A glass fiber cloth weaving tension self-adaptive regulation system and method

Abstract

The application discloses a glass fiber cloth weaving tension self-adaptive regulation and control system and method, and particularly relates to the field of textile machinery control, and is used for solving the problem of tension response divergence of multiple looms. The system is characterized in that: under a unified time scale, the tension observation and the executed displacement of each loom are collected, the contact section is divided according to the guide roller heald reed tooth path to generate the tension observation track and the contact mark, the tension state mapping is generated based on the contact mark and the beat sequence, the loom grouping is selected according to the tension state mapping to generate the grouping reference, the grouping setting table is generated, the real tension and the grouping reference are compared in the beat window to calculate the unwinding ratio and the beat phase drift degree, the working condition atlas is input to generate the ruling coefficient, the contact mark correction or the automatic regrouping is executed according to the ruling coefficient to recover the track data and optimize the model, the real tension and the quality alarm are gathered according to batches to verify and solidify the grouping, and the updated grouping setting table is output.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of textile machinery control, more particularly, the present application relates to a glass fiber cloth weaving tension adaptive control system and method. BACKGROUND

[0002] The workshop is controlled by the central control to issue tension targets and tempos, and multiple looms follow the closed loop according to the local sensing and execution unit. The warp yarn runs continuously along the guide roller, the heald and the reed teeth, and the contact and sliding state changes with the working condition, and the friction and bending difference between the edge and the middle part is obvious. Some machines only sample at a single cross section, combine the contact component and the stretching component into a tension reading; different batches of raw silk, sizing residues and roller surface conditions bring perceptual differences; there is a time delay and drift in the command transmission and execution response. The central control hopes to obtain consistent stress with a unified target, but the machine perception meaning and execution characteristics are not consistent, and it is difficult to convert the unified setting into consistent weaving response at the production line level.

[0003] The contradiction lies in the consistency of the central setting and the divergence of the machine response, and the root cause is not parameter integration, but structural mismatch in measurement meaning and execution state. Single cross section sampling mixes contact component and stretching component into closed loop, low frequency shows slow deviation, high frequency shows overshoot and fall; the central control side lacks cross-machine tension state normalization and online identification, the setting is not grouped according to the working condition, and the quality monitoring does not carry out closed loop correction on the stress trajectory. The result is that the same target lands in different real stress and fabric performance on different machines, and there is a long-term gap between group control target and edge execution.

[0004] In order to solve the above problems, a technical scheme is provided. SUMMARY

[0005] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide a glass fiber cloth weaving tension adaptive control system and method, which unifies the collection of tension data and displacement information of each machine, divides the contact section to generate trajectory and mark, separates the contact and stretching component to form state mapping, groups the machines and sets the reference, compares the real tension to calculate the deviation parameter in the tempo window, executes correction or regrouping according to the judgment result, and verifies and updates the setting according to the batch, to solve the problems raised in the background art.

[0006] To achieve the above purpose, the present application provides the following technical scheme:

[0007] A glass fiber cloth weaving tension adaptive control method, comprising the steps of:

[0008] Collecting the tension observation and execution displacement of each machine under a unified time scale, and dividing the contact section according to the guide roller, heald and reed path to generate tension observation trajectory and contact mark;

[0009] The contact component and the stretching component are separated based on the sequence relationship between the contact mark and the beat, a tension state mapping of each machine is generated, and the beat association is maintained;

[0010] Machine grouping is completed according to the tension state mapping, a reference trajectory of each group is selected as a group reference, a group setting table is generated, and the corresponding relationship between the machine and the group reference is recorded;

[0011] The real tension and the group reference are compared in the beat window, the unwinding ratio and the beat phase drift degree are calculated, the decision coefficient is generated by inputting the working condition atlas, the contact mark correction or automatic regrouping is selected according to the decision coefficient, and the trajectory data is recovered according to the correction result for optimization model;

[0012] The real tension and the quality alarm are aggregated and verified by batch, the machine solidification grouping is stabilized, the group reference is saved, the fluctuation machine enters the iterative review process, and the updated group setting table is output for the next cycle start.

[0013] In a preferred embodiment, the tension observation data and the execution displacement data of each machine are collected under a unified time scale, the contact section is divided according to the guide roller heald reed path, the displacement sensor records the displacement change of the execution element in the start-stop fine adjustment stage, and the tension observation trajectory and the contact mark are generated.

[0014] In a preferred embodiment, the collected tension observation data and execution displacement data are time-synchronized, the data time sequence is adjusted by using linear interpolation method, and all machine data is aligned on a unified time axis to form a continuous time sequence.

[0015] In a preferred embodiment, the contact component and the stretching component are separated based on the sequence relationship between the contact mark and the beat, the tension observation trajectory is divided into time periods by time window sliding window technology, each time period corresponds to a tension change trend, and the contact component and the stretching component are extracted by using wavelet transform.

[0016] In a preferred embodiment, the separated contact component and stretching component are normalized to make the contact component and the stretching component of each machine have the same dimension scale, and the tension state mapping is formed by the time-synchronized execution displacement data, and the beat association is maintained.

[0017] In a preferred embodiment, machine grouping is completed according to the tension state mapping, the K-means algorithm is used for clustering the tension state mapping, the normalized contact component and the stretching component are used as features to calculate the similarity of the machine, and the machines with similar behaviors are divided into the same group.

[0018] In a preferred embodiment, a reference trajectory is selected as a group reference for each clustering result, a group reference is generated based on the tension state mapping of the cluster group and the cluster center, a cluster setting table is generated to record the corresponding relationship between the machine and the group reference, and the corresponding machine is issued.

[0019] In a preferred embodiment, the real tension is compared with the group reference in the beat window, the unwinding ratio and the beat phase drift degree are calculated, the contact component is obtained independently of the stretching component through wavelet transform, the ratio of contact work and stretching work is calculated as the unwinding ratio, and the phase drift degree is quantified using time sequence correlation analysis.

[0020] The unwinding ratio and the beat phase drift degree are input into the working condition atlas for projection comparison, a decision coefficient is calculated through a support vector machine algorithm, and contact mark correction or automatic re-clustering is selected for execution according to the decision coefficient, and trajectory data is recovered for optimization model.

[0021] In a preferred embodiment, the unwinding ratio and the beat phase drift degree are input into the working condition atlas for projection comparison, a decision coefficient is calculated through a support vector machine algorithm, and contact mark correction or automatic re-clustering is selected for execution according to the decision coefficient, and trajectory data is recovered.

[0022] A glass fiber cloth weaving tension self-adaptive control system, comprising:

[0023] A data acquisition module: under a unified time scale, the tension observation and the execution displacement of each machine are collected, the contact section is divided according to the guide roller heald reed tooth path, and the tension observation trajectory and the contact mark are generated;

[0024] A component separation module: based on the contact mark and the beat sequence, the contact component and the stretching component are separated, the tension state mapping of each machine is generated, and the beat association is maintained;

[0025] A machine clustering module: according to the tension state mapping, the machine is clustered, a reference trajectory of each group is selected as a group reference, a cluster setting table is generated, and the corresponding relationship between the machine and the group reference is recorded;

[0026] A deviation calculation module: in the beat window, the real tension is compared with the group reference, the unwinding ratio and the beat phase drift degree are calculated, and the decision coefficient is generated by inputting the working condition atlas;

[0027] A decision correction module: according to the decision coefficient, contact mark correction or automatic re-clustering is selected for execution, and trajectory data is recovered according to the correction result for optimization model;

[0028] A batch verification module: the real tension and the quality alarm are aggregated and verified according to the batch, the machine is stabilized and clustered, the group reference is saved, and the fluctuating machine enters an iterative review process;

[0029] A setting update module: an updated cluster setting table is output for starting in the next period.

[0030] The technical effects and advantages of the glass fiber cloth weaving tension self-adaptive regulation system and method of the present application are as follows:

[0031] The present application provides a time sequence consistent basis for component separation through data acquisition and contact section division under a unified time scale, ensures that the tension observation trajectory and the contact mark accurately reflect the warp path dynamics; further separates the contact and stretching components based on the relationship between the mark and the beat, forms a tension state mapping, realizes the structural normalization of the measurement meaning; relies on this mapping to complete the machine grouping and group reference setting, issues unified parameters to adapt to the differences in working conditions; at the same time, the real tension and the reference are compared in real time in the beat window, the deviation parameters are calculated to generate the decision coefficient, the contact mark correction or regrouping is triggered, and the data optimization model is recovered to promote online identification and closed-loop checking; finally, the stable machines are verified and iterated according to the batch, and the updated setting table is output to start the next cycle. By bridging the mismatch between the center setting and the edge execution, the tension response consistency of multiple looms is improved, the real force deviation and the difference in cloth appearance are reduced, and the adaptive regulation capability and quality stability of the glass fiber cloth weaving production line are optimized as a whole. BRIEF DESCRIPTION OF DRAWINGS

[0032] Figure 1 The flowchart of the glass fiber cloth weaving tension self-adaptive regulation method of the present application is shown.

[0033] Figure 2 The structure diagram of the glass fiber cloth weaving tension self-adaptive regulation system of the present application is shown. DETAILED DESCRIPTION

[0034] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0035] Embodiment 1: Figure 1 The glass fiber cloth weaving tension self-adaptive regulation method of the present application is given, which comprises:

[0036] S1: under a unified time scale, collect the tension observation and execution displacement of each machine, and divide the contact section according to the path of the guide roller, the heald and the reed teeth to generate the tension observation trajectory and the contact mark.

[0037] S2: based on the relationship between the contact mark and the beat, separate the contact component and the stretching component, generate the tension state mapping of each machine, and keep the beat associated.

[0038] S3: According to the tension state mapping, complete the machine grouping, select each group reference trajectory as the group reference, generate the grouping setting table, and record the corresponding relationship between the machine and the group reference.

[0039] S4: Compare the real tension with the group reference in the beat window, calculate the unwinding ratio and the beat phase drift degree, input the working condition atlas to generate the decision coefficient, select to execute the contact mark correction or automatic re-grouping according to the decision coefficient, and recycle the trajectory data according to the correction result for optimization model.

[0040] S5: According to the batch, aggregate and verify the real tension and quality alarm, stabilize the machine solidification grouping and save the group reference, and output the updated grouping setting table for the next period start.

[0041] In the weaving process of glass fiber cloth, the workshop issues tension targets and beats through the central control system, and multiple looms follow according to the machine sensing and execution unit closed loop. The warp yarn runs along the guide roller, heald and reed teeth, causing the contact and slip state to switch with the working condition. The friction and bending difference between the edge and the middle part is obvious, causing the machine perception meaning and execution characteristics to be inconsistent, and the unified setting is difficult to convert into consistent weaving response. This design aims to solve the core contradiction of machine response divergence. Through the collection of basic data to support the unified basis for subsequent processing, it ensures the accuracy and comparability of cross-machine tension observation, thereby improving the consistent weaving response of the overall production line.

[0042] The background technology reveals the core contradiction of the central setting consistency and the machine response divergence, which is rooted in the structural mismatch between the measurement meaning and the execution state. Step S1 collects the tension observation and execution displacement of each machine under the unified time scale, and divides the contact section according to the guide roller, heald and reed path, to generate the tension observation trajectory and contact mark, thereby providing basic data for subsequent separation of contact component and stretching component. Through data collection under the unified time scale, time sequence deviation can be eliminated, ensuring the cross-machine consistency of the input data in subsequent steps, and avoiding the structural mismatch caused by the perception difference.

[0043] The specific processing technology logic of step S1 is as follows:

[0044] Step S1.1: Data collection.

[0045] Considering the dynamic changes of contact and slip state, as well as the challenges brought by the edge friction difference, this collection method captures the mechanical details on the warp path through high-precision sampling, thereby providing reliable mixed force data basis for subsequent analysis, avoiding the limitations of traditional single section sampling.

[0046] Under the unified time scale, the tension observation data and the execution displacement data of each machine are collected. The tension observation data is recorded by the tension sensor of each machine in real time, which is the force applied to the warp yarn during the process of passing through the guide roller, heald and reed, including the mixed value of contact friction force and tensile force; the execution displacement data is recorded by the displacement sensor, which records the displacement change of the execution element of each machine in the starting, stopping and fine tuning stages. The data collection time accuracy is controlled at the microsecond level to capture the instantaneous correspondence of tension fluctuation and operation relationship.

[0047] The tension observation value T obs (t) and the execution displacement value D exec (t) are collected by the tension sensor and the displacement sensor of each machine synchronously triggered in a collection cycle, and the tension observation value T obs (t) is sampled once every microsecond. exec (t) represents the tension observation mixed force at time t, and D

[0048] Step S1.2: Time calibration.

[0049] In view of the fact that the instruction delay and drift will amplify high-frequency overshoot and low-frequency offset, the calibration adopts linear interpolation to align the multi-machine data, so as to establish a unified time sequence framework, accurately associate the tension change and the execution action, and avoid response divergence.

[0050] The collected tension observation data and execution displacement data are synchronously time calibrated to ensure that the data of all machines are aligned on the same time axis. If there is a time difference in the collection process, the linear interpolation method is used to adjust the data time sequence. For the time offset Δt between any two machines, the adjusted tension observation value is calculated by the formula wherein is the adjusted tension observation value, t ′ is the target time point under the unified time scale, and Δt is the offset amount of the specific machine relative to the unified time scale; similarly, the same formula is applied to the execution displacement value wherein is the adjusted execution displacement value. After time calibration, the tension observation data and the execution displacement data of all machines are aligned under the unified time scale, forming a continuous time sequence.

[0051] Step S1.3: Section division and generation.

[0052] In view of the specific requirement of contact start and end on the warp yarn path, the displacement threshold is used to identify the section, so as to structure the observation data into trajectory and marker, which lays a path-related accurate input for the separation of contact components and resolves the influence of different batch perception differences.

[0053] Based on the time-calibrated data, the contact section is divided according to the path of the guide roller, the heald and the reed teeth, and the tension observation trajectory and the contact marker are generated. The tension observation trajectory is composed of the adjusted tension observation data sequence, and the contact marker identifies the contact start and end points of the warp yarn on the path of the guide roller, the heald and the reed teeth. By executing the displacement data to identify the path position, the contact section is divided, for example, when the displacement value exceeds the preset threshold, it is marked as the contact start; the tension observation trajectory is generated as the contact marker is generated as a binary sequence C m (t), where C m (t) = 1 indicates that it is in the contact section at time t, and C m (t) = 0 indicates a non-contact section.

[0054] Step S1 generates the tension observation trajectory and the contact marker under the unified time scale through data acquisition, time calibration and section division, provides basic data for separating the contact component and the stretching component based on the contact marker and the sequence of the beat, and ensures the consistency of the meaning of the tension observation between machines.

[0055] Step S1 generates the tension observation trajectory and the contact marker under the unified time scale, providing time-consistent basic data for separating the contact component and the stretching component, and step S2 separates the contact component and the stretching component based on the contact marker and the sequence of the beat, generates the tension state mapping of each machine, and keeps the beat association, thereby preparing for the subsequent group setting and group reference generation.

[0056] The specific processing technical logic of step S2 is as follows:

[0057] Step S2.1: Time period segmentation.

[0058] Since the tension observation trajectory mixes dynamic events, and the beat determines the weaving rhythm, segmentation uses a sliding window to capture independent trends, thereby converting continuous data into an analyzable event sequence, facilitating accurate positioning of component extraction.

[0059] Based on the tension observation trajectory and the contact marker generated in step S1, by comparing the tension observation trajectory change of different machines, the tension observation trajectory is segmented into several time periods by using the time window sliding window technology, and each time period represents an independent event in the weaving process, corresponding to a tension change trend. The specific setting of the window width is a fixed time length, for example, corresponding to a beat period; the tension observation trajectory The sliding window is applied, and the transition point of the contact marker C m (t) in each window is identified as the event boundary to generate a time period sequence where k is the time period index, and The start and end time points of the kth time period, respectively.

[0060] Step S2.2: Component separation.

[0061] In view of the high-frequency characteristics of contact friction and the low-frequency characteristics of tensile force, the separation method decomposes the signal frequency band by means of wavelet transform, thereby independently extracting components, resolving the mixed input problem caused by single-section sampling, and improving the accuracy of tension control.

[0062] In each time period, according to the change of the tension observation trajectory, the contact component and the tensile component are extracted from the tension observation trajectory by wavelet transform. k The wavelet transform is performed on the tension observation trajectory subsequence in each time period S where T stretch (t) is the tensile component at time t, c J,j is the Jth approximation coefficient, φ J,j (t) is the Jth scale function, and J is the maximum decomposition order; the high-frequency detail coefficient corresponds to the contact component where T contact (t) is the contact component at time t, d i,j is the ith detail coefficient, ψ i,j (t) is the ith wavelet function.

[0063] Step S2.3: Normalization processing.

[0064] Considering that the dimensional differences of different machines will interfere with similarity comparison, normalization unifies the scale by amplitude mapping, thereby ensuring the comparability of components in cross-machine analysis and directly supporting the generation of subsequent group benchmarks.

[0065] The separated contact component and tensile component are normalized to make the contact component and tensile component of each machine have the same dimension and scale. In each time period S k , the normalized value of the contact component sequence is calculated as where T is the normalized contact component at time t, and and are the minimum and maximum values of the contact component in the time period S k ; similarly, the normalized value of the tensile component sequence is calculated as where T is the normalized tensile component at time t, and and are the minimum and maximum values of the tensile component in the time period S k .

[0066] Step S2.4: State mapping generation.

[0067] Based on the need for beat association, component and displacement data are integrated into a mapping, thereby providing behavioral characteristics for clustering and effectively bridging the structural mismatch of inconsistent execution characteristics.

[0068] The normalized contact and tension components, along with the time-calibrated execution displacement data from step S1, are used to construct the tension state mapping for each machine, maintaining its correlation with the cycle time. For each machine, data is combined across all time periods to form a multidimensional sequence of tension state mappings. Where M state This represents the tension state mapping of the machine, with each element corresponding to the normalized contact component, normalized tension component, and adjusted execution displacement value at time t. The correspondence between each time period and the cycle time is also marked to maintain the correlation.

[0069] Step S2 generates a tension state map by segmenting time periods, separating components, normalizing, and mapping states. Based on the relationship between contact markers and beats, the contact component and tension component are separated to generate a tension state map for each machine while maintaining beat association. This provides behavioral similarity features for step S3 to group machines based on the tension state map, ensuring consistency in weaving response.

[0070] Step S2 generates a tension state mapping for each machine and maintains the cycle time correlation, providing behavioral similarity features for group setting. Step S3 completes machine grouping based on the tension state mapping, selects a reference trajectory for each group as the group benchmark, generates a group setting table, and records the correspondence between the machine and the group benchmark, which is used for unified parameter distribution and subsequent verification.

[0071] The specific processing logic of step S3:

[0072] Step S3.1: Grouping machines.

[0073] Because tension state mapping captures the unique behavior patterns of each machine, this clustering identifies similar groups through clustering algorithms, thereby transforming perceived differences into manageable categories. This directly addresses the problem of inconsistent execution characteristics and improves the accuracy of control.

[0074] Based on the tension state mapping generated in step S2, the machines are grouped according to their tension characteristics and behavioral similarity. The K-means algorithm is used to cluster the tension state mappings, grouping machines with similar behaviors into the same group. The tension state mapping M of each machine is then... state As input feature vectors, each vector consists of normalized contact components. Normalized stretching component With the adjusted execution displacement value Sequence composition in time period; initialize cluster center number, preset initial value according to total number of machines; iteratively calculate similarity between machines, use Euclidean distance formula

[0075] where d euclid (M i ,M j ) is the Euclidean distance between the tension state mapping of machine i and machine j, is the normalized contact component of machine i at time t, is the normalized contact component of machine j at time t, is the normalized stretching component of machine i at time t, is the normalized stretching component of machine j at time t, is the adjusted execution displacement value of machine i at time t, is the adjusted execution displacement value of machine j at time t; assign machines to the nearest center by minimizing the distance, and update the center until convergence to generate the clustering result.

[0076] Step S3.2: group reference selection.

[0077] In view of the need to represent a typical tension pattern after clustering, the center trajectory in the calculation group is selected as the reference, thereby adapting to subtle differences, bridging the structural mismatch between the control target and the machine response, and ensuring the effectiveness of the setting and issuance.

[0078] For each cluster result formed group, a reference trajectory is selected as the group reference, which represents the tension behavior pattern of the group of machines in a specific working state, based on the cluster center of the tension state mapping of the machines in the clustering group. In each group, the cluster center of the tension state mapping of the machine is calculated as the reference trajectory, and the cosine similarity is used to optimize the center position, and the formula is

[0079]

[0080] where s cos (M g ,M p ) is the cosine similarity of the group reference mapping M g and the mapping M p of the machine in the group, is the normalized contact component of the group reference at time t, is the normalized contact component of the machine in the group at time t, is the normalized stretching component of the group reference at time t, is the normalized stretching component of the machine in the group at time t, adjusted execution displacement value of the group reference at time t, adjusted execution displacement value of the group reference at time t,

[0081] Step S3.3: Grouping setting table generation.

[0082] In view of the demand for unified parameter issuance, a corresponding relationship is recorded and a table is issued, so as to realize the balanced landing of group control target and reduce the gap between real stress and cloth performance.

[0083] A grouping setting table is generated, and the corresponding relationship between the machine and the group reference is recorded, which is used for unified parameter issuance and subsequent verification. The grouping result and the group reference are combined into a table structure, each row containing the machine identification, the grouping index and the corresponding group reference reference trajectory sequence; the grouping setting table is automatically scheduled and issued to the corresponding machine as the tension target.

[0084] Step S3 completes machine grouping, group reference selection, and grouping setting table generation according to the tension state mapping, selects each group reference trajectory as a group reference, generates a grouping setting table, and records the corresponding relationship between the machine and the group reference, which provides a grouping adaptation basis for step S4 to compare the real tension and the group reference in the beat window, and ensures the balanced execution of the tension target of the production line.

[0085] Step S3 generates a grouping setting table and records the corresponding relationship between the machine and the group reference, which provides a grouping adaptation basis for unified parameter issuance. Step S4 compares the real tension and the group reference in the beat window, calculates the unwinding ratio and the beat phase drift degree, inputs the working condition atlas to generate the ruling coefficient, selects the execution contact mark correction or automatic regrouping according to the ruling coefficient, and recovers the trajectory data for model optimization according to the correction result, so as to realize online identification and closed-loop checking.

[0086] The specific processing technology logic of step S4 is as follows:

[0087] Step S4.1: Unwinding ratio and drift degree calculation.

[0088] In view of the fact that the real tension may deviate from the group reference, the deviation is quantified through the work ratio value and the phase analysis, so as to capture the specific performance of high-frequency overshoot and low-frequency deviation, and provide accurate dynamic indicators for decision-making, avoiding the accumulation of response divergence.

[0089] In each time period, the tension observation trajectory generated by step S1 is compared with the group reference of step S3. First, the tension change related to the contact component is extracted, the change amplitude of contact work is calculated, the contact component independent of the stretching component is obtained through wavelet transform, and then the ratio of contact work to stretching work is calculated as the unwinding ratio. At the same time, the drift degree of tension change relative to the beat target in each time period is calculated, and the phase drift degree is quantified using time sequence correlation analysis. In the beat window, the normalized contact component is compared with the adjusted execution displacement value to calculate the contact work integral where W contact is the contact work in the kth time period, and the integral is based on displacement differentiation; similarly, the normalized stretching component is compared with the adjusted execution displacement value to calculate the stretching work integral where W stretch is the stretching work in the kth time period; the unwinding ratio is obtained by formula where R decouple is the unwinding ratio; for the drift degree, the phase shift is calculated using the cross-correlation function, and the formula is where φ drift is the beat phase drift degree, τ is the time lag variable, B group (t) is the reference tension value of the group reference at time t, and the integral covers the beat window.

[0090] Step S4.2: Work condition map projection.

[0091] For comprehensive evaluation of multi-dimensional parameters, the unwinding ratio and drift degree are mapped onto a preset map, and coefficients are generated through machine learning, which helps to integrate historical and current states, and provides operable decision basis for the problem of lacking online identification.

[0092] The calculated unwinding ratio and beat phase drift degree are combined and input into the work condition map for projection comparison, and a decision coefficient is calculated through a support vector machine algorithm. The decision coefficient is based on historical data, work condition patterns, and the current machine state. The work condition map is constructed as a two-dimensional space, with the horizontal axis as the unwinding ratio R decouple and the vertical axis as the beat phase drift degree φ drift ; the current point is projected onto the preset historical trajectory, a support vector machine classifier is used to generate a decision coefficient for deciding whether to perform contact mark correction or automatic regrouping. The formula is C judge = w1R decouple + w2φ drift + b, where C judgewherein w is the support vector coefficient of the unwinding ratio (obtained by training with historical data), w2 is the support vector coefficient of the beat phase shift degree (obtained by training with historical data), and b is the bias term (obtained by training with historical data); a positive value of the coefficient indicates a correction requirement, and a negative value indicates a re-grouping requirement.

[0093] wherein the support vector machine algorithm is a supervised learning method mainly used for classification and regression problems, and the core thereof is to find an optimal hyperplane to maximize the margin between different class data points, thereby achieving a robust decision boundary. The SVM algorithm is applied to the working condition map projection, and a classifier is constructed by training historical data (including the unwinding ratio, the beat phase shift degree, and corresponding machine response labels such as “need to correct” or “need to re-group” of previous batches). The algorithm first maps the input features (unwinding ratio and beat phase shift degree) to a high-dimensional space (if it is a nonlinear SVM, a kernel function such as a radial basis function is used), and then optimizes the hyperplane parameters to minimize the classification error and maximize the margin between support vectors. The optimization process uses a quadratic programming solver based on the Lagrange multiplier method to ensure that only data points located on the margin boundary (i.e., support vectors) affect the final model.

[0094] The support vector coefficient (i.e., the weight vector w, denoted as w1 and w2 in the present application, corresponding to the coefficients of the unwinding ratio and the beat phase shift degree, respectively) is determined through the training process. Specifically, in the SVM training phase, a historical data set (such as the unwinding ratio-shift degree pairs and their labels of the past 100 batches) is used for optimization, and the goal is to solve the dual problem: maximize the Lagrange function wherein α i is the Lagrange multiplier, y i is the sample label, K(x i ,x j ) is the kernel function, and x i is the feature vector (unwinding ratio and shift degree). After training, the support vector coefficient w is determined by the support vectors (samples with α i > 0): w = ∑ i∈SV α i y i x i , wherein SV represents the support vector set. For example, in a simplified training set, assuming that the historical data includes 10 sample points, of which 3 are support vectors (marginal points), then w1 and w2 are calculated by the weighted linear combination of these support vectors, and the specific values are iteratively optimized by a quadratic programming solver (such as the SMO algorithm) to ensure the model generalization ability.

[0095] The determination of the bias term b is also derived from the training optimization process, after solving the dual problem, through the KKT condition (Karush-Kuhn-Tucker condition) calculation: for any support vector x k , b = y k -wx k , and take the average to improve stability. In the present application, the bias term is trained based on historical data, for example, if the training set shows that when the unwinding ratio is 1.0 and the drift degree is 0.5, it corresponds to the "need to correct" label, then the bias term is adjusted to offset the decision boundary, to ensure that the current machine state (such as high frequency overshoot) is correctly classified. For example, assuming a two-dimensional data set: sample 1 (unwinding ratio = 0.8, drift degree = 0.3, label = +1), sample 2 (unwinding ratio = 1.2, drift degree = 0.7, label = -1), after training, the support vector coefficient can be w1 = 2.5, w2 = -1.8, and the bias term b = 0.4, so that the decision coefficient C judge = 2.5x0.8 - 1.8x0.3 + 0.4 > 0, indicating that the contact mark correction is needed.

[0096] Step S4.3: decision shunting.

[0097] Because the decision coefficient reflects the severity of the deviation, this shunting triggers the corresponding action according to the threshold, and recovers the data, ensuring that the production line quickly adapts when the working condition is switched, and resolving the structural mismatch of the closed-loop verification of quality monitoring.

[0098] According to the decision coefficient, "contact mark correction" or "automatic re-clustering" is selected to be executed, and the trajectory data is recovered according to the correction result for model optimization. A threshold is set, if the decision coefficient C judge > 0, then the contact mark correction is triggered, and the starting and ending points of the contact mark C m (t) are matched with the cluster reference; if the decision coefficient C judge ≤ 0, then the automatic re-clustering is triggered, and the clustering process of step S3 is re-run; after correction, the updated tension observation trajectory and cluster reference difference data are recovered and input into the model optimization loop to update the support vector coefficient.

[0099] Step S4 calculates the unwinding ratio and drift degree by comparing the real tension with the cluster reference in the beat window, generates a decision coefficient by projecting the working condition atlas, selects "contact mark correction" or "automatic re-clustering" according to the decision coefficient, and recovers the trajectory data according to the correction result for model optimization, to provide a dynamic verification basis for step S5 batch aggregation verification, ensuring the iterative review of fluctuating machines and the consistency of production line tension.

[0100] Step S4 calculates the adjudication coefficient and performs correction, and collects trajectory data for model optimization, providing an online identification basis for dynamic verification. Step S5 performs convergence verification of real tension and quality alarms in batches, stabilizes the machine grouping and saves the group benchmark, and the fluctuating machine enters the iterative review process, and outputs the updated grouping setting table for the next cycle to start, thereby realizing the closed-loop verification of quality monitoring and the adaptive allocation for the next cycle.

[0101] The specific processing logic of step S5:

[0102] Step S5.1: Tension behavior tracking and comparison.

[0103] Given that the production line needs to be continuously monitored to capture changes in operating conditions, deviations can be identified by comparing with the benchmark in real time. This allows for early intervention in the issue of actual force drop and strengthens the normalization of tension states across machines.

[0104] During production, the tension behavior of each machine is continuously tracked and compared with its cluster benchmark. The current tension observation trajectory for each machine is recorded. The path similarity is calculated by comparing the trajectory sequence with the reference trajectory sequence of the corresponding group benchmark, step by step, using the dynamic time warping algorithm. The formula is as follows: in The dynamic time warping similarity between the tension observation trajectory and the group benchmark is given. t i For the time index of the tension observation trajectory, t j B is the time index for the group benchmark. group (t j ) is the group reference at time t j Reference tension value at the location; when the similarity is below the threshold, it is marked as a potential fluctuation.

[0105] Step S5.2: Batch aggregation verification.

[0106] For the overall evaluation at the end of the batch, verify the consistency of the aggregated data analysis to quantify the differences in fabric performance, and check the gap in edge execution from the perspective of group control to ensure the comprehensiveness of the verification.

[0107] At the end of each batch, quality alarm information and tension data are summarized to analyze the tension consistency of each machine. The tension observation trajectories of all machines within this batch are compared with the group benchmark, and a consistency score is calculated using the integral deviation formula. Among them U consist The score represents the tension consistency of the machine within this batch, with the score covering the entire batch time range. (B) group(t) is the reference tension value at time t corresponding to the group reference; the score is adjusted in combination with quality warning information (such as fabric defect report).

[0108] Step S5.3: Warning issuance and iterative review.

[0109] Because deviation exceeding the range will amplify response divergence, the mechanism of issuing a warning and incorporating review is isolated for fluctuating machines, directly responding to the core contradiction of lacking closed-loop review, and promoting balanced execution between machines.

[0110] If the tension consistency score of some machines exceeds the set tolerance range, a quality warning is issued, and it is incorporated into the iterative review process. Compare the consistency score U consist with the preset threshold value, and if it exceeds, trigger a quality warning to the central control; map the tension state of the fluctuating machine M state with the contact mark C m (t) input iterative review, adjust the group reference using gradient descent optimization, the formula is wherein is the updated group reference, η is the learning rate, is the gradient of the loss function with respect to the group reference, and the loss function is defined as the mean square error of the fluctuating machine observation and the reference.

[0111] Step S5.4: Update output.

[0112] Considering that historical comparison can improve adaptive ability, the update outputs new settings by optimizing the table, ensuring that the next cycle starts to bridge the structural mismatch of the ungrouped setting delivery, realizing continuous production line optimization.

[0113] By comparing with historical batch data, the working condition atlas is updated and the grouping setting table is optimized; in the next round of start, according to the latest grouping setting table and group reference, each machine is optimized and allocated again. Compare the unwinding ratio R decouple and the beat phase drift degree φ drift of the current batch with that of the historical batch; update the projection parameters of the working condition atlas; optimize the grouping setting table based on the iterative review results, record the updated machine and group reference corresponding to it; output the table for central control to deliver in the next cycle, realizing machine allocation.

[0114] Step S5 tracks and compares tension behavior, batch aggregation verification, warning issuance and iterative review, and update output, aggregates and verifies real tension and quality warnings by batch, stabilizes machine solidification and saves group reference, and fluctuating machines enter the iterative review process, and output the updated grouping setting table for the next cycle start, to ensure the long-term consistency of group control targets and edge execution in the glass fiber fabric weaving production line.

[0115] Example 2:Figure 2 The application provides a glass fiber cloth weaving tension self-adaptive control system, which comprises the following components.

[0116] A data acquisition module: under a unified time scale, the tension observation and the executed displacement of each machine are collected, the contact section is divided according to the guide roller heald reed tooth path, the tension observation track and the contact mark are generated.

[0117] A component separation module: based on the contact mark and the beat sequence, the contact component and the stretching component are separated, the tension state mapping of each machine is generated, and the beat correlation is maintained.

[0118] A machine grouping module: the machine grouping is completed according to the tension state mapping, the reference track of each group is selected as a group reference, the grouping setting table is generated, and the machine and the group reference corresponding relationship are recorded.

[0119] A deviation calculation module: the real tension and the group reference are compared in the beat window, the untwining ratio and the beat phase drift degree are calculated, and the decision coefficient is generated by inputting the working condition atlas.

[0120] A decision correction module: according to the decision coefficient, the contact mark correction or automatic re-grouping is selected, and according to the correction result, the track data is recycled for optimization model.

[0121] A batch verification module: the real tension and the quality alarm are aggregated and verified according to the batch, the machine is stabilized and solidified, the group reference is saved, and the fluctuating machine enters the iterative review process.

[0122] A setting update module: the updated grouping setting table is output for starting in the next period.

[0123] The above formulas are dimensionless and the numerical values are calculated, the formula is obtained by collecting a large amount of data to simulate the nearest real situation, and the preset parameters in the formula are set by the person skilled in the art according to the actual situation.

[0124] It should be noted that the system of the application can be deployed on the device itself to realize embedded application, or can be run on PC or other terminal with user interface, so as to meet various hardware environments and use requirements.

[0125] The above only describes some exemplary embodiments of the application by way of illustration, without doubt, for ordinary skilled in the art, the described embodiments can be modified in various ways without departing from the spirit and scope of the application. Therefore, the above figures and description are illustrative in nature and should not be understood as limiting the scope of protection of the claims of the application.

[0126] It has to be noted that, in the present document, the terms "first", "second", etc. merely serve the purpose of differentiation and do not require or imply any kind of ordering or sequence of the entities or actions associated therewith. Furthermore, the terms "comprising", "containing", etc. are to be interpreted as non- exclusive in the sense that a process, method, article, or apparatus that comprises a list of elements is not necessarily limited to those elements, but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises... a", "contains... a", etc. does not, without further restriction, exclude the presence of additional identical elements in the process, method, article, or apparatus.

[0127] The above description is only specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present application, which should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for adaptive control of weaving tension of glass fiber cloth, characterized in that, Including the following steps: Tension observation and execution displacement of each machine are collected under a unified time scale, and contact sections are divided according to the paths of guide rollers, heddles and reed teeth to generate tension observation trajectories and contact marks; Based on the relationship between contact marks and cycle time, the contact component and tension component are separated to generate a tension state mapping for each machine while maintaining cycle time correlation. Based on the tension state mapping, the machine tools are grouped, and the reference trajectory of each group is selected as the group reference. A group setting table is generated, and the correspondence between the machine tools and the group reference is recorded. In the beat window, compare the actual tension with the group benchmark, calculate the untangling ratio and beat phase drift, input the working condition map to generate the decision coefficient, select to perform contact mark correction or automatic regrouping based on the decision coefficient, and collect trajectory data based on the correction result for model optimization. The actual tension and quality alarms are aggregated and verified in batches. The stable machines are solidified into groups and the group benchmark is saved. The fluctuating machines enter the iterative review process and output the updated group setting table for the next cycle to start.

2. The method for adaptive control of fiberglass cloth weaving tension according to claim 1, characterized in that: Tension observation data and displacement data of each machine are collected under a unified time scale. The contact section is divided according to the path of the guide roller heddle and reed teeth. The displacement change of the actuator during the start-up, stop and fine-tuning stages is recorded by the displacement sensor, and tension observation trajectory and contact mark are generated.

3. The method for adaptive control of fiberglass cloth weaving tension according to claim 2, characterized in that: Synchronous time calibration is performed on the collected tension observation data and the executed displacement data. The data timing is adjusted using a linear interpolation method to ensure that all machine data are aligned on a unified time axis, forming a continuous time series.

4. The adaptive tension control method for glass fiber cloth weaving according to claim 1, characterized in that: Based on the relationship between contact markers and beat sequence, the contact component and tension component are separated. The tension observation trajectory is divided into time periods using the time window sliding window technique. Each time period corresponds to the tension change trend. Wavelet transform is applied to extract the contact component and tension component.

5. The adaptive tension control method for glass fiber cloth weaving according to claim 4, characterized in that: The separation contact component and tension component are normalized to ensure that the contact component and tension component of each machine have the same dimension. This is then used to form a tension state mapping with the time-calibrated displacement data to maintain cycle time correlation.

6. The method for adaptive control of weaving tension of glass fiber cloth according to claim 1, characterized in that: Machine groups are formed based on tension state mapping. The K-means algorithm is used to cluster the tension state mapping. The normalized contact component and tension component are used as features to calculate machine similarity and group machines with similar behavior into the same group.

7. The method for adaptive control of fiberglass cloth weaving tension according to claim 6, characterized in that: For each clustering result, a reference trajectory is selected as the group baseline. Based on the tension state mapping of the machine tools within the group, the group baseline is generated. A group setting table is generated to record the correspondence between the machine tools and the group baseline, and then sent to the corresponding machine tools.

8. The method for adaptive control of fiberglass cloth weaving tension according to claim 1, characterized in that: Compare the actual tension with the group benchmark within the beat window, calculate the unwrapping ratio and beat phase drift, obtain the contact component as independent of the tension component through wavelet transform, calculate the ratio of contact work to tension work as the unwrapping ratio, and use time-series correlation analysis to quantify the phase drift. The unwinding ratio and the beat phase drift are input into the working condition map for projection comparison. The decision coefficient is calculated by the support vector machine algorithm. Based on the decision coefficient, the contact mark correction or automatic regrouping is selected, and the trajectory data is recovered to optimize the model.

9. The method for adaptive control of fiberglass cloth weaving tension according to claim 1, characterized in that: The unwinding ratio and the beat phase drift are input into the working condition map for projection comparison. The decision coefficient is calculated by the support vector machine algorithm. Based on the decision coefficient, the contact mark correction or automatic regrouping is selected to retrieve the trajectory data.

10. A glass fiber cloth weaving tension adaptive control system, used to implement the glass fiber cloth weaving tension adaptive control method according to any one of claims 1-9, characterized in that, include: Data acquisition module: Collects tension observation and execution displacement of each machine under a unified time scale, divides the contact section according to the path of the guide roller heddles and reed teeth, and generates tension observation trajectory and contact mark; Component separation module: Based on the contact mark and the sequence of beats, the contact component and the tension component are separated to generate a tension state mapping for each machine and maintain the beat association; Machine grouping module: Based on the tension state mapping, the machine is grouped, a reference trajectory for each group is selected as the group reference, a grouping setting table is generated, and the correspondence between the machine and the group reference is recorded. Deviation calculation module: Compare the actual tension with the group benchmark in the beat window, calculate the unwinding ratio and beat phase drift, and generate the adjudication coefficient by inputting the working condition map; Decision correction module: Selects to perform contact marker correction or automatic regrouping based on the decision coefficient, and recovers trajectory data based on the correction results for model optimization; Batch verification module: aggregates and verifies the actual tension and quality alarms by batch, stabilizes the machine tools, solidifies the group and saves the group benchmark, and the fluctuating machine tools enter the iterative review process; Setting Update Module: Outputs the updated cluster setting table for the next cycle to start.

Images