Superstrong wear-resistant yarn production control method and system

Abstract

The invention discloses a super-strong wear-resistant yarn production control method and system, and relates to the technical field of automatic control and intelligent manufacturing of textile machinery, and the system comprises an upper industrial personal computer, a lower computer PLC control module, a motor driving module, a yarn section high-speed imaging unit, an infrared friction thermal imaging sensor and a dynamic parameter analyzer; the upper industrial personal computer is respectively connected with the lower computer PLC control module and the dynamic parameter analyzer; the lower computer PLC control module is connected with the motor driving module, the yarn section high-speed imaging unit and the infrared friction thermal imaging sensor. According to the production control method and system for the super-strong wear-resistant yarn, through organic combination of three-level verification and iterative optimization, the system can continuously control the uniformity of the cross section structure of the yarn, the stability of the twist degree and the fluctuation of the tension within the target range, and therefore stable output of the wear-resistant performance of the yarn is ensured.

Description

Technical Field

[0001] This invention relates to the field of automatic control and intelligent manufacturing technology of textile machinery, specifically to a method and system for controlling the production of ultra-wear-resistant yarn. Background Technology

[0002] With the rapid development of the textile industry, yarn products are constantly evolving towards high performance, multi-functionality, and differentiation. Among them, ultra-abrasion-resistant yarns, due to their excellent mechanical properties and durability, are widely used in industrial textiles, outdoor equipment, high-performance clothing, and special protective applications. These yarns typically use high-strength fiber raw materials and combine them with specific spinning processes to give them tensile strength, abrasion resistance, and fatigue resistance, meeting the performance requirements of extreme operating environments. Therefore, how to efficiently and stably produce ultra-abrasion-resistant yarns of consistent quality has become an important research direction in the textile manufacturing field.

[0003] Currently, the production of ultra-abrasion-resistant yarns mainly relies on the modification and upgrading of traditional ring spinning machines or twisting machines. In actual production, the abrasion resistance of yarn depends not only on the characteristics of the fiber raw material itself, but also on the draft ratio, twist distribution, tension control, and the precision of the multi-roller coordinated motion during the spinning process. Insufficient control precision can easily lead to uneven yarn structure, large twist fluctuations, and an increase in local weak loops, thus significantly reducing the yarn's abrasion resistance and service life. However, the control precision and stability of existing domestic spinning equipment under high-speed operation still need improvement, making it difficult to meet the requirements of precise control of process parameters for high-performance yarns.

[0004] Publication number "CN108776463A" describes "An Intelligent Fancy Yarn Control System and Method." This solution, through the collaborative work of a host industrial control computer and multiple slave PLC control modules, combined with digital signal processing and motor drive modules, achieves precise control of mechanisms such as the middle and back rollers and hollow spindles, enabling the production of composite fancy yarns with diverse structures and rich shapes. This technical solution has made significant progress in improving the automation level and processing accuracy of spinning equipment, providing strong support for the intelligent production of fancy yarns. However, this solution mainly focuses on the appearance effect and structural diversity control of fancy yarns, and has not yet addressed how to improve the intrinsic mechanical properties of yarns, especially abrasion resistance, through fine adjustment of process parameters. In the production of ultra-abrasion-resistant yarns, more attention needs to be paid to raw material ratio, twist distribution, tension fluctuation, and dynamic compensation control through multi-motor collaboration to ensure uniform yarn cross-sectional structure, stable twist, and consistent strength, thereby obtaining excellent abrasion resistance characteristics.

[0005] In summary, existing spinning control systems still have certain limitations in achieving efficient and stable production of ultra-abrasion-resistant yarns. Therefore, there is an urgent need to develop an intelligent control method and system specifically for the production of ultra-abrasion-resistant yarns. This system would optimize process parameter matching through precise monitoring and dynamic control of the entire spinning process, improve the uniformity of yarn structure and the stability of mechanical properties, and meet the stringent requirements of the high-end market for abrasion-resistant yarn products. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for controlling the production of ultra-abrasion-resistant yarn, so as to solve the problems mentioned in the background art.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a method and system for controlling the production of ultra-wear-resistant yarn, including an upper industrial control computer, a lower PLC control module, a motor drive module, a high-speed imaging unit for yarn cross-section, an infrared friction thermal imaging sensor, and a dynamic parameter analyzer; The upper-level industrial control computer is connected to the lower-level PLC control module and the dynamic parameter parser, respectively. The lower-level PLC control module is connected to the motor drive module, the yarn cross-section high-speed imaging unit, and the infrared friction thermal imaging sensor, respectively. The motor drive module is connected to the Roller servo motor; The host industrial control computer has a built-in dynamic wear resistance coefficient prediction model and a yarn structure stability evaluation model. The dynamic parameter parser interacts with the dynamic abrasion resistance coefficient prediction model and the yarn structure stability evaluation model, respectively. The dynamic parameter parser is also connected to the lower-level PLC control module, and is used to receive real-time data uploaded by the lower-level PLC control module, and send the parsed control commands to the lower-level PLC control module.

[0008] The host industrial control computer, acting as the decision-making core, establishes bidirectional data communication with the lower-level PLC control module. The dynamic parameter parser completes information interaction and command issuance with each module through multiple data interfaces, realizing the coordinated linkage of hardware and software modules.

[0009] Preferably, the dynamic wear resistance coefficient prediction model is a prediction model based on a BP neural network, whose input parameters include process parameters, real-time state parameters and raw material characteristic parameters, and whose output parameter is the instantaneous wear resistance coefficient prediction value; The process parameters include draft ratio, twist coefficient, core yarn tension, and overfeed ratio; The real-time status parameters include the difference in rotational speed between the middle and rear rollers, the fluctuation rate of the rotational speed of the hollow spindle, and the ellipticity of the yarn cross-section. The difference in rotational speed between the middle and rear rollers is calculated by the lower-level PLC control module based on the feedback signal from the roller encoder. The fluctuation rate of the rotational speed of the hollow spindle is calculated by the lower-level PLC control module based on the feedback signal from the hollow spindle encoder. The ellipticity of the yarn cross-section is obtained by image processing of the image acquired by the high-speed imaging unit of the yarn cross-section. The raw material characteristic parameters include fiber length dispersion and breaking strength variation coefficient, which are obtained by the host industrial control computer from a preset database based on the input raw material batch information.

[0010] The prediction model adopts a three-layer BP neural network architecture. The input layer contains nine nodes corresponding to various parameters. The model determines the weight coefficients through offline training and fixes them. It can perform forward propagation calculations at a fixed frequency to output predicted values.

[0011] Preferably, the yarn structure stability assessment model is an assessment model constructed based on the support vector regression algorithm. Its input parameters are the feature parameters extracted after image processing of multiple consecutive frames of yarn cross-section images acquired by the high-speed imaging unit of the yarn cross-section, and the output parameter is the structural stability index. The characteristic parameters include yarn diameter variation coefficient, fiber arrangement entropy, and hairiness index dynamic change rate; The yarn diameter variation coefficient is calculated by the ratio of the standard deviation to the mean of the yarn diameter in multiple consecutive frames of images; The fiber arrangement entropy is calculated based on the image gray-level co-occurrence matrix; The dynamic change rate of the feather index is calculated from the fluctuation range of the number of feathers in multiple consecutive frames of images.

[0012] The evaluation model uses Gaussian radial basis function as kernel function, determines support vectors and weight coefficients through offline training, and completes the quantitative evaluation of structural stability based on continuously acquired image feature parameters.

[0013] Preferably, the dynamic parameter parser has a built-in three-level data verification module; The three-level data verification module includes a first-level verification unit, a second-level verification unit, and a third-level verification unit; The first-level verification unit is connected to the dynamic abrasion resistance coefficient prediction model and the yarn structure stability evaluation model, respectively, and is used to receive the instantaneous abrasion resistance coefficient prediction value and the structure stability index, and calculate the absolute value of the difference between the two. The first-level verification unit has a built-in first preset threshold. When the absolute value of the difference is less than or equal to the first preset threshold, the first-level verification unit sends the instantaneous abrasion resistance coefficient prediction value to the second-level verification unit. When the absolute value of the difference is greater than the first preset threshold, the first-level verification unit sends a parameter correction instruction to the dynamic abrasion resistance coefficient prediction model. The dynamic abrasion resistance coefficient prediction model adjusts the set values ​​of the twist coefficient and core yarn tension according to the parameter correction instruction and then recalculates the instantaneous abrasion resistance coefficient prediction value until the absolute value of the difference is less than or equal to the first preset threshold or the number of triggers reaches a preset number.

[0014] The first-level verification unit is equipped with a comparator and a counter. The parameter correction command will adjust the twist coefficient and core yarn tension according to a fixed ratio. When the number of triggers reaches the upper limit, the system will issue an alarm signal and switch the control mode.

[0015] Preferably, the second-level verification unit is connected to the historical best process library built into the host industrial control computer; The second-level verification unit receives the instantaneous wear resistance coefficient prediction value after verification by the first-level verification unit, and searches in the historical optimal process library for similar process combinations whose Euclidean distance from the current process parameter combination is less than or equal to a second preset threshold. It reads the actual wear resistance index corresponding to the similar process combination and calculates the average deviation between the instantaneous wear resistance coefficient prediction value and the actual wear resistance index. The second-level verification unit has a built-in third preset threshold. When the average deviation is less than or equal to the third preset threshold, the second-level verification unit sends the instantaneous wear resistance coefficient prediction value to the third-level verification unit. When the average deviation is greater than the third preset threshold, the second-level verification unit extracts the raw material characteristic parameters in the similar process combination that differ from the current raw material characteristic parameters by more than a fourth preset threshold. Based on a preset raw material sensitivity coefficient table, the second-level verification unit corrects the weight coefficients of the raw material characteristic parameters in the dynamic wear resistance coefficient prediction model and sends the corrected weight coefficients to the dynamic wear resistance coefficient prediction model, triggering the dynamic wear resistance coefficient prediction model to recalculate the instantaneous wear resistance coefficient prediction value.

[0016] The second-level verification unit normalizes the combination of process parameters and calculates the Euclidean distance. The weight corrector then performs linear interpolation correction of the model weight coefficients according to the relative differences in the raw material characteristic parameters and the raw material sensitivity coefficient table.

[0017] Preferably, the third-level verification unit is connected to the infrared triboelectric thermal imaging sensor; The infrared friction thermal imaging sensor is used to monitor the temperature rise rate of the friction area between the yarn and the guide hook in real time, and sends the temperature rise rate to the third-level verification unit. The third-level verification unit has a built-in conversion formula between temperature rise rate and wear resistance coefficient, and calculates the thermal imaging feedback wear resistance coefficient based on the temperature rise rate. The third-level verification unit receives the instantaneous wear resistance coefficient prediction value after verification by the second-level verification unit, and calculates the absolute value of the difference between the instantaneous wear resistance coefficient prediction value and the thermal imaging feedback wear resistance coefficient. The third-level verification unit has a built-in fifth preset threshold. When the absolute value of the difference is less than or equal to the fifth preset threshold, the third-level verification unit outputs the instantaneous wear resistance coefficient prediction value as the final control target value to the dynamic parameter parser. When the absolute value of the difference is greater than the fifth preset threshold, the third-level verification unit sends the thermal imaging feedback wear resistance coefficient as the target value to the dynamic parameter parser.

[0018] The calculated value of the wear resistance coefficient based on thermal imaging feedback will be limited to a range of 0 to 100. The third-level verification unit will count the deviation exceeding the limit events, and continuous exceeding of the limit will trigger the sensor-related inspection prompts.

[0019] Preferably, the dynamic parameter parser has a built-in particle swarm optimization module; The particle swarm optimization module is connected to the third-level verification unit. When the thermal imaging feedback abrasion resistance coefficient is received as the target value, the particle swarm optimization module takes the thermal imaging feedback abrasion resistance coefficient as the target and uses the draft ratio, twist coefficient, core yarn tension and overfeed ratio as optimization variables to solve a new set of process parameters in reverse using the particle swarm optimization algorithm. The dynamic parameter parser converts the new combination of process parameters into motion control commands and sends them to the lower-level PLC control module. The lower-level PLC control module controls the roller servo motor to run according to the motion control command, and restarts the calculation of the dynamic wear resistance coefficient prediction model after a preset time interval.

[0020] The population size and number of iterations in the particle swarm optimization module are fixed values. The algorithm uses the absolute value of the difference between the predicted value and the target value as the fitness function. The combination of process parameters obtained by solving the algorithm will be output after boundary truncation.

[0021] Preferably, the host industrial control computer has a built-in historical best process library and a self-learning module; The historical optimal process library is used to store combinations of process parameters verified by laboratory wear resistance tests and their corresponding actual wear resistance indices. The self-learning module is connected to the dynamic parameter parser and the historical optimal process library, respectively, and is used to associate and store the final process parameter combination determined after each iteration of optimization and its corresponding instantaneous wear resistance coefficient prediction value to the historical optimal process library.

[0022] The historical best process library is a structured database. The stored data records are timestamped and have raw material batch labels. The self-learning module performs a data deduplication check before writing data.

[0023] Preferably, the frame rate of the high-speed imaging unit for yarn cross-section is not less than one thousand frames per second, and it is used to acquire continuous cross-sectional images during the yarn forming process; The infrared friction thermal imaging sensor is installed in the friction area between the yarn and the yarn guide hook before the yarn is wound, and is used to collect the surface temperature distribution of the yarn in real time and calculate the temperature rise rate.

[0024] The high-speed imaging unit for yarn cross-section, in conjunction with a backlight, enables clear acquisition of the yarn outline. The infrared friction thermal imaging sensor integrates ambient temperature compensation and emissivity correction circuits, which can reduce the impact of environmental interference on the detection results.

[0025] A method for controlling the production of ultra-abrasion-resistant yarn, applied to a system, includes the following steps: S1: Set the initial process parameters and input the raw material characteristic parameters through the host industrial control computer; S2: The instantaneous wear resistance coefficient is predicted based on the initial process parameters and raw material characteristic parameters using a dynamic wear resistance coefficient prediction model; S3: Calculate the structural stability index based on the image acquired by the high-speed imaging unit of the yarn cross-section using the yarn structure stability assessment model; S4: The instantaneous wear resistance coefficient prediction value is verified step by step through the three-level data verification module built into the dynamic parameter parser, and the process parameters are dynamically adjusted or the optimization algorithm is triggered to recalculate based on the verification results during the verification process. S5: Convert the process parameter combination corresponding to the final determined instantaneous wear resistance coefficient prediction value into motion control instructions, and send them to the motor drive module via the lower-level PLC control module to control the operation of the roller servo motor; S6: After a preset time interval, repeat steps two through five to achieve closed-loop iterative control.

[0026] Each step is executed at a fixed time frequency. After the instruction is issued, the lower-level PLC control module will adjust the output frequency of the motor drive module to achieve precise control of the speed of the roller servo motor.

[0027] This invention provides a method and system for controlling the production of ultra-abrasion-resistant yarn. It has the following beneficial effects: This method and system for controlling the production of ultra-abrasion-resistant yarn achieves closed-loop control of abrasion resistance performance throughout the yarn production process by constructing a dual-model collaborative architecture of a dynamic abrasion resistance coefficient prediction model and a yarn structure stability evaluation model, combined with a three-level data verification mechanism built into the dynamic parameter parser. During operation, the system employs a dynamic abrasion resistance coefficient prediction model that calculates the instantaneous abrasion resistance coefficient prediction based on process parameters, real-time state parameters, and raw material characteristic parameters. A yarn structure stability assessment model independently evaluates the yarn structure uniformity based on images acquired by a high-speed imaging unit of the yarn cross-section. The difference between the two outputs is cross-validated by a first-level verification unit, effectively avoiding prediction biases that may exist with a single model. A second-level verification unit searches the historical best process library for similar process combinations and their actual abrasion resistance indices, screening the instantaneous abrasion resistance coefficient prediction for historical data matching. When there are differences between the raw material characteristic parameters and historical data, the weighting coefficients of the prediction model are automatically adjusted to better reflect the current raw material characteristics. A third-level verification unit introduces an infrared friction thermal imaging sensor to monitor the temperature rise rate of the yarn friction area in real time. The thermal imaging feedback abrasion resistance coefficient is compared with the instantaneous abrasion resistance coefficient prediction. When the deviation exceeds a preset range, a particle swarm optimization module is triggered to solve for new process parameter combinations, achieving dynamic adjustment of the production process. Through the organic combination of the above three-level verification and iterative optimization, the system can continuously control the uniformity of the yarn cross-sectional structure, the stability of the twist, and the fluctuation of the tension within the target range, thereby ensuring the stable output of the yarn's abrasion resistance performance.

[0028] This method and system for controlling the production of ultra-abrasion-resistant yarn uses a self-learning module to associate and store the final combination of process parameters determined after each round of iterative optimization with their corresponding instantaneous abrasion resistance coefficient prediction values ​​in a historical optimal process library. After obtaining the offline measured abrasion resistance index, the prediction model is incrementally retrained, allowing the accuracy of the dynamic abrasion resistance coefficient prediction model to gradually improve with the accumulation of production data. Within a five-second iteration cycle, the system completes a closed-loop process from data acquisition, model calculation, three-level verification to command issuance, achieving real-time and precise control of the roller servo motor speed. Compared with traditional production methods relying on offline detection and manual adjustments, this system not only improves the consistency of yarn product quality but also significantly shortens the debugging cycle of process parameters for new varieties, providing a reproducible and traceable technical path for the intelligent production of high-performance yarns. Attached Figure Description

[0029] Figure 1 This is a flowchart illustrating a method for controlling the production of ultra-abrasion-resistant yarn according to the present invention. Figure 2 This is a data flow diagram between modules of a method and system for controlling the production of ultra-abrasion-resistant yarn according to the present invention. Detailed Implementation

[0030] 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.

[0031] Please see Figure 1 and Figure 2 The present invention provides a technical solution: a method and system for controlling the production of ultra-wear-resistant yarn, which includes an upper industrial control computer, a lower PLC control module, a motor drive module, a high-speed imaging unit for yarn cross-section, an infrared friction thermal imaging sensor, and a dynamic parameter analyzer. As the decision-making core of the system, the host industrial control computer establishes bidirectional data communication connections with the lower-level PLC control module and the dynamic parameter parser. The host industrial control computer is pre-configured with a dynamic wear resistance coefficient prediction model and a yarn structure stability evaluation model. These two models are embedded in the processor of the host industrial control computer in the form of software program modules. The lower-level PLC control module serves as the field control unit. Its signal input terminals are connected to the data output terminals of the high-speed yarn cross-section imaging unit and the infrared friction thermal imaging sensor, respectively, to receive cross-sectional image data and friction area temperature data during the yarn forming process. The control signal output terminal of the lower-level PLC control module is connected to the input terminal of the motor drive module, and the output terminal of the motor drive module is connected to the control terminal of the roller servo motor to drive the roller servo motor to run at the set speed. The dynamic parameter parser is an independent logic processing unit that integrates data verification logic and optimization algorithms. Its first data interface is connected to the data output of the dynamic abrasion coefficient prediction model, and its second data interface is connected to the data output of the yarn structure stability assessment model. This interface simultaneously receives the instantaneous abrasion coefficient prediction value from the dynamic abrasion coefficient prediction model and the structural stability index from the yarn structure stability assessment model. The third data interface is connected to the data output of the lower-level PLC control module, receiving the roller speed feedback signal, yarn cross-sectional ellipticity calculation value, and real-time temperature rise rate data uploaded by the lower-level PLC control module. The fourth data interface is connected to the instruction input of the lower-level PLC control module, converting the process parameter combination generated after three levels of data verification and iterative optimization into digital control instructions and sending them to the lower-level PLC control module. The lower-level PLC control module adjusts the speed of the roller servo motor through the motor drive module according to the received digital control instructions, thereby achieving closed-loop control of the draft ratio, twist coefficient, and core yarn tension during yarn production. The high-speed imaging unit for yarn cross-section uses a high-speed industrial camera with a frame rate of no less than 1,000 frames per second. Its lens is aimed at the yarn channel in the yarn forming area to continuously acquire cross-sectional contour images of the yarn in motion and transmit the image data to the lower PLC control module in real time. The infrared friction thermal imaging sensor uses a non-contact infrared temperature probe. Its detection area is aligned with the friction area where the yarn contacts the guide hook before winding. It is used to collect the temperature distribution data of the friction area in real time and calculate the temperature rise rate per unit time. The temperature rise rate data is uploaded to the dynamic parameter analyzer via the lower-level PLC control module. The dynamic abrasion resistance coefficient prediction model adopts a three-layer BP neural network architecture. The input layer contains nine input nodes, corresponding to the draft ratio, twist coefficient, core yarn tension, overfeed ratio, difference in speed between the middle and back rollers, fluctuation rate of the speed of the hollow spindle, ellipticity of the yarn cross section, dispersion of fiber length, and coefficient of variation of breaking strength, respectively. The hidden layer contains ten nodes and uses the ReLU activation function. The output layer is a single node that outputs the instantaneous abrasion resistance coefficient prediction value. The value of the instantaneous abrasion resistance coefficient prediction value ranges from 0 to 100. The higher the value, the better the abrasion resistance performance of the yarn. In each iteration, the dynamic abrasion resistance coefficient prediction model obtains the latest process parameters and real-time state parameters from the dynamic parameter parser and completes the forward propagation calculation based on the preset model weight coefficients. The yarn structure stability assessment model employs a support vector regression algorithm. Its input parameters are three feature parameters extracted after image processing from fifty consecutive frames of yarn cross-section images acquired by the high-speed imaging unit: yarn diameter variation coefficient, fiber arrangement entropy, and hairiness index dynamic change rate. The yarn diameter variation coefficient is calculated by the ratio of the standard deviation to the mean of the yarn diameter in the fifty consecutive frames of images. The fiber arrangement entropy is calculated based on the texture features of the image gray-level co-occurrence matrix. The hairiness index dynamic change rate is calculated by dividing the difference between the maximum and minimum hair counts in the fifty consecutive frames of images by the average value. The output of the yarn structure stability assessment model is a structural stability index, which also ranges from 0 to 100. A higher value indicates better uniformity of the yarn's internal structure. The yarn structure stability assessment model performs a calculation every two seconds and sends the result to the dynamic parameter parser. The dynamic parameter parser's built-in three-level data verification module verifies the instantaneous abrasion coefficient prediction value step by step according to a preset logical order. The first-level verification unit calculates the absolute value of the difference between the instantaneous abrasion coefficient prediction value and the structural stability index, and compares the absolute value of the difference with the first preset threshold of eight. When the absolute value of the difference is less than or equal to eight, the first-level verification unit transmits the instantaneous abrasion coefficient prediction value to the second-level verification unit. When the absolute value of the difference is greater than eight, the first-level verification unit triggers a rollback mechanism and sends a parameter correction command to the dynamic abrasion coefficient prediction model. The dynamic abrasion coefficient prediction model increases or decreases the current set values ​​of twist coefficient and core yarn tension by a preset step size according to the parameter correction command, and then recalculates the instantaneous abrasion coefficient prediction value. The recalculated instantaneous abrasion coefficient prediction value is compared with the structural stability index again. This rollback process is executed a maximum of three times. If the absolute value of the difference is still greater than eight after three times, the dynamic parameter parser sends an alarm signal to the host industrial control computer and switches to manual adjustment mode. The second-level verification unit establishes a data connection with the historical optimal process library built into the upper-level industrial control computer. The historical optimal process library stores one thousand sets of process parameter combinations verified by laboratory wear resistance tests and their corresponding actual wear resistance indices. The second-level verification unit receives the instantaneous wear resistance coefficient prediction value verified by the first-level verification unit, and searches the historical optimal process library for similar process combinations with an Euclidean distance of less than or equal to five with the current process parameter combination. It reads the actual wear resistance index corresponding to these similar process combinations and calculates the average deviation between the instantaneous wear resistance coefficient prediction value and these actual wear resistance indices. When the average deviation is less than or equal to 10%, the second-level verification unit transmits the instantaneous wear resistance coefficient prediction value to the third-level verification unit. When the average deviation is greater than 10%, the second-level verification unit extracts the raw material characteristic parameters in the similar process combinations whose fiber length dispersion differs from the current fiber length dispersion by more than 20%, and performs linear interpolation correction on the weight coefficients corresponding to fiber length dispersion and breaking strength variation coefficient in the dynamic wear resistance coefficient prediction model according to the preset raw material sensitivity coefficient table. The corrected weight coefficients are sent to the dynamic wear resistance coefficient prediction model, triggering the dynamic wear resistance coefficient prediction model to recalculate the instantaneous wear resistance coefficient prediction value. The third-level verification unit receives the temperature rise rate data uploaded by the infrared friction thermal imaging sensor and calculates the thermal imaging feedback wear resistance coefficient according to the preset conversion formula between temperature rise rate and wear resistance coefficient. The conversion formula is that the thermal imaging feedback wear resistance coefficient equals 100 minus 25 multiplied by the temperature rise rate, where the temperature rise rate is in degrees Celsius per second. The third-level verification unit receives the instantaneous wear resistance coefficient prediction value after verification by the second-level verification unit and calculates the absolute value of the difference between the instantaneous wear resistance coefficient prediction value and the thermal imaging feedback wear resistance coefficient. The absolute value of the difference is compared with the fifth preset threshold value. When the absolute value of the difference is less than or equal to five, the third-level verification unit outputs the instantaneous wear resistance coefficient prediction value as the final control target value to the instruction generation module of the dynamic parameter resolver. When the absolute value of the difference is greater than five, the third-level verification unit sends the thermal imaging feedback wear resistance coefficient as the target value to the particle swarm optimization module built into the dynamic parameter resolver. After receiving the target value, the particle swarm optimization module takes the abrasion resistance coefficient fed back by thermal imaging as the optimization target and the draft ratio, twist coefficient, core yarn tension and overfeed ratio as optimization variables. It uses the particle swarm optimization algorithm to solve in reverse to obtain a new combination of process parameters. The population size of the particle swarm optimization algorithm is set to fifty, the number of iterations is set to one hundred, and the fitness function is the absolute value of the difference between the instantaneous abrasion resistance coefficient prediction value output by the dynamic abrasion resistance coefficient prediction model under the current combination of process parameters and the target value. After the algorithm iteration is completed, the combination of process parameters that minimizes the fitness function value is output as the optimization result. The dynamic parameter parser converts the optimized new combination of process parameters into digital control instructions, which are then sent to the lower-level PLC control module via the fourth data interface. The lower-level PLC control module adjusts the output frequency of the motor drive module according to the digital control instructions, thereby controlling the roller servo motor to run at the new speed. After completing the entire process from data acquisition to instruction issuance, the system waits for five seconds before restarting the next round of data acquisition and model calculation, forming a continuous closed-loop iterative control. At the same time, the dynamic parameter parser sends the process parameter combination finally determined in each iteration and its corresponding instantaneous wear resistance coefficient prediction value to the self-learning module of the host industrial control computer. The self-learning module associates and stores these data in the historical optimal process library for data matching and weight correction of the second-level verification unit in the subsequent production process.

[0032] The dynamic abrasion resistance coefficient prediction model is specifically embedded in the processor of the host industrial control computer in the form of a software program module. The prediction model is constructed using a three-layer BP neural network architecture, which is used to map the parameters of multiple dimensions that affect abrasion resistance in the yarn production process into a quantitative index that can be calculated in real time. The input layer of the dynamic abrasion resistance coefficient prediction model contains nine input nodes, which correspond to the draft ratio, twist coefficient, core yarn tension, overfeed ratio, difference in speed between middle and back rollers, fluctuation rate of hollow spindle speed, ellipticity of yarn cross section, dispersion of fiber length, and coefficient of variation of breaking strength, respectively. Among them, the draft ratio, twist coefficient, core yarn tension, and overfeed ratio are process parameters. These four process parameters are read by the host computer according to the initial process formula set by the operator and input into the dynamic abrasion resistance coefficient prediction model. The speed difference between the middle and rear rollers, the speed fluctuation rate of the hollow spindle, and the ellipticity of the yarn cross-section are real-time status parameters. The speed difference between the middle and rear rollers is calculated by the lower-level PLC control module based on the pulse signals fed back by the encoders installed on the shaft ends of the middle and rear rollers. The actual speed of the middle and rear rollers per unit time is calculated by a high-speed counter, and then the absolute value of the speed difference is calculated. The speed fluctuation rate of the hollow spindle is calculated by the lower-level PLC control module based on the pulse signals fed back by the encoders installed on the shaft ends of the hollow spindles. The speed values ​​within a preset time period are continuously collected, and the ratio of the standard deviation to the average value of the speed within that time period is calculated. The ellipticity of the yarn cross-section is obtained by the high-speed imaging unit of the yarn cross-section acquiring continuous cross-sectional images of the yarn forming area. The lower-level PLC control module performs edge extraction and ellipse fitting processing on the images, calculates the ratio of the major axis to the minor axis of the fitted ellipse, and uploads this ratio as the ellipticity of the yarn cross-section to the dynamic parameter resolver in real time. The dynamic parameter resolver then forwards the ellipticity of the yarn cross-section to the dynamic abrasion resistance coefficient prediction model. Fiber length dispersion and tensile strength coefficient of variation are raw material characteristic parameters. These two raw material characteristic parameters are obtained by the host industrial control computer based on the raw material batch information input by the operator, which retrieves the measured statistical values ​​of the fiber in the preset raw material database. The preset raw material database stores the length distribution data and tensile strength test data of fibers from different suppliers and different batches. Fiber length dispersion is represented by the coefficient of variation of fiber length distribution, and tensile strength coefficient of variation is represented by the coefficient of variation of single fiber tensile strength test value. The hidden layer of the dynamic wear resistance coefficient prediction model contains ten nodes. The activation function of the hidden layer nodes adopts the linear rectified function, which is used to perform nonlinear transformation and feature extraction on the parameters passed from the input layer. The output layer of the dynamic abrasion resistance coefficient prediction model contains one node. The output layer uses a linear activation function to map the features extracted from the hidden layer to an instantaneous abrasion resistance coefficient prediction value. The numerical range of the instantaneous abrasion resistance coefficient prediction value is set from 0 to 100. The higher the value, the better the yarn abrasion resistance performance predicted by the model. In each iteration, the dynamic abrasion coefficient prediction model obtains the latest process parameters and real-time state parameters from the dynamic parameter resolver and performs forward propagation calculations based on the current model weight coefficients. The model weight coefficients are determined in advance through offline training. The offline training process uses 5,000 sets of experimental data as training samples. Each set of training samples includes the draft ratio, twist coefficient, core yarn tension, overfeed ratio, difference in speed between middle and back rollers, fluctuation rate of hollow spindle speed, ellipticity of yarn cross section, dispersion of fiber length, coefficient of variation of breaking strength, and the corresponding actual abrasion test value. The training objective is to minimize the mean square error between the instantaneous abrasion coefficient prediction value output by the model and the actual abrasion test value. After the training converges, a set of fixed weight coefficients is obtained and fixed in the dynamic abrasion coefficient prediction model. During real-time production, the dynamic wear resistance coefficient prediction model performs a forward propagation calculation every 0.5 seconds, converting the input parameters at the current moment into an instantaneous wear resistance coefficient prediction value, and sending the prediction value to the dynamic parameter parser for subsequent verification and processing by the three-level data verification module.

[0033] The yarn structure stability assessment model is specifically embedded in the processor of the host industrial control computer in the form of a software program module. The assessment model is constructed using the support vector regression algorithm, which is used to convert the continuous multi-frame yarn cross-section images acquired by the high-speed imaging unit of the yarn cross-section into a quantitative index of the uniformity of the internal structure of the yarn. The high-speed imaging unit for yarn cross-section uses a high-speed industrial camera with a frame rate of no less than 1,000 frames per second. The camera lens is installed perpendicular to the yarn channel axis and is used with a backlight to obtain clear images of the yarn outline. The high-speed imaging unit for yarn cross-section acquires a sequence of cross-sectional images of the yarn in motion using a continuous image acquisition method and transmits the image data to the lower-level PLC control module in real time. The lower-level PLC control module has a built-in image preprocessing unit. The image preprocessing unit performs grayscale processing, median filtering for noise reduction, and Canny edge detection on each received frame of image in sequence to extract the outer contour edge point set of the yarn cross-section. Then, the edge point set is fitted to an ellipse using the least squares method to obtain the equivalent diameter and ellipticity parameters of the yarn cross-section. The input parameters of the yarn structure stability assessment model are three feature parameters extracted after image processing from fifty consecutive frames of yarn cross-section images acquired by the high-speed imaging unit of the yarn cross-section: yarn diameter variation coefficient, fiber arrangement entropy, and hairiness index dynamic change rate. The yarn diameter variation coefficient is calculated by the lower-level PLC control module. The specific steps are as follows: For each frame in fifty consecutive frames of images, the equivalent diameter of the yarn in that frame is calculated based on the major axis and minor axis obtained by ellipse fitting. The equivalent diameter is defined as the arithmetic mean of the major axis and minor axis. After obtaining fifty equivalent diameter values, the ratio of the standard deviation to the mean of these fifty values ​​is calculated. The ratio is multiplied by one hundred and then used as the yarn diameter variation coefficient for output. The fiber arrangement entropy is calculated based on the image gray-level co-occurrence matrix. The specific steps are as follows: The lower-level PLC control module extracts the pixel gray-level matrix from the region of interest of the yarn cross-section image, sets the pixel spacing to two pixels and the direction to the horizontal direction, constructs the gray-level co-occurrence matrix, normalizes the gray-level co-occurrence matrix to obtain the probability matrix, calculates the product of the negative logarithm of each element in the probability matrix and the probability, and sums them to obtain the fiber arrangement entropy value. This value reflects the randomness and disorder of the fiber arrangement within the yarn cross-section. The dynamic change rate of the hairiness index is calculated by the lower-level PLC control module. The specific steps are as follows: For each frame in fifty consecutive frames of images, count the number of hairs that extend outward from the main outline of the yarn beyond the preset length threshold to obtain fifty hairiness count values. Calculate the difference between the maximum and minimum values ​​of these fifty hairiness count values, and then divide it by the average value of these fifty hairiness count values. The resulting ratio is output as the dynamic change rate of the hairiness index. The yarn structure stability assessment model performs a calculation every two seconds, that is, after collecting fifty consecutive new images, the model calculation is triggered once. The lower computer PLC control module sends the calculated yarn diameter variation coefficient, fiber arrangement entropy and hairiness index dynamic change rate as input vectors to the yarn structure stability assessment model. The yarn structure stability assessment model adopts the support vector regression algorithm. The model is pre-determined through offline training. The training sample set contains one thousand feature vectors composed of the above three feature parameters and corresponding manual scores of yarn structure stability. The manual scores are comprehensively evaluated by textile experts based on the uniformity, hair distribution and fiber arrangement regularity of the microscopic images of the yarn cross section. The score range is set from 0 to 100. The training process uses the Gaussian radial basis kernel function. The penalty parameter and kernel function parameter are optimized through grid search. After the training converges, the support vectors, weight coefficients and bias terms of the support vector regression model are obtained and fixed in the assessment model. When the yarn structure stability assessment model is running, it will form a feature vector from the three received feature parameters, map it to a high-dimensional feature space through a kernel function, calculate the weighted sum with the support vector and add a bias term to obtain the output structure stability index. The value range of the structure stability index is also 0 to 100. The higher the value, the better the uniformity of the yarn internal structure, that is, the smaller the fluctuation of the yarn diameter, the orderly arrangement of fibers, and the stable number of hairs. After the yarn structure stability assessment model is calculated, the structural stability index is sent to the dynamic parameter parser for the first-level verification unit to use as a comparison benchmark in the subsequent verification process.

[0034] The dynamic parameter parser has a built-in three-level data verification module, which is integrated into the processor of the dynamic parameter parser in the form of a solid logic circuit or an embedded software program. It is used to perform multi-dimensional cross-verification of the instantaneous wear resistance coefficient prediction value output by the dynamic wear resistance coefficient prediction model. The three-level data verification module specifically includes a first-level verification unit, a second-level verification unit, and a third-level verification unit. The three verification units perform verification operations sequentially in a preset serial order. The first-level verification unit establishes communication connections with the data output terminals of the dynamic abrasion coefficient prediction model and the yarn structure stability evaluation model, respectively, to simultaneously receive the instantaneous abrasion coefficient prediction value output by the dynamic abrasion coefficient prediction model every 0.5 seconds and the structural stability index output by the yarn structure stability evaluation model every two seconds. The first-level verification unit is equipped with a first comparator and a first counter. The first comparator is used to calculate the absolute value of the difference between the instantaneous wear resistance coefficient prediction value and the structural stability index received at the same time, and compare the absolute value of the difference with a first preset threshold, which is set to eight. When the absolute value of the difference is less than or equal to eight, the first comparator outputs the first enable signal, marks the instantaneous wear resistance coefficient prediction value as passed at the first level of verification, and sends it to the second level of verification unit through the data forwarding interface of the first level of verification unit. When the absolute value of the difference is greater than eight, the first comparator outputs the first trigger signal to the first counter. The first counter records the trigger event and accumulates the count. At the same time, the first-level verification unit generates a parameter correction instruction, which is sent to the dynamic wear resistance coefficient prediction model through the internal data bus of the dynamic parameter parser. The parameter correction instruction includes a twist coefficient correction amount and a core yarn tension correction amount. The twist coefficient correction amount is an increase of 5% of the current twist coefficient setting value, and the core yarn tension correction amount is a decrease of 5% of the current core yarn tension setting value. After receiving the parameter correction instruction, the dynamic abrasion resistance coefficient prediction model increases and decreases the current setting values ​​of twist coefficient and core yarn tension by the corresponding correction amounts, respectively. Then, based on the corrected process parameter combination, combined with other real-time state parameters and raw material characteristic parameters at the current moment, it re-executes the forward propagation calculation to obtain a new instantaneous abrasion resistance coefficient prediction value, and sends the new instantaneous abrasion resistance coefficient prediction value to the first-level verification unit again. The first-level verification unit recalculates the absolute value of the difference between the new instantaneous abrasion resistance coefficient prediction value and the structural stability index, and compares it with the first preset threshold of eight. If the absolute value of the difference is still greater than eight, the above parameter correction and recalculation process is repeated. Each time the correction is performed, the twist coefficient is increased by five percent based on the current value, and the core yarn tension is decreased by five percent based on the current value, until the absolute value of the difference is less than or equal to eight, or the cumulative triggering number of the first counter reaches three. When the cumulative trigger count of the first counter reaches three times and the absolute value of the difference between the predicted instantaneous abrasion resistance coefficient and the structural stability index obtained from the last recalculation is still greater than eight, the first-level verification unit generates an alarm signal. This alarm signal is sent to the human-machine interface of the upper industrial control computer through the dynamic parameter parser, prompting the operator that there is a persistent large deviation between the dynamic abrasion resistance coefficient prediction model and the yarn structural stability assessment model in the current yarn production process, requiring manual intervention to check the raw material batch or equipment operating status. At the same time, the system pauses the automatic control mode and switches to the manual adjustment mode, waiting for the operator to confirm or reset the process parameters. In manual adjustment mode, operators can directly adjust the set values ​​of draft ratio, twist coefficient, core yarn tension and overfeed ratio through the host industrial control computer, and the system no longer performs automatic iterative optimization; When the first-level verification unit successfully controls the absolute value of the difference to be less than or equal to eight, the first-level verification unit takes the predicted value of the instantaneous wear resistance coefficient obtained from the last calculation as the first-level verification result and sends it to the second-level verification unit for subsequent processing through the data forwarding interface.

[0035] The second-level verification unit is integrated into the processor of the dynamic parameter parser in the form of a fixed logic circuit or embedded software program. This second-level verification unit establishes a bidirectional data communication connection with the historical best process library built into the host industrial control computer. The historical best process library is a structured database deployed on the solid-state drive of the host industrial control computer. This database stores one thousand sets of process parameter combinations verified by laboratory abrasion resistance tests and their corresponding actual abrasion resistance indices. Each set of data records includes at least the draft ratio, twist coefficient, core yarn tension, overfeed ratio, difference in speed between middle and back rollers, fluctuation rate of hollow spindle speed, ellipticity of yarn cross section, dispersion of fiber length, coefficient of variation of breaking strength, and the corresponding actual abrasion resistance index. The actual abrasion resistance index is obtained by measuring the number of frictions until the yarn sample breaks under constant pressure using an offline Martindale abrasion tester and then normalizing the result. The value range is set to 0 to 100. The second-level verification unit receives the instantaneous abrasion resistance coefficient prediction value sent after verification by the first-level verification unit, and at the same time reads the process parameter combination at the current moment from the internal data bus of the dynamic parameter parser. The process parameter combination at the current moment includes at least the values ​​of four dimensions: draft ratio, twist coefficient, core yarn tension, and overfeed ratio. The second-level verification unit is equipped with an Euclidean distance calculator, a similar sample extractor, an average deviation calculator, and a weight corrector. The Euclidean distance calculator first normalizes the four dimensions of the process parameter combination at the current moment. The normalization formula is to subtract the minimum value of each dimension in the historical best process library from the value of each dimension and then divide by the range of that dimension to obtain four dimensionless numbers between 0 and 1, which constitute the current process feature vector. The Euclidean distance calculator then traverses each set of data records in the historical best process library and performs the same normalization process on the draft ratio, twist coefficient, core yarn tension and overfeed ratio in each set of records to obtain historical feature vectors. The Euclidean distance between the current process feature vector and each historical feature vector is then calculated. The similar sample extractor presets a second preset threshold of five. The similar sample extractor selects all data records with an Euclidean distance of less than or equal to five from the historical best process library as similar process combinations, and extracts the actual wear resistance index corresponding to these similar process combinations and sends it to the average deviation calculator. After receiving the actual wear resistance index corresponding to similar process combinations, the average deviation calculator calculates the arithmetic mean of these actual wear resistance indices, then calculates the absolute value of the difference between the instantaneous wear resistance coefficient prediction value and the arithmetic mean, and divides the absolute value by the arithmetic mean to obtain the relative deviation value. The average deviation calculator has a preset third threshold of 10%. When the relative deviation value is less than or equal to 10%, the average deviation calculator determines that the current instantaneous wear resistance coefficient prediction value conforms to the historical pattern, generates a second-level verification pass signal, and sends the instantaneous wear resistance coefficient prediction value to the third-level verification unit through the data forwarding interface. When the relative deviation value is greater than 10%, the average deviation calculator generates a deviation over-limit signal and triggers the weight corrector to start. After receiving the deviation exceeding the limit signal, the weight corrector extracts the fiber length dispersion and breaking strength coefficient of variation corresponding to all similar process combinations from the historical optimal process library, calculates the average value of these two raw material characteristic parameters in the similar process combinations, and reads the fiber length dispersion and breaking strength coefficient of variation at the current moment from the dynamic parameter parser. It calculates the absolute value of the difference between the current fiber length dispersion and the average fiber length dispersion of similar process combinations, and the absolute value of the difference between the current breaking strength coefficient of variation and the average breaking strength coefficient of variation of similar process combinations. The absolute values ​​of these two differences are divided by the average value of the corresponding characteristics of the similar process combinations to obtain two relative difference values. The weight corrector internally presets a fourth preset threshold of 20%. When the relative difference value of fiber length dispersion is greater than 20% or the relative difference value of breaking strength coefficient of variation is greater than 20%, the weight corrector determines that there is a significant difference between the current raw material characteristics and the raw material characteristics of historical similar process combinations. The weight corrector has a pre-set raw material sensitivity coefficient table, which is stored in the form of a lookup table. The table records the correction coefficient of the fiber length dispersion weight coefficient in the dynamic abrasion resistance coefficient prediction model corresponding to each percentage change in fiber length dispersion, and the correction coefficient of the fracture strength variation coefficient weight coefficient corresponding to each percentage change in fracture strength variation coefficient. The weight corrector queries the raw material sensitivity coefficient table to obtain the corresponding correction coefficient based on the two relative difference values ​​calculated. It then multiplies the correction coefficient by the original weight coefficients corresponding to the fiber length dispersion and the coefficient of variation of breaking strength in the current dynamic abrasion resistance coefficient prediction model to obtain the corrected weight coefficient. The weight corrector sends the corrected weight coefficients to the dynamic wear resistance coefficient prediction model through the internal data bus of the dynamic parameter resolver, and triggers the dynamic wear resistance coefficient prediction model to re-execute the forward propagation calculation using the corrected weight coefficients while keeping other input parameters unchanged at the current moment, so as to obtain a new instantaneous wear resistance coefficient prediction value. The dynamic wear resistance coefficient prediction model recalculates the instantaneous wear resistance coefficient prediction value and sends it back to the first-level verification unit. The first-level verification unit performs a first-level verification on the new instantaneous wear resistance coefficient prediction value. After the verification is passed, it is sent back to the second-level verification unit. The second-level verification unit repeats the above Euclidean distance calculation, similar sample extraction and average deviation calculation process until the relative deviation value is less than or equal to 10% or the number of iterations reaches the preset upper limit of two. If the relative deviation is still greater than 10% after two iterations, the second-level verification unit generates a second-level verification failure signal and sends the signal to the alarm module of the dynamic parameter parser. The dynamic parameter parser then pushes a prompt message to the human-machine interface of the upper-level industrial control computer, informing the operator that the current process parameter combination has a consistently low matching degree with historical data, and suggesting that the raw material batch be re-evaluated or the process parameters be manually optimized.

[0036] The third-level verification unit is integrated into the processor of the dynamic parameter parser in the form of a solidified logic circuit or embedded software program. This third-level verification unit establishes a real-time communication connection with the data output terminal of the infrared friction thermal imaging sensor. The infrared friction thermal imaging sensor is a non-contact infrared temperature probe. Its installation position is aligned with the friction area where the yarn contacts the guide hook before winding. The optical axis of the probe is perpendicular to the yarn running direction. The temperature measurement field diameter is set to three millimeters to ensure coverage of the entire friction contact point. The infrared friction thermal imaging sensor continuously collects the surface temperature data of the friction area at a sampling frequency of ten frames per second, and uploads the temperature data to the dynamic parameter analyzer in real time through the analog input port of the lower-level PLC control module. After receiving the temperature data, the lower-level PLC control module is equipped with a temperature rise rate calculator. This calculator calculates the temperature rise rate every five consecutive frames of temperature data. The specific calculation method is to extract the highest and lowest temperatures from the five consecutive frames of temperature values, calculate the difference between the highest and lowest temperatures, and then divide it by the time span corresponding to these five frames of data to obtain the temperature rise rate per unit time. The unit of the temperature rise rate is degrees Celsius per second. The lower-level PLC control module sends the calculated temperature rise rate to the third-level verification unit through the data bus. The third-level calibration unit has a pre-set conversion formula between the temperature rise rate and the abrasion resistance coefficient. The conversion formula is that the thermal imaging feedback abrasion resistance coefficient is equal to 100 minus 25 multiplied by the temperature rise rate. The coefficient 25 was determined through a large number of calibration experiments in the early stage. The calibration experiment process is to collect the temperature rise rate of the yarn friction area under different process conditions, and at the same time, to perform offline Martindale abrasion resistance test on the yarn sample to obtain the actual abrasion resistance index. The temperature rise rate and the actual abrasion resistance index are linearly regressed to obtain the regression coefficient 25. The third-level verification unit receives the instantaneous wear resistance coefficient prediction value sent after verification by the second-level verification unit, and at the same time receives the current temperature rise rate uploaded by the lower-level PLC control module. It calculates the thermal imaging feedback wear resistance coefficient according to the conversion formula. The value range of the thermal imaging feedback wear resistance coefficient is limited to 0 to 100. When the calculated value is less than 0, it is taken as 0, and when it is greater than 100, it is taken as 100. The third-level verification unit is equipped with a second comparator and a second counter. The second comparator is used to calculate the absolute value of the difference between the instantaneous wear resistance coefficient prediction value and the thermal imaging feedback wear resistance coefficient, and compare the absolute value of the difference with a fifth preset threshold, which is set to five. When the absolute value of the difference is less than or equal to five, the second comparator outputs a third enable signal, marking the instantaneous wear resistance coefficient prediction value as passed at the third level of verification, and sending it to the instruction generation module as the final control target value through the data output interface of the dynamic parameter parser. When the absolute value of the difference is greater than five, the second comparator outputs a second trigger signal. This trigger signal is sent to the second counter for event recording on the one hand, and encapsulates the thermal imaging feedback wear resistance coefficient as the target value into a data packet on the other hand, and sends it to the particle swarm optimization module through the internal data bus of the dynamic parameter resolver. The second counter records the number of events where the absolute value of the difference is greater than five. When the absolute value of the difference is greater than five three times in a row, the second counter generates a verification failure signal and sends it to the alarm module of the dynamic parameter parser. The dynamic parameter parser then pushes a prompt message to the host industrial control computer, informing the operator that there is a continuous deviation between the current thermal imaging feedback and the model prediction, and suggesting that the cleanliness of the infrared sensor lens be checked or the sensor be calibrated. After receiving the abrasion resistance coefficient as the target value from the thermal imaging feedback, the particle swarm optimization module immediately starts the reverse solution process. With this target value as the optimization guide, and with the draft ratio, twist coefficient, core yarn tension and overfeed ratio as optimization variables, it recalculates a new set of process parameters and sends the new set of process parameters to the lower-level PLC control module for execution. During the operation of the particle swarm optimization module, the third-level verification unit remains in standby mode, waiting for the next round of instantaneous wear resistance coefficient prediction value, which is recalculated by the dynamic wear resistance coefficient prediction model and re-entered after passing the first and second-level verifications, as well as the temperature rise rate uploaded in real time by the infrared friction thermal imaging sensor, in preparation for a new round of third-level verification comparison.

[0037] The dynamic parameter parser has a built-in particle swarm optimization module, which is embedded in the processor of the dynamic parameter parser as an embedded software algorithm and establishes a one-way data communication connection with the data output of the third-level verification unit. When the absolute value of the difference between the instantaneous wear resistance coefficient prediction value and the thermal imaging feedback wear resistance coefficient is greater than the fifth preset threshold, the third-level verification unit encapsulates the thermal imaging feedback wear resistance coefficient as the target value into an optimization trigger data packet, and sends it to the input end of the particle swarm optimization module through the internal data bus of the dynamic parameter parser. The particle swarm optimization module is activated immediately after receiving the optimization trigger data packet. It parses the specific value of the thermal imaging feedback abrasion coefficient from the data packet as the optimization target. At the same time, it reads the current set values ​​of the four process parameters—draft ratio, twist coefficient, core yarn tension, and overfeed ratio—from the internal data bus of the dynamic parameter resolver as the benchmark for particle swarm initialization. The particle swarm optimization module has a preset population size of fifty particles and an upper limit of one hundred iterations. The position vector of each particle consists of four dimensions, corresponding to the draft ratio, twist coefficient, core yarn tension, and overfeed ratio. The velocity vector of each particle is initialized to zero. The position vector is initialized by randomly generating fifty different initial positions within a range of plus or minus ten percent of the current process parameter setting value. In each iteration, the particle swarm optimization module sends the position vector of each particle, i.e., a set of candidate process parameters, to the dynamic wear resistance coefficient prediction model. The dynamic wear resistance coefficient prediction model performs forward propagation calculations using the set of candidate process parameters combined with the real-time state parameters and raw material characteristic parameters at the current moment, and outputs a corresponding instantaneous wear resistance coefficient prediction value. The particle swarm optimization module receives the instantaneous wear resistance coefficient prediction value and calculates the absolute value of the difference between the instantaneous wear resistance coefficient prediction value and the optimization target, i.e., the wear resistance coefficient fed back by thermal imaging, as the fitness function value of the particle. The particle swarm optimization module is equipped with a particle updater. The particle updater records the individual's historical best position and global best position based on the fitness function value of each particle. It uses the standard particle swarm velocity update formula and position update formula to iteratively update the velocity and position of all particles. The velocity update formula includes an inertia weight coefficient set to 0.729, and both the individual learning factor and the social learning factor are set to 1.49445. After the position update, the positions that exceed the preset process parameter range are truncated. Each time the particle swarm optimization module completes an iteration, it checks whether the current global optimal fitness function value is less than the preset convergence threshold of 0.5, or whether the number of iterations has reached the upper limit of one hundred. If either condition is met, the iteration is terminated, and the values ​​of the four process parameters corresponding to the current global optimal position are output as the optimization result. The particle swarm optimization module sends the optimized new combination of process parameters, namely the draft ratio, twist coefficient, core yarn tension, and overfeed ratio, to the instruction generation module through the data output interface of the dynamic parameter parser. The instruction generation module converts this combination of process parameters into digital control instructions according to a predetermined communication protocol and sends them to the instruction buffer of the lower-level PLC control module through the fourth data interface. After receiving the new digital control instructions, the lower-level PLC control module parses the set values ​​of each process parameter and outputs the corresponding pulse frequency signal and direction signal to the motor drive module through its high-speed data output port. The motor drive module adjusts the speed of the roller servo motor according to the received signals, so that the actual running draft ratio, twist coefficient, core yarn tension and overfeed ratio are adjusted to the new set values. After the lower-level PLC control module completes the execution of this instruction, its internal timer starts counting down. The preset timing duration is five seconds. After the timing ends, the lower-level PLC control module automatically sends a restart signal to the dynamic parameter parser. Upon receiving the restart signal, the dynamic parameter parser re-collects the latest real-time status parameters from the yarn cross-section high-speed imaging unit, the infrared friction thermal imaging sensor, and the lower-level PLC control module. It then triggers the dynamic abrasion coefficient prediction model to re-execute the forward propagation calculation with the new combination of process parameters and the latest real-time status parameters as input, starting a new round of instantaneous abrasion coefficient prediction and three-level data verification process.

[0038] The host industrial control computer has a built-in historical optimal process library and a self-learning module. The historical optimal process library is a relational database deployed on the solid-state drive of the host industrial control computer. This database stores process parameter combinations verified by laboratory abrasion resistance tests and their corresponding actual abrasion resistance indices. Each set of data records includes at least the draft ratio, twist coefficient, core yarn tension, overfeed ratio, difference in speed between middle and back rollers, fluctuation rate of hollow spindle speed, ellipticity of yarn cross section, dispersion of fiber length, coefficient of variation of breaking strength, and the corresponding actual abrasion resistance index. The actual abrasion resistance index is obtained by measuring the number of frictions until the yarn sample breaks under constant pressure using an offline Martindale abrasion tester and then normalizing the results. The value range is set to 0 to 100. The database also adds timestamp tags and raw material batch tags to each set of data records for multi-dimensional filtering during subsequent retrieval. The initial data for the historical best process library comes from previous process experiments and production accumulation. The database size is set to one thousand sets of data records. When the number of data records exceeds one thousand sets, the earliest data record is overwritten using the first-in-first-out principle to maintain the timeliness of the database and the stability of the storage space. The self-learning module is integrated into the processor of the host industrial computer in the form of an embedded software program. The self-learning module establishes a bidirectional data communication connection with the data output terminal of the dynamic parameter parser and the data read / write interface of the historical best process library. After the system completes a full closed-loop iterative control cycle, the dynamic parameter parser encapsulates the final combination of process parameters determined in this iteration and the corresponding instantaneous abrasion coefficient prediction value into a data packet and sends it to the self-learning module. The data packet contains at least the final set values ​​of four process parameters: draft ratio, twist coefficient, core yarn tension, and overfeed ratio, as well as the final output instantaneous abrasion coefficient prediction value after three levels of data verification. After receiving the data packet, the self-learning module first reads the real-time status parameters recorded during this iteration from the dynamic parameter parser. The real-time status parameters include the difference in speed between the middle and back rollers, the fluctuation rate of the speed of the hollow spindle, and the average value of the ellipticity of the yarn cross section. At the same time, it reads the raw material characteristic parameters corresponding to the current production batch from the dynamic parameter parser. The raw material characteristic parameters include the fiber length dispersion and the coefficient of variation of breaking strength. The self-learning module combines all the above parameters into a complete data record and adds the current system time as a timestamp label and the current raw material batch number as a batch label to the data record. Before writing data records into the historical best process library, the self-learning module first performs a data deduplication check. That is, it uses the four process parameters in the current data record—draft ratio, twist coefficient, core yarn tension, and overfeed ratio—as search conditions to check whether there are existing records in the historical best process library with an Euclidean distance of less than one. If they exist, they are determined to be duplicate data, and the writing operation is abandoned. If there is no duplicate data, the new data record is written into the historical best process library. When a batch of yarn is produced, the operator takes offline samples of the yarn and sends them to the laboratory for Martindale abrasion resistance testing to obtain the actual abrasion resistance index of the yarn. The operator then inputs the actual abrasion resistance index through the human-machine interface of the upper industrial control computer and associates it with the corresponding production time period and raw material batch label. After receiving the actual wear resistance index input by the operator, the self-learning module searches for the corresponding data record in the historical best process library using the production time period and raw material batch label as search conditions, and writes the actual wear resistance index into the actual wear resistance index field of the data record, thus completing the association mapping from online predicted value to offline measured value. As production batches increase, the amount of actual wear resistance index data accumulated in the historical best process library gradually increases. The self-learning module periodically triggers the model retraining process, which involves extracting all data records with actual wear resistance index labels from the historical best process library as training sample sets. The process parameters, real-time status parameters, and raw material characteristic parameters in these data records are used as inputs, and the actual wear resistance index is used as the output to incrementally retrain the weight coefficients of the dynamic wear resistance coefficient prediction model. After training, the updated weight coefficients are sent to the dynamic wear resistance coefficient prediction model to update the model parameters, so that the prediction accuracy of the dynamic wear resistance coefficient prediction model gradually improves with the accumulation of production data. After each model parameter update, the self-learning module automatically records the difference in weight coefficients before and after the update and generates a log file, which is stored on the hard drive of the host industrial control computer for subsequent quality traceability and process analysis.

[0039] The high-speed imaging unit for yarn cross-section specifically uses a CMOS industrial camera with a frame rate of no less than 1,000 frames per second. This industrial camera is equipped with a high-resolution fixed-focus lens and a coaxial parallel light source. The camera is fixed in front of the yarn channel between the front roller outlet and the yarn guide hook of the spinning machine by a custom mounting bracket. The optical axis of the camera is perpendicular to the yarn running axis. The distance between the front end of the lens and the yarn is set to 50 mm to ensure image clarity. The coaxial parallel light source is installed on the opposite side of the camera, and the direction of the light is at a 90-degree angle to the optical axis of the camera to form dark field illumination to enhance the contrast between the yarn edge and the background. The high-speed imaging unit for yarn cross-section operates in continuous trigger mode, acquiring one frame of image every millisecond. The image resolution is set to 640 pixels by 480 pixels. Each frame of image covers a field of view with a width of 5 mm and a height of 4 mm, which is sufficient to include the complete cross-sectional outline of a single yarn and the fuzzy areas on both sides. The yarn cross-section high-speed imaging unit is connected to the high-speed image input port of the lower-level PLC control module through the CameraLink interface, and transmits the acquired raw image data to the lower-level PLC control module in real time in the form of digital signals. The lower-level PLC control module integrates a field-programmable gate array chip as an image coprocessor. The coprocessor performs real-time preprocessing on each received image frame. The preprocessing process includes first performing median filtering on the original image to remove salt-and-pepper noise, then using the Sobel operator to perform edge detection to extract the yarn contour, then connecting the broken edge points through morphological closing operation, and finally outputting a binarized yarn contour image for subsequent feature parameter extraction. The infrared friction thermal imaging sensor specifically adopts a non-contact infrared thermopile temperature probe. The temperature measurement range of this probe is set from 0 degrees Celsius to 200 degrees Celsius, the spectral response range is from 8 micrometers to 14 micrometers, the response time is less than 50 milliseconds, and the temperature measurement accuracy is ±1.5 degrees Celsius. The infrared friction thermal imaging sensor is mounted on the last set of yarn guide hooks before the yarn is wound using an adjustable angle universal bracket. The sensor detection window is aligned with the contact point between the yarn and the ceramic surface of the yarn guide hook. The detection distance is set to 15 mm to ensure that the field of view completely covers the friction area, and the field of view diameter is set to 3 mm to eliminate interference from the surrounding environmental radiation. The infrared friction thermal imaging sensor integrates an ambient temperature compensation circuit and an emissivity correction circuit. The preset emissivity value for the yarn material is 0.95. The sensor continuously outputs the instantaneous temperature value of the friction area at a sampling frequency of 10 frames per second. The output signal type is a 4 to 20 mA analog current signal, which is transmitted to the analog input module of the lower-level PLC control module through a shielded cable. The analog input module of the lower-level PLC control module converts the received 4 to 20 mA current signal into a digital value of 0 to 4,000, and then linearly maps it to the actual temperature value according to the preset temperature measurement range. The unit is degrees Celsius, and the temperature value is retained to one decimal place. The lower-level PLC control module is equipped with a temperature rise rate calculation program. This program triggers a calculation every 0.5 seconds after collecting five consecutive frames of temperature values. During the calculation, the maximum and minimum values ​​of these five frames of temperature values ​​are extracted. The maximum value is subtracted from the minimum value to obtain the temperature difference value. Then, the difference is divided by the time span of 0.5 seconds to obtain the temperature rise rate per unit time. The temperature rise rate is in degrees Celsius per second. The calculation results are stored in floating-point format and prepared for uploading. The lower-level PLC control module uses its data bus to package the image feature parameters preprocessed by the high-speed imaging unit of the yarn cross-section and the temperature rise rate data calculated by the infrared friction thermal imaging sensor, and uploads them to the dynamic parameter parser at regular intervals according to the preset communication protocol. This data is then used by the yarn structure stability assessment model and the third-level verification unit in subsequent processing.

[0040] A method for controlling the production of ultra-abrasion-resistant yarn, applied to a system, includes the following steps: S1: Set the initial process parameters and input the raw material characteristic parameters through the host industrial control computer; S2: The instantaneous wear resistance coefficient is predicted based on the initial process parameters and raw material characteristic parameters using a dynamic wear resistance coefficient prediction model; S3: Calculate the structural stability index based on the image acquired by the high-speed imaging unit of the yarn cross-section using the yarn structure stability assessment model; S4: The instantaneous wear resistance coefficient prediction value is verified step by step through the three-level data verification module built into the dynamic parameter parser, and the process parameters are dynamically adjusted or the optimization algorithm is triggered to recalculate based on the verification results during the verification process. S5: Convert the process parameter combination corresponding to the final determined instantaneous wear resistance coefficient prediction value into motion control instructions, and send them to the motor drive module via the lower-level PLC control module to control the operation of the roller servo motor; S6: After a preset time interval, repeat steps two through five to achieve closed-loop iterative control.

[0041] It should be further explained that this method achieves closed-loop iterative control of abrasion resistance in yarn production through the collaborative work of the upper industrial control computer, the lower PLC control module, the dynamic parameter parser, and the embedded prediction and evaluation models. In step one, the operator inputs the initial process parameters through the human-machine interface of the host industrial control computer. The initial process parameters include at least the draft ratio, twist coefficient, core yarn tension and overfeed ratio. At the same time, the operator inputs the raw material characteristic parameters of the current production batch, including fiber length dispersion and breaking strength variation coefficient. The host industrial control computer stores the initial process parameters and raw material characteristic parameters and sends them to the dynamic abrasion resistance coefficient prediction model and dynamic parameter parser, respectively. In step two, the dynamic abrasion resistance coefficient prediction model performs forward propagation calculation based on the received initial process parameters and raw material characteristic parameters, combined with the three real-time status parameters uploaded in real time by the lower-level PLC control module: the difference in speed between the middle and rear rollers, the fluctuation rate of the speed of the hollow spindle, and the ellipticity of the yarn cross section. It outputs an instantaneous abrasion resistance coefficient prediction value and sends this instantaneous abrasion resistance coefficient prediction value to the dynamic parameter parser. In step three, the yarn structure stability assessment model obtains three feature parameters—the yarn diameter variation coefficient, fiber arrangement entropy, and dynamic change rate of hairiness index—from the lower-level PLC control module based on fifty consecutive frames of yarn cross-section images acquired by the high-speed imaging unit of the yarn cross-section. After image preprocessing, the model performs support vector regression calculation, outputs a structural stability index, and sends the structural stability index to the dynamic parameter parser. In step four, the three-level data verification module built into the dynamic parameter parser verifies the instantaneous abrasion coefficient prediction value step by step. The first-level verification unit calculates the absolute value of the difference between the instantaneous abrasion coefficient prediction value and the structural stability index. When the absolute value is greater than the first preset threshold of eight, the first-level verification unit triggers the rollback mechanism and sends a parameter correction command to the dynamic abrasion coefficient prediction model. The dynamic abrasion coefficient prediction model adjusts the twist coefficient and core yarn tension and recalculates the instantaneous abrasion coefficient prediction value until the absolute value of the difference is less than or equal to eight or the number of rollbacks reaches three. When the absolute value of the difference is still greater than eight after three rollbacks, the dynamic parameter parser sends an alarm signal to the host industrial control computer and pauses the automatic control mode. After the first-level verification passes, the second-level verification unit retrieves similar process combinations from the historical best process library that have an Euclidean distance of less than or equal to five from the current process parameter combination. It reads the actual wear resistance index corresponding to these similar process combinations and calculates the average deviation between the instantaneous wear resistance coefficient prediction value and these actual wear resistance indices. When the average deviation is greater than 10% of the third preset threshold, the second-level verification unit extracts the raw material characteristic parameters from the similar process combinations that differ from the current raw material characteristic parameters by more than 20% of the fourth preset threshold. Based on the preset raw material sensitivity coefficient table, it performs linear interpolation correction on the weight coefficients of fiber length dispersion and breaking strength variation coefficient in the dynamic wear resistance coefficient prediction model, triggering the dynamic wear resistance coefficient prediction model to recalculate the instantaneous wear resistance coefficient prediction value until the average deviation is less than or equal to 10%. After the second-level verification passes, the third-level verification unit receives the temperature rise rate uploaded by the infrared friction thermal imaging sensor, calculates the thermal imaging feedback abrasion resistance coefficient according to the preset conversion formula, and calculates the absolute value of the difference between the instantaneous abrasion resistance coefficient prediction value and the thermal imaging feedback abrasion resistance coefficient. When the absolute value is greater than the fifth preset threshold, the third-level verification unit sends the thermal imaging feedback abrasion resistance coefficient as the target value to the particle swarm optimization module. The particle swarm optimization module uses the thermal imaging feedback abrasion resistance coefficient as the target, and uses the draft ratio, twist coefficient, core yarn tension and overfeed ratio as optimization variables. It uses the particle swarm optimization algorithm to solve in reverse to obtain a new set of process parameter combinations, and sends the new set of process parameter combinations to the instruction generation module of the dynamic parameter parser. Once all three levels of verification pass, the dynamic parameter parser will output the process parameter combination corresponding to the final determined instantaneous wear resistance coefficient prediction value to the instruction generation module. In step five, the instruction generation module converts the received process parameter combination into digital control instructions according to the preset communication protocol, and sends them to the instruction buffer of the lower-level PLC control module through the fourth data interface of the dynamic parameter parser. After parsing the instructions, the lower-level PLC control module outputs the corresponding pulse frequency signal and direction signal to the motor drive module through its high-speed data output port. The motor drive module adjusts the speed of the roller servo motor according to the received signals, so that the actual running draft ratio, twist coefficient, core yarn tension and overfeed ratio are adjusted to the new set values. In step six, the timer inside the lower-level PLC control module starts timing after the instruction is executed. The timing duration is preset to five seconds. After the timing ends, the lower-level PLC control module sends a restart signal to the dynamic parameter parser. After receiving the restart signal, the dynamic parameter parser re-collects the latest real-time status parameters from the high-speed imaging unit of the yarn cross-section, the infrared friction thermal imaging sensor, and the lower-level PLC control module. This triggers the dynamic abrasion resistance coefficient prediction model to re-execute the forward propagation calculation with the current process parameter combination and the latest real-time status parameters as input, starting a new round of iterative control from step two to step five. This cycle repeats continuously to achieve continuous closed-loop optimization control of the yarn abrasion resistance performance.

[0042] This invention achieves closed-loop control of the abrasion resistance performance throughout the yarn production process by constructing a dual-model collaborative architecture of a dynamic abrasion resistance coefficient prediction model and a yarn structure stability evaluation model, combined with a three-level data verification mechanism built into the dynamic parameter parser.

[0043] During operation, the system employs a dynamic abrasion resistance coefficient prediction model that calculates the instantaneous abrasion resistance coefficient prediction based on process parameters, real-time state parameters, and raw material characteristic parameters. A yarn structure stability assessment model independently evaluates the yarn structure uniformity based on images acquired by a high-speed imaging unit of the yarn cross-section. The difference between the two outputs is cross-validated by a first-level verification unit, effectively avoiding prediction biases that may exist with a single model. A second-level verification unit searches the historical best process library for similar process combinations and their actual abrasion resistance indices, screening the instantaneous abrasion resistance coefficient prediction for historical data matching. When there are differences between the raw material characteristic parameters and historical data, the weighting coefficients of the prediction model are automatically adjusted to better reflect the current raw material characteristics. A third-level verification unit introduces an infrared friction thermal imaging sensor to monitor the temperature rise rate of the yarn friction area in real time. The thermal imaging feedback abrasion resistance coefficient is compared with the instantaneous abrasion resistance coefficient prediction. When the deviation exceeds a preset range, a particle swarm optimization module is triggered to solve for new process parameter combinations, achieving dynamic adjustment of the production process.

[0044] Through the organic combination of the above three-level verification and iterative optimization, the system can continuously control the uniformity of the yarn cross-sectional structure, the stability of the twist, and the fluctuation of the tension within the target range, thereby ensuring the stable output of the yarn's abrasion resistance performance.

[0045] Furthermore, this invention uses a self-learning module to associate and store the final combination of process parameters determined after each round of iterative optimization and its corresponding instantaneous wear resistance coefficient prediction value into the historical optimal process library. After obtaining the offline measured wear resistance index, the prediction model is incrementally retrained, so that the accuracy of the dynamic wear resistance coefficient prediction model gradually improves with the accumulation of production data.

[0046] This system completes a closed-loop process from data acquisition, model calculation, three-level verification to command issuance within a five-second iteration cycle, achieving real-time and precise control of the roller servo motor speed. Compared with traditional production methods that rely on offline detection and manual adjustments, this system not only improves the consistency of yarn product quality but also significantly shortens the debugging cycle of process parameters for new varieties, providing a reproducible and traceable technical path for the intelligent production of high-performance yarns.

[0047] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0048] 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 high-wear-resistant yarn production control system, characterized in that, It includes a host industrial computer, a slave PLC control module, a motor drive module, a high-speed imaging unit for yarn cross-section, an infrared friction thermal imaging sensor, and a dynamic parameter analyzer; The upper-level industrial control computer is connected to the lower-level PLC control module and the dynamic parameter parser, respectively. The lower-level PLC control module is connected to the motor drive module, the yarn cross-section high-speed imaging unit, and the infrared friction thermal imaging sensor, respectively. The motor drive module is connected to the Roller servo motor; The host industrial control computer has a built-in dynamic wear resistance coefficient prediction model and a yarn structure stability evaluation model. The dynamic parameter parser interacts with the dynamic abrasion resistance coefficient prediction model and the yarn structure stability evaluation model, respectively. The dynamic parameter parser is also connected to the lower-level PLC control module, and is used to receive real-time data uploaded by the lower-level PLC control module and send the parsed control commands to the lower-level PLC control module. The dynamic wear resistance coefficient prediction model is a prediction model based on a BP neural network. Its input parameters include process parameters, real-time state parameters, and raw material characteristic parameters, and its output parameter is the instantaneous wear resistance coefficient prediction value. The yarn structure stability assessment model is an assessment model built based on the support vector regression algorithm. Its input parameters are the feature parameters extracted after image processing of multiple consecutive frames of yarn cross-section images acquired by the high-speed imaging unit of the yarn cross-section, and the output parameter is the structural stability index. The dynamic parameter parser has a built-in three-level data verification module; The three-level data verification module includes a first-level verification unit, a second-level verification unit, and a third-level verification unit; The first-level verification unit is connected to the dynamic abrasion resistance coefficient prediction model and the yarn structure stability evaluation model, respectively, and is used to receive the instantaneous abrasion resistance coefficient prediction value and the structure stability index, and calculate the absolute value of the difference between the two. The first-level verification unit has a first preset threshold. When the absolute value of the difference is less than or equal to the first preset threshold, the first-level verification unit sends the instantaneous wear resistance coefficient prediction value to the second-level verification unit. When the absolute value of the difference is greater than the first preset threshold, the first-level verification unit sends a parameter correction instruction to the dynamic abrasion resistance coefficient prediction model. The dynamic abrasion resistance coefficient prediction model adjusts the set values ​​of twist coefficient and core yarn tension according to the parameter correction instruction and then recalculates the instantaneous abrasion resistance coefficient prediction value until the absolute value of the difference is less than or equal to the first preset threshold or the number of triggers reaches the preset number. The second-level verification unit is connected to the historical best process library built into the host industrial control computer; The second-level verification unit receives the instantaneous wear resistance coefficient prediction value after verification by the first-level verification unit, and searches in the historical optimal process library for similar process combinations whose Euclidean distance from the current process parameter combination is less than or equal to a second preset threshold. It reads the actual wear resistance index corresponding to the similar process combination and calculates the average deviation between the instantaneous wear resistance coefficient prediction value and the actual wear resistance index. The second-level verification unit has a built-in third preset threshold. When the average deviation is less than or equal to the third preset threshold, the second-level verification unit sends the instantaneous wear resistance coefficient prediction value to the third-level verification unit. When the average deviation is greater than the third preset threshold, the second-level verification unit extracts the raw material characteristic parameters in the similar process combination that differ from the current raw material characteristic parameters by more than a fourth preset threshold. Based on a preset raw material sensitivity coefficient table, the second-level verification unit corrects the weight coefficients of the raw material characteristic parameters in the dynamic wear resistance coefficient prediction model and sends the corrected weight coefficients to the dynamic wear resistance coefficient prediction model, triggering the dynamic wear resistance coefficient prediction model to recalculate the instantaneous wear resistance coefficient prediction value. The third-level verification unit is connected to the infrared tribothermographic sensor. The infrared friction thermal imaging sensor is used to monitor the temperature rise rate of the friction area between the yarn and the guide hook in real time, and sends the temperature rise rate to the third-level verification unit. The third-level verification unit has a built-in conversion formula between temperature rise rate and wear resistance coefficient, and calculates the thermal imaging feedback wear resistance coefficient based on the temperature rise rate. The third-level verification unit receives the instantaneous wear resistance coefficient prediction value after verification by the second-level verification unit, and calculates the absolute value of the difference between the instantaneous wear resistance coefficient prediction value and the thermal imaging feedback wear resistance coefficient. The third-level verification unit has a built-in fifth preset threshold. When the absolute value of the difference is less than or equal to the fifth preset threshold, the third-level verification unit outputs the instantaneous wear resistance coefficient prediction value as the final control target value to the dynamic parameter parser. When the absolute value of the difference is greater than the fifth preset threshold, the third-level verification unit sends the thermal imaging feedback wear resistance coefficient as the target value to the dynamic parameter parser. The dynamic parameter parser has a built-in particle swarm optimization module; The particle swarm optimization module is connected to the third-level verification unit. When the thermal imaging feedback abrasion resistance coefficient is received as the target value, the particle swarm optimization module takes the thermal imaging feedback abrasion resistance coefficient as the target and uses the draft ratio, twist coefficient, core yarn tension and overfeed ratio as optimization variables to solve a new set of process parameters in reverse using the particle swarm optimization algorithm. The dynamic parameter parser converts the new combination of process parameters into motion control commands and sends them to the lower-level PLC control module. The lower-level PLC control module controls the roller servo motor to run according to the motion control command, and restarts the calculation of the dynamic wear resistance coefficient prediction model after a preset time interval; The host industrial control computer has a built-in historical best process library and a self-learning module; The historical optimal process library is used to store combinations of process parameters verified by laboratory wear resistance tests and their corresponding actual wear resistance indices. The self-learning module is connected to the dynamic parameter parser and the historical optimal process library, respectively, and is used to associate and store the final process parameter combination determined after each iteration of optimization and its corresponding instantaneous wear resistance coefficient prediction value into the historical optimal process library.

2. The ultra-wear-resistant yarn production control system according to claim 1, characterized in that: The process parameters include draft ratio, twist coefficient, core yarn tension, and overfeed ratio; The real-time status parameters include the difference in speed between the middle and rear rollers, the fluctuation rate of the hollow spindle speed, and the ellipticity of the yarn cross-section. The difference in speed between the middle and rear rollers is calculated by the lower-level PLC control module based on the feedback signal from the roller encoder. The fluctuation rate of the hollow spindle speed is calculated by the lower-level PLC control module based on the feedback signal from the hollow spindle encoder. The ellipticity of the yarn cross-section is obtained by image processing of the image acquired by the high-speed imaging unit of the yarn cross-section. The raw material characteristic parameters include fiber length dispersion and breaking strength variation coefficient, which are obtained by the host industrial control computer from a preset database based on the input raw material batch information.

3. The ultra-wear-resistant yarn production control system according to claim 1, characterized in that: The characteristic parameters include yarn diameter variation coefficient, fiber arrangement entropy, and hairiness index dynamic change rate; The yarn diameter variation coefficient is calculated by the ratio of the standard deviation to the mean of the yarn diameter in multiple consecutive frames of images; The fiber arrangement entropy is calculated based on the image gray-level co-occurrence matrix; The dynamic change rate of the feather index is calculated from the fluctuation range of the number of feathers in multiple consecutive frames of images.

4. The ultra-wear-resistant yarn production control system according to claim 1, characterized in that: The high-speed imaging unit for yarn cross-section has a frame rate of no less than one thousand frames per second and is used to acquire continuous cross-sectional images during the yarn forming process. The infrared friction thermal imaging sensor is installed in the friction area between the yarn and the yarn guide hook before the yarn is wound, and is used to collect the surface temperature distribution of the yarn in real time and calculate the temperature rise rate.

5. A method for controlling the production of ultra-abrasion-resistant yarn, applied to the system described in any one of claims 1 to 4, characterized in that, Includes the following steps: S1: Set the initial process parameters and input the raw material characteristic parameters through the host industrial control computer; S2: The instantaneous wear resistance coefficient is predicted based on the initial process parameters and raw material characteristic parameters using the dynamic wear resistance coefficient prediction model; S3: Calculate the structural stability index based on the image acquired by the high-speed imaging unit of the yarn cross-section using the yarn structure stability assessment model; S4: The instantaneous wear resistance coefficient prediction value is verified step by step through the three-level data verification module built into the dynamic parameter parser, and the process parameters are dynamically adjusted or the optimization algorithm is triggered to recalculate based on the verification results during the verification process. S5: Convert the process parameter combination corresponding to the final determined instantaneous wear resistance coefficient prediction value into motion control instructions, and send them to the motor drive module via the lower-level PLC control module to control the operation of the roller servo motor; S6: After a preset time interval, repeat steps two through five to achieve closed-loop iterative control.

Images