Intelligent self-adaptive dynamic weaving control mechanism of textile machine

Abstract

The invention provides a dynamic weaving control mechanism of an intelligent self-adaptive textile machine, relates to the technical field of self-adaptive control systems, and fundamentally solves the industrial problem that complex three-dimensional shaping and top-level surface texture cannot be obtained at the same time by realizing coupling control of yarn microfiber orientation and macrothermodynamic state. Quantified yarn comprehensive state indexes are innovatively constructed, and a control strategy is promoted to prospective risk avoidance and active quality protection from post compensation. Besides, according to the scheme, highly-integrated multi-task cooperative control logic is adopted, for example, fiber shaping and stress compensation are completed at the same time through a single actuator, finally, a cross-cycle self-evolution closed loop based on finished product quality is achieved, equipment learning and iteration capacity is given, a solidified execution tool is improved into an intelligent manufacturing platform, and maximization of the product value is ensured.

Description

Technical Field

[0001] This invention relates to the field of adaptive control system technology, specifically to a dynamic weaving control mechanism for an intelligent adaptive textile machine. Background Technology

[0002] In the production of high-end seamless sweaters, especially when using luxurious but delicate blended yarns such as cashmere and silk to create styles with complex three-dimensional structures (such as one-piece shoulders and elbows), in order to achieve precise three-dimensional shaping, the yarn must be subjected to varying stretching and relaxation. This generates intense frictional heat, causing the arrangement of cashmere and silk short fibers inside the yarn to become disordered, thus severely damaging the final luster and delicate feel of the fabric.

[0003] When using physically sensitive and expensive specialty yarns (such as cashmere-silk blends, functional composite filaments, etc.) to manufacture fabrics with significant three-dimensional curves (such as one-piece shoulders, elbows, or ergonomic structures), existing control systems reveal fundamental flaws such as limited sensing dimensions, decoupled control strategies, and a lack of cross-cycle adaptive capabilities. First, the core contradiction of existing control logic lies in the fact that, in order to achieve precise three-dimensional shaping, dynamic and non-constant stretching and relaxation of the yarn must be applied, which contradicts the traditional control objective of pursuing "constant tension." This drastic change in stress gradient inevitably generates significant frictional heat effects between the yarn and the guide components and knitting needles; this localized, transient "thermal-mechanical coupling" phenomenon cannot be effectively detected by existing systems. Secondly, this unmonitored and unmanaged composite stress can cause irreversible damage to the microstructure of the yarn, especially for short fiber blended yarns, which can lead to disordered orientation of internal fibers and directly degrade the surface gloss and delicate feel of the finished product. At the same time, the combined effect of thermal effects and mechanical stress can induce "stress memory" in the fiber polymer chain, causing unpredictable dimensional shrinkage and deformation of the fabric during finishing or use, which seriously affects the product's premium feel and durability. Existing systems lack the ability to simultaneously perceive the macroscopic thermodynamic state and microscopic structural state of yarn. Their control strategies are essentially "open-loop" or "single-variable closed-loop," failing to predict and mitigate the comprehensive degradation risk of yarn at its source. These systems typically execute fixed weaving procedures and lack the ability to iteratively optimize their internal decision-making logic based on the final product quality. This makes them unable to adapt to subtle differences in raw materials from different batches and unable to achieve self-evolution of the process through accumulated production experience. The key technical challenges that need to be addressed are how to achieve online perception of the multi-physical field state of yarn during weaving, quantitative prediction of degradation risk, coordinated execution of multi-objective control, and cross-cycle intelligent optimization based on the final product quality.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent adaptive textile machine dynamic knitting control mechanism to solve the problems mentioned in the background art. To achieve the above objective, this invention provides the following technical solution: The intelligent adaptive textile machine dynamic knitting control mechanism specifically includes: Data acquisition unit: used to enable the controller to synchronously acquire multiple physical state parameters characterizing the macroscopic thermodynamic state and microscopic structural state of the yarn during the weaving process through the fusion sensing module; Comprehensive index calculation unit: used to calculate a comprehensive yarn status index that can characterize the current comprehensive deterioration risk of the yarn based on the multiple physical state parameters and through a preset weighted fusion algorithm; Optimization target determination unit: Based on the yarn comprehensive state index and a weaving task vector characterizing the requirements of the current weaving task, it dynamically generates a composite optimization target that includes two sub-targets: stress control and fiber shaping, through a pre-trained high-level policy network. Instruction parsing unit: used to parse the composite optimization target into a set of time-synchronized linkage control instructions through a low-level execution network, so as to collaboratively achieve compensation for yarn stress and shaping of fiber orientation; The optimization unit is used to receive an offline quality assessment parameter that characterizes the final quality of the finished fabric after a complete weaving task is completed, and to iteratively optimize the decision logic of the high-level policy network by using the deviation between the offline quality assessment parameter and the weaving task vector before weaving.

[0006] Compared with the prior art, the beneficial effects of the present invention are: 1. The control dimension is lowered to the microscopic physical level of fiber orientation and coupled with the macroscopic thermodynamic state of the yarn. Through synergistic regulation, the microscopic arrangement of fibers can be actively managed while shaping complex three-dimensional forms. This solves the fundamental contradiction in the industry that has long existed between "pursuing three-dimensional shaping" and "maintaining surface texture," resulting in finished products that have both excellent morphological stability and top-notch surface gloss.

[0007] 2. By integrating multi-dimensional physical parameters, a "Yarn Comprehensive Status Index" was created that can dynamically quantify the overall risk of yarn degradation. This transforms the control system from traditional passive compensation to proactive, forward-looking risk avoidance, enabling intervention before irreversible damage occurs to the yarn's microstructure. This not only effectively protects the intrinsic quality of high-value raw materials such as cashmere and silk but also elevates product quality control to a new level of prevention and prediction.

[0008] 3. Through the instruction parsing unit, physical actuators (such as electromagnetic fields) are given dual or even multiple tasks, simultaneously performing microscopic fiber shaping and macroscopic stress compensation. This highly integrated collaborative control logic improves the system's control efficiency and resource utilization, while reducing hardware complexity and cost.

[0009] 4. By iteratively optimizing the high-level decision-making network using the quality of the final product, a cross-cycle intelligent closed loop oriented towards the value of the final product was constructed. This mechanism enables the equipment to "learn" from production experience, autonomously adapt to the differences in the characteristics of different batches of raw materials, and continuously optimize the weaving process. This not only ensures the long-term stability of product quality but also elevates the equipment from a fixed execution tool into a self-evolving intelligent manufacturing platform, significantly enhancing the company's core competitiveness and economic benefits. Attached Figure Description

[0010] Figure 1 This is a schematic diagram illustrating the control mechanism of the present invention. Figure 2 This is a schematic diagram of the control mechanism of the present invention; Figure 3 This is a schematic diagram of the logic of the comprehensive index calculation unit of the present invention; Figure 4 This is a schematic diagram of the optimization target determination unit of the present invention; Figure 5 This is a schematic diagram of the instruction parsing unit logic of the present invention; Figure 6 This is a schematic diagram of the optimization unit logic of the present invention; In the diagram: 1. Electromagnetic actuator; 2. Pneumatic actuator. Detailed Implementation

[0011] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0012] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0013] Example 1: Please see Figures 1 to 6 This invention provides a technical solution: an intelligent adaptive textile machine dynamic knitting control mechanism, a controller applied to the control mechanism, and the control mechanism specifically includes: Data acquisition unit: used to enable the controller to synchronously acquire multiple physical state parameters characterizing the macroscopic thermodynamic state and microscopic structural state of the yarn during the weaving process through the fusion sensing module; Further explanation: The controller synchronously drives the high-frequency piezoelectric tension exciter and the terahertz wave transceiver array through a fusion sensing module. Under the same time reference, it acquires the dynamic creep recovery rate, which characterizes the macroscopic viscoelasticity of the yarn, and the dielectric constant anisotropy map, which characterizes the microscopic fiber arrangement of the yarn. The controller determines the dynamic creep recovery rate by calculating the phase hysteresis between the stress excitation applied by the high-frequency piezoelectric tension exciter and the sensed strain response. The dielectric constant anisotropy map is reconstructed by analyzing the attenuation and phase shift differences of mutually perpendicular orthogonally polarized waves emitted by the terahertz wave transceiver array after penetrating the yarn. The controller is also used to: drive a miniature infrared thermal imaging array to obtain an instantaneous infrared thermal imaging temperature rise gradient characterizing the yarn's frictional heating state by performing time-dependent differential operations on the temperature values ​​of the same yarn micro-element in continuous thermal imaging frames; and reconstruct a space charge density distribution cloud map characterizing the yarn's static charge accumulation by polling multiple non-contact electrostatic sensors deployed along the yarn channel and inversely solving the Poisson equation based on the collected electric field strength values. The specific implementation details of the data acquisition unit are as follows: 1.1) The instantaneous infrared thermal imaging temperature rise gradient is denoted as... The controller instructs the miniature infrared thermal imaging array to perform continuous thermal imaging of the yarn segment from the yarn guide outlet to the first knitting needle at a preset frame rate. By tracking the temperature change of the same yarn micro-element in the continuous thermal imaging frames, the controller performs a time-dependent differential operation on the temperature change to determine the instantaneous infrared thermal imaging temperature rise gradient. The controller is further configured to: before performing the differential operation, use a Kalman filter to filter the temperature value sequence of the yarn micro-element to eliminate the interference of random thermal noise; and the preset frame rate is not less than 100Hz to ensure that the instantaneous temperature fluctuation caused by a single yarn-needle interaction can be captured.

[0014] The controller obtains the instantaneous infrared thermal imaging temperature rise gradient through a preset calculation process, which is broken down as follows: Step 1, Yarn micro-element tracking and temperature sequence acquisition: Defining the tracking area: The controller defines a rectangular region of interest located between the yarn guide outlet and the first knitting needle within the field of view of the miniature infrared thermal imaging array. Micro-element identification and tracking: The controller employs a tracking algorithm based on optical flow. In the k-th frame image, the controller identifies a yarn micro-element. The pixel coordinates are then determined. Subsequently, in the (k+1)th frame, the controller predicts and locates the new position of the yarn micro-element by calculating the optical flow field in the pixel neighborhood. .

[0015] Temperature value extraction: For each successfully tracked yarn micro-element, the controller extracts the temperature value of its pixel location in each frame of the image, forming a discrete raw temperature sequence, denoted as . Where k is the frame number.

[0016] Original temperature sequence This refers to the surface temperature of the tracked yarn element at time k. It is obtained using a miniature infrared thermal imaging array, with the value directly measured by the array's sensors, and the unit is Kelvin.

[0017] The second step is the filtering of the temperature sequence: To improve the accuracy of gradient calculation, the controller filters the acquired raw temperature sequence. A one-dimensional Kalman filter is applied to smooth the noise.

[0018] State prediction: The controller predicts the current temperature value based on the filtered temperature value from the previous moment.

[0019] State Update: The controller combines the actual measured temperature value and the predicted value at the current moment to calculate the optimal estimated temperature value for the current moment, i.e., the filtered temperature value, denoted as . .

[0020] Filtered temperature value This refers to the optimal estimate of the yarn's infinitesimal temperature after Kalman filtering and smoothing. Its value is determined using the Kalman filtering algorithm. For the calculation of the instantaneous temperature rise gradient: the controller is based on the filtered temperature value. The backward difference method is used to calculate the instantaneous infrared thermal imaging temperature rise gradient.

[0021] 1. Differential calculation: The controller takes the difference between the filtered temperature value at the current time and the previous time to obtain the temperature change.

[0022] 2. Gradient Determination: The controller divides the temperature change by the time interval between two frames to obtain the instantaneous infrared thermal imaging temperature rise gradient, denoted as... The calculation method is described as follows: the instantaneous infrared thermal imaging temperature rise gradient Its value is derived from the filtered temperature value at the current moment. Subtract the filtered temperature value from the previous moment The difference is then divided by the frame period of the miniature infrared thermal imaging array. get.

[0023] Frame period The frame rate is the time interval between two consecutive frames, and is the reciprocal of the frame rate. It is obtained through system preset parameters, and its value is determined based on hardware performance. In this embodiment, if the frame rate is 100Hz, then... It takes 0.01 seconds.

[0024] 1.2) The dynamic creep recovery rate is denoted as... The controller applies a preset, sinusoidal stress excitation to the yarn by driving a high-frequency piezoelectric tension exciter, and simultaneously measures the yarn's strain response using a high-precision laser displacement sensor to obtain the dynamic creep recovery rate, characterizing the yarn's viscoelasticity. The controller calculates the dynamic creep recovery rate by determining the phase lag angle of the strain response signal relative to the stress excitation signal, based on a preset yarn viscoelasticity model. The amplitude of the stress excitation is set between 1% and 5% of the yarn's nominal tension to ensure the measurement is performed within the yarn's linear viscoelastic range and does not affect the normal weaving process. Furthermore, the preset yarn viscoelasticity model is a Zener model that simultaneously characterizes elastic and viscous behavior. The controller obtains the dynamic creep recovery rate through a preset calculation process, which is broken down as follows: The first step is the synchronous acquisition of excitation-response signals: 1. Applying stress excitation: The controller drives the high-frequency piezoelectric tension exciter to apply a sinusoidal tension to the yarn passing through it. This tension signal, i.e., the stress excitation signal, can be expressed as a function of time.

[0025] 2. Strain Response Measurement: Synchronous with the applied excitation, the controller instructs the high-precision laser displacement sensor to measure the minute change in yarn length caused by tension variation, i.e., strain. This strain signal, the strain response signal, is also a function of time. The second step is the calculation of the phase lag angle: The controller employs a cross-correlation method based on Fast Fourier Transform (FFT) to accurately calculate the phase difference between the two signals. 1. Signal Transformation: The controller performs Fast Fourier Transform on the time series of stress excitation signal and strain response signal acquired within a complete cycle to obtain their frequency domain representations, including amplitude and phase information.

[0026] 2. Phase difference calculation: The controller extracts the phase angles of the two signals at the excitation frequency point and subtracts the two to obtain the phase lag angle, denoted as . Phase lag angle This means that the strain response of the yarn lags behind the phase difference of the stress excitation. It is obtained by performing FFT analysis on the excitation and response signals, and the numerical unit is radians. The third step is the conversion of the dynamic creep recovery rate: the controller uses the calculated phase lag angle... The dynamic creep recovery rate is calculated based on the preset Zener viscoelastic model.

[0027] 1. Calculate the loss tangent: The controller first calculates the phase lag angle. The tangent value, which is called the loss tangent in dynamic mechanical analysis. It directly reflects the ratio of energy dissipation to energy storage of a material.

[0028] 2. Determination of Dynamic Creep Recovery Rate: According to the Zener model, the recovery characteristics of a material are related to its energy storage capacity; the controller converts the loss tangent value into the dynamic creep recovery rate through a preset mapping function, denoted as... The calculation method is described as follows: the dynamic creep recovery rate Its value is determined by a preset maximum recovery rate. Subtract the maximum response rate With loss tangent The product of these two factors yields the maximum recovery rate. : This refers to the recovery rate of an ideal elastic material, which is set to 1 in this embodiment; it is a dimensionless system preset parameter. Because and All of these are dimensionless scalars, therefore the final dynamic creep recovery rate is... It is also a dimensionless scalar. The dimensions of both sides of the formula are consistent, which is in line with the physical definition. 1.3) The dielectric anisotropy diagram is denoted as... The controller drives the terahertz wave transceiver array to obtain a dielectric anisotropy diagram characterizing the microscopic arrangement order of fibers inside the yarn. The controller instructs the terahertz wave transceiver array to emit two orthogonally polarized terahertz waves with mutually perpendicular polarization directions to penetrate the cross section of the yarn. The controller reconstructs the dielectric anisotropy diagram by analyzing the differences in attenuation and phase shift of the two waves after penetration. The terahertz wave transceiver array is configured as an array with multiple transceiver channels to achieve synchronous scanning of different positions on the yarn cross-section; furthermore, before reconstruction, the controller employs a constraint optimization algorithm based on the known yarn diameter to improve the spatial resolution and accuracy of the reconstructed image; the specific implementation details are as follows: The controller obtains the dielectric constant anisotropy diagram through a preset calculation process, which is broken down as follows: Step 1, acquisition of orthogonal polarization signals: 1. Orthogonal wave transmission: The controller drives the terahertz wave transceiver array to transmit two terahertz waves simultaneously or in a time-division manner. One beam is horizontally polarized (H-pol), and the other is vertically polarized (V-pol). The two beams together irradiate and penetrate the cross-section of the yarn. Here, pol represents the "polarization direction," with H and V corresponding to the horizontal and vertical directions, respectively. 2. Signal Reception and Parameter Extraction: The receiving array synchronously receives two polarized waves after they penetrate the yarn. The controller extracts the attenuation coefficient of the horizontally polarized wave from the received signals. and phase shift and the attenuation coefficient of vertically polarized waves. and phase shift The attenuation coefficient is the degree of amplitude attenuation of the terahertz wave after it penetrates the yarn. It is obtained by comparing the amplitude of the received signal with the amplitude of the transmitted signal; the value is a dimensionless scalar or expressed in dB. The phase shift is the phase change of the terahertz wave after it penetrates the yarn. It is obtained by comparing the phase of the received signal with the phase of the transmitted signal; the unit of value is radians. The second step is the calculation of the real and imaginary parts of the dielectric constant: the dielectric constant is a complex number; its real part is related to the phase shift, and its imaginary part is related to the attenuation. The controller calculates the attenuation for each pixel on the yarn cross-section. Based on the attenuation and phase shift values ​​corresponding to the pixel, the real and imaginary parts of the dielectric constant are calculated respectively.

[0029] 1. Imaginary Part Calculation: The controller converts the attenuation coefficient into the imaginary part of the dielectric constant, denoted as... The calculation method is expressed as follows: the imaginary part of the dielectric constant. From this pixel Attenuation coefficient corresponding to polarization direction Multiply by a system calibration constant get.

[0030] 2. Real Part Calculation: The controller converts the phase shift into the real part of the dielectric constant, denoted as... The calculation method is described as follows: Real part of dielectric constant The value is determined by this pixel. Phase shift corresponding to polarization direction Multiply by a system calibration constant Then, the vacuum permittivity is added to obtain the system calibration constant. , This refers to the scaling factor that converts a measured physical quantity into a dielectric constant value, and it is related to the terahertz wave frequency and the average yarn thickness. It is obtained through calibration by measuring standard dielectric materials before the equipment leaves the factory. The third step is the reconstruction of the anisotropy map: the controller combines the calculated complex dielectric constant values ​​of each pixel to form the final image. The controller creates a two-dimensional matrix, where each pixel... A two-dimensional vector is stored, with its two components representing the dielectric constant value under horizontal polarization and the dielectric constant value under vertical polarization at that point, respectively. The two-dimensional matrix storing the dielectric constant information of each pixel in the two orthogonal directions constitutes the dielectric constant anisotropy map, denoted as [image of the vector]. The dielectric constant anisotropy diagram It is a two-dimensional matrix, where each element Each is a complex vector, which is generated by the pixel. Horizontal polarization permittivity and vertical polarization permittivity constitute.

[0031] 1.4) The space charge density distribution cloud map is denoted as... ; The controller polls multiple non-contact electrostatic sensors deployed along the yarn channel to obtain a space charge density distribution cloud map characterizing the degree of electrostatic charge accumulation along the yarn path. Specifically, the controller collects the electric field strength values ​​measured by the multiple non-contact electrostatic sensors at their respective locations and uses these electric field strength values ​​as boundary conditions to perform an inverse solution of the Poisson equation for the spatial region surrounding the yarn, thereby reconstructing the space charge density distribution cloud map. The multiple non-contact electrostatic sensors are arranged in an array around the yarn channel to provide electric field strength data in multiple directions in three-dimensional space. Furthermore, the controller employs a numerical calculation method based on the finite element method during the inverse solution to adapt to complex sensor layouts and boundary geometries.

[0032] The controller obtains the space charge density distribution cloud map through a preset calculation process, which is broken down as follows: Step 1, acquisition of electric field intensity values ​​at multiple points: 1. Sensor Polling: The controller sequentially sends read commands to N1 non-contact electrostatic sensors deployed around the yarn channel according to a preset order and frequency. 2. Data Acquisition: Each sensor measures the three-dimensional electric field intensity vector at its location and sends it to the controller. The controller integrates the data from all sensors to form a set containing N1 electric field intensity vectors. : This means the location of the i1th sensor. The electric field strength is measured directly by a non-contact electrostatic sensor, with the value measured in volts per meter (V / m). The second step is discretization of the solution domain: the controller uses the finite element method to divide the solution space; 1. Define the solution domain: The controller defines a three-dimensional virtual space containing all sensors and related yarn segments as the solution domain. 2. Mesh generation: The controller divides this solution domain into a large number of tiny, irregular tetrahedral elements, forming a finite element mesh. 3. Inverse solution based on the Poisson equation: The controller solves the Poisson equation numerically on the discretized mesh. To deduce the nodal charge density value ;in It is the electric field divergence; It is the vacuum permittivity, also known as the free-space permittivity or electrical constant. It describes the ability of a vacuum to respond to an electric field; in this embodiment... The value is 8.854 × 10 -12 Farads per meter; 1. Constructing a Linear Equation System: Based on the fundamental principles of the finite element method, the controller transforms the Poisson equation into a large linear equation system A1x=b1 over the entire solution domain. Here, the unknown vector x represents the charge density value of each node in the mesh, matrix A1 is determined by the mesh geometry and physical laws, and vector b1 is derived from the electric field intensity vector acquired in the first step. Constructed as boundary conditions. 2. Solving the equations: The controller uses an iterative solver to solve the linear equations, obtaining the charge density value of each grid node; in this embodiment, the iterative solver is the conjugate gradient method; node charge density This refers to the charge density of the j1-th node in the finite element mesh. Its value is obtained by solving the aforementioned linear equations, and the unit is coulombs per cubic meter. The fourth step involves the controller visualizing the obtained node charge density values ​​to form the spatial charge density distribution cloud map. 1. Data Interpolation: The controller interpolates the discrete node charge densities... The data undergoes three-dimensional interpolation to obtain a continuous charge density field. 2. Contour Map Determination: This data field, characterizing the continuous distribution of charge density in the three-dimensional space surrounding the yarn, is the spatial charge density distribution contour map, denoted as... The space charge density distribution cloud map It is a three-dimensional data field function. The value of this function at any point (x, y, z) in space represents the charge density of that node, and its value is derived from the charge density of all nodes. It is obtained by interpolating in three-dimensional space.

[0033] Comprehensive index calculation unit: used to calculate a comprehensive yarn status index that can characterize the current comprehensive deterioration risk of the yarn based on the multiple physical state parameters and through a preset weighted fusion algorithm; Further explanation: The controller performs principal component analysis (PCA) on production data containing multiple historical physical state parameters and corresponding defect labels. Offline, it determines the principal component vector (PCA) among these physical state parameters that contributes most to the fabric defect, and assigns the components of this PCA vector as the contribution weights of each physical state parameter. In real-time calculation, the controller uses these contribution weights to weightedly fuse the normalized physical state parameters to generate the yarn comprehensive condition index. The core innovation described above lies in using PCA, a data-driven method, to objectively and quantitatively reveal the "primary contradiction" or "core degradation pattern" leading to fabric defects from historical data. This makes the final "yarn comprehensive condition index" physically a projection of the current yarn condition onto the "most dangerous" degradation direction, thereby improving the index's predictive ability for potential risks and its physical interpretability.

[0034] Before performing the weighted fusion, the controller first extracts the principal direction dispersion from the dielectric constant anisotropy map as the fiber disorder, and extracts its peak value from the space charge density distribution cloud map as the electrostatic intensity. Then, the four parameters—instantaneous infrared thermal imaging temperature rise gradient, dynamic creep recovery rate, fiber disorder, and electrostatic intensity—are mapped to a unified numerical range. The specific implementation details are as follows: 2.1) The controller calculates the comprehensive state index of the yarn through a preset calculation process, which is broken down as follows: First step, offline learning and determination of contribution weights: 1. Constructing a training dataset: The controller accesses the historical database and constructs a training dataset containing multiple samples. Each sample consists of two parts: one part is a feature vector, which contains the time series of normalized instantaneous infrared thermal imaging temperature rise gradient, dynamic creep recovery rate, fiber disorder, and electrostatic intensity within a specific time window before the appearance of a defect in a certain historical batch; the other part is the defect label corresponding to the time window, with a value of 1 indicating that a defect was eventually produced and 0 indicating that no defect was produced; in this embodiment, the specific time window is initially set to 3 seconds; 2. Performing principal component analysis: The controller performs principal component analysis (PCA) on the feature vector set of all samples labeled "produced defect" in the training dataset. 3. Extracting principal component vectors: The controller calculates the covariance matrix of the feature vector set and solves for its eigenvalues ​​and corresponding eigenvectors. The controller selects the eigenvector corresponding to the largest eigenvalue. 4. Determining Contribution Weights: The controller normalizes the selected feature vector so that the sum of its components is 1. The four components of this normalized feature vector are then determined as the contribution weights of the four physical state parameters, denoted as the temperature rise gradient weights. Creep recovery rate weight Fiber disorder weight and electrostatic strength weight .

[0035] Contribution weight The meaning is the degree of contribution of each physical state parameter to the overall risk of weaving defects. It is obtained through offline learning of principal component analysis on historical production data, and consists of a set of dimensionless scalars whose sum is 1. The second step is the real-time calculation of the yarn's overall state index: this step is executed cyclically in real-time by the controller during the operation of the textile machine. 1. Feature extraction and normalization: The controller obtains the instantaneous infrared thermal imaging temperature rise gradient from the data acquisition unit in real time. and dynamic creep recovery rate The controller processes the real-time dielectric constant anisotropy diagram. Image analysis is performed to calculate the dispersion of the principal direction, thus obtaining the fiber disorder. The controller analyzes the real-time acquired space charge density distribution cloud map. The analysis was performed, and the peak value was extracted to obtain the electrostatic strength. The controller will use the above four parameters. , , , After normalization, a set of dimensionless parameters with values ​​in the range [0,1] is obtained, which are denoted as the normalized temperature rise gradients. Normalized creep recovery rate Normalized fiber disorder Normalized electrostatic strength The controller acquires four raw physical parameters. , , , Then, a risk normalization process is performed to map parameters with different physical meanings to a dimensionless parameter in the range [0,1] that characterizes the risk level. This process is divided into two methods based on the physical properties of the parameter: for the instantaneous infrared thermal imaging temperature rise gradient of the positive risk parameter... Fiber disorder electrostatic strength The controller normalizes the real-time measurements of these parameters using a linear forward mapping function. This function maps their respective ideal values ​​to a risk value of 0 and their respective preset risk threshold upper limits to a risk value of 1. This yields the normalized temperature rise gradient. Normalized fiber disorder and normalized electrostatic strength For the dynamic creep recovery rate of the reverse risk parameter. The controller normalizes the real-time measured value of this parameter using a linear inverse mapping function. This function maps its ideal value to a risk value of 0 and its preset lower risk threshold to a risk value of 1. This yields the normalized creep recovery rate. Through the above adjustments, all normalized parameters have a unified meaning: 0 represents a risk-free ideal state, and 1 represents a deteriorated state that reaches the risk threshold. The better the yarn elasticity, The closer it gets to 1, the higher its corresponding normalized creep recovery rate. It will approach 0 more and more. The worse the yarn elasticity, The closer it gets to 0, the higher its corresponding normalized creep recovery rate. It will get closer and closer to 1; 2. Weighted Fusion Calculation: The controller multiplies each normalized parameter by its contribution weight obtained from offline learning in the first step, and then sums them to obtain the final yarn overall state index, denoted as... The calculation method is described as follows: The yarn comprehensive condition index Its value is derived from the normalized temperature rise gradient. Multiply by temperature rise gradient weight Adding the normalized creep recovery rate Multiplied by creep recovery rate weight In addition to normalized fiber disorder Multiply by fiber disorder weight In addition, normalized electrostatic strength Multiplied by electrostatic strength weight It was obtained later.

[0036] Since all normalization parameters and contribution weights are dimensionless scalars, the final yarn overall condition index is... It is also a dimensionless scalar with a value range between [0,1]. Both sides of the formula have the same dimension, being dimensionless. This indicator intuitively quantifies the current overall deterioration risk of the yarn; the larger the value, the closer the yarn condition is to the "typical pattern" that historically led to weaving defects, and the higher the potential risk.

[0037] 2.2) Further explanation of the technical effect of the comprehensive index calculation unit: The calculation of the comprehensive yarn condition index The core of the algorithm is the linear weighted sum model; this model originates from linear algebra in mathematics and is the most basic and widely used classic model in the fields of multi-attribute decision-making and data fusion.

[0038] Calculating the overall condition index of yarn In the algorithm, the normalized temperature rise gradient Normalized creep recovery rate Normalized fiber disorder and normalized electrostatic strength The input layers are arranged in parallel, and they collectively serve as independent variables. Four contribution weights. As adjustment coefficients, each is multiplied by its corresponding input parameter, reflecting the strength of the influence of each input parameter on the final result. Calculation logic and trend analysis: The comprehensive condition index of this yarn. In the algorithm, each normalization parameter and its corresponding weight have a multiplicative relationship, which is a linear gain relationship. Once the weights are determined, the output value increases linearly in proportion to the input. Different terms have an additive relationship, indicating that the degradation effects of each physical parameter are linearly superimposed. This design is based on the premise that, within a certain range, the contributions of degradation risks from different sources to the final result are independent and additive.

[0039] Trend Analysis: Final Yarn Overall Condition Indicators It exhibits a positive linear correlation with any normalized input parameter. That is, when other parameters remain constant, The increase will lead to an increase in the overall condition index of the yarn. The linear increase is as follows. To illustrate the operability and technical effects of this invention, a specific embodiment is provided below. Through principal component analysis of historical data, the contribution weights obtained from offline learning are: , , , This set of weights indicates that, in this production scenario, the main cause of weaving defects is the deterioration of the yarn's thermodynamic properties. and Dominant, followed by microstructure electrostatic effect Minimal impact. The table below shows the process of real-time monitoring, calculation, and result analysis of six yarn samples under different working conditions: Parameter name / unit Sample A: Normal yarn Sample B: Slight friction Sample C: Decreased elasticity Sample D: Loose structure Sample E: Severe static electricity Sample F: Composite Deterioration Normalized temperature gradient 0.05 0.3 0.05 0.05 0.05 0.6 Normalized creep recovery rate 0.1 0.1 0.4 0.1 0.1 0.5 Normalized fiber disorder 0.08 0.08 0.08 0.5 0.08 0.7 Normalized electrostatic strength 0.12 0.12 0.12 0.12 0.9 0.8 Yarn overall condition index 0.088 0.18 0.168 0.144 0.15 0.61 Sample A: All normalized parameters are at a low level, resulting in the calculated overall yarn condition index. A value of 0.088, close to 0, indicates a healthy yarn condition and low risk. Samples B, C, D, and E: simulated single-dimensional degradation, respectively. Although their respective degradation parameters are high (e.g., the normalized electrostatic strength of sample E),... (It is 0.9), but because its corresponding weight is low, the final yarn overall condition index All values ​​are at a low to medium level (0.144~0.180), indicating that the risk of slight degradation in a single dimension is manageable. Sample F: simulates the situation where multiple dimensions simultaneously experience severe degradation; although no single item reaches the limit value of 1, due to multiple high-weight parameters ( , At the same time, it is at a relatively high level, and the calculated comprehensive yarn condition index The value reached 0.610, significantly higher than other samples, clearly indicating a high-risk condition and the possibility of weaving defects. This yarn overall condition index... The range of values ​​is limited to the interval [0,1].

[0040] When the comprehensive condition index of yarn The closer the output is to 0, the lower the risk of overall yarn deterioration; when the overall yarn condition index... Approaching 0, there exists a situation where the combined calculated value of all weighted terms approaches 0. Since all weights are positive, this requires all normalized parameters to approach 0. Physically, this indicates that the yarn's temperature rise gradient, creep recovery loss, fiber disorder, and static electricity accumulation are all in an ideal or near-ideal healthy state. Therefore, the design of this calculation formula reasonably maps multi-dimensional health states to the low-value range of the indicators.

[0041] When the comprehensive condition index of yarn The closer the output is to 1, the higher the risk of overall yarn deterioration, and the closer the condition pattern is to the main deterioration pattern that historically led to fabric defects. Yarn Overall Condition Index The value of the normalization parameter, which has a large contribution weight, tends to 1. This means that the overall calculated value of all weighted terms may approach 1. For example, in the above embodiment, the yarn overall condition index... Approaching 1 mainly means ( , The values ​​must simultaneously approach 1. Technically, this means the system not only detected degradation but also accurately identified that the current state evolution path highly matches the main historical pattern of "thermo-mechanical coupling degradation" that led to fabric defects. It can not only measure the magnitude of risk but also implicitly identify the "type" of risk, providing deeper information for subsequent precise control.

[0042] Optimization target determination unit: Based on the yarn comprehensive state index and a weaving task vector characterizing the requirements of the current weaving task, it dynamically generates a composite optimization target that includes two sub-targets: stress control and fiber shaping, through a pre-trained high-level policy network. Further explanation: The high-level strategy network is configured to output a set of dynamic weight coefficients based on the combined input of the yarn comprehensive state index and the weaving task vector; the controller then uses these dynamic weight coefficients to linearly weight multiple preset sub-objective functions that correspond to different physical dimensions to construct the composite optimization objective; The core innovation of the optimization target determination unit in this embodiment lies in establishing a decision-making mechanism that intelligently integrates macroscopic process design intentions (represented by the weaving task vector) with microscopic real-time yarn conditions (represented by the comprehensive yarn condition index). Through a high-level policy network, a dynamic trade-off between the two is achieved: it can intelligently adjust the "aggressiveness" of achieving the process target based on the current yarn health condition. For example, even if the task requires high gloss, if the yarn condition is close to the edge of deterioration, the network will automatically reduce the weight of fiber shaping, prioritizing production stability.

[0043] 3.1) The weaving task vector includes two dimensions: target gloss level and target three-dimensional shape retention; the multiple preset sub-objective functions include a stress control sub-objective function aimed at minimizing predicted residual stress, and a fiber shaping sub-objective function aimed at maximizing fiber arrangement order; the specific implementation of the optimization target determination unit is as follows: The controller generates the composite optimization objective through a preset calculation process, which is broken down as follows: The first step is the construction and standardization of the input vector: 1. Obtaining input data: The controller obtains the comprehensive yarn status index from the comprehensive index calculation unit in real time, and loads the weaving task vector corresponding to the current weaving area from the process file.

[0044] 2. Yarn Overall Condition Index: A dimensionless scalar with a value range between [0,1]. A higher value indicates a greater risk of overall yarn deterioration. This parameter is provided by the overall index calculation unit. Weaving Task Vector: A two-dimensional vector composed of the target gloss level and the target 3D shape retention. Target Gloss Level: A dimensionless scalar set by the process designer according to product requirements, ranging from 1 to 5. A higher value indicates a higher requirement for fabric surface gloss. It is obtained by reading a preset process design file. Target 3D Shape Retention: A dimensionless scalar, also set by the process designer, ranging from 1 to 5. A higher value indicates a higher requirement for fabric 3D shape stability. It is obtained by reading a preset process design file. 3. Standardization Processing: The controller performs linear normalization on the target gloss level and the target 3D shape retention, mapping their value ranges to the [0,1] interval, resulting in normalized gloss requirements and normalized shape retention requirements.

[0045] The second step, generation of dynamic weight coefficients: 1. Network inference: The controller feeds three standardized values—the yarn overall condition index, the normalized gloss requirement, and the normalized shape retention requirement—as a three-dimensional input vector into the pre-trained high-level policy network. In this embodiment, the high-level policy network is a feedforward neural network. 2. Output weights: The high-level policy network performs inference calculations on the input vector and outputs two dimensionless weight coefficients: stress control weights. and fiber shaping weight Stress control weight This refers to the degree of importance the control strategy places on the stress control sub-objective under the current state and task. It is obtained from the output of the high-level policy network, with a numerical range of [0,1]. Fiber shaping weights. This refers to the degree of importance the control policy places on the fiber shaping sub-objective under the current state and task. It is obtained from the output of the high-level policy network, with a numerical range of [0,1]. Weight constraint: The network design ensures that the sum of its two output weights is always equal to 1, i.e. .

[0046] The third step is to construct the composite optimization objective: 1. Define sub-objective functions. The controller has two built-in sub-objective functions: stress control sub-objective. This is a function used to evaluate the effectiveness of stress control. Its value is negatively correlated with the predicted residual stress in the yarn; that is, the smaller the residual stress, the larger the function value. (Fiber shaping sub-objective) : A function used to evaluate the effect of fiber arrangement. Its value is positively correlated with the degree of orderliness of fiber arrangement, that is, the more ordered the fiber arrangement, the larger the value of the function.

[0047] 2. Weighted Fusion: The controller multiplies the dynamic weight coefficients generated in the second step with the corresponding sub-objective functions, and then sums them to obtain the final composite optimization objective, denoted as... Its construction method is described as follows: The composite optimization objective Its value is determined by the stress-controlled weight. Multiply by stress control sub-objective Plus fiber shaping weight Multiply by fiber shaping sub-target The result was obtained later. Since the weighting coefficients are dimensionless, the composite optimization objective... Physical dimensions and sub-objective functions and The dimensions of both sides are consistent, representing reward values. The dimensions of both sides of this formula are consistent.

[0048] 3.2) Further explanation of the technical effects of the optimization target determination unit: It integrates macroscopic process intentions and microscopic real-time status at the strategic decision-making level. The optimization focus at different stages was anticipated before the task began, achieving proactive and forward-looking planning. The "normalized gloss requirement" is identified as... The "normalized conformity requirement" is marked as The high-level policy network is a multi-input, dual-output nonlinear function approximator, its technology originating from feedforward neural networks in the field of artificial intelligence. This network is trained to learn a highly complex mapping function. The algorithm used to ultimately construct the composite optimization objective is based on a linear weighted sum model, derived from multi-objective optimization theory in mathematics, used to weigh multiple conflicting objectives. In this decision-making process, there are three input parameters. , and The parallel input layers together constitute the "context" for decision-making. This set of context vectors is fed into the higher-level policy network, which serves as the core processing engine. The output of the higher-level policy network is two weight parameters. and This set of weights, positioned at the intermediate decision-making level, then acts as adjustment coefficients, affecting downstream sub-objective functions and ultimately converging into a single composite optimization objective.

[0049] The functional relationships learned by high-level policy networks are highly non-linear, but their training objectives give their behavior a clear logical trend. The output weights... and There are complex, interdependent relationships between the three input parameters. Not only with Positive correlation, and also affected Constraints, when Even when height increases high, The increase will also be suppressed or even reversed. This inherent interaction effect is the key to achieving intelligent trade-offs, going beyond simple linear rules.

[0050] To illustrate the operability and technical effects of this invention, six typical weaving scenarios are used as examples to demonstrate how a high-level policy network dynamically adjusts weights based on input to generate the optimal control policy guidance.

[0051] Parameter name / unit Scene 1: Regular plain weave fabric Scene 2: High-gloss silk surface Scene 3: 3D Shoulder Shaping Scenario 4: High gloss but risky Scenario 5: Shaping with risks Scenario 6: Basic Knitting under High Risk Yarn overall condition index 0.1 0.1 0.1 0.7 0.7 0.9 Normalized gloss requirements 0.2 0.9 0.2 0.9 0.2 0.1 Normalization conformity requirements 0.2 0.2 0.9 0.2 0.9 0.1 Stress control weights (output) 0.5 0.15 0.85 0.8 0.95 1 Fiber shaping weight (output) 0.5 0.85 0.15 0.2 0.05 0 Scenario 1: Routine task, in good condition. The system adopts a balanced strategy, with each component having a weight of 0.50, taking both aspects into account.

[0052] Scene 2 and Scene 4: Both aim for high gloss. When the yarn is in good condition (Scenario 2), The system focuses 85% of its optimization efforts on fiber shaping. However, when the yarn condition deteriorates (Scenario 4), The system will intelligently reverse the strategy, shifting 80% of the focus to lower-risk stress control and reserving only 20% for fiber shaping. This reflects the priority given to "ensuring production reliability," with clear and quantifiable judgment criteria.

[0053] Scenario 3 and Scenario 5: Both aim for high form retention. When the condition is good (Scenario 3), the system allocates 85% of its focus to stress control (the primary means of achieving shaping). When the condition deteriorates (Scenario 5), the system becomes more conservative, increasing the weight of stress control to 95%, almost completely abandoning the pursuit of fiber luster, in order to ensure the stable shaping of the core three-dimensional form. Scenario 6: Under the extremely high risk of yarn breakage ( Regardless of task requirements, the system executes the most conservative "safety first" strategy, placing all (100%) optimization objectives on stress control to ensure uninterrupted production. The weights output by the high-level strategy network... and Its range is strictly limited to the interval [0,1], and the sum of the two is always 1. Here, stress controls the weight. Let's take an example to analyze: when As the output approaches 0, the control strategy tends to pursue the ultimate fiber arrangement effect to improve gloss; Reasoning: Approaching 0 indicates Approaching 1. This situation requires two conditions to be met simultaneously: 1) the overall yarn condition index. A value within a low range indicates healthy yarn material and a demand for higher performance; 2) Normalized gloss requirements A high numerical value indicates that the current process intent is clearly geared towards fiber shaping. This reasoning suggests that the system will only adopt this optimization strategy when both yarn condition and task requirements are met.

[0054] when As the output approaches 1, the control strategy tends towards conservative stress control to ensure production stability and shape. Reasoning: Approaching 1 means Approaching 0. This situation is mainly driven by two inputs: 1) Overall yarn condition index. A high value indicates a high risk of yarn deterioration. In this case, the system will ignore the task requirements and force a conservative strategy to avoid faults such as yarn breakage; or 2) Normalized shape preservation requirements. It is in a high numerical range because the main physical means of achieving three-dimensional shaping is precise stress control. At this time, the strategy focus naturally shifts towards... Tilting. This reasoning demonstrates how the system can make clear and correct decision-making switches based on risk or core task requirements. It should be noted that: yarn overall condition index... The low and high value ranges are set based on the actual usage environment and the expert group. In this embodiment, the low value range is initially set to be less than 0.2 and the high value range is greater than 0.8.

[0055] Instruction parsing unit: used to parse the composite optimization target into a set of time-synchronized linkage control instructions through a low-level execution network, so as to collaboratively achieve compensation for yarn stress and shaping of fiber orientation; Further explanation: The low-level execution network is configured to: based on the weight coefficients contained in the composite optimization objective, which correspond to stress control and fiber shaping respectively, decompose the single objective value into control command vectors containing multiple sub-task components that act on the electromagnetic execution unit 1 and the pneumatic execution unit 2 respectively, so as to collaboratively achieve compensation for yarn stress and shaping of fiber orientation; The core innovation of the instruction parsing unit lies in establishing a mechanism that intelligently decomposes a single, abstract, composite optimization objective into a set of multi-dimensional physical action instructions applied to the same yarn and executed synchronously. Through a low-level execution network, a unified optimization objective value is deconstructed into multiple functionally different but physically coordinated action components generated by the same set of electromagnetic-pneumatic joint execution units. This solves the fundamental technical problem of the difficulty in coordinating macroscopic stress regulation and microscopic fiber shaping due to their different execution principles, achieving a precise and harmonious mapping from a "single objective" to "composite actions."

[0056] 4.1) The electromagnetic actuator 1 includes an integrated yarn electromagnetic shaper, which is fixed at the exit of the yarn feeder on a seamless flat knitting machine. The integrated yarn electromagnetic shaper includes a multi-pole programmable micro electromagnetic coil array installed around the yarn channel. Specifically, a ring-shaped multi-pole programmable micro electromagnetic coil array composed of multiple micro coils is directly integrated at the exit of each yarn feeder in the ion nozzle array. The yarn passes through the center of this ring-shaped electromagnetic field at the last moment before leaving the yarn feeder and being caught by the knitting needle. The rationale for this design is that it represents the final and most critical control point before the yarn enters the weaving triangle area. Microscopic shaping and stress pre-compensation of the yarn at this point are most direct and can immediately affect the coils that are about to be formed. The yarn feeder itself moves at high speed back and forth on the yarn guide track with the machine head. Integrating the integrated electromagnetic yarn shaper onto the yarn feeder means that it can always follow the yarn being woven, achieving "close-fitting" control of each moving yarn. In this embodiment, the power supply and control signals of the integrated electromagnetic yarn shaper are transmitted from the yarn guide track through a flexible cable or slip ring contacts.

[0057] 4.2) The pneumatic actuator 2 includes a yarn path electrostatic neutralization and heat management module, which is installed below the yarn guide rail of the flat knitting machine. The yarn path electrostatic neutralization and heat management module includes an adaptive ion curtain system composed of an array of ion nozzles. Specifically, one or more rows of independently controllable micro ion nozzles are installed directly below the yarn guide rail or fixed to the inside of the machine head cover, with their spray direction pointing towards the yarn path between the lower yarn nozzle and the needle bed. The rationale for this configuration is as follows: Global coverage: From the moment the yarn emerges from the yarn nozzle until it is caught by the knitting needle, it passes through an open space, which is the main area where static electricity and heat accumulate. By installing an ion nozzle array below the yarn guide rail, an "ion curtain" can be formed that covers the entire width of the knitting head movement (i.e., the entire knitting width) and is dynamically adjustable.

[0058] Non-contact global control: No matter where the yarn nozzle moves, the yarn below it is always within the effective range of this air curtain. The controller can precisely activate several ion nozzles below the yarn nozzle according to the real-time position of the yarn nozzle, forming a localized and enhanced "laminar cooling air jacket" that moves with the yarn nozzle, while managing the electrostatic environment of the entire area; The controller generates the linkage control command through a preset calculation process, which is broken down as follows: Step 1: Input Data Acquisition and Target Decomposition 1. Input Data Acquisition: The controller acquires the composite optimization objective from the optimization objective determination unit in real time, as well as the stress control weight and fiber shaping weight used when generating the objective; 2. Composite Optimization Objective: The higher the value, the higher the overall demand or reward for optimizing the control system. Stress Control Weight: A dimensionless weight coefficient with a value range of [0,1]. The higher the value, the higher the importance attached to stress control in the composite optimization objective. Fiber Shaping Weight: A dimensionless weight coefficient with a value range of [0,1]. The higher the value, the higher the importance attached to fiber shaping in the composite optimization objective. Second Step: Calculation of Sub-task Command Components: Based on the input weights, the controller allocates the overall optimization intensity represented by the composite optimization objective to four specific sub-task command components. A weighted mapping method is used to ensure that the input parameters and output components are all proportional. The "control command vector containing multiple sub-task components" specifically includes: electromagnetic shaping command component, electromagnetic stress command component, electrostatic neutralization command component, and active thermal management command component; The controller has four pre-defined mapping functions, each defined by two numerical constants: a "steepness" and a "center point". These constants are used to non-linearly map an intermediate calculated value to a target range of zero to one.

[0059] The first mapping function constant has a preset kurtosis of 5.0 and a preset center point of 0.5. The second mapping function constant has a preset kurtosis of 5.0 and a preset center point of 0.5. The third mapping function constant has a preset kurtosis of 4.0 and a preset center point of 0.2. The fourth mapping function constant has a preset kurtosis of 4.5 and a preset center point of 0.3. Within each control cycle, the controller calculates the four instruction components sequentially according to the following procedure: 4.21) Calculating the Electromagnetic Shaping Command Component: 1. Calculating the Weighted Input Value: Obtain the values ​​of the "Composite Optimization Target" and the "Fiber Shaping Weight," and calculate the product of these two values ​​to obtain a temporary weighted input value. 2. Applying the First Mapping Function for Transformation: First, calculate the difference between the weighted input value obtained in the previous step and the preset center point (value 0.5) of the first mapping function. Then, multiply this difference by the preset steepness (value 5.0) of the first mapping function to obtain an intermediate result. Next, take the negative of this intermediate result to obtain an exponent term. Then, calculate the exponent term raised to the power of the natural constant e to obtain a power operation result. Add this power operation result to the value 1.0 to obtain a denominator value. Finally, divide the value 1.0 by the denominator value; the quotient is the final electromagnetic shaping command component.

[0060] The electromagnetic shaping command component controls a set of electromagnetic actuators to actively construct the three-dimensional microstructure of the fabric by non-contactly moving, positioning, or bending conductive / magnetic fibers through the generation of precise magnetic fields. It directly relates to the shape retention and structural complexity of the final product. The electromagnetic shaping command component's value range is [0,1]. An electromagnetic shaping command component value of 0 indicates that the electromagnetic shaping actuators are completely off, applying no shaping force. This is the system's baseline or "free" state, where the fibers are only affected by natural stress and gravity. Electromagnetic shaping command component values ​​between (0,1) indicate that the command value is proportional to the output shaping force. For example, 0.5 represents 50% of the maximum designed shaping force. This range allows for fine, smooth adjustments to the fibers and is key to achieving complex, gradient surfaces or fine textures. An electromagnetic shaping command component value of 1 indicates the maximum value, meaning the actuators operate at 100% of their maximum designed power, applying the strongest shaping force. This state is used for rapid, large-amplitude shape construction or when overcoming significant fiber stiffness.

[0061] 4.22) Calculating the Electromagnetic Stress Command Component: 1. Calculating the Weighted Input Value: Obtain the values ​​of the "Composite Optimization Target" and the "Stress Control Weight," and calculate their product to obtain a temporary weighted input value. 2. Applying the Second Mapping Function for Transformation: First, calculate the difference between the weighted input value obtained in the previous step and the preset center point of the second mapping function (its value is 0.5). Then, multiply this difference by the preset steepness of the second mapping function (its value is 5.0) to obtain an intermediate result. Next, take the negative of this intermediate result to obtain an exponent term. Then, calculate the exponent term raised to the power of the natural constant e to obtain a power operation result. Add this power operation result to the value 1.0 to obtain a denominator value. Finally, divide the value 1.0 by the denominator value; the quotient is the final electromagnetic stress command component.

[0062] The electromagnetic stress command component is used to control another set of electromagnetic actuators to apply precise tensile stress (tension) to the fibers. By affecting the straightness of the fibers, it is directly related to the fabric's density, elasticity, dimensional stability, and luster. The electromagnetic stress command component ranges from [0,1]; a value of 0 indicates that the electromagnetic actuators do not apply additional tension. The fibers are in their natural tension state under the current weave path. Values ​​between (0,1) indicate that the command value is directly proportional to the applied additional tension. By dynamically adjusting within this range, the yarn tension can be precisely controlled, for example, by reducing stress in weave curves and increasing stress in straight areas to obtain a uniform fabric surface. A value of 1 indicates that the maximum additional tension that the system can provide is applied. This state is used to manufacture high-density, low-stretch fabrics or to ensure the dimensional stability of critical structural components.

[0063] 4.23) Calculating the electrostatic neutralization command component: 1. Obtaining the input value: Directly use the value of the "composite optimization target" as the input value. 2. Applying the third mapping function for transformation: First, calculate the difference between the input value and the preset center point of the third mapping function (its value is 0.2). Then, multiply the difference by the preset kurtosis of the third mapping function (its value is 4.0) to obtain an intermediate result. Next, take the negative of the intermediate result to obtain an exponent term. Then, calculate the exponent term raised to the power of the natural constant e to obtain a power operation result. Add the power operation result to the value 1.0 to obtain a denominator value. Finally, divide the value 1.0 by the denominator value; the quotient is the final electrostatic neutralization command component.

[0064] The static neutralization command component is used to control the operating intensity of the ion generator or other static elimination devices. Its purpose is to eliminate the static charge generated by fibers during high-speed movement and friction, preventing fiber entanglement, deviation from the path, or dust adsorption on the finished product due to electrostatic adsorption. The static neutralization command component value range is [0,1]; a static neutralization command component value of 0: the command is zero, indicating that the static neutralization device is off. This state can be used to save energy when operating at low speeds or using materials that are not prone to generating static electricity. A static neutralization command component value between (0,1): the command value is directly proportional to the ion production rate (or neutralization efficiency) of the ion generator. The system can output a compliant command value based on the real-time monitored static electricity level or prior knowledge based on material properties, achieving efficient and energy-saving static electricity management. A static neutralization command component value of 1: the command is at its maximum value, indicating that the static neutralization device operates at maximum power. This state is needed when weaving at high speeds or handling materials that are highly prone to generating static electricity to ensure that static electricity is completely and reliably eliminated.

[0065] 4.24) Calculating the Active Thermal Management Command Component: 1. Obtaining the Input Value: Directly use the value of the "Composite Optimization Objective" as the input value. 2. Applying the Fourth Mapping Function for Transformation: First, calculate the difference between the input value and the preset center point of the fourth mapping function (its value is 0.3). Then, multiply this difference by the preset kurtosis of the fourth mapping function (its value is 4.5) to obtain an intermediate result. Next, take the negative of this intermediate result to obtain an exponent term. Then, calculate the exponent term raised to the power of the natural constant e to obtain a power operation result. Add this power operation result to the value 1.0 to obtain a denominator value. Finally, divide the value 1.0 by the denominator value; the quotient is the final active thermal management command component.

[0066] The active thermal management command component controls a bidirectional thermal management unit (such as a thermoelectric module based on the Peltier effect), which can both heat and cool the fiber processing area. Its purpose is to precisely control the temperature of the fiber material to influence its phase transformation, plasticity, curing rate, or crystallinity. The active thermal management command component value range is [-1, 1]; an active thermal management command component value of (0, 1) indicates a heating mode. The value is proportional to the heating power. A value of 1 represents maximum heating power, used for rapid heating or melting and solidification of the material. An active thermal management command component value of 0 indicates that the thermal management unit is off, in a thermally neutral state, neither actively heating nor actively cooling. An active thermal management command component value of [-1, 0) indicates a cooling mode. The absolute value is proportional to the cooling power. A value of -1 represents maximum cooling power, used for rapid cooling, quenching, or inhibiting chemical reactions to "freeze" the fiber to a specific form. All four command components are dimensionless scalars generated from intermediate calculations, and their values ​​are proportional to the execution intensity of the corresponding subtask. The mapping function is a simple linear or exponential function used to scale the calculation results to a range of instructions suitable for the driver to execute. The third step is the generation of the final linkage control command: the controller combines the calculated subtask command components into the final linkage control command sent to the physical execution unit.

[0067] 1. Generate linkage electromagnetic control commands The instruction is a vector containing two components; it is described as follows: the linked electromagnetic control instruction, the value of which is composed of an electromagnetic shaping instruction component and an electromagnetic stress instruction component through vector synthesis. The amplitude and phase of this synthesized instruction vector are used to drive the integrated yarn electromagnetic shaping device to generate a composite electromagnetic field that can apply both torque and thrust. Its physical dimensions are ultimately converted into amperes or volts by the drive module.

[0068] 2. Generate linkage pneumatic control commands This instruction is a vector containing two components. Specifically, it is described as follows: the linked pneumatic control instruction's value is composed of an electrostatic neutralization instruction component and an active thermal management instruction component, synthesized as a vector. The amplitude and direction of this synthesized instruction vector are used to drive the adaptive ion curtain system, controlling the concentration, velocity, and angle of its ion wind. Its physical dimensions are ultimately converted to Pascals or Liters per Minute by the drive module. Through this process, the controller precisely and practically decomposes and maps a single, high-level optimization objective into a set of low-level, multi-dimensional, time-synchronized physical execution instructions.

[0069] 4.1) In one specific embodiment, when the textile machine is about to transition from planar body parts to complex three-dimensional shoulder knitting, the low-level execution network performs the following key technical linkage and collaborative control processes: 1. Cooperative Action – Dual Tasks of Electromagnetic Fields: The low-level execution network is based on the composite optimization objective. Generate and send linkage electromagnetic control commands To the multi-pole programmable micro electromagnetic coil array, to perform: Task 1, Microscopic Shaping: This task involves driving the multi-pole programmable micro-electromagnetic coil array to generate a precise, rotating electromagnetic field. This field utilizes dielectric force to apply minute torques to fibers with different dielectric constants within the yarn, inducing them to rearrange into a more compact and ordered structure. This process is driven by the dielectric constant anisotropy diagram. Real-time data is used for closed-loop feedback. Task 2, Macroscopic Auxiliary Stress Compensation: Simultaneously, the low-level execution network performs closed-loop feedback based on the composite optimization objective. The goal of medium stress control is to calculate the pre-relaxation force required to offset subsequent stretching, and to instruct the electromagnetic field to generate an additional weak, non-contact thrust along the yarn movement direction while inducing fiber alignment, so as to achieve zero-contact, zero-friction pre-compensation for yarn stress. 2. Cooperative Action – Dual Tasks of Ion Wind: The low-level execution network is based on the composite optimization objective. Generate and send linkage pneumatic control commands The ion nozzle array is directed to perform: Task 1, electrostatic neutralization: The ion nozzle array is driven to generate a controlled ion wind to neutralize static electricity generated in the yarn due to friction and electromagnetic fields. This process is illustrated by the space charge density distribution cloud map. The real-time data is used for closed-loop feedback. Task 2, Active Thermal Management: Simultaneously, the lower-level execution network performs closed-loop feedback based on the instantaneous infrared thermal imaging temperature rise gradient. Based on real-time data, the flow rate and angle of the ion wind are adjusted to form a laminar flow cooling air jacket that wraps around the yarn, actively stabilizing the yarn temperature at the preset optimal viscoelastic working point to suppress its material memory effect.

[0070] 3. Precise coordination of the mechanical system: Since the electromagnetic actuator 1 has pre-completed some stress compensation and thermal management, the lower-level actuator network fine-tunes the control commands of the mechanical actuator (active yarn feeder and take-up roller), so that it only needs to perform smaller and smoother adjustment actions; the mechanical actuator includes an active yarn feeder and a take-up roller; Further explanation of the "instruction parsing unit": The "electromagnetic shaping instruction component" is identified as... The "electromagnetic stress command component" is identified as... The "static neutralization instruction component" is identified as... The "active thermal management command component" is identified as... The core algorithm used in this instruction parsing unit is derived from decoupling control in control theory and basis decomposition in signal processing. The low-level execution network is a trained inverse model that learns how to convert a desired "effect" (from...) into... and weight , (Definition) Derive the "cause" (i.e., physical execution instructions) required to produce this effect by working backwards. and The method of assigning a single objective value to different components according to weights is a direct application of the weighted projection concept in linear algebra. In this analytical process, high-level decision parameters , , Located at the instruction input layer, these elements collectively define the "optimization task" at the current moment. These inputs are fed into the lower-level execution network, which serves as the core parsing engine. The network internally computes and generates... , , and These are intermediate instruction components. These components are ultimately combined into... and It resides at the physical instruction output layer and is sent directly to the driver.

[0071] To illustrate the operability and technical effects of this invention, the six typical weaving scenarios mentioned above are used as examples to demonstrate how this instruction parsing unit transforms high-level decisions into specific, quantifiable execution instructions. All instruction components are normalized to the [0,1] interval after passing through a mapping function, representing an execution intensity of 0% to 100%.

[0072] In Scene 2 (High Gloss), High and high Precisely converted into high strength Simultaneously, because high-intensity electromagnetic interaction generates static electricity and heat, the system automatically generates an equally high-intensity... and This coordinated action demonstrates the system's proactive compensation for associated physical effects. Dynamic shift of command focus: Comparing scenarios two and four, when risk increases ( (The rise) has led to a shift in the focus of high-level decision-making from Transferred to At that time, this unit executed this transfer. It plummeted from 0.77 to 0.16, while The energy of the command was precisely redistributed from the "shaping" task to the "stress compensation" task, with the judgment criteria being clear and quantifiable, as the value surged from 0.14 to 0.64.

[0073] Logical consistency of coordinated commands: coordinated pneumatic commands Two components and Its strength and composite optimization objectives A direct positive correlation. Because any optimization action involving strength (whether shaping or stress compensation) generates electrostatic and thermal disturbances, requiring pneumatic units for environmental stabilization. Therefore, in In the highest scenes two and three, the intensities of these two components are the highest at 0.9, while... In the lowest scenario, Scenario 6, the intensity is the lowest at 0.40, which is logically clear and self-consistent.

[0074] Final instruction synthesis: Linked electromagnetic instruction and linkage pneumatic commands The overall strength reflects the total workload of the corresponding execution unit. For example, in scenarios two and three, although the internal task composition of the electromagnetic instructions is different, their overall strength is 0.78, indicating that the system requires high-intensity operation of the electromagnetic units under both different tasks, only the mode of operation differs. All four sub-task instruction components... , , and The output range is normalized to the [0,1] interval. For the electromagnetic shaping command component, electromagnetic stress command component, and electrostatic neutralization command component, the closer their execution intensity is to 0, the weaker the corresponding subtask execution intensity becomes, until it is not executed at all; for the active thermal management command component, the closer its execution intensity is to 0, the closer it is to the thermal neutral state. Reasoning: Based on electromagnetic stress command components For example, the condition for it to approach 0 is the composite optimization objective. A value approaching 0 indicates that the system is in a standby state or has no optimization requirements, or that the stress control weights are in a state of equilibrium. A value approaching 0 indicates that high-level decisions deem stress compensation unnecessary or inappropriate, such as during low-speed, low-tension knitting stages where yarn damage is insensitive. This logic ensures that energy is not wasted on unnecessary tasks. The closer the absolute value of any instruction component is to 1, the greater the intensity of its corresponding subtask execution, until the subtask's maximum designed execution capacity is reached. Reasoning: Neutralize instruction components with electrostatics. For example, the condition for it to approach 1 is the composite optimization objective. Approaching its maximum value. This indicates that when the system requires the most intensive optimization operations, regardless of whether the focus is on shaping or stress control, it is accompanied by the strongest risk of electrostatic generation. Therefore, the pneumatic unit needs to perform electrostatic neutralization at maximum power to maintain a stable environment in the yarn channel. This reasoning demonstrates the system's ability to quantitatively manage physical byproducts, ensuring the smooth execution of core tasks.

[0075] The optimization unit is used to receive an offline quality assessment parameter that characterizes the final quality of the finished fabric after a complete weaving task is completed, and to iteratively optimize the decision logic of the high-level policy network by using the deviation between the offline quality assessment parameter and the weaving task vector before weaving.

[0076] Further explanation: The optimization unit is configured to perform adaptive optimization based on the physical properties of the final product after a complete weaving task cycle; the weaving task vector is the actual measurement result and the preset output value representing the desired process target; by converting the deviation vector between the offline quality evaluation parameters and the weaving task vector into a scalarized meta-reward signal, a directional adjustment instruction for the decision-making logic within the high-level strategy network is generated; the directional adjustment instruction iteratively updates the parameters of the high-level strategy network in a cross-cycle, closed-loop feedback manner. It should be noted that the core innovation of the targeted adjustment instruction mechanism lies in its direct mapping of abstract, high-dimensional process target deviations into specific, executable fine-tuning of the strategy network weights. This enables the system to learn autonomously from the "gap between results and targets," continuously optimizing its ability to weigh and decide on different optimization objectives when facing similar future tasks. The offline quality assessment parameters are derived from measurements of finished fabrics by external physical testing equipment. These offline quality assessment parameters include a set of quantified physical indicators, which include at least one or more of surface optical gloss or dimensional stability. 5.1) The optimization unit is decomposed and executed according to the following calculation process: First step, data acquisition and alignment: The optimization unit first acquires two core input data: one is "offline quality assessment parameters" and the other is "weaving task vector".

[0077] The offline quality assessment parameter is a multi-dimensional vector, with each dimension corresponding to a quantifiable physical performance indicator of the finished fabric, such as luster and dimensional stability. This parameter is obtained by offline testing of the finished product using an external fabric analyzer, and through standardization, the values ​​of each dimension of the offline quality assessment parameter are mapped to a unified range of [0,1]. The weaving task vector is a multi-dimensional vector with the same dimensions as the "offline quality assessment parameter," set by the operator or upper-level system before the task begins. It represents the ideal target values ​​for various physical performance characteristics expected to be achieved in this weaving task, and its dimensions are also within the [0,1] range. The directional adjustment instructions specifically include the following: 5.2) The second step is to calculate the target result deviation vector, which is the first layer of direction of the directional adjustment instruction. The optimization unit generates a "target result deviation vector" by subtracting the "weaving task vector" from the "offline quality assessment parameters". Each component of this vector represents the gap between the "expected value" and the "actual value" of the corresponding physical performance, and its positive or negative sign indicates the direction of the deviation. 5.3) The third step is to calculate the overall quality deviation score: The multi-dimensional deviations are integrated into a single scalar signal driving optimization. The optimization unit calculates the "overall quality deviation score" using a weighted summation method. The absolute value of each component in the "target result deviation vector" is multiplied by a preset, corresponding "quality dimension importance weight," and then all products are summed to obtain the "overall quality deviation score." The quality dimension importance weight represents the second-level direction of the directional adjustment instruction; it is a weight vector with the same dimensions as the deviation vector, where each component value is greater than zero, and the sum of all components is 1. It represents the importance of different physical performance indicators, set subjectively or according to production needs, in the final quality assessment. This weight vector is configured by the system administrator based on the core requirements of the current production category. The core requirements of this embodiment are: for silk scarves, luster has the highest weight; for structural components, dimensional stability has the highest weight; ensuring that deviations in different physical dimensions can contribute to the final optimization direction to different degrees according to their business importance, and since all weights are positive, ensuring that deviations in any dimension will have a positive and cumulative contribution to the "overall quality deviation score".

[0078] 5.4) Fourth step: Generating and applying the policy network update gradient: The optimization unit uses the calculated "overall quality deviation score" as the core input and transforms it into a "policy network update gradient" through a preset, monotonically increasing mapping function. The direction of this gradient vector is determined by the historical gradient information and the sign of the current deviation vector, and its magnitude is proportional to the "overall quality deviation score". Finally, this "policy network update gradient" is used to adjust the values ​​of the "higher-level policy network parameters" to complete one iteration of optimization; the output of the directional adjustment command is represented by the "policy network update gradient"; mathematically, it is expressed as... .in, It is the adjusted "high-level policy network parameters". This refers to the "high-level strategy network parameters" before adjustment; These are empirical coefficients, determined by an expert group using fuzzy hierarchical analysis; in this embodiment, the mapping function is a linear function or an exponential function. It is the policy network that updates the gradient; 5.5) High-level strategy network parameters include: network output weights and biases; in this embodiment, the network output weights are stress control weights. and fiber shaping weight The update process follows the basic principle of gradient descent, where parameters are fine-tuned in the opposite direction of the gradient. This ensures that when the network encounters a similar "weaving task vector" again, its output will bring the final "offline quality assessment parameters" closer to the target. This step ensures that the greater the deviation, the greater the adjustment to the high-level policy network parameters, which aligns with the basic logic of learning and optimization.

[0079] 5.6) Further explanation of the technical effects of the optimization unit: In high-end textile processes, "how to balance stress and shaping based on objectives (such as luster and shape retention) and state" is a highly complex expert knowledge that is difficult to express with explicit rules. This solution, through an end-to-end optimization loop, allows the high-level strategy network to autonomously learn and internalize this implicit knowledge from the final, macroscopic finished product quality deviations. Its effectiveness and adaptability far exceed any system based on fixed rules or explicit models. By adjusting the "importance weights of quality dimensions," the system can optimize its core decision preferences for different products. The beneficial effect of this invention lies in establishing a cross-cycle optimization closed loop that quantitatively compares the physical characteristics of the final finished product with the initial process objectives. This allows the system to autonomously learn from production results and generate targeted update gradients to iteratively optimize the internal parameters of the high-level strategy network. Thus, in long-term production practice, it continuously improves its ability to accurately map task requirements and system states to optimal trade-off strategies.

[0080] Identify "Weaving Task Vector" as , its origin Composition; Identify "Offline Quality Assessment Parameters" as The "target result deviation vector" is identified as... The "importance weight of the quality dimension" is marked as The "overall quality deviation score" is marked as... The "Policy Network Update Gradient" is marked as ; Identify "High-level policy network parameters" as .

[0081] The core algorithm used in this optimization unit is derived from meta-learning or offline reinforcement learning in the field of machine learning. Specifically, the entire weaving cycle is regarded as a "decision-execution" cycle, and the final "overall quality deviation score" is... This is considered a cost function directly related to the decision-making quality of the high-level policy network. The calculation of this... The weighted summation method employed is a direct application of the calculation of the weighted L1 norm in linear algebra, used to quantify weighted biases in multidimensional space. This score is ultimately used... The process of updating network parameters draws on the gradient-descent method from optimization theory, with the goal of adjusting the parameters of the high-level policy network within the feedforward neural network. To minimize the expected cost function. In this cross-cycle optimization process, the initially set process objective... It contains and Measurement results with the external physical world Located at the data input and result feedback layer, this is the starting and ending point of the entire optimization process. Both layers generate a deviation vector at the deviation calculation layer. This vector and its preset weights The interactions at the deviation quantization layer generate a standardized "overall quality deviation score". . As a driving signal, it is transformed into "policy network updates gradient" in the gradient generation layer. .final, The training parameters of the high-level policy network are directly applied at the application layer. .

[0082] To illustrate the operability and technical effects of this invention, we will now take the production of six batches of products with different requirements as an example to demonstrate how this optimization unit quantifies the evaluation results and drives the learning of the high-level policy network. and Both are two-dimensional vectors, representing [glossiness, stability] respectively.

[0083] Causal chain of decisions and outcomes: In batch two, to achieve the goal of high gloss... The initial high-level policy network outputs a trade-off policy. , This strategy resulted in insufficient gloss. This unit calculates the larger deviation scores. This score will drive the adjustment of high-level policy network parameters. Updated. In the table "Batch Two: First Production of High-Gloss Products", ;and ; The first component, "+0.3", clearly indicates one direction: the gloss level is below standard and needs to be improved. The second component, "-0.1", clearly indicates another direction: the stability exceeds the target and can be appropriately reduced.

[0084] The system calculates the overall deviation score. At that time, the contribution of gloss deviation ( The contribution of stability bias is much greater than that of stability bias. Although there are deviations in two directions, the final generated "policy network update gradient" The primary driving force behind the adjustment instructions will come from the highly weighted direction of "enhancing gloss." The main objective of the adjustment is "directed" to address the problem of insufficient gloss.

[0085] therefore, and It is the original, multi-dimensional source of "direction," which represents the "direction" of the directional adjustment command; Based on the overall deviation score mentioned above Calculations show that in the gradient descent algorithm of machine learning, the gradient is a vector that points to the "higher-level policy network parameters" in the parameter space. "The direction in which losses can be increased most quickly. Therefore, using..." This instruction points to the direction that most effectively reduces the current "weighted, directional" bias. The output of the directional adjustment instruction is through... The representation, whose "direction" originates from the target result deviation vector. The multidimensional deviation signal, and through After focusing and prioritizing, the final result is a specific vector in the network parameter space. This vector precisely guides the "high-level policy network parameters". "Adjustments will be made to correct the most significant, specific decision-making errors exposed during this production run. In batch three, when faced with the same task again..." At that time, the optimized network output a more aggressive, shaping-oriented strategy. This better decision led to better results. This results in the "overall quality deviation score". The value decreased significantly from 0.26 to 0.09. This clearly and quantitatively demonstrates that this optimization unit successfully improved the decision-making ability of the high-level policy network. Weight-guided optimization direction: When the production objective switches from batch two (high gloss) to batch four (high stability), even if the initial decision-making error is similar ( (The absolute values ​​of the components are similar), but if a "glossy priority" weighting is adopted... The system's penalty for errors in batch two ( ) is much larger than that of batch four ( This allows for prioritizing the learning of how to improve gloss. The core concept in this invention is the "overall quality deviation score." Its output range is designed and normalized to the interval [0,1); when As the output approaches 0, the trade-off strategy output by the higher-level policy network... The closer it is to the optimal solution, the higher the parameters of its internal high-level policy network. The more stable the system, the less need for strategy adjustments; Reasoning: The only condition for the value to approach 0 is that the target result deviation vector... All components approach 0. This physically means that the trade-off policy output by the high-level policy network... The entire production process it guides, and its final output The preset goal was achieved precisely. At this point, the system will generate an update gradient with an amplitude close to zero. Maintain existing high-level strategy network parameters The fact that it remains unchanged demonstrates the rationale that the system will retain its learned successful strategies after achieving its goal.

[0086] when As the output approaches 1, the trade-off strategy output by the high-level policy network... The more severe the defects, the more important it is to optimize the internal high-level policy network parameters. The larger the penalty update, the better. Reasoning: Approaching 1 means that the target result deviation vector It reached its maximum deviation in one or more dimensions and was assigned a higher quality dimension importance weight. In terms of dimensions, this physically represents the parameters of the current high-level policy network. Decisions generated This led to a production failure. At this point, the system will generate an update gradient with the largest amplitude. For high-level policy network parameters A significant adjustment is made to ensure that a distinctly different and superior trade-off strategy is adopted when similar tasks are encountered in the future. This demonstrates the rationale behind the system's design of undergoing profound "reflection" and fundamental "correction" after experiencing major failures.

[0087] It should be noted that separate "overall quality deviation scores" are set. stability threshold and critical threshold The stability threshold in this embodiment It is a positive number approaching 0, initially set to 0.05. Critical threshold. It is a number close to 1, initially set to 0.8. Stability threshold. and critical threshold In practical applications, adaptive adjustments are made using fuzzy hierarchical analysis, which will not be elaborated upon here. When the "overall quality deviation score" The value of satisfies When this value approaches zero, the deviation is defined as "approaching zero"; deviations within this range are considered to be within the normal fluctuations or measurement error range of the production process. This indicates that the current high-level strategy network parameters... The generated trade-off strategy The production results under its guidance are consistent with the preset goals. The match is highly accurate. Within this range, the "policy network update gradient"... The amplitude was forcibly set to zero.

[0088] When the "overall quality deviation score" The value of satisfies When this value approaches 1, it is defined as "approaching 1". Deviation within this range is considered a production failure. This indicates that the current high-level policy network parameters... The generated trade-off strategy has fundamental flaws in addressing the current task, leading to a discrepancy between the final product quality and the target. Within this range, the "policy network updates the gradient." The amplitude through To determine; to ensure the "overall quality deviation score". The closer it is to 1, the greater the increase in the update magnitude.

[0089] When the "overall quality deviation score" The value of satisfies At this point, the deviation is defined as a routine deviation; deviations within this range are considered predictable production deviations that are part of routine optimization. This is the main working area for the system to perform incremental learning and continuous improvement.

[0090] Within this range, the "policy network update gradient" The magnitude and the "overall quality deviation score" The value is linearly proportional to the deviation. The larger the deviation, the linearly larger the adjustment range. The update range is greater than The update range.

[0091] It should be noted that all calculation formulas in this application employ regression analysis, including but not limited to machine learning algorithms, to deeply analyze the collected parameters and identify their natural trends and interrelationships. Specialized software, such as Python's Scikit-learn library or the R language, is used to automatically generate mathematical models that match the data. Then, cross-validation and other methods are used to objectively evaluate the model performance, and continuous feedback and optimization are combined to ensure that the created formulas truly reflect the inherent laws of the data, thereby guaranteeing their effectiveness and accuracy. In all calculation formulas in this application, the parameters in each formula undergo dimensionless processing within a consistent range to ensure that different physical quantities are compared on the same scale; dimensionless processing techniques include, but are not limited to, min-max-normalization and Z-score standardization. The technical solution of this invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random-access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments of this invention.

[0092] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. Intelligent self-adapting textile machine dynamic weaving control mechanism, a controller applied to the control mechanism, characterized in that, The control mechanism specifically comprises: a data acquisition unit configured to enable the controller to synchronously acquire, through a fusion perception module, a plurality of physical state parameters representing the macroscopic thermodynamic state and the microscopic structural state of the yarn during weaving; a comprehensive index calculation unit configured to calculate, based on the plurality of physical state parameters, a yarn comprehensive state index representing the current comprehensive deterioration risk of the yarn through a preset weighted fusion algorithm; an optimization target determination unit configured to dynamically generate, based on the yarn comprehensive state index and a weaving task vector representing the current weaving task requirement, a composite optimization target including two sub-targets of stress control and fiber shaping through a pre-trained high-level strategy network; an instruction analysis unit configured to analyze the composite optimization target into a set of time-synchronized linkage control instructions through a low-level execution network to cooperatively achieve compensation of the yarn stress and shaping of the fiber orientation; an optimization unit configured to receive, after completion of a complete weaving task, an offline quality evaluation parameter representing the final quality of the finished fabric, and utilize the deviation between the offline quality evaluation parameter and the weaving task vector before weaving to iteratively optimize the decision logic of the high-level strategy network.

2. The intelligent self-adapting textile machine dynamic weave control mechanism of claim 1, wherein: The controller synchronously drives the high-frequency piezoelectric tension exciter and the terahertz wave transceiver array through the fusion perception module, and acquires, under the same time reference, a dynamic creep recovery rate representing the macroscopic viscoelasticity of the yarn and a dielectric constant anisotropy map representing the microscopic fiber arrangement of the yarn; wherein the controller determines the dynamic creep recovery rate by calculating the phase lag between the stress excitation applied by the high-frequency piezoelectric tension exciter and the sensed strain response; and reconstructs the dielectric constant anisotropy map by analyzing the attenuation and phase shift difference of mutually perpendicular orthogonal polarized waves penetrating the yarn after being emitted by the terahertz wave transceiver array; The controller is further configured to drive a miniature infrared thermal imaging array, acquire an instantaneous infrared thermal imaging temperature rise gradient representing the frictional heat generation state of the yarn by performing a time differential operation on the temperature values of the same yarn microelement in consecutive thermal imaging frames, and reconstruct a spatial charge density distribution cloud map representing the degree of static charge accumulation of the yarn by polling a plurality of non-contact electrostatic sensors deployed along the yarn channel and inversely solving Poisson's equation based on the collected electric field strength values. The controller determines, in an offline manner, a principal component vector having the largest contribution degree to the defect by performing principal component analysis on production data including a plurality of historical physical state parameters and corresponding defect labels, and determines the components of the principal component vector as the contribution weights of the physical state parameters; 3. The intelligent self-adapting textile machine dynamic weave control mechanism of claim 2, wherein: In real-time calculation, the controller performs weighted fusion on the normalized physical state parameters using the contribution weights to generate the yarn comprehensive state index; ​ The controller first extracts the main direction dispersion of the dielectric constant anisotropy as the fiber disorder degree and extracts the peak value of the space charge density distribution as the electrostatic strength from the dielectric constant anisotropy map before the weighted fusion; then, the four parameters of the instantaneous infrared thermal imaging temperature rise gradient, the dynamic creep recovery rate, the fiber disorder degree and the electrostatic strength are mapped to a unified numerical interval.

4. The intelligent self-adapting textile machine dynamic weave control mechanism of claim 3, wherein: The high-level policy network is configured to output a set of dynamic weight coefficients according to the combined input of the yarn comprehensive state index and the knitting task vector; the controller then linearly weights a plurality of preset sub-objective functions corresponding to different physical dimensions respectively by using the set of dynamic weight coefficients to construct the composite optimization target; The dynamic weight coefficients correspond to stress control weight and fiber shaping weight of stress control and fiber shaping respectively; The knitting task vector includes two dimensions of target gloss level and target three-dimensional shape retention; the plurality of preset sub-objective functions include a stress control sub-objective function aiming to minimize the predicted residual stress and a fiber shaping sub-objective function aiming to maximize the fiber arrangement order.

5. The intelligent self-adapting textile machine dynamic weave control mechanism of claim 4, wherein: The low-level execution network is configured to: according to the stress control weight and the fiber shaping weight contained in the composite optimization target, decompose the single target value into a control instruction vector containing a plurality of sub-task components acting on the electromagnetic execution unit and the pneumatic execution unit respectively, to cooperatively realize the compensation of yarn stress and the shaping of fiber orientation; The electromagnetic execution unit includes an integrated yarn electromagnetic shaper fixed at the outlet of the needle mouth of the seamless flat knitting machine.

6. The intelligent self-adapting textile machine dynamic weave control mechanism of claim 5, wherein: The pneumatic execution unit includes a yarn path static neutralization and thermal management module installed below the guide rail of the flat knitting machine; The yarn path static neutralization and thermal management module includes an adaptive ion wind curtain system composed of an ion wind nozzle array; The "control instruction vector containing a plurality of sub-task components" specifically includes electromagnetic shaping instruction component, electromagnetic stress instruction component, static neutralization instruction component and active thermal management instruction component, and the effective value range of electromagnetic shaping instruction component, electromagnetic stress instruction component and static neutralization instruction component is limited in the interval [0, 1]; for the active thermal management instruction component, the effective value range is in the interval [-1, 1].

7. The intelligent self-adapting textile machine dynamic weave control mechanism of claim 6, wherein: Linkage electromagnetic control instructions are generated; the linkage electromagnetic control instructions are composed of electromagnetic shaping instruction component and electromagnetic stress instruction component in a vector synthesis manner; Linkage pneumatic control instructions are generated; the linkage pneumatic control instructions are composed of static neutralization instruction component and active thermal management instruction component in a vector synthesis manner; For electromagnetic shaping instruction component, electromagnetic stress instruction component and static neutralization instruction component, the closer the execution intensity approaches to 0, the weaker the corresponding sub-task execution intensity, until complete non-execution; for active thermal management instruction component, the closer the execution intensity approaches to 0, the closer to the thermal neutral state. When the absolute value of any instruction component tends to approach 1, the corresponding sub-task execution intensity is greater, until the maximum design execution capacity of the sub-task is reached.

8. The intelligent self-adapting textile machine dynamic weave control mechanism of claim 7, wherein: The optimization unit is configured to perform adaptive optimization based on the physical properties of the final product after a complete weaving task cycle ends. The weaving task vector is the actual measurement result and the preset output value representing the expected process target. The deviation vector between the offline quality evaluation parameter and the weaving task vector is converted into a scalarized meta-reward signal to generate directional adjustment instructions for the internal decision logic of the high-level policy network. The directional adjustment instructions iteratively update the high-level policy network parameters in a cross-cycle, closed-loop feedback manner. The offline quality evaluation parameter is a multi-dimensional vector, and the values of each dimension are mapped to the uniform interval [0, 1]. The weaving task vector is a multi-dimensional vector with the same dimensions as the "offline quality evaluation parameter"; the values of each dimension are in the interval [0, 1].

9. The intelligent self-adaptive textile machine dynamic weaving control mechanism according to claim 8, wherein: The first layer direction of the directional adjustment instructions is characterized by a target result deviation vector; the optimization unit generates a "target result deviation vector" by performing vector subtraction between the "weaving task vector" and the "offline quality evaluation parameter"; The absolute value of each component in the "target result deviation vector" is multiplied by the corresponding "quality dimension importance weight", and then all the products are added to obtain the "overall quality deviation score"; The quality dimension importance weight is the second layer direction of the directional adjustment instructions; The calculated "overall quality deviation score" is used as the core input to convert it into a "policy network update gradient" through a preset mapping function; The "policy network update gradient" is used to adjust the values of the "high-level policy network parameters" to complete an iterative optimization; the output of the directional adjustment instructions is represented by the "policy network update gradient".

Images