast When the color boundary exceeds the contrast threshold, retrieve the robotic arm identifier ID. Arm And summarize the amplitude of the movement A, the frequency of the movement F, and the start time of the immersion movement T. s and the end time T of the immersion action e， The independent adjustment action parameters are obtained, where the average saturation D in region i is... i′ The average saturation D of region j is less than j′ At that time, the robotic arm located on one side of region i′ is denoted as ID. Arm Otherwise, the robotic arm on the j′ side is denoted as ID. Arm .

[0099] It should be noted that the summation of independently adjustable motion parameters and clamping angles must maintain a consistent instruction set structure to prevent calculation conflicts; and the viscosity value V isc The effects on the frequency and amplitude of movements need to be calibrated experimentally to balance penetration prevention and motion stability.

[0100] Based on the above technical scheme, by introducing the weighted fusion mechanism of support score and boundary avoidance score, combining mathematical models (such as exponential function to calculate Ssupport and minimum distance to calculate Sboundary), the candidate clamp point pairs are comprehensively evaluated, and the point pair with the highest comprehensive score is selected as the optimal solution, and the coordinates (PAopt, PBopt) are output, so that the adaptive adjustment model is connected with the scene perception module, receives data and performs calculation to generate an instruction set, so that the double robot arms can maximize dye adsorption (through support score) and prevent color penetration (through boundary avoidance score) when clamping, and improve dyeing quality. Avoiding the fixed clamping point or simple rule selection of the existing robot arm system, which leads to lack of multi-objective optimization, resulting in fabric wrinkles, insufficient dye adsorption or color cross contamination, especially in the case of poor edge definition in color dyeing. At the same time, by integrating the material-dye adsorption relationship model and the micro-motion model, using the nonlinear mapping function calculated by the historical data to train the parameters, the robot arm can perform independent fine tuning while ensuring synchronous overall motion, so that the robot arm action matches the dye characteristics and prevents penetration.

[0101] In a possible implementation manner of the embodiment of the present application, the double robot arm driving module can be implemented through the following steps 501 to 504, which are described in detail as follows: Step 501, parse the executable cooperative work instruction set, and filter the motion parameters of the double robot arms for synchronization, denoted as reference cooperative motion layer instructions, the reference cooperative motion layer instructions including overall immersion speed V t , overall pulling speed V I , and overall immersion depth D. The independent action parameters of the single-sided robot arm are denoted as independent adjustment instructions, the independent adjustment instructions including adjustment action type T action , action amplitude A, action frequency F, execution start time T s , execution end time T e , and acting robot arm identifier ID Arm .

[0102] Among them, the reference cooperative motion layer instructions are a set of core motion parameters that the double robot arms need to strictly synchronize to execute. The overall immersion speed V t indicates the constant speed when the double robot arms as a whole unit immerse in the dye vat. The overall pulling speed V I indicates the constant speed when the double robot arms as a whole unit pull out of the dye vat. The overall immersion depth D indicates the preset depth of the double robot arms cooperatively controlling the fabric to immerse in the dye vat. The independent adjustment instructions are a set of fine tuning action parameters of the single-sided robot arm based on the reference motion. The adjustment action type T action indicates the specific form of fine tuning action such as swing or jitter. The action amplitude A indicates the displacement size of fine tuning action. The action frequency F indicates the periodic repetition rate of fine tuning action. The execution start time T sTime indicating the start of the fine-tuning action. Execution end time T e Time indicating the termination of the fine-tuning action. Acting arm identifier ID Arm Indicates the number of the arm designated to perform the fine-tuning action.

[0103] In some implementations, the executable collaborative operation instruction set is parsed, and the parameters that need to be synchronized by the dual-arm are screened out as the reference collaborative motion layer instruction, which is directly used to generate the ideal synchronous motion trajectory of the dual-arm, thereby ensuring that the dual-arm operates as a whole unit in a macroscopic manner; then the independent action parameters of the single-arm are recorded as the independent adjustment instruction, which is dynamically superimposed on the trajectory of the corresponding arm according to the timestamp and the arm identifier, and is used to control the specific arm to perform fine-tuning action while maintaining the overall motion. Finally, the motion parameters of the dual-arm after instruction fusion are ensured to be coordinated and consistent, avoiding conflicts.

[0104] It should be noted that the stability of the reference collaborative motion layer instruction is prioritized, and real-time conflict detection and safety arbitration are required for the superposition of the independent adjustment instruction, thereby preventing the fine-tuning action from affecting the safety of the overall motion or causing damage to the fabric.

[0105] Step 502, based on the reference collaborative motion layer instruction, generate the ideal synchronous motion trajectory of the dual-arm end effector in three-dimensional space. The ideal synchronous motion trajectory is calculated by a kinematics inverse solution algorithm, and the output is a target angle sequence θ target(t） of each joint of the dual-arm, where t is the time variable.

[0106] The dual-arm end effector is the end tool of the arm, used to directly hold or operate the fabric. The ideal synchronous motion trajectory in three-dimensional space is the collaborative motion path planned for the dual-arm end effector in a three-dimensional coordinate system, ensuring the consistency of the dual-arm motion. The kinematics inverse solution algorithm is a mathematical method for calculating the angles of each joint from the target position and attitude of the end effector. The target angle sequence θ target(t） is a sequence of the angles of each joint of the dual-arm changing with time t, used to accurately control the joint motion.

[0107] In some implementations, based on the overall immersion speed V t , the overall pulling speed V I and the overall dip depth D in the reference collaborative motion layer instruction, the ideal synchronous motion trajectory of the dual-arm end effector in three-dimensional space is planned, thereby limiting the continuous path of the end effector from the starting point to the target point; then through the kinematics inverse solution algorithm (considering the geometric parameters such as the length of the connecting rod and the joint limit of the arm), each point on the trajectory is converted into the target angle of each joint of the dual-arm, thereby outputting the target angle sequence θ target(t） of each joint of the dual-arm. Finally, the target angle sequence θ target(t）Drive the joint servo system of the mechanical arm to realize the synchronous motion control of the end effector.

[0108] Step 503, obtain the adjustment action type T action , amplitude A and frequency F, and calculate the trajectory offset Δθ (t） by the micro-motion model. Superimpose the trajectory offset Δθ (t） on the joint target angle sequence of the corresponding mechanical arm to obtain the adjusted joint angle sequence θ adjusted（t） .

[0109] The micro-motion model refers to a mathematical model for calculating the trajectory offset according to the action type, amplitude and frequency parameters. The trajectory offset Δθ (t） refers to the joint angle offset of the mechanical arm caused by micro-motion. The adjusted joint angle sequence θ adjusted(t） refers to the new angle sequence obtained by superimposing the trajectory offset on the joint target angle sequence.

[0110] In some implementations, the parameters of the adjustment action type T action , amplitude A and frequency F are obtained, and then a periodic function describing the periodic change pattern of the micro-motion is generated according to the frequency F with time t as the variable. Thus, by processing the adjustment action type T action , amplitude A and frequency F and the periodic function by the micro-motion model, the action parameters are mapped to a specific trajectory offset Δθ (t） , and the calculated trajectory offset Δθ (t） is superimposed on the joint target angle sequence of the corresponding mechanical arm by point-by-point addition. The adjusted joint angle sequence θ adjusted(t） is obtained, which is used for precise control of the motion of the mechanical arm.

[0111] It should be noted that the superimposition process must be strictly synchronized in time to avoid jumps or conflicts in the angle sequence; at the same time, the adjusted θ adjusted(t） needs to be checked for feasibility to prevent exceeding the joint motion range of the mechanical arm and to ensure smooth and safe motion.

[0112] The micro-motion model includes: By extracting the effective micro-motion segments for preventing color penetration from the end trajectory data of the mechanical arm in historical successful dyeing cases, and performing principal component analysis on the effective micro-motion segments, the association between the action primitives and the dyeing effect evaluation indicators is established, and the action primitive library is obtained.

[0113] According to the adjustment action type T actionSelect the corresponding target action primitive from the action primitive library, map the action amplitude A to the amplitude scaling factor of the action primitive, map the action frequency F to the angular frequency of the action primitive, and obtain the corresponding trajectory offset Δθ from the action primitive library. (t） .

[0114] Among them, effective micro-motion segments to prevent color penetration refer to robotic arm motion segments extracted from historical data that effectively avoid cross-contamination of dyes. Motion primitives are the basic units or templates of micro-motions, representing specific motion patterns. Dyeing effect evaluation indicators are parameters that quantify dyeing quality, such as edge sharpness or color uniformity. The motion primitive library is a database storing various motion primitives and their relationships with effect indicators. Target motion primitives are based on T... action The specific motion template selected from the library. The amplitude scaling factor is a scaling factor used to adjust the amplitude of the motion primitive. The angular frequency is the mapping value of the motion frequency F in the angular domain, in radians per second.

[0115] In some implementations, principal component analysis is performed on micro-motion segments that have been proven to effectively prevent color penetration by analyzing the end-effector trajectory data of robotic arms in historical successful staining cases. This reduces data dimensionality and extracts key motion features. Simultaneously, a quantitative correlation is established between these motion features and staining effect evaluation indicators (such as edge sharpness), constructing a motion primitive library containing various optimized motion modes. This allows for fine-tuning based on the real-time specified adjustment motion type T. action The process involves selecting a target motion primitive from the motion primitive library; mapping the real-time parameter motion amplitude A to the amplitude scaling factor of the target motion primitive, i.e., converting A into an adjustment ratio for the primitive's amplitude; simultaneously mapping the motion frequency F to the primitive's angular frequency, i.e., determining the angular velocity of the primitive's motion; and finally, based on the scaled amplitude and angular frequency, obtaining the corresponding pre-calculated trajectory offset Δθ from the motion primitive library. (t） This describes the change in joint angle that should be superimposed on the reference motion.

[0116] Step 504: Sequence the joint angles θ adjusted（t） The angular velocity and angular acceleration of each joint are calculated, and a pulse width modulation signal is generated through a servo drive algorithm.

[0117] In some implementations, the adjusted joint angle sequence θ is obtained. adjusted(t） And θ is obtained through numerical differentiation methods (such as backward difference or central difference algorithms). adjusted(t）The angular velocity of each joint is calculated. The numerical differentiation method is used again to calculate the angular acceleration sequence from the angular velocity sequence to ensure the calculation accuracy and reduce the cumulative error. The servo drive algorithm receives the angle, angular velocity and angular acceleration parameters, combines the preset control law (such as proportional-integral-derivative control), calculates the torque or position correction required by the motor, and finally converts the correction into a pulse width modulation signal to drive the servo motor by adjusting the duty cycle and frequency of the pulse to realize accurate tracking motion of the joint.

[0118] According to the technical scheme, the cooperative operation instruction set is divided into a reference layer and an adjustment layer by screening the synchronization parameters and the independent parameters, so that the dual-robot arm can not only maintain the dyeing depth and speed stability as a whole unit, but also can adjust the dyeing depth and speed of each arm through the identification ID Arm to solve the problem of task conflict of the dual-robot arm and realize the effect of freely switching between the synchronous operation and the independent operation according to different scenes. In addition, the motion trajectory generation structure converts the reference instruction into an ideal trajectory in the three-dimensional space through the inverse kinematics algorithm, and outputs a target angle sequence θ target(t） , so as to ensure the accurate and reproducible path of the end effector, eliminate the trajectory planning error, and improve the dyeing consistency. In addition, the historical data extraction action primitive maps T action , A and F into a trajectory offset Δθ (t） , and superimposes the trajectory offset Δθ target(t） on θ adjusted(t） to obtain θ i , which can effectively prevent the dynamic compensation of color penetration and adaptively match the motion of the robot arm with the dye characteristics. Finally, the servo drive structure calculates the angular velocity and angular acceleration through the numerical differentiation, and generates a PWM signal to drive the motor, thereby forming a closed-loop control link to ensure the high-frequency response and anti-interference ability of the motion parameters.

[0119] In a possible implementation manner of the embodiment of the present application, the adaptive updating module can be implemented through the following steps 601 to 605, which are described in detail as follows: Step 601, obtaining residual region information extracting the area area S i and the average brightness C i of the region i, through calculating the total area S total , and calculating the average brightness variance V arc of all regions by calculating the average brightness Ci' of all regions. Wherein C i ' is a three-dimensional vector representing the color value in the LAB color space, and the variance calculation is performed on each component L, a and b respectively.

[0120] Wherein, the average brightness variance V arcThe average brightness C of all areas is obtained by averaging the variances of each component. i The variance of ' is used to measure the uniformity of color distribution.

[0121] In some implementations, information about the remaining area is obtained. Extract the area S of each region. i ′ and use the summation formula S total =∑S i Calculate the total area S of the region. total This yields the overall coverage measure of the patterned area; then, the average brightness C of each area is calculated. i ′, due to C i ' is a three-dimensional vector in the LAB color space. The variances of the L component values, a component values, and b component values ​​for all regions are calculated separately. The arithmetic mean of these three variances is then taken to obtain the average lightness variance V. arc This allows for a quantitative assessment of the uniformity of color distribution.

[0122] Step 602: Obtain the total area S of the historically successful staining cases. total and mean lightness variance V arc And the corresponding optimal speed ratio and amplitude ratio, which are used to obtain the coloring uniformity f through the trained neural network. v and edge sharpness f a .

[0123] Among them, the optimal speed ratio is the best ratio parameter between the overall immersion speed and lifting speed of the dual robotic arms in historical cases. The amplitude ratio is the best ratio parameter for independently adjusting the amplitude of motion in historical cases. Maximizing dyeing uniformity f v This is a function used to optimize staining uniformity metrics. Edge sharpness f a It is a function used to optimize the color edge sharpness index.

[0124] In some implementations, data on historical successful dyeing cases (including the total area S of each case) are retrieved from a dyeing process database. total and mean lightness variance V arc The corresponding optimal speed ratio and amplitude ratio are used as the training set, and the total area S in the set is taken as the training set. total and mean lightness variance V arc Using the optimal speed ratio and amplitude ratio as input features and the target labels as target labels, a neural network model is trained using supervised learning. The network weights are adjusted via backpropagation to minimize prediction error. The resulting trained neural network can then predict the prediction error based on the new total region area S. total and mean lightness variance V arc The value directly outputs the maximum staining uniformity f.v and edge sharpness f a The function value is obtained, thus providing optimized parameters for real-time control.

[0125] Step 603: The overall immersion velocity V in the reference co-motion layer command. t and overall lifting speed V I pass and After making corrections, the overall immersion velocity V is obtained. t ′ and overall lifting speed V I ′.

[0126] In some implementations, the original overall immersion velocity V is obtained from the baseline cooperative motion layer command. t and the overall lifting speed V1; through the function f v Combined with the total area S total and mean lightness variance V arc Calculate a uniform correction factor; then multiply the correction factor by the original velocity, i.e., using the formula... The corrected immersion velocity V is obtained. t ′, and through formula The corrected lifting speed V is obtained I ′.

[0127] Step 604: Correct the motion amplitude A in the independent adjustment command layer command. The corrected amplitude A′ is obtained.

[0128] Step 605: Summarize the corrected overall immersion velocity V t Overall lifting speed V I The hierarchical collaborative operation instruction set is obtained from the amplitude A' and is fed back to the dual-arm drive module to be recalculated into the final motion parameters of each joint of the dual-arm, driving the dual-arm to perform motion.

[0129] In some implementations, the corrected overall immersion velocity V is... t Overall lifting speed V I The overall immersion speed Vt′ and overall lifting speed VI′ are summarized and integrated into the hierarchical collaborative operation instruction set, so that the speed parameters in the baseline collaborative motion layer instruction are updated for the overall immersion speed Vt′ and the overall lifting speed VI′, and the amplitude A′ is updated for the amplitude parameters in the independent adjustment instruction layer instruction. The updated instruction set is then fed back to the dual robotic arm drive module. After receiving the instruction set, the dual robotic arm drive module recalculates the instructions through its internal algorithm, converts the macroscopic motion parameters into specific motion parameters for each joint of the dual robotic arms, and sends them to the servo control system to drive the dual robotic arms to perform motion, thereby achieving collaborative immersion and fine-tuning.

[0130] Although the present application has been described in connection with certain specific features and embodiments thereof, it is to be understood that it is intended to cover all modifications and variations of this application which are within the scope of the appended claims and their equivalents. Accordingly, the description and drawings are to be regarded as illustrative in nature and not as restrictive. It is intended that all such modifications and variations are included within the scope of the present application as defined by the following claims and their equivalents.

Claims

1. A dual-robotic arm control system applicable to different dyeing scenarios, characterized in that, include: The scene perception module collects image and physical information of the fabric to be dyed and determines the dyeing type, which is either solid color or patterned color. The central collaborative control and decision-making module is connected to the scene perception module and is used to call the corresponding collaborative strategy according to the dyeing type to parse and generate a layered collaborative operation instruction set for the dual robotic arms, including the target motion trajectory, posture, speed, and immersion depth. The collaborative strategy is a two-layer framework including an immersion posture optimization model for the solid color type and a pattern area adaptive adjustment model for the patterned type. The immersion posture optimization model is an algorithm that calculates the optimal gripping point, gripping angle, and immersion posture of the dual robotic arms by analyzing the fabric material, dye viscosity, drying speed, and real-time temperature, with the optimization objectives of maximizing the amount of dye adsorbed per unit area of ​​the fabric and minimizing wrinkle deformation. The pattern area adaptive adjustment model is an algorithm that dynamically calculates and adjusts the gripping position, angle, and timing and amplitude of the dual robotic arms by analyzing the color distribution, area boundary, and color depth information of the pattern, with the optimization objective of the edge clarity between different color areas. The dual robotic arm drive module is connected to the central collaborative control and decision-making module. It is used to receive the hierarchical collaborative operation instruction set and calculate it into the final motion parameters of each joint of the dual robotic arm. It drives the dual robotic arms to be controlled according to the scene. When synchronized, the dual robotic arms operate as a whole unit. When local optimization is needed, the dual robotic arms independently perform fine-tuning actions. An adaptive update module, connected to the central collaborative control and decision-making module, is used to store dyeing process parameter records to form a dyeing database; Based on the comparative analysis of historical data and real-time monitoring data, the collaborative strategy algorithm is optimized and updated, and the updated strategy is fed back to the central collaborative control and decision-making module.

2. The dual-robotic arm control system for different dyeing scenarios according to claim 1, characterized in that, The scene perception module includes a perception layer and a decision layer: The sensing layer is used to collect image information of the fabric to be dyed and physical information including material composition information, viscosity information of the dye in the vat, and real-time temperature information. The decision layer preprocesses and extracts features from the image and physical information collected by the perception layer, calculates the variance of the image color feature distribution, and if the variance is less than or equal to a preset variance threshold, the coloring type is determined to be a solid color, and the material composition information, viscosity information, and real-time temperature information are sent to the central collaborative control and decision module; if the variance is greater than the preset variance threshold, the coloring type is determined to be patterned, and the image segmentation submodule is activated to segment the surface image, identify the color and depth of different regions, and synchronously send the color, depth, and image coordinate information of each region, along with the material composition information, viscosity information, and real-time temperature information, to the central collaborative control and decision module.

3. The dual-robotic arm control system for different dyeing scenarios according to claim 2, characterized in that, The image segmentation submodule includes: The preprocessed RGB image of the fabric surface to be dyed is converted from the RGB color space to the LAB color space, and the color vector of each pixel is extracted. And calculate the average color difference ΔC between it and its 8 neighboring pixels. {avg} 8-neighbor pixels refer to 8 pixels that are directly adjacent in the horizontal, vertical, and diagonal directions. The preprocessed RGB image of the fabric surface to be dyed is converted to grayscale, and the Sobel operator is used to calculate the gradient magnitude G of each pixel in the x and y directions. (x,y） And the average color difference ΔC {avg｝ and gradient magnitude G (x,y） Weighted fusion is performed to obtain the edge probability value P for each pixel. edge ; A mapping table between the sensitivity coefficient k and the fabric material vector M and the target dyeing effect level E is established using historical data and stored in the database of the scene perception module. Obtain the basic segmentation threshold T base And based on the material vector M of the fabric to be dyed and the target dyeing effect level E, the mapping table is queried to obtain the current sensitivity coefficient k, and then... Calculate the segmentation threshold T seg , where γ is the adjustment factor; Using marginal probability value P edge Greater than the dynamic segmentation threshold T seg Using the pixels as initial edge points, a region growing algorithm is used for image segmentation to form n candidate regions R. n , where n is the total number of candidate regions; Calculate the average brightness C of each pixel within all candidate regions. i Average saturation D i and contour coordinates point P i and area S i The area of ​​all candidate regions is denoted as the area set, where i is the index number of the region. Filter out the area S within the area set of the regions i Candidate regions smaller than a preset area threshold will have their remaining area information... The output is sent to the central collaborative control and decision-making module, where i′ is the index of the filtered region. These represent the area, average brightness, average saturation, and outline coordinates of the filtered region i, respectively.

4. The dual-robotic arm control system for different dyeing scenarios according to claim 3, characterized in that, The immersion posture optimization model includes a data input layer, a relation modeling layer, and an optimization calculation layer: The data input layer is used to receive the fabric material vector M and dye viscosity value V from the scene perception module. isc Real-time temperature value T emp and historical adsorption rate data; wherein the fabric material vector M is a multidimensional vector M=[m1,m2,…,m] of the percentage content of various material components such as cotton, linen, silk, and chemical fiber measured by a near-infrared spectroscopy sensor. n ], and satisfy ∑m n =100%; the data input layer is also used to receive the dye concentration value [Dye] from the dyeing vat concentration sensor and the elastic modulus distribution parameter E corresponding to the current fabric material vector M. i ; The relationship modeling layer acquires historical data to train a material-dye adsorption relationship model, and then processes the data from the input layer through this model. The adsorption rate coefficient K was obtained. ads And output, where f is a nonlinear mapping function obtained by training a neural network, used to characterize the adsorption rate characteristics of dye on fabric under different material, viscosity and temperature conditions; Obtain the effective dyeing area A of the fabric eff After normalization with the dynamic weighting factor ω, it is combined with the adsorption rate coefficient K. ads An optimization objective function is constructed to maximize the amount of dye adsorbed per unit area of ​​the fabric and minimize the wrinkle deformation. Wherein Q ads The total amount of adsorbed dye is obtained by passing through The calculation yielded that the S wrinkle The degree of wrinkle deformation is determined by The calculation yields Δx i Let be the displacement of the i-th mesh element in the finite element model; The optimization calculation layer solves the objective function Z through an optimization algorithm to obtain the optimal combination of gripping point coordinates, gripper angle and immersion posture that minimizes the Z value, generates an executable set of collaborative operation instructions, and outputs them to the high-precision dual robotic arm drive module.

5. A dual-robotic arm control system for different dyeing scenarios according to claim 4, characterized in that, The adaptive adjustment model includes: Obtain the remaining area information output by the scene perception module. Extract the contour coordinates of point P i And solve for the coordinates of all contour points P. i The minimum bounding rectangle of the convex hull is used to obtain the minimum bounding rectangle R of the fabric. k points are uniformly sampled on the four sides of the minimum bounding rectangle R to form candidate clamping points. All candidate clamping points on opposite sides of the minimum bounding rectangle R are then grouped into all possible pairs of opposite edges, resulting in a set of candidate clamping point pairs. Each candidate clamping point (P) in this set... A P B P represents a candidate gripping point of one side of the robotic arm. A And the candidate gripping point P of the other robotic arm B ; Through the contour coordinate point P i Calculate the centroid of region i', and connect the centroid of region i' to the clamping point to obtain the length L of the calculation line. Obtain the line length L connecting the centroid of region i' to the clamping point. （PA，PB） vertical distance d i ',pass Calculate the support score S support , where exp is an exponential function; Extract the color boundary point E of region i′ to obtain the clamping point P. A Distance d to the nearest color boundary point E AE and clamping point P B Distance d to the nearest color boundary point E BE， pass Calculate the boundary avoidance score S boundary ; The boundary avoidance score S boundary and support score S support The weighted fusion is used to obtain the clamping point comprehensive score S. total The clamping point is comprehensively scored S total The maximum value is denoted as the candidate clamping point pair, which is then used as the optimal clamping point pair. The coordinates of the clamping points (P) are output. Aopt ,P Bopt） ; Based on the optimal clamping point P Aopt and P Bopt Calculate the clamping angle θ along the tangent direction of the fabric boundary. A and θ B The tangent direction is obtained by calculating the differential of the contour points near the clamping point; Based on remaining region information and dye viscosity value V isc Calculate the independent adjustment action parameters required to prevent color penetration, and combine the independent adjustment action parameters with the clamping angle θ. A and θ B The commands are compiled into an executable set of collaborative operation instructions and output to the high-precision dual robotic arm drive module.

6. A dual-robotic arm control system for different dyeing scenarios according to claim 5, characterized in that, The calculation of the independent adjustment parameters required to prevent color penetration includes: Based on dye viscosity value V isc Calculate the motion frequency F and motion amplitude A, and preset the start time T of the dyeing motion based on the dyeing process parameters. s and the end time T of the immersion action e ; Each color boundary is identified based on the contour intersection of two adjacent regions i′ and j′, and then... Calculate the contrast C of each color boundary. contrast ; At contrast C contrast When the color boundary exceeds the contrast threshold, retrieve the robotic arm identifier ID. Arm And summarize the amplitude of the movement A, the frequency of the movement F, and the start time of the immersion movement T. s and the end time T of the immersion action e， The independent adjustment action parameters are obtained, where the average saturation D in region i is... i′ The average saturation D of region j is less than j′ At that time, the robotic arm located on one side of region i′ is denoted as ID. Arm Otherwise, the robotic arm on the j′ side is denoted as ID. Arm .

7. A dual-robotic arm control system for different dyeing scenarios according to claim 6, characterized in that, The dual robotic arm drive module includes: The executable set of cooperative operation instructions is analyzed, and the motion parameters of the two robotic arms synchronized are selected and recorded as the reference cooperative motion layer instructions. The reference cooperative motion layer instructions include the overall immersion speed V. t Overall lifting speed V I The overall immersion depth D; the independent motion parameters of a single robotic arm are recorded as independent adjustment commands, the independent adjustment commands including the adjustment motion type T. action Amplitude of movement (A), frequency of movement (F), execution start time (T) s Execution end time T e and the function of robotic arm identification ID Arm ; Based on the baseline coordinated motion layer instructions, an ideal synchronous motion trajectory of the dual robotic arm end effectors in three-dimensional space is generated; the ideal synchronous motion trajectory is calculated by an inverse kinematics algorithm, and the output is the target angle sequence θ of each joint of the dual robotic arms. target(t） , where t is a time variable; Get the adjustment action type T action The periodic function of amplitude A and frequency F is used to calculate the trajectory offset Δθ using a micro-motion model. (t） and through The trajectory offset Δθ (t） By superimposing this onto the target joint angle sequence of the corresponding robotic arm, the adjusted joint angle sequence θ is obtained. adjusted（t） ; joint angle sequence θ adjusted（t） The angular velocity and angular acceleration of each joint are calculated, and a pulse width modulation signal is generated through a servo drive algorithm.

8. A dual-robotic arm control system for different dyeing scenarios according to claim 7, characterized in that, The micro-motion model includes: By extracting effective micro-motion segments to prevent color penetration from the end-effector trajectory data of the robotic arm in historical successful dyeing cases, principal component analysis was performed on the effective micro-motion segments to establish the correlation between motion primitives and dyeing effect evaluation indicators, thus obtaining a motion primitive library. According to the type of adjustment action T action Select the corresponding target action primitive from the action primitive library, map the action amplitude A to the amplitude scaling factor of the action primitive, map the action frequency F to the angular frequency of the action primitive, and obtain the corresponding trajectory offset Δθ from the action primitive library. (t） .

9. A dual-robotic arm control system for different dyeing scenarios according to claim 8, characterized in that, The adaptive update module includes: Obtain the remaining area information Extract the area S of region i i ′ and average brightness C i ',pass Calculate the total area S total The variance of the average brightness Ci′ across all regions is calculated to obtain the average brightness variance V. arc ; where C i ′ is a three-dimensional vector representing the color value in the LAB color space. The variance is calculated by averaging each component L, a, and b separately. Obtain the total area S of historically successful staining cases. total and mean lightness variance V arc And the corresponding optimal speed ratio and amplitude ratio, which are used to obtain the coloring uniformity f through the trained neural network. v and edge sharpness f a ; The overall immersion velocity V in the baseline co-motion layer command t and overall lifting speed V I pass and After making corrections, the overall immersion velocity V is obtained. t ′ and overall lifting speed V I ′; The action amplitude A in the independent adjustment command layer is corrected. The corrected amplitude A′ is obtained; Summarized and corrected overall immersion velocity V t Overall lifting speed V I The hierarchical collaborative operation instruction set is obtained from the amplitude A' and is fed back to the dual-arm drive module to be recalculated into the final motion parameters of each joint of the dual-arm, driving the dual-arm to perform motion.

10. A dual-robotic arm control device applicable to different dyeing scenarios, characterized in that, The device includes: a communication unit and a processing unit; The communication unit is used by the scene perception module to collect image information and physical property information of the fabric to be dyed, to collect surface image information of the fabric to be dyed through a high-resolution industrial camera, to collect material composition information of the fabric through a near-infrared spectral sensor, to collect viscosity information of the dye in the dyeing vat through a viscosity sensor, and to collect real-time temperature information of the dye in the dyeing vat through a temperature sensor. The processing unit is used to determine the dyeing type, which is either solid color or patterned color. The central collaborative control and decision-making module is connected to the scene perception module and is used to call the corresponding collaborative strategy according to the dyeing type to parse and generate a layered collaborative operation instruction set for the dual robotic arms, including the target motion trajectory, posture, speed, and immersion depth. The collaborative strategy is a two-layer framework including an immersion posture optimization algorithm for the solid color type and a pattern area adaptive adjustment model for the patterned type. The immersion posture optimization algorithm is an algorithm that calculates the optimal gripping point, gripping angle, and immersion posture of the dual robotic arms by analyzing the fabric material, dye viscosity, drying speed, and real-time temperature, with the optimization objectives of maximizing the amount of dye adsorbed per unit area of ​​the fabric and minimizing wrinkle deformation. The pattern area adaptive adjustment model is an algorithm that dynamically calculates and adjusts the gripping position, angle, and timing and amplitude of the dual robotic arms by analyzing the color distribution, area boundary, and color depth information of the pattern, with the optimization objective of the edge clarity between different color areas. The dual robotic arm drive module is connected to the central collaborative control and decision-making module. It is used to receive the hierarchical collaborative operation instruction set and calculate it into the final motion parameters of each joint of the dual robotic arm. It drives the dual robotic arms to be controlled according to the scene. When synchronized, the dual robotic arms operate as a whole unit. When local optimization is needed, the dual robotic arms independently perform fine-tuning actions. An adaptive update module, connected to the central collaborative control and decision-making module, is used to store dyeing process parameter records to form a dyeing database; and based on the comparative analysis of historical data and real-time monitoring data, optimizes and updates the collaborative strategy algorithm, and feeds the updated strategy back to the central collaborative control and decision-making module.

Images