Self-adaptive control method for weaving tension of glass fiber cloth

Abstract

The invention discloses a glass fiber cloth weaving tension self-adaptive regulation and control method, and relates to the technical field of textile weaving automation control, and the method comprises the steps: building a phase tension prediction model based on a phase tension reference curve, carrying out the iterative updating through error back propagation, and generating a trained phase tension prediction model; acquiring a real-time phase tension sequence through synchronous sampling, inputting the real-time phase tension sequence into the trained phase tension prediction model to execute forward calculation, and outputting a real-time tension deviation sequence; segmenting the real-time tension deviation sequence, extracting a mechanical disturbance sequence and a phase structure disturbance sequence at the same time, and generating a tension compensation demand through cross-correlation positioning and recursive least square identification. According to the invention, the precision and robustness of tension self-adaptive regulation and control are achieved, and the weaving quality and production efficiency of the glass fiber cloth are improved.

Description

Technical Field

[0001] This invention relates to the field of automated control technology in textile weaving, and in particular to an adaptive control method for the weaving tension of glass fiber cloth. Background Technology

[0002] In glass fiber weaving, the stability of warp tension directly affects fabric quality and production efficiency. Conventional tension control methods mainly rely on mechanical braking devices or closed-loop control structures based on real-time sensor feedback. These methods monitor tension fluctuations during the weaving cycle and drive the warp feed motor for responsive adjustment. This technical field focuses on maintaining the mechanical balance of the warp yarns during the shedding, weft insertion, and beat-up stages to ensure stable weft position, and is one of the core components of automated control in textile machinery.

[0003] However, traditional methods are mostly lagging adjustments, lacking the ability to predict tension trends and struggling to cope with dynamic changes under high-speed weaving. Furthermore, existing solutions typically fail to distinguish between tension fluctuations caused by mechanical disturbances and those caused by phase structure, resulting in a single compensation strategy that cannot achieve fine-grained decoupled control for different disturbance sources, thus affecting the adaptability and accuracy of tension regulation. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a method for adaptive control of the weaving tension of glass fiber cloth to solve the problems of response lag and unclear identification of disturbance sources.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] This invention provides an adaptive tension control method for fiberglass fabric weaving, comprising: calculating stage tension values ​​based on warp yarn density parameters and weaving structure parameters; generating a phase tension reference curve through phase unfolding and environmental temperature and humidity compensation; constructing a phase tension prediction model based on the phase tension reference curve and iteratively updating it through error backpropagation to generate a trained phase tension prediction model; acquiring a real-time phase tension sequence through synchronous sampling and inputting it into the trained phase tension prediction model for forward calculation, outputting a real-time tension deviation sequence; segmenting the real-time tension deviation sequence, simultaneously extracting the mechanical disturbance sequence and the phase structure disturbance sequence, and generating a tension compensation requirement through cross-correlation positioning and recursive least squares identification; performing phase weighted allocation and over-limit back-off redistribution on the tension compensation requirement to generate warp release compensation torque and tension roller additional load, and generating a collaborative control output through collaborative gain fine-tuning, and updating the phase tension prediction model in reverse.

[0008] As a preferred embodiment of the adaptive tension control method for glass fiber cloth weaving described in this invention, the step of calculating the stage tension value based on the warp yarn density parameter and weaving structure parameter is as follows:

[0009] Based on the warp yarn density parameters and weaving structure parameters, and using the loom spindle phase signal as the basis for dividing the weaving cycle, the cycle is segmented to generate a set of weaving cycle stages.

[0010] The warp length variation is obtained from the set of weaving cycle stages, and the stage tension value is calculated by combining the warp linear density parameter and the fiber axial tension relationship.

[0011] As a preferred embodiment of the adaptive control method for fiberglass cloth weaving tension described in this invention, the step of generating a phase tension reference curve through phase unfolding and environmental temperature and humidity compensation is as follows:

[0012] Based on the phase signal of the loom spindle, the stage tension value is subjected to phase expansion and discrete sampling processing to generate the initial phase tension reference curve and phase tension sequence.

[0013] Based on environmental temperature and humidity parameters, the initial phase tension reference curve is deterministically compensated using a compensation coefficient table to generate the phase tension reference curve.

[0014] As a preferred embodiment of the adaptive control method for fiberglass cloth weaving tension described in this invention, the steps of constructing a phase tension prediction model based on the phase tension reference curve and iteratively updating it through error backpropagation to generate a trained phase tension prediction model are as follows:

[0015] A phase tension sample set is established based on the phase tension reference curve and historical phase tension sequence;

[0016] The input layer is constructed based on the phase tension sample set to map the sample vector. The long short-term memory neural network structure is used to build the long short-term memory network layer to capture the temporal dynamic features of tension. The fully connected layer is built through the linear fully connected structure to fuse the features and map them to the output space. The output layer is constructed based on the predicted phase tension nodes.

[0017] A phase tension prediction model is constructed by using a deep neural network to perform directed interlayer connections between the input layer, long short-term memory network layer, fully connected layer, and output layer through sequential forward connections.

[0018] The phase tension sample set is input into the phase tension prediction model, and the connection weights are iteratively updated through the error backpropagation algorithm until the prediction error converges, thus generating the trained phase tension prediction model.

[0019] As a preferred embodiment of the adaptive control method for the weaving tension of glass fiber cloth according to the present invention, the step of obtaining the real-time phase tension sequence through synchronous sampling is as follows:

[0020] Tension detection signals are collected from the glass fiber warp path;

[0021] The tension detection signal is subjected to phase synchronization sampling processing to obtain the real-time phase tension sequence.

[0022] As a preferred embodiment of the adaptive tension control method for glass fiber cloth weaving described in this invention, the input trained phase tension prediction model performs forward calculation and outputs a real-time tension deviation sequence, as follows:

[0023] The real-time phase tension sequence is input into the trained phase tension prediction model to perform forward computation and generate the predicted phase tension sequence.

[0024] The phase-by-phase difference between the predicted phase tension sequence and the real-time phase tension sequence is calculated to generate the real-time tension deviation sequence.

[0025] As a preferred embodiment of the adaptive tension control method for glass fiber cloth weaving described in this invention, the steps of segmenting the real-time tension deviation sequence and extracting the mechanical disturbance sequence and the phase structure disturbance sequence are as follows:

[0026] Based on the phase signal of the loom spindle, the real-time tension deviation sequence is phase-synchronized and segmented to form a phase window sequence, and instantaneous displacement and rotation speed values ​​are collected.

[0027] The mechanical disturbance sequence and the phase structure disturbance sequence are extracted from the phase window sequence using exponential moving average filtering and bandpass filtering.

[0028] As a preferred embodiment of the adaptive tension control method for glass fiber cloth weaving described in this invention, the step of generating the tension compensation requirement through cross-correlation positioning and recursive least squares identification is as follows:

[0029] The radius time delay parameter is obtained by cross-correlation positioning of the mechanical disturbance sequence and the instantaneous value of the rotation speed, and the radius disturbance value of the warp bobbin is calculated by recursive least squares identification.

[0030] The friction delay parameters are obtained by cross-correlation positioning of the mechanical disturbance sequence and the instantaneous displacement value, and the friction disturbance value of the guide roller is calculated by recursive least squares identification.

[0031] A phase window weight set is generated based on the phase structure disturbance sequence, the warp bobbin radius disturbance value, and the guide roller friction disturbance value. Weighted synthesis is then performed according to the phase window weight set to generate the tension compensation requirement.

[0032] As a preferred embodiment of the adaptive tension control method for glass fiber cloth weaving described in this invention, the steps of performing phase-weighted allocation and over-limit back-off redistribution on the tension compensation demand to generate the warp release compensation torque and the additional load on the tension roller are as follows:

[0033] A phase window weighted allocation is performed on the warp compensation component in the tension compensation demand, and under the constraint of the upper limit of the warp motor torque, the over-limit allocation value is saturated and back-off redistributed to generate the warp compensation torque.

[0034] The tension roller compensation component in the tension compensation demand is weighted by phase window, and under the constraint of the upper limit of the tension roller additional load, the over-limit allocation value is saturated and redistributed to generate the tension roller additional load.

[0035] The beneficial effects of this invention are as follows: Through the phase tension prediction model, intelligent prediction and dynamic correction of tension evolution trend within the weaving cycle are realized; by segmenting the real-time tension deviation sequence and extracting the mechanical disturbance sequence and phase structure disturbance sequence, the decoupling identification and accurate compensation of multi-source disturbances are realized, achieving the accuracy and robustness of tension adaptive control, and improving the weaving quality and production efficiency of glass fiber cloth. Attached Figure Description

[0036] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 A flowchart for an adaptive tension control method for fiberglass cloth weaving.

[0038] Figure 2 This is a flowchart for tension coordination control and model updating.

[0039] Figure 3 A flowchart for building and training a phase tension prediction model.

[0040] Figure 4 This is a flowchart for perturbation sequence extraction and identification. Detailed Implementation

[0041] 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.

[0042] 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.

[0043] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0044] Reference Figures 1-4 As one embodiment of the present invention, this embodiment provides a method for adaptive control of the weaving tension of glass fiber cloth, comprising the following steps:

[0045] S1. Calculate the stage tension value based on the warp yarn density parameter and weaving structure parameter, and generate the phase tension reference curve through phase unfolding and environmental temperature and humidity compensation.

[0046] Based on the warp yarn density parameters and weaving structure parameters, and using the loom spindle phase signal as the basis for dividing the weaving cycle, the cycle is segmented to generate a set of weaving cycle stages.

[0047] Furthermore, based on the warp yarn linear density parameters and weaving structure parameters, the phase signal of the loom spindle is collected synchronously, and a complete weaving cycle is defined from 0° to 360°. The phase position of the plain weave interlacing point (e.g., 90° and 270°) is extracted as a fixed boundary, and the ratio of linear density to baseline linear density (e.g., 30tex) is used as the warp stiffness coefficient. The spindle phase signal is divided into five sub-intervals (shearing, weft insertion, beat-up, warp feed, and take-up, for example, the shedding stage is 0° to 90° and the weft insertion stage is 90° to 150°) according to the fixed boundary and combined with the warp stiffness coefficient, and arranged in chronological order to form a set of weaving cycle stages.

[0048] It should be noted that the warp linear density parameters are obtained by measuring the warp sample, including linear density (e.g., 50 tex) and glass fiber warp cross-sectional area (e.g., 0.0196 mm²); the weaving structure parameters are read from the weaving process sheet, including weave type (e.g., plain weave), warp density (e.g., 28 ends / cm), weft density (e.g., 24 ends / cm), and interlacing point phase position (e.g., 90° to 150° of the opening stage).

[0049] The warp length variation is obtained from the set of weaving cycle stages, and the stage tension value is calculated by combining the warp linear density parameter and the fiber axial tension relationship.

[0050] Furthermore, the starting and ending phases of each stage are read from the weaving cycle stage set, and the stage phase duration is obtained by combining the phase signal of the loom spindle; based on the phase position of the interlacing point in the weaving structure parameters, the effective elongation section length is determined as the initial length of the warp yarn, and the warp length change is obtained by reading the warp feed and take-up displacement within the stage phase duration, for example, 0.15 mm; based on the cross-sectional area and elastic modulus of the glass fiber warp yarn in the warp yarn linear density parameters, the stage tension value is calculated according to the fiber axial tension relationship and written into the stage tension value set.

[0051] The expression for calculating the stage tension value is:

[0052] ;

[0053] in, The first in the set of weaving cycle stages The stage tension value for each stage, in Newtons (N); Index for weaving cycle stages; The elastic modulus of the glass fiber warp is read from the material parameter table (including the supply specification and material test report) according to the material grade of the warp batch, for example, the value is 74000N / mm²; The cross-sectional area of ​​the glass fiber warp yarn is in mm². This represents the change in warp length, expressed in mm. This is the initial length of the warp yarn.

[0054] Based on the phase signal of the loom spindle, the stage tension value is subjected to phase expansion and discrete sampling processing to generate the initial phase tension reference curve and phase tension sequence.

[0055] Furthermore, the starting phase, ending phase, and tension value of each stage in the tension value set are read. Starting from 0° of the main axis phase signal, the tension value is linearly mapped to a period from 0° to 360° according to the phase ratio. The mapped tension value is discretely sampled at fixed phase intervals (e.g., 5°) to obtain a set of phase tension points. Cubic spline interpolation is used to continuously fit the set of phase tension points to generate an initial phase tension reference curve. The sampled tension values ​​are arranged in phase order to form a phase tension sequence.

[0056] Based on environmental temperature and humidity parameters, the initial phase tension reference curve is deterministically compensated using a compensation coefficient table to generate the phase tension reference curve.

[0057] Furthermore, ambient temperature and humidity values ​​are read from environmental sensors as ambient temperature and humidity parameters. Using a reference ambient temperature (e.g., 25℃) and a reference ambient humidity (e.g., 50%RH) as fixed references, temperature offset and humidity offset are obtained respectively. For example, when the ambient temperature is 32℃ and the ambient humidity is 68%RH, the temperature offset is 7℃ and the humidity offset is 18%RH. The temperature compensation coefficient and humidity compensation coefficient are read from the compensation coefficient table, and the compensation increment is obtained phase by phase and superimposed on the corresponding tension point. All the compensated tension points are connected in phase order to generate a phase tension reference curve.

[0058] It should be noted that the compensation coefficient table is set based on the measured phase tension drift data of the same batch of glass fiber warp yarn under different environmental temperature and humidity parameters. It includes temperature range, humidity range, temperature compensation coefficient, humidity compensation coefficient and warp yarn batch number. The exemplary temperature compensation coefficient ranges from 0.002 / ℃ to 0.005 / ℃, which represents the relative change rate of tension caused by a unit temperature change. The exemplary humidity compensation coefficient ranges from 0.0008 / %RH to 0.0015 / %RH, which represents the relative change rate of tension caused by a unit humidity change.

[0059] S2. Based on the phase tension baseline curve, construct a phase tension prediction model and iteratively update it through error backpropagation to generate a trained phase tension prediction model.

[0060] A phase tension sample set is established based on the phase tension reference curve and historical phase tension sequence.

[0061] Furthermore, based on the phase tension reference curve and combined with historical phase tension sequences (e.g., 100 cycles), using the same phase tension point in the phase tension reference curve as a reference, the positions of each set of historical phase tension sequences and the corresponding phase points of the phase tension reference curve are aligned to form a reference phase tension vector and a historical phase tension vector. These are then sequentially slid and spliced ​​together according to time to generate a phase tension sample set.

[0062] The input layer is constructed by mapping sample vectors based on the phase tension sample set. A long short-term memory neural network structure is used to build a long short-term memory network layer to capture the temporal dynamic features of tension. A fully connected layer is built through a linear fully connected structure to fuse features and map them to the output space. The output layer is constructed based on the predicted phase tension nodes.

[0063] Furthermore, the baseline phase tension vector and historical phase tension vector in the phase tension sample set are mapped to the input layer nodes according to their corresponding phases (e.g., 10 time steps × 72-dimensional phase features), and the number of input layer nodes is set to be consistent with the vector dimension. A long short-term memory (LSTM) network layer is built using a long short-term memory (LSTM) neural network structure, and the weight matrices of the forget gate, input gate, and output gate are set (e.g., 72 × 128), and the number of hidden layer units is set (e.g., 128). A fully connected layer is built using a linear fully connected structure, and the output vector of the LSTM network layer is connected to the neurons of the fully connected layer by weights. An output layer is constructed based on the predicted phase tension nodes, and the number of output layer nodes is set to be consistent with the length of the phase tension sequence (e.g., 72), which serves as the structural basis of the phase tension prediction model.

[0064] A phase tension prediction model is constructed by using a deep neural network to perform directed interlayer connections between the input layer, long short-term memory network layer, fully connected layer, and output layer through sequential forward connections.

[0065] Furthermore, the input layer maps the phase tension sample vector to the input nodes; the long short-term memory network layer calculates the gating signal through the forget gate, input gate, and output gate using the sigmoid activation function; the candidate cell state is mapped through the tanh activation function; the hidden layer unit state is updated by weighted sum of the gating signal and the candidate state, and the temporal dynamic features of tension are captured in the hidden unit; the fully connected layer fuses the output of the hidden layer and maps it to the output space; the output layer transforms the predicted phase tension sequence through a linear activation function, and forward propagation is used between layers to complete the model construction.

[0066] It should be noted that connection weights refer to the parameter matrix between neurons in adjacent layers of a deep neural network, including the weight matrix from the input layer to the gated units of the long short-term memory network layer, the recursive weight matrix of the hidden units of the long short-term memory network layer, the weight matrix from the output of the hidden layer to the fully connected layer, and the weight matrix from the fully connected layer to the output layer.

[0067] The phase tension sample set is input into the phase tension prediction model, and the connection weights are iteratively updated through the error backpropagation algorithm until the prediction error converges, thus generating the trained phase tension prediction model.

[0068] Furthermore, the phase tension sample set is sequentially input into the phase tension prediction model, and forward propagation is performed to obtain the training output phase tension sequence. The mean square error between the training output phase tension sequence and the reference phase tension vector is calculated. If the mean square error is greater than the convergence threshold, the connection weight gradient is calculated through the error backpropagation algorithm. The Adam optimizer is used to update the connection weights and bias vectors with a learning rate (e.g., 0.001). The forward propagation calculation and weight update process are repeated until the mean square error is less than the convergence threshold, at which point the iteration stops, and the trained phase tension prediction model is generated.

[0069] It should be noted that the backpropagation algorithm refers to an iterative optimization method that uses the mean squared error as the loss function to obtain gradients and update connection weights layer by layer from the output layer to the input layer; the convergence threshold is set according to the inflection point of the mean squared error decrease curve after normalization of the phase tension sample set (e.g., the first 100 cycles), and the exemplary value range is usually 0.001 to 0.01; the Adam optimizer refers to an adaptive moment estimation optimization algorithm that adaptively adjusts the learning rate of each weight parameter (e.g., 0.001) according to the first moment estimation (e.g., 0.9) and the second moment estimation (e.g., 0.999).

[0070] S3. Obtain the real-time phase tension sequence through synchronous sampling, input it into the trained phase tension prediction model to perform forward calculation, and output the real-time tension deviation sequence.

[0071] Tension detection signals are collected from the glass fiber warp path.

[0072] Furthermore, a strain gauge tension sensor is installed at the gap between the guide rollers in the glass fiber warp path. The sampling frequency is set (e.g., 1000Hz), and the phase signal of the loom spindle (e.g., 0° to 360°) is used as the synchronous trigger reference. The instantaneous value of the warp tension is collected at fixed phase intervals (e.g., 5°) as the tension detection signal.

[0073] The tension detection signal is subjected to phase synchronization sampling processing to obtain the real-time phase tension sequence.

[0074] Furthermore, using the phase signal of the loom spindle as the synchronous trigger reference, the timestamps of each sampling point in the tension detection signal are mapped to the corresponding phase angles. The mapped tension values ​​are resampled at fixed phase intervals (e.g., 5°), and the resampled tension values ​​are arranged in ascending phase order to form a real-time phase tension sequence.

[0075] The real-time phase tension sequence is input into the trained phase tension prediction model to perform forward computation and generate the predicted phase tension sequence.

[0076] Furthermore, the real-time phase tension sequence is mapped to the input layer nodes to form an input vector. The input vector is multiplied with the connection weights (e.g., a 72×128 weight matrix) and a bias vector is superimposed to form a superimposed vector. The superimposed vector is then passed to the input of the forget gate in the long short-term memory network layer. The forget gate, input gate, and output gate are gating signals calculated using the sigmoid activation function. The candidate cell states are mapped using the tanh activation function. The hidden layer unit states are updated by weighting the gating signals and candidate states and are connected to the fully connected layer neurons through the output gate weight matrix. Finally, the sequence is passed to the output layer nodes through a linear activation function to generate the predicted phase tension sequence.

[0077] The phase-by-phase difference between the predicted phase tension sequence and the real-time phase tension sequence is calculated to generate the real-time tension deviation sequence.

[0078] Furthermore, the phase tension difference (e.g., 0.7N) is calculated phase by phase by phase between the predicted phase tension sequence and the real-time phase tension sequence according to the phase index. The phase tension difference is then arranged in ascending phase order to form a real-time tension deviation sequence, which serves as the input for phase synchronization segmentation and mechanical disturbance sequence extraction.

[0079] S4. The real-time tension deviation sequence is segmented, and the mechanical disturbance sequence and phase structure disturbance sequence are extracted. The tension compensation requirement is generated by cross-correlation positioning and recursive least squares identification.

[0080] Based on the phase signal of the loom spindle, the real-time tension deviation sequence is phase-synchronized and segmented to form a phase window sequence, and instantaneous displacement and rotation speed values ​​are collected.

[0081] Furthermore, the tension difference value of the phase node in the real-time tension deviation sequence (e.g., 0.7N) is read. Using the phase signal of the loom spindle as the synchronization reference, the phase window is divided according to a fixed phase interval (e.g., 30°) and arranged in ascending phase order to form a phase window sequence. The instantaneous displacement value (e.g., 0.05mm) output by the tension roller displacement sensor and the instantaneous speed value (e.g., 120rpm) output by the warp drive encoder are collected. The instantaneous displacement value and the instantaneous speed value are aligned according to the phase window index and written into the phase window sequence as input objects for the extraction of mechanical disturbance sequence and phase structure disturbance sequence.

[0082] The mechanical disturbance sequence and the phase structure disturbance sequence are extracted from the phase window sequence using exponential moving average filtering and bandpass filtering.

[0083] Furthermore, an exponential moving average filter is applied to the tension difference in the phase window sequence, with a smoothing coefficient set (e.g., 0.1). Low-frequency components are obtained through a recursive formula, and the mechanical disturbance sequence is extracted. Simultaneously, a bandpass filter is applied to the phase window sequence, with the passband frequency range set to be synchronized with the weaving spindle speed (e.g., 120 rpm). Periodic components with the same frequency as the shedding, weft insertion, and beat-up actions are extracted, and the phase structure disturbance sequence is extracted. The mechanical disturbance sequence and the phase structure disturbance sequence are aligned and output according to the phase index, serving as inputs for cross-correlation localization and recursive least squares identification.

[0084] It should be noted that the smoothing coefficient is set according to the tension difference fluctuation amplitude of the phase window sequence (e.g., the first 50 cycles), and the exemplary value range is usually 0.05 to 0.2. It is used to control the tracking response speed of the exponential moving average filter to the low-frequency component of mechanical disturbance. The mechanical disturbance sequence characterizes the tension fluctuation caused by mechanical factors and is used to identify the mechanical disturbance source and achieve decoupling compensation. The phase structure disturbance sequence is used to accurately identify the tension fluctuation caused by the process phase to avoid compensation inaccuracy.

[0085] The radius time delay parameter is obtained by cross-correlation positioning of the mechanical disturbance sequence and the instantaneous value of the rotation speed, and the radius disturbance value of the warp bobbin is calculated by recursive least squares identification.

[0086] Furthermore, based on the tension difference and instantaneous rotational speed of the phase nodes in the mechanical disturbance sequence, cross-correlation calculations are performed according to the phase index alignment to obtain the radius correlation coefficient corresponding to each time delay index. The time delay index corresponding to the peak value of the radius correlation coefficient is selected as the radius time delay parameter (e.g., 3 phase nodes), and the radius disturbance state equation is established. With the radius time delay parameter as the initial estimate, the radius estimate is iteratively updated according to the error gradient using the recursive least squares algorithm to generate the warp bobbin radius disturbance value (e.g., 0.02 mm), which is used to quantify the influence of radius change on tension and provide parameter basis for accurate calculation of warp compensation torque.

[0087] It should be noted that the radius correlation coefficient refers to the linear correlation between the tension difference of the mechanical disturbance sequence and the instantaneous value of the rotational speed under different time delay indices, with an exemplary value range of -1 to 1; the peak value of the radius correlation coefficient refers to the correlation coefficient value with the largest absolute value in the cross-correlation calculation result, corresponding to the position of the strongest synchronization time delay between the mechanical disturbance and the rotational speed signal.

[0088] The friction delay parameters are obtained by cross-correlation positioning of the mechanical disturbance sequence and the instantaneous displacement value, and the friction disturbance value of the guide roller is calculated by recursive least squares identification.

[0089] Furthermore, based on the tension difference and instantaneous displacement value of the phase nodes in the mechanical disturbance sequence, cross-correlation calculation is performed according to the phase index alignment to obtain the friction correlation coefficient corresponding to each time delay index; the time delay index corresponding to the peak value of the friction correlation coefficient is selected as the friction time delay parameter (e.g., 2 phase nodes), and the friction disturbance state equation is established. With the friction time delay parameter as the initial estimate, the friction estimate is iteratively updated according to the error gradient using the recursive least squares algorithm to generate the guide roller friction disturbance value (e.g., 0.03 N·m), which is used to quantify the interference of friction factors on tension and provide parameter support for the additional load compensation of the tension roller.

[0090] It should be noted that the friction correlation coefficient is set based on the synchronicity of the tension difference in the mechanical disturbance sequence and the instantaneous value of the tension roller displacement in the phase window sequence (e.g., the first 50 cycles), with an exemplary value range of -1 to 1.

[0091] A phase window weight set is generated based on the phase structure disturbance sequence, the warp bobbin radius disturbance value, and the guide roller friction disturbance value. Weighted synthesis is then performed according to the phase window weight set to generate the tension compensation requirement.

[0092] Furthermore, the tension difference of the current phase node in the phase structure disturbance sequence is extracted and its absolute value is taken. The ratio of the warp bobbin radius disturbance value to the reference radius tolerance is processed by a normalization function. The ratio of the guide roller friction disturbance value to the reference friction threshold is mapped by a projection function. The absolute value of the tension difference, the normalized radius ratio, and the projected friction ratio are nonlinearly fused to obtain the phase window weight coefficient of the current phase node (e.g., 3.0), which is written into the phase window weight set. According to the phase window weight set, the tension difference in the real-time tension deviation sequence and the weight coefficient in the phase window weight set are weighted and modulated. The time-domain integral of all weighted tension differences is performed to obtain the tension compensation requirement, which serves as the basis for the allocation of warp release torque and tension roller load, thereby achieving precise compensation for multi-source disturbances.

[0093] It should be noted that the reference radius tolerance refers to the process tolerance value (e.g., 0.01 mm) that allows fluctuations in the radius of the warp yarn bobbin, serving as a reference for normalizing the radius disturbance value; the reference friction threshold is set based on the friction characteristics of the guide roller bearing and the measured data of the sliding friction of the warp yarn, and is obtained through the statistical analysis of tension fluctuations during the no-load operation phase, with an exemplary value range of 0.005 N·m to 0.02 N·m.

[0094] The expression for calculating the phase window weighting coefficient is:

[0095] ;

[0096] in, The first in the phase window weight set The weight coefficients of each phase node; For phase index; It is a multidimensional coupled operation function used to achieve nonlinear fusion of tension difference, radius gain, and friction gain; The first in the phase structure perturbation sequence The tension difference between each phase node; This is the normalization function; This represents the disturbance value of the warp yarn bobbin radius; The baseline radius tolerance serves as a reference for normalizing the radius disturbance value. This is a projection function used to map friction disturbance values ​​to a reference friction threshold range; This represents the frictional disturbance value of the guide roller. The reference friction threshold is used as a reference for projecting friction disturbance values.

[0097] S5. Perform phase-weighted allocation and over-limit back-off redistribution on the tension compensation demand to generate the release compensation torque and tension roller additional load, and generate the collaborative control output through collaborative gain fine-tuning to update the phase tension prediction model in reverse.

[0098] A phase window weighted allocation is performed on the warp compensation component in the tension compensation demand, and under the constraint of the upper limit of the warp motor torque, the over-limit allocation value is saturated and back-off redistributed to generate the warp compensation torque.

[0099] Furthermore, the tension compensation demand is read from the warp compensation component, and weighted mapping is performed on each phase node according to the phase window weight set to obtain the phase allocation tension value (e.g., 15.0N), which is then converted into the torque demand value (e.g., 0.8N·m) through the lever arm conversion coefficient; the torque demand value exceeding the upper limit of the warp motor torque is identified, and the over-limit torque demand value (referring to the torque demand value exceeding the upper limit of the warp motor torque) is saturated and truncated to obtain the trimmed torque value. The torque over-limit difference (referring to the difference between the over-limit torque demand value and the upper limit of torque) is preferentially redistributed to the unsaturated phase nodes on the left and right adjacent unsaturated nodes according to the weight ratio, generating the warp compensation torque sequence.

[0100] It should be noted that the lever arm conversion factor is the effective working radius of the tension roller, which is measured by the physical radius of the tension roller. The exemplary value range is usually 0.05m to 0.15m. The upper limit of the torque of the unwinding motor refers to the maximum torque value that the unwinding motor is allowed to output under rated operating conditions. It is read from the parameters on the motor nameplate, for example, 0.6 N·m. Saturation clipping refers to applying nonlinear saturation constraints to the over-limit value that exceeds the upper limit of the torque of the unwinding motor. The Clip function is used to cut the over-limit torque demand value to the upper limit range of the torque.

[0101] The tension roller compensation component in the tension compensation demand is weighted by phase window, and under the constraint of the upper limit of the tension roller additional load, the over-limit allocation value is saturated and redistributed to generate the tension roller additional load.

[0102] Furthermore, the tension roll compensation component in the tension compensation demand is read, and a multidimensional coupling mapping is performed on the phase nodes according to the phase window weight set to obtain the phase allocation load value (e.g., 8.0N), which is then converted into an additional load demand value (e.g., 0.5N·m). Additional load demand values ​​exceeding the upper limit of the tension roll additional load are identified, and nonlinear saturation constraints are applied to the over-limit additional load demand values ​​to obtain the trimmed load value. The load over-limit difference is back-allocated to the neighboring unsaturated nodes according to the phase window weight gradient distribution to generate the tension roll additional load sequence.

[0103] It should be noted that the upper limit of the tension roller additional load refers to the maximum additional load value that the tension roller is allowed to apply under rated operating conditions, which is read from the mechanical structure parameters of the tension roller, for example, 0.4 N·m.

[0104] Using the real-time tension deviation sequence as an error signal, the coordinated gain fine-tuning of the discharge compensation torque and the additional load of the tension roller is performed to generate a coordinated control output.

[0105] Furthermore, the tension difference at the phase node in the real-time tension deviation sequence is used as an error feedback signal. The proportional scaling and phase shift (meaning moving the compensation window index in the hysteresis direction) are performed on the warp compensation torque (e.g., 0.6 N·m) and the tension roller additional load (e.g., 0.4 N·m) to complete the coordinated gain fine-tuning. Based on the phase window weight coefficient and the error gradient, multi-dimensional coupling operation is performed to correct the distribution weight of the warp compensation torque and the tension roller additional load in real time, and generate the coordinated control output.

[0106] Based on the coordinated control output, the deviation between the real-time phase tension sequence and the predicted phase tension sequence is written back to the phase tension sample set, and error backpropagation is performed to update the connection weights of the phase tension prediction model.

[0107] Furthermore, based on the output of coordinated control, the phase-by-phase deviation value between the real-time phase tension sequence and the predicted phase tension sequence is obtained. The deviation value is written to the end of the phase tension sample set according to the phase index to form an online update sample. Error backpropagation is performed using the online update sample to calculate the gradient of the connection weights. The Adam optimizer is used to update the connection weights and bias vectors with a learning rate (e.g., 0.001) to complete the online iterative update of the phase tension prediction model, which serves as the basis for the execution of tension prediction in the next weaving cycle.

[0108] In summary, this invention achieves intelligent prediction and dynamic correction of tension evolution trends within the weaving cycle through a phase tension prediction model; by segmenting the real-time tension deviation sequence and extracting the mechanical disturbance sequence and phase structure disturbance sequence, it achieves decoupling identification and precise compensation of multi-source disturbances, thereby achieving the accuracy and robustness of tension adaptive control and improving the weaving quality and production efficiency of glass fiber cloth.

[0109] 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. A method for adaptive control of weaving tension of glass fiber cloth, characterized in that, include: Based on the warp yarn density parameters and weaving structure parameters, the stage tension value is calculated, and a phase tension reference curve is generated through phase unfolding and environmental temperature and humidity compensation. Based on the phase tension baseline curve, a phase tension prediction model is constructed and iteratively updated through error backpropagation to generate a trained phase tension prediction model. The real-time phase tension sequence is obtained by synchronous sampling and input into the trained phase tension prediction model to perform forward calculation, and the real-time tension deviation sequence is output. The real-time tension deviation sequence is segmented, and the mechanical disturbance sequence and phase structure disturbance sequence are extracted. The tension compensation requirement is generated by cross-correlation positioning and recursive least squares identification. Phase-weighted allocation and over-limit back-off redistribution are performed on the tension compensation demand to generate the warp compensation torque and tension roller additional load. The coordinated control output is generated through coordinated gain fine-tuning and the phase tension prediction model is updated in reverse.

2. The adaptive tension control method for glass fiber cloth weaving as described in claim 1, characterized in that, The steps for calculating the stage tension value based on the warp yarn density parameter and weaving structure parameter are as follows: Based on the warp yarn density parameters and weaving structure parameters, and using the loom spindle phase signal as the basis for dividing the weaving cycle, the cycle is segmented to generate a set of weaving cycle stages. The warp length variation is obtained from the set of weaving cycle stages, and the stage tension value is calculated by combining the warp linear density parameter and the fiber axial tension relationship.

3. The adaptive tension control method for glass fiber cloth weaving as described in claim 1, characterized in that, The phase tension reference curve is generated through phase unwrapping and environmental temperature and humidity compensation, and the steps are as follows: Based on the phase signal of the loom spindle, the stage tension value is subjected to phase expansion and discrete sampling processing to generate the initial phase tension reference curve and phase tension sequence. Based on environmental temperature and humidity parameters, the initial phase tension reference curve is deterministically compensated using a compensation coefficient table to generate the phase tension reference curve.

4. The adaptive tension control method for glass fiber cloth weaving as described in claim 3, characterized in that, The step of constructing a phase tension prediction model based on the phase tension baseline curve and iteratively updating it through error backpropagation to generate a trained phase tension prediction model is as follows: A phase tension sample set is established based on the phase tension reference curve and historical phase tension sequence; The input layer is constructed based on the phase tension sample set to map the sample vector. The long short-term memory neural network structure is used to build the long short-term memory network layer to capture the temporal dynamic features of tension. The fully connected layer is built through the linear fully connected structure to fuse the features and map them to the output space. The output layer is constructed based on the predicted phase tension nodes. A phase tension prediction model is constructed by using a deep neural network to perform directed interlayer connections between the input layer, long short-term memory network layer, fully connected layer, and output layer through sequential forward connections. The phase tension sample set is input into the phase tension prediction model, and the connection weights are iteratively updated through the error backpropagation algorithm until the prediction error converges, thus generating the trained phase tension prediction model.

5. The adaptive tension control method for glass fiber cloth weaving as described in claim 4, characterized in that, The steps for obtaining the real-time phase tension sequence through synchronous sampling are as follows: Tension detection signals are collected from the glass fiber warp path; The tension detection signal is subjected to phase synchronization sampling processing to obtain the real-time phase tension sequence.

6. The adaptive tension control method for glass fiber cloth weaving as described in claim 1 or 5, characterized in that, The input-trained phase tension prediction model performs forward computation and outputs a real-time tension deviation sequence, as follows: The real-time phase tension sequence is input into the trained phase tension prediction model to perform forward computation and generate the predicted phase tension sequence. The phase-by-phase difference between the predicted phase tension sequence and the real-time phase tension sequence is calculated to generate the real-time tension deviation sequence.

7. The adaptive tension control method for glass fiber cloth weaving as described in claim 1, characterized in that, The steps for segmenting the real-time tension deviation sequence and extracting the mechanical disturbance sequence and phase structure disturbance sequence are as follows: Based on the phase signal of the loom spindle, the real-time tension deviation sequence is phase-synchronized and segmented to form a phase window sequence, and instantaneous displacement and rotation speed values ​​are collected. The mechanical disturbance sequence and the phase structure disturbance sequence are extracted from the phase window sequence using exponential moving average filtering and bandpass filtering.

8. The adaptive tension control method for glass fiber cloth weaving as described in claim 7, characterized in that, The process of generating the tension compensation requirement through cross-correlation localization and recursive least squares identification is as follows: The radius time delay parameter is obtained by cross-correlation positioning of the mechanical disturbance sequence and the instantaneous value of the rotation speed, and the radius disturbance value of the warp bobbin is calculated by recursive least squares identification. The friction delay parameters are obtained by cross-correlation positioning of the mechanical disturbance sequence and the instantaneous displacement value, and the friction disturbance value of the guide roller is calculated by recursive least squares identification. A phase window weight set is generated based on the phase structure disturbance sequence, the warp bobbin radius disturbance value, and the guide roller friction disturbance value. Weighted synthesis is then performed according to the phase window weight set to generate the tension compensation requirement.

9. The adaptive tension control method for glass fiber cloth weaving as described in claim 1 or 8, characterized in that, The steps for performing phase-weighted allocation and over-limit retraction reallocation on the tension compensation demand to generate the warp compensation torque and additional load on the tension roller are as follows: A phase window weighted allocation is performed on the warp compensation component in the tension compensation demand, and under the constraint of the upper limit of the warp motor torque, the over-limit allocation value is saturated and back-off redistributed to generate the warp compensation torque. The tension roller compensation component in the tension compensation demand is weighted by phase window, and under the constraint of the upper limit of the tension roller additional load, the over-limit allocation value is saturated and redistributed to generate the tension roller additional load.

10. The adaptive tension control method for glass fiber cloth weaving as described in claim 1, characterized in that, The steps for generating a coordinated control output through coordinated gain fine-tuning and then updating the phase tension prediction model in reverse are as follows: Using the real-time tension deviation sequence as an error signal, the coordinated gain fine-tuning of the release compensation torque and the additional load of the tension roller is performed to generate a coordinated control output. Based on the output of coordinated regulation, the deviation between the real-time phase tension sequence and the predicted phase tension sequence is written back to the phase tension sample set, and error backpropagation is performed to update the connection weights of the phase tension prediction model.

Images