Oxford cloth warping stage tension gradient section control method

Abstract

This invention relates to the field of specific computational models in textile engineering, specifically a segmented control method for tension gradient during the warping stage of Oxford cloth. The specific implementation steps include: first, real-time monitoring of the physical tension of N parallel warp yarn segments on the warping machine and forming a feedback vector; second, constructing a segmented weighted graph model of the warp beam describing the physical coupling relationship between the N tension execution points; then, using an ant colony optimization algorithm for iterative optimization; next, constructing candidate tension gradient profiles and quantifying and evaluating all profiles, updating the probability weight matrix based on the evaluation results; finally, outputting the currently found optimal tension gradient profile, achieving dynamically optimized tension gradient distribution. This invention, by employing an ant colony optimization algorithm combined with dynamic monitoring and multi-segment control, can achieve the effects of reducing scrap rate, improving overall equipment efficiency, reducing labor costs, and increasing production efficiency.

Description

Technical Field

[0001] This invention relates to the field of specific calculation models in textile engineering, specifically a method for segmented control of tension gradient during the warping stage of Oxford cloth. Background Technology

[0002] With the increasing complexity of modern manufacturing processes, such as piecewise control problems of tension gradients during the warping stage, a new type of computational challenge has emerged that transcends the scope of traditional control theory. The core of these challenges is no longer simply maintaining the stability of a single physical quantity, but rather finding an optimal combination solution within a multidimensional parameter space composed of a vast number of possibilities. As research into high-performance materials and special-function products deepens, engineers have discovered that in certain specific application scenarios, an absolutely uniform parameter distribution is not the optimal solution. Traditional computational and control models are designed to adjust and converge around one or more preset, fixed target values. These models, including the classic PID controller, perform excellently in handling reactive regulation problems such as "maintaining the system state at setpoint A." However, these traditional models struggle to find a set of parameter combinations, consisting of multiple different values, that maximize the overall performance of the final product from a solution space composed of a vast number of possibilities.

[0003] Therefore, the fundamental limitation of existing technologies lies in the lack of a specific computational model capable of effectively handling such combinatorial optimization problems; the ant colony algorithm is a prime example. By simulating the mechanism of communication and positive feedback using pheromones during foraging, the ant colony algorithm allows a large number of computational agents to collaboratively explore a complex solution space. This method inherently possesses the ability to handle complex combinatorial optimization problems, gradually discovering paths leading to the global optimum through iterative learning. It can perform efficient heuristic searches within a vast solution space, thus providing a feasible and efficient approach to solving the aforementioned combinatorial optimization challenges.

[0004] To address this, a segmented control method for tension gradient during the warping stage of Oxford cloth is proposed. Summary of the Invention

[0005] The purpose of this invention is to provide a method for segmented control of tension gradient during the warping stage of Oxford cloth, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A method for segmented control of tension gradient during the warping stage of Oxford cloth includes:

[0008] The physical tension of N parallel warp yarn sections on the warping machine is monitored in real time. The N measured values ​​are formed into a feedback vector representing the current process state. A predefined target tension gradient profile is received as a preset control target. A warp beam segment weighted graph model describing the physical coupling relationship between the N tension execution points is constructed.

[0009] In the controller, an ant colony optimization algorithm is used for iterative optimization based on the graph model. In each iteration, an artificial ant colony composed of multiple computing agents is deployed as a population. Each agent makes a series of probabilistic choices at each segment stage of the graph model based on a dynamically updated probability weight matrix and a static heuristic information matrix to construct candidate tension gradient profiles. After all populations have completed construction, the candidate tension gradient profiles are all quantitatively evaluated through a preset function, and the probability weight matrix is ​​updated according to the evaluation results.

[0010] When the iteration terminates, the currently found optimal tension gradient profile is output and transmitted to an array of N tension actuators, which drive the actuators to apply tension to each warp section, thereby achieving a dynamically optimized tension gradient distribution.

[0011] Preferably, the specific implementation process for acquiring the real-time status feedback vector of warp tension includes:

[0012] The tension sensor array is polled at a preset sampling frequency to obtain the real-time tension measurement value of the corresponding warp section, and the N measurement values ​​are combined into an N-dimensional real-time tension vector.

[0013] Preferably, the construction process of the meridian segmented weighted graph model includes:

[0014] The N warp segments are defined as N consecutive stages of the graphical model to map the entire width of the Oxford cloth; multiple nodes are created in each stage, where each node represents a discrete candidate tension set point applied to the k-th segment; and weighted edges are established between nodes in adjacent stages, with weights calculated based on the expected contribution to the physical properties of the final Oxford cloth, to find the adjustment path of the optimal tension gradient profile.

[0015] Preferably, the specific implementation process of the deployment, in which an artificial ant colony composed of multiple computing agents serves as the population, includes:

[0016] At the start of each iteration, all computational agents in the population are initialized and placed in the initial stage of the graph model. Each agent, based on shared pheromones and heuristic information, independently and in parallel performs probabilistic path selection in N consecutive stages of the graph model, and simultaneously constructs a set of candidate tension gradient profiles with structural diversity. The collective exploration behavior of the population is coordinated through a positive feedback mechanism, in which the evaluation results of multiple profiles constructed by the entire population are used together to update the pheromones, guiding the path selection tendencies of the agents in subsequent iterations, and realizing a systematic search for the global optimal solution.

[0017] Preferably, the specific implementation process of making the probabilistic selection based on the dynamically updated probability weight matrix and the static heuristic information matrix includes:

[0018] The dynamically updated probability weight matrix is ​​a pheromone matrix that can store and evaporate pheromones, reflecting the population's accumulated optimization experience; the static heuristic information matrix encodes tension gradient ranges set for forming the basket weave structure unique to Oxford cloth and yarn critical tension thresholds for the interlacing characteristics of grouped warp yarns and single weft yarns.

[0019] Preferably, the specific implementation process of constructing the candidate tension gradient profile includes:

[0020] At each stage k in the graph model, an agent selects an edge from its current node to a node in stage k according to a probability transition rule, where the probability of selecting a particular edge is proportional to a weighted function of the pheromone intensity associated with the particular edge and the heuristic information value encoding knowledge of Oxford cloth production; the agent constructs a complete path through all stages by repeating this process N times, which represents a candidate tension gradient profile designed to form the Oxford cloth basket weave.

[0021] Preferably, the specific implementation process of updating the probability weight matrix includes:

[0022] All existing weight values ​​in the matrix are multiplied by an evaporation coefficient less than 1, and each candidate tension gradient profile constructed by all agents in the current iteration is evaluated according to a preset physical performance evaluation function. The physical performance function quantifies and scores the profile based on the predicted final fabric physical uniformity index and the safety margin of yarn tension and critical tension threshold in each segment. Based on the evaluation results, the weight values ​​on the path corresponding to the best-performing candidate profile are enhanced.

[0023] Preferably, the specific implementation process of the tension gradient distribution includes:

[0024] When the iteration terminates, the final determined optimal tension gradient profile is converted into an N-dimensional digital command vector, and the command vector is sent to the actuator controller through a wireless communication network. The controller instructs the actuator array to apply precise and differentiated tension to the N warp sections on the warping machine, thereby stably forming the non-uniform tension gradient required for production.

[0025] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0026] 1. In terms of production efficiency, this invention dynamically adjusts tension through real-time monitoring and ant colony optimization algorithms to prevent yarn breakage due to improper tension. This reduces downtime during production, improves warping and weaving efficiency, lowers scrap rates, reduces the need for manual intervention, reduces labor costs, and makes the production process smoother.

[0027] 2. In terms of improving product quality, this method dynamically optimizes the warp tension gradient by pre-setting different target tension gradient profiles, precisely controlling the physical properties of different areas of the fabric. This significantly reduces common fabric defects found in traditional uniform tension control, such as slack warp yarns or uneven arrangement. This precise control ensures a smoother fabric surface and a more consistent structure, thereby developing high-quality Oxford cloth with specific hand feel, specific functionality, or convenience for specific downstream finishing processes. It is particularly suitable for the production of high-quality textiles.

[0028] 3. Regarding cost savings in production, this invention utilizes an ant colony optimization algorithm for dynamic tension control, ensuring that every tension applied to the yarn is precise and necessary. This refined management prevents energy waste caused by excessive or fluctuating tension at the source, directly reducing energy consumption of components such as drive motors. Simultaneously, by minimizing yarn overstretching and breakage, it significantly reduces raw material loss and eliminates unnecessary labor costs associated with yarn breakage. It also reduces mechanical impact and wear on key components such as tension sensors, yarn guides, and transmission systems in warping and weaving machines, effectively extending the overall lifespan of the equipment, reducing spare parts replacement frequency and maintenance costs, and ultimately achieving comprehensive optimization of production costs. Attached Figure Description

[0029] Figure 1 This is a flowchart of a segmented control method for tension gradient during the warping stage of Oxford cloth, as described in this invention.

[0030] Figure 2 This is a flowchart illustrating the design process for constructing the weighted graph model of the meridian segmentation as described in this invention.

[0031] Figure 3 This is a schematic diagram of obtaining candidate tension gradient profiles by performing a probabilistic selection decision loop for a single computational agent as described in this invention. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] Please see Figures 1 to 3 This invention relates to a method for segmented control of tension gradient during the warping stage of Oxford cloth, the specific implementation steps of which are as follows:

[0034] The physical tension of N parallel warp yarn sections on the warping machine is monitored in real time. The N measured values ​​are formed into a feedback vector representing the current process state. A predefined target tension gradient profile is received as a preset control target. A warp beam segment weighted graph model describing the physical coupling relationship between the N tension execution points is constructed.

[0035] In the controller, an ant colony optimization algorithm is used for iterative optimization based on the graph model. In each iteration, an artificial ant colony composed of multiple computing agents is deployed as a population. Each agent makes a series of probabilistic choices at each segment stage of the graph model based on a dynamically updated probability weight matrix and a static heuristic information matrix to construct candidate tension gradient profiles. After all populations have completed construction, the candidate tension gradient profiles are all quantitatively evaluated through a preset function, and the probability weight matrix is ​​updated according to the evaluation results.

[0036] When the iteration terminates, the currently found optimal tension gradient profile is output and transmitted to an array of N tension actuators, which drive the actuators to apply tension to each warp section, thereby achieving a dynamically optimized tension gradient distribution.

[0037] The technical solution of the present invention will be further described in detail below with reference to specific embodiments.

[0038] Example 1

[0039] This application discloses a segmented control method for the tension gradient during the warping stage of Oxford cloth. For the segmented control process of the tension gradient during the warping stage of Oxford cloth, please refer to... Figure 1The specific implementation steps of the method proposed in this invention include: Step S1, real-time monitoring of the physical tension on the warping machine to form a feedback vector representing the current process state; Step S2, receiving a predefined target tension gradient profile and constructing a segmented weighted graph model of the warp beam; Step S3, using an ant colony optimization algorithm for iterative optimization, deploying multiple computational agents in each iteration; Step S4, each agent makes probabilistic selections at each segment stage of the graph model to construct candidate tension gradient profiles; Step S5, after construction, quantifying and evaluating using a preset function, and updating the probability weight matrix based on the evaluation results; Step S6, when the iteration terminates, outputting the optimal tension gradient profile to achieve dynamically optimized tension gradient distribution.

[0040] Furthermore, the physical tension of N parallel warp yarn sections on the warping machine is monitored in real time, and the N measured values ​​are formed into a feedback vector representing the current process status; corresponding to step S1 above, the specific implementation process includes:

[0041] Based on the requirements of high tension in the central section and low tension in the edge section of the Oxford cloth basket weave, and to ensure the uniformity and strength of the fabric, a target tension gradient profile that meets the needs of this production is predefined as a control target.

[0042] A high-precision tension sensor array is deployed across the entire width of the warping machine to accurately capture the warp tension. The warp beam is conceptually divided into N consecutive parallel segments, where N is a sufficiently large integer to ensure that subtle changes in the tension gradient can be captured at high resolution. Each segment is monitored by at least one dedicated tension sensor. Various advanced sensor technologies can be employed here, depending on specific application requirements and cost considerations; the choice of sensor directly determines the quality of the feedback data.

[0043] The central controller polls the sensor array at a preset high sampling frequency to acquire real-time tension measurements for each warp segment. These N scalar measurements are aggregated in real time into an N-dimensional feedback vector. This vector constitutes a comprehensive quantitative snapshot of the physical state of the current warping process. This comprehensive quantitative snapshot reflects the tension state of the entire warp at any given moment and is the most important feedback signal for the entire control system. Traditional average value feedback cannot detect local tension deviations that lead to defects. The N-dimensional snapshot, on the other hand, facilitates the control system's real-time monitoring of the tension distribution details across the entire yarn surface, providing the necessary data foundation for subsequent differentiated high-resolution control.

[0044] By monitoring tension values ​​in real time, the problem of "tension blind spots" in traditional systems, which assume uniform tension or only measure average values, is solved, thus failing to detect local tension deviations that cause defects. A precise, comprehensive, and real-time warp tension map is constructed, laying the necessary data foundation for subsequent intelligent and differentiated control.

[0045] Furthermore, a predefined target tension gradient profile is received as a preset control target, and a piecewise weighted graph model describing the physical coupling relationship between N tension execution points is constructed; corresponding to step S2 above, the specific implementation process includes:

[0046] The physical width of the warp beam of the warping machine is divided into a series of N consecutive stages, each corresponding to one of N monitored warp yarn segments. Multiple discrete nodes are created within each stage. Each node represents a discrete candidate tension setpoint that can be applied to that segment. The range and resolution of these discrete tension values ​​are determined based on the specific physical characteristics of the yarn and the performance of the actuator. Directed edges are established between each node in the previous stage and every two discrete nodes in the next stage. Different weight values ​​for each edge are calculated based on its expected contribution to the physical properties of the final Oxford cloth. The weight of the edge represents the "goodness" or "attractiveness" of the decision. For example, if a local gradient change between two tension nodes is known to contribute to an ideal basket weave structure in Oxford cloth, the edge has a lower weight; conversely, if the gradient change is more likely to increase the risk of yarn breakage or defects, the edge has a higher weight. By using a weighted algorithm, a high-dimensional, continuous, nonlinear physical control problem in the field of control is successfully transformed into a discrete combinatorial optimization problem with mature solution algorithms in the field of computer science. This transformation facilitates the application of powerful metaheuristic algorithms such as ant colony optimization.

[0047] The system receives a predefined target tension gradient profile. This profile is not a rigid instruction, but rather an idealized tension distribution guide for a specific Oxford cloth variety based on historical data or expert knowledge. This instruction is primarily used to initialize heuristic information in the ant colony optimization algorithm.

[0048] Studies have shown a complex and variable relationship between warp tension and physical properties such as fabric crimp. For example, higher warp tension in the central region of the loom can sometimes lead to higher warp crimp, possibly due to the "pull-in effect" of the weft yarn during beat-up. Traditional linear control models cannot capture and optimize this complex physical phenomenon. However, by performing a modeling step, a complex, multivariable continuous control problem can be transformed into a structured, discrete combinatorial optimization problem. This abstraction allows combinatorial optimization algorithms, which are difficult to apply directly to the original physical system, to be directly incorporated into the model. This model can encode complex, nonlinear fabric mechanics principles, providing a framework for directly embedding expert process knowledge into the control logic through edge weights.

[0049] The weighted graph model of this invention, through its edge weight assignment, can accurately encode this empirical, nonlinear process knowledge. The weights of the edges connecting the central high-tension sections can be set to reflect the actual effect of this increased buckling rate. This graph model is not only a discretization of the problem space, but also a refined expression of the physical process of weaving, enabling subsequent optimization algorithms to find the optimal solution that truly conforms to physical reality.

[0050] Furthermore, an ant colony optimization algorithm is used for iterative optimization, and in each iteration, an artificial ant colony composed of multiple computing agents is deployed as the population; corresponding to step S3 above, the specific implementation process includes:

[0051] At the beginning of each optimization iteration cycle, the controller deploys an artificial ant colony consisting of M independent computing agents as a population on the graph model constructed in step S2, where M is an integer preset according to the problem complexity and computing resources.

[0052] All M computation agents are logically initialized and placed at the virtual starting point of the graph model before entering the first warp segment. These computation agents together constitute an artificial ant colony for exploring the solution space.

[0053] The initialization process ensures that all agents in each iteration start from the same point, exploring the complete tension profile path from one end of the bandwidth to the other, ensuring a complete exploration without omissions. Deploying the population allows multiple agents to work in parallel and simultaneously generate a set of structurally diverse candidate solutions. Each agent follows shared rules, independently constructing a path on the graph model. Unlike traditional single-point search algorithms, this collective exploration behavior forms the basis for subsequent systematic searching for the global optimum.

[0054] By deploying a population of multiple parallel computing agents, a distributed, parallel search framework is established. This approach fundamentally improves the efficiency and robustness of finding the global optimum. Compared with traditional single-point, serial search methods, it can explore a wider range of complex solution spaces, laying the foundation for avoiding getting trapped in local optima.

[0055] Furthermore, each agent, based on a dynamically updated probability weight matrix and a static heuristic information matrix, performs a series of probabilistic selections at various stages of the graph model to construct candidate tension gradient profiles; corresponding to step S4 above, the specific implementation process includes:

[0056] Each computational agent independently and in parallel constructs a candidate solution, i.e., a complete path, across N stages of the graph model. In each stage, all computational agents located at different nodes choose their next node to visit according to a probabilistic transition rule. The probability of choosing a specific path is influenced by both the pheromone trajectory and heuristic information. The pheromone trajectory is the concentration of "pheromone" attached to the edge connecting two nodes. The "pheromone" concentration value is stored in a dynamically updated probability weight matrix, representing the cumulative optimization experience of the entire ant colony and indicating the success rate of this particular state transition in past iterations. The heuristic information is a static value associated with this state transition, representing proven advanced process knowledge in the domain; this value is stored in a static heuristic information matrix.

[0057] The computational agent's path selection at each decision point is not based on randomness or a single greedy criterion, but is guided by the synergy of two complementary and different core information matrices.

[0058] The dynamically updated probability weight matrix reflects the algorithm's learning ability. Essentially, it's a dynamic repository of collective experience accumulated by the entire computational agent population during past iterations of optimization. This matrix doesn't store fixed rules, but rather evolving probabilistic preferences reflecting the population's optimization experience. Each weight value in the matrix corresponds to a specific edge connecting two adjacent stage nodes in the graph model, and its value represents the "attractiveness" or "success history" of that path transition. During algorithm execution, if a path is proven to be part of a high-quality final solution, the weight value of the corresponding edge is strengthened; conversely, it decays over time. This matrix guides the computational agent to prioritize paths historically proven to be more likely to lead to the global optimum, effectively utilizing existing successful experiences.

[0059] The static heuristic information matrix is ​​a static knowledge base that is constructed once before the algorithm runs and remains unchanged throughout the optimization process. Its contents encode prior domain knowledge, physical constraints, and process rules already established regarding Oxford cloth production.

[0060] Based on the study of Oxford cloth's basket weave structure, the tension values ​​and their gradient ranges known to contribute to ideal fabric feel and appearance are designated as "preferred zones" with high heuristic information values. According to the mechanical properties of the yarn material, any path selection that might cause the yarn tension to approach its critical breaking strength is assigned extremely low heuristic information values, thus establishing a safety barrier at the algorithm level to prevent yarn breakage. Integrating historical production data and fabric defect analysis results, significant heuristic penalties are applied to specific tension jumps or distribution patterns that have been proven to be strongly correlated with common fabric defects, guiding the agent to proactively avoid these known "risk zones" from the outset. This matrix can provide directional guidance in the early stages of algorithm exploration, ensuring that the computational agent's search behavior is within a reasonable and promising solution space from the beginning.

[0061] The combined use of a dynamic probability weight matrix and a static heuristic information matrix brings significant and beneficial technical effects to the intelligent optimization process. The static heuristic matrix pre-loads deterministic process knowledge, such as optimal tension ranges and defect risk zones, guiding the search direction from the initial optimization stage and avoiding ineffective calculations within known poor solution regions. The dynamic probability weight matrix records and reinforces paths proven effective in iterations, leading the exploration towards better solutions and demonstrating the system's self-learning and adaptive capabilities. This dual-matrix mechanism utilizes accumulated successful experience, making the exploration more comprehensive and complete. By employing probabilistic selection, it effectively avoids the local optimum trap caused by relying solely on heuristic information, endowing the system with high flexibility to adapt to different production requirements, ultimately achieving optimization results unattainable by traditional single-direction methods.

[0062] After N consecutive probabilistic selections, each computational agent constructs a complete path that runs through all stages. Each such path represents a candidate tension gradient profile designed to form an Oxford basket organization. The collective behavior of the ant colony generates a set of M candidate solutions with structural diversity in each iteration.

[0063] The process of a single computational agent constructing a candidate profile is a sequential decision-making process that proceeds N times consecutively from the first stage to the Nth stage of the graph model.

[0064] The process begins with the agent being placed at the starting point of the graph model. At each segmentation stage, the agent needs to choose a path from its current node to the next stage node. The controller comprehensively evaluates all possible paths from the current node to all available nodes in the next stage and calculates a comprehensive attractiveness for each path. This attractiveness is a weighted combination of dynamic probability weights and static heuristic information; paths with higher comprehensive attractiveness are more likely to be selected by the agent. Randomness is also introduced, meaning even paths with slightly lower attractiveness have a chance of being selected. Each agent in the population is influenced by both the calculated comprehensive attractiveness and randomness in its path selection, ensuring the algorithm can escape local optima and explore potentially better, albeit less intuitive, solutions.

[0065] The agent repeatedly executes this "evaluation-probabilistic selection-transfer" decision loop N times. Each time a selection is completed, the target tension value for the next warp segment is determined. After N selections, the agent has established a complete path across all stages on the graphical model. This path, consisting of N sequentially connected nodes, is physically mapped precisely to a complete candidate tension gradient profile covering all N segments of the warping machine. In the same iteration, all computational agents in the population independently and in parallel complete this construction process, synchronously generating a set of structurally diverse candidate solutions, providing rich input data for subsequent evaluation and optimization steps.

[0066] By having each computational agent construct a complete N-dimensional solution sequentially and stage by stage, the ant colony optimization algorithm ensures that each candidate tension gradient profile generated is a coherent and holistic solution, rather than a disordered set of discrete tension setpoints. This structured construction process considers the relationships between adjacent segments at each step, guaranteeing the physical realizability of the final profile. Furthermore, since each agent's path selection is a probabilistic event based on shared information but performed independently, the entire population can generate a set of candidate tension gradient profiles with high structural diversity in parallel. By constructing a large number of candidate tension gradient profiles, rich samples are provided for subsequent evaluation and learning steps, effectively preventing the algorithm from prematurely converging to local optima and significantly increasing the likelihood of finding the truly globally optimal tension distribution.

[0067] Furthermore, after all populations have been constructed, the candidate tension gradient profiles will all be quantized and evaluated using a preset function, and the probability weight matrix will be updated based on the evaluation results; corresponding to step S5 above, the specific implementation process includes:

[0068] After all M computational agents have completed path construction, each of these M candidate profiles will be quantitatively evaluated through a pre-defined, multi-objective evaluation function and assigned a fitness score. This function, acting as a "digital twin," evaluates each profile's performance in terms of expected fabric physical uniformity, safety margins for yarn tension and critical thresholds in each segment, and predicted fabric breaking strength, based on a weighted composite index. It integrates multiple key performance indicators defining the final quality of Oxford cloth into a single, optimizable evaluation value. In actual production, different quality objectives often conflict. This evaluation function allows for intelligent trade-offs between different objectives by adjusting the weights of each indicator. Manufacturers can flexibly customize optimization objectives according to different order requirements, such as prioritizing "strength" for industrial fabrics and "uniformity" for high-end fabrics, enhancing production flexibility.

[0069] The probability weight matrix is ​​updated based on the fitness scores of all M candidate profiles. After each iteration, the pheromone value on all edges in the graph model is multiplied by an evaporation coefficient less than 1 to simulate the natural decay of pheromones. This decay mechanism helps the algorithm eliminate interference from poorly performing paths, preventing premature convergence to suboptimal solutions. Paths corresponding to the best-performing candidate profiles receive additional pheromone deposition. The amount of deposited pheromone is proportional to the fitness score of that profile. This positive feedback mechanism eliminates interference from poorly performing paths, strengthens promising paths, and provides more biased guidance for subsequent computation of the agent to search these high-quality regions in the solution space.

[0070] A robust self-learning mechanism is formed through the use of an "evaluation-update" closed loop. The system iterates to continuously deepen its understanding of the optimal tension profile and adapts to specific yarn, equipment conditions, and fabric types. It allows for holistic, collaborative optimization of multiple, sometimes even conflicting, quality objectives—something traditional single-variable control systems cannot achieve.

[0071] Furthermore, when the iteration terminates, the currently found optimal tension gradient profile is output and transmitted to an array of N tension actuators, driving the actuators to apply tension to each warp segment, thereby achieving a dynamically optimized tension gradient distribution; corresponding to step S6 above, the specific implementation process includes:

[0072] The iterative optimization process described above will continue until a preset termination condition is met. This condition can be a fixed number of iterations, a time limit, or when the improvement of the optimal solution falls below a certain threshold for several consecutive iterations. When the iteration terminates, the algorithm outputs the globally optimal tension gradient profile P found throughout all iterations.

[0073] This P-vector, composed of N target tension values, is converted into an N-dimensional digital command vector. Each element of the vector is a digital signal that drives the corresponding actuator to generate the required tension. This command vector is transmitted via a highly reliable industrial communication protocol or a stable wireless network to an array of N high-performance actuators, one actuator corresponding to each warp segment. Upon receiving the command vector, the actuator controller synchronously drives each of the N actuators to its designated setpoint. This physically creates a precise, non-uniform tension gradient on the warp surface, determined by an ant colony optimization algorithm. The high-speed response of the actuators not only enables the system to establish this static profile but also allows for dynamic updates in response to changes in machine speed or other disturbances, thus maintaining optimal tension throughout the entire warping process.

[0074] By employing an array of tension actuators to apply tension to each warp segment, the results of the preceding optimization process are fully realized, achieving a dynamically optimized tension gradient distribution. This physically creates a complex, optimized tension distribution that traditional mechanical or simple electronic systems cannot achieve. This high-fidelity transformation from optimal blueprint to physical reality results in quantifiable improvements in fabric quality, including a significant reduction in defect rates, enhanced mechanical properties, and improved consistency in the production process.

[0075] Example 2

[0076] A textile factory A, producing high-quality Oxford cloth, has a warping machine that processes warp bundles with a total width of 1.5 meters, which can be divided into N=10 parallel warp sections, each containing approximately 500 individual warp yarns. To produce Oxford cloth with specific physical properties, the process requires creating a specific tension gradient profile during warping: the warp tension in the central section needs to be maintained at a higher level of approximately 47 g / f, gradually decreasing towards the outer edges to a minimum of approximately 25 g / f. This invention achieves dynamic and precise control of this complex tension gradient through a closed-loop control system integrating sensing, intelligent optimization, and precise execution, aiming to reduce the yarn breakage rate by 2% from 5% and increase the fabric breaking strength by 0.3 kN.

[0077] A central controller polls an array of 10 high-precision tension sensors deployed on a 1.5-meter-wide warp machine at a preset high sampling frequency of 100Hz. Each sensor independently monitors the instantaneous physical tension of its corresponding warp section, which contains approximately 500 warp yarns, obtaining a set of readings [26.1, 27.5, 35.2, 45.8, 47.1, 46.9, 36.0, 28.1, 25.5, 24.9] cN. The controller combines these 10 measurements into a 10-dimensional feedback vector. Simultaneously, the controller receives a predefined target tension gradient profile, set according to the process requirements of the Oxford basket weave, with a specific gradient of approximately 47 cN in the central section and approximately 25 cN in the edge sections. After acquiring the real-time status and control objective, the system transforms this physical optimization problem into a structured mathematical problem for easy solution within the computational domain.

[0078] A segmented weighted graph model of the warp beam is constructed, which abstracts 10 parallel warp yarn segments into 10 consecutive stages in the graph model, and creates a series of nodes representing discrete candidate tension setpoints within each stage. Directed weighted edges are established between nodes in adjacent stages, and the weight of the edge is calculated based on the expected contribution of the local decision to the final physical properties of the Oxford cloth. This model successfully transforms the complex tension optimization problem into finding an optimal path from the first stage to the tenth stage on the weighted graph. The tension values ​​represented by the 10 nodes traversed along the path constitute a complete candidate tension gradient profile.

[0079] Subsequently, the system activates the ant colony optimization algorithm as its core optimization engine to iteratively explore the vast solution space defined by the graph model. At the start of each iteration, the controller deploys a population of M=100 computational agents, acting as artificial ants. These agents independently and in parallel perform probabilistic path selection in 10 consecutive stages of the graph model, thereby simultaneously constructing a set of candidate tension gradient profiles with high structural diversity. The dynamically updated probability weight matrix reflects the population's accumulated optimization experience; the static heuristic information matrix encodes prior domain knowledge, such as the preferred tension gradient range set for forming the unique basket weave in Oxford cloth production, and the critical tension threshold set to prevent yarn breakage. Each agent's decision-making process is based on a probabilistic transition rule that combines key information from both matrices.

[0080] Once all agents in the ant colony have completed path construction, the system enters the evaluation and learning phase. All generated candidate tension gradient profiles are quantitatively scored using a pre-defined physical performance evaluation function. This function acts as a "digital twin," comprehensively evaluating each profile's performance in terms of expected fabric physical uniformity, safety margins of yarn tension and critical thresholds in each segment, and predicted fabric breaking strength. The evaluation results are then used to update the shared probability weight matrix: on the one hand, all existing pheromones are multiplied by an evaporation coefficient of ρ = 0.1 for attenuation; on the other hand, the pheromones on the path corresponding to the best-performing candidate profile are enhanced.

[0081] Finally, when the maximum number of iterations (200) is reached or the solution achieves quality convergence, the preset iteration termination condition is met, and the optimization process stops. The system outputs the optimal tension gradient profile with the highest evaluation score found throughout the optimization process. This optimal profile is converted into a 10-dimensional digital command vector and sent to the controller of the actuator array via a highly reliable industrial wireless communication network. The controller then instructs an array of 10 high-performance asynchronous servo motors to apply precise, differentiated tension to 10 warp sections on the warping machine, stably forming the non-uniform tension gradient required for producing high-quality Oxford cloth during the production process.

[0082] Example 3

[0083] The method involves constructing a segmented weighted graph model of the warp axis that describes the physical coupling relationship between N tension execution points. The specific implementation process includes:

[0084] First, the various stages in the diagram are established. The numerous warp yarn areas that are actually arranged are defined as a series of successive diagram stages within the simulation framework. By dividing the warp yarn segments into stages, the actual operation of "depicting the spatial tension distribution" can be equated in the simulation system to "finding a path from the starting point to the end point in the diagram", which facilitates the use of optimization algorithms for control in subsequent steps.

[0085] Considering both practical ease of operation and computational efficiency, several nodes are constructed in each stage. Each node represents a discretized alternative tension value applied to the corresponding warp yarn region. For example, based on specific textile process specifications, an ideal tension range for a certain cotton yarn is pre-set. To achieve precise and controllable tension adjustment, this continuous tension range is divided into equal intervals. If a small increment is set, numerous nodes will appear within each stage, each representing a specific alternative tension value within that range. The entire graph includes a large number of state nodes, providing a broad option space for subsequent path selection. By defining different nodes, it is ensured that the model can capture subtle tension changes, thereby providing robust data support for the optimization process.

[0086] Weighted edges are constructed between nodes in adjacent stages. An edge is created by connecting nodes in the previous stage to nodes in the next stage. Each edge corresponds to a specific decision chain: after applying a specific tension to the current warp area, another tension is immediately applied to the next adjacent area. The weight of each edge is precisely calculated based on the expected contribution of this decision to the final physical properties of the Oxford cloth, reflecting the merits of each path. For example, if practical experience or experimental data in textile processes reveals that maintaining a small and specific tension difference between adjacent yarn areas helps form a smoother, more structurally stable basket weave, then the weight of the connection between two tension value nodes that meet this condition will be assigned a higher value to encourage the algorithm to favor this type of tension transition. If a tension combination might cause yarn breakage or fabric defects, its weight will be significantly reduced. This parameter helps in the probabilistic selection of the ant colony optimization algorithm in subsequent applications. Each computational agent needs to rely on this weight when exploring the optimal tension path in the graph, thereby achieving a dynamically optimized tension gradient distribution. By introducing weights for each edge, the model can not only identify feasible tension combinations, but also assess their profound impact on the quality of the final product, providing a data foundation for future model updates.

[0087] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for segmented control of tension gradient during the warping stage of Oxford cloth, characterized in that, include: The physical tension of N parallel warp yarn sections on the warping machine is monitored in real time. The N measured values ​​are formed into a feedback vector representing the current process state. A predefined target tension gradient profile is received as a preset control target. A warp beam segment weighted graph model describing the physical coupling relationship between the N tension execution points is constructed. In the controller, an ant colony optimization algorithm is used for iterative optimization based on the graph model; in each iteration, an artificial ant colony composed of multiple computing agents is deployed as the population. Each agent makes a series of probabilistic choices at each segment stage of the graph model based on a dynamically updated probability weight matrix and a static heuristic information matrix, thereby constructing candidate tension gradient profiles. After all populations are constructed, the candidate tension gradient profiles will be quantified and evaluated using a preset function, and the probability weight matrix will be updated based on the evaluation results. When the iteration terminates, the currently found optimal tension gradient profile is output and transmitted to an array of N tension actuators, which drive the actuators to apply tension to each warp section, thereby achieving a dynamically optimized tension gradient distribution.

2. The method for segmented control of tension gradient during the warping stage of Oxford cloth according to claim 1, characterized in that, The specific implementation process of acquiring the real-time state feedback vector of warp tension includes: polling the tension sensor array at a preset sampling frequency to obtain the instantaneous tension measurement value of the corresponding warp section, and combining the N measurement values ​​into an N-dimensional real-time tension vector.

3. The method for segmented control of tension gradient during the warping stage of Oxford cloth according to claim 1, characterized in that, The construction process of the warp beam segmented weighted graph model includes: defining the N warp yarn segments as N consecutive stages of the graph model to map the entire width of the Oxford cloth; creating multiple nodes in each stage, where each node represents a discrete candidate tension set point applied to the k-th segment; and establishing weighted edges between nodes in adjacent stages, with weights calculated based on the expected contribution to the physical properties of the final Oxford cloth, to find the adjustment path of the optimal tension gradient profile.

4. The method for segmented control of tension gradient during the warping stage of Oxford cloth according to claim 1, characterized in that, The specific implementation process of deploying an artificial ant colony composed of multiple computational agents as a population includes: at the beginning of each iteration, initializing all computational agents in the population and placing them in the initial stage of the graph model; each agent, based on shared pheromones and heuristic information, independently and in parallel performs probabilistic path selection in N consecutive stages of the graph model, and simultaneously constructs a set of candidate tension gradient profiles with structural diversity; the collective exploration behavior of the population is coordinated through a positive feedback mechanism, wherein the evaluation results of multiple profiles constructed by the entire population will be used together to update the pheromones, guiding the path selection tendencies of the agents in subsequent iterations, and realizing a systematic search for the global optimal solution.

5. The method for segmented control of tension gradient during the warping stage of Oxford cloth according to claim 1, characterized in that, The specific implementation process of the probabilistic selection based on the dynamically updated probability weight matrix and the static heuristic information matrix includes: the dynamically updated probability weight matrix is ​​a pheromone matrix that can store and evaporate pheromones and reflects the accumulated optimization experience of the population; the static heuristic information matrix encodes the tension gradient range set for forming the basket weave structure unique to Oxford cloth and the yarn critical tension threshold for the interlacing characteristics of grouped warp yarns and single weft yarns.

6. The method for segmented control of tension gradient during the warping stage of Oxford cloth according to claim 1, characterized in that, The specific implementation process of constructing the candidate tension gradient profile includes: at each stage k of the graph model, an agent selects an edge from its current node to a node in stage k according to a probability transition rule, wherein the probability of selecting a specific edge is proportional to the weighted function of the pheromone intensity associated with the specific edge and the heuristic information value encoding the knowledge of Oxford cloth production; the agent constructs a complete path through all stages by repeating this process N times, and this path represents a candidate tension gradient profile designed to form the Oxford cloth basket weave.

7. The method for segmented control of tension gradient during the warping stage of Oxford cloth according to claim 1, characterized in that, The specific implementation process of updating the probability weight matrix includes: multiplying all existing weight values ​​in the matrix by an evaporation coefficient less than 1, and evaluating each candidate tension gradient profile constructed by all agents in the current iteration according to a preset physical performance evaluation function. The physical performance evaluation function quantifies and scores the profile based on the predicted final fabric physical uniformity index and the safety margin of yarn tension and critical tension threshold in each section; and enhancing the weight values ​​on the path corresponding to the best-performing candidate profile according to the evaluation results.

8. The method for segmented control of tension gradient during the warping stage of Oxford cloth according to claim 1, characterized in that, The specific implementation process of the tension gradient distribution includes: when the iteration terminates, the finally determined optimal tension gradient profile is converted into an N-dimensional digital command vector, and the command vector is sent to the actuator controller through a wireless communication network. The controller controls the actuator array to apply precise and differentiated tension to N warp sections on the warping machine, thereby stably forming the non-uniform tension gradient required for production.

Images