Light ray textile intelligent regulation system

Abstract

The application discloses a light textile intelligent regulation and control system, and relates to the technical field of light textile intelligent regulation and control.The system comprises a collection module, a calculation module and a regulation and control module.The collection module uses a multi-spectral stroboscopic light source array, a switchable polaroid assembly and two heterotopic synchronous trigger image sensors to obtain yarn surface multi-band reflectivity difference, reflectivity phase difference and parallel / vertical polarization response.The calculation module performs height compensation, reconstructs yarn three-dimensional morphology, extracts normalized deviation as an interference confidence factor, and corrects yarn cross-section characteristic quantity.The regulation and control module adopts fractional order active disturbance rejection control and tension fluctuation feedforward compensation, combines adaptive weight evolution and multi-source heterogeneous fusion, and outputs regulation and control instructions.The system solves the problems of low semi-transparent yarn three-dimensional reconstruction accuracy, poor adaptive heterofiber identification and poor viscoelastic drafting control, and significantly improves the accuracy and robustness of yarn quality online monitoring.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for optical textiles, and in particular to an intelligent control system for optical textiles. Background Technology

[0002] Yarn quality is a core indicator in the textile industry, with yarn cross-sectional uniformity, foreign fiber content, and hairiness directly affecting subsequent weaving quality. To achieve online monitoring and closed-loop control of yarn quality, the industry has developed various optical detection and automatic control technologies.

[0003] Existing yarn optical inspection systems typically employ single-band or dual-band light sources in conjunction with one or two image sensors to reconstruct the yarn outline based on stereo vision principles. However, in actual production, yarns are often made of semi-transparent materials (such as cotton, linen, and viscose fibers), where light is scattered and absorbed within the yarn. This results in a systematic deviation in the elevation distribution obtained from stereo parallax calculations, making it difficult to accurately reproduce the true three-dimensional shape of the yarn. To address the elevation distortion caused by semi-transparent materials, some solutions attempt empirical compensation using differences in reflectivity across different bands. However, these compensation functions are often static linear or simple nonlinear, failing to consider the dynamic changes of the yarn during high-speed travel and the twisting and orientation distribution of fibers on the yarn surface, thus limiting the accuracy of the compensation.

[0004] Furthermore, foreign fibers (such as colored fibers and plastic flakes) adhering to the yarn surface are a key defect affecting yarn quality. Existing detection methods are mostly based on the reflectance or transmittance thresholds of a single spectral channel, which are easily affected by fluctuations in the yarn's background material and ambient light interference, leading to high false alarm and false negative rates. Although some systems have introduced multispectral fusion strategies, the weights of each spectral channel are usually fixed or only determined empirically, which cannot adapt to the differences in optical properties of different batches and materials of yarn, and is even more difficult to deal with the joint optimization problem of non-convex false alarm and false negative.

[0005] In yarn drafting control, traditional PID controllers struggle to handle the viscoelastic properties of yarn materials and the large transmission delay between the detection point and the actuator, easily leading to speed overshoot or steady-state oscillations. In recent years, Active Disturbance Rejection Control (ADRC) has been introduced into the textile field; however, existing ADRC models are all integer-order, unable to accurately match the fractional-order viscoelastic constitutive relationship of the yarn, and lack active feedforward compensation for tension fluctuations, resulting in insufficient suppression of sudden speed changes.

[0006] On the other hand, optical detection windows are susceptible to contamination from fly shavings and dust. Existing air curtain protection devices mostly use fixed airflow speeds, which are not dynamically correlated with yarn speed, resulting in reduced protection effectiveness under high yarn speed conditions. At the same time, factors such as ambient light fluctuations and unstable air curtain pressure reduce the confidence level of optical measurements, while existing fusion methods rely only on static environmental parameters, ignoring dynamic operating conditions such as speed fluctuations and polarization stability. Summary of the Invention

[0007] To address the above issues, this invention achieves high-precision reconstruction of the three-dimensional morphology of yarn, adaptive identification of foreign fibers, and fractional-order active disturbance rejection control of the drafting process by coordinating the work of the acquisition module, calculation module, and control module.

[0008] The technical solution is a light-based intelligent control system for textiles, comprising: the acquisition module including a multispectral stroboscopic light source array, two image sensors that are synchronously triggered in different locations, and a switchable polarizer assembly. The multispectral stroboscopic light source array projects first, second, and third center band detection beams in the yarn running channel in a time-division manner. The switchable polarizer assembly alternately acquires spatial scattering response images in parallel polarization and perpendicular polarization directions during each exposure frame. The two image sensors synchronously acquire spatial scattering response images of the yarn surface under each detection beam and different polarization directions, providing a multidimensional light field data stream with spatial coordinates, sampling time sequence, spectral labels, and polarization state labels to the calculation module.

[0009] The calculation module performs stereo parallax calculation based on the multidimensional light field data stream to obtain the initial elevation distribution of the yarn profile. It extracts the reflectivity difference between the first and second band channels and the reflectivity phase difference between the third and first band channels. Based on the parallel and perpendicular polarization responses, it calculates the surface fiber orientation polarization degree. The reflectivity difference, phase difference, and polarization degree are input into a nonlinear compression mapping function based on the yarn dynamic torsion-polarization joint model to obtain elevation compensation values ​​related to the instantaneous yarn torsion angle, surface fiber orientation angle, and real-time travel speed. These elevation compensation values ​​are then superimposed onto the initial... The three-dimensional shape of the yarn profile is reconstructed on the elevation distribution. The elevation compensation value is non-negative, and its rate of approaching the upper limit value increases with the increase of the real-time travel speed of the yarn. It is used to correct the initial elevation distribution upward to approximate the three-dimensional shape of the yarn. The normalized deviation of the reflectance of each band and polarization channel relative to the yarn background reflectance benchmark is extracted as the interference confidence factor characterizing the probability of foreign fiber adhesion. The equivalent geometric features of the yarn main profile are fitted according to the three-dimensional shape. The equivalent geometric features are proportionally reduced and corrected by the interference confidence factor to obtain the corrected yarn cross-sectional feature quantity.

[0010] The control module includes a fractional-order active disturbance rejection control submodule and a tension fluctuation feedforward compensator. The tension fluctuation feedforward compensator predicts the amount of yarn tension fluctuation at future moments and outputs a feedforward compensation signal based on the rate of change of the rotational speed of the upstream drafting mechanism and the transmission delay of the yarn from the detection station to the drafting mechanism. The fractional-order active disturbance rejection control submodule runs an expanded state observer embedded with a fractional-order calculus operator to track the fractional-order lumped disturbance during the yarn drafting process, and generates a control command to control the rotational speed adjustment of the yarn drafting mechanism based on the fractional-order lumped disturbance, the feedforward compensation signal, and the cross-sectional characteristic deviation after the predicted feedforward time constant.

[0011] Furthermore, the nonlinear compression mapping function employs a torsion-polarization joint model to reconstruct the three-dimensional topographic coordinates. The following intermediate variables are given:

[0012] Define normalized reflectance difference and torsion-polarization joint factor for:

[0013]

[0014]

[0015] but:

[0016]

[0017] in, For the initial elevation distribution, , These are the measured reflectivities for the first and second bands, respectively. The phase difference between the reflectance of the third and first wavebands reflects the instantaneous twist angle of the fibers on the yarn surface at that position. For the surface fiber orientation polarization degree, This is a translucency correction factor related to the yarn material. This refers to the real-time travel speed of the yarn. This is the speed coupling adjustment coefficient. The polarization enhancement coefficient, To prevent the loss of small quantities; among which Reflecting the relative difference in reflectivity between the two bands, the value ranges from [-1, 1]. The intensity of elevation compensation is adjusted by combining torsion angle cosine and polarization enhancement.

[0018] Furthermore, the control module includes a fractional-order active disturbance rejection control submodule and a tension fluctuation feedforward compensator;

[0019] The tension fluctuation feedforward compensator predicts the amount of yarn tension fluctuation at future moments based on the rate of change of the rotational speed of the upstream drafting mechanism and the transmission delay of the yarn from the detection station to the drafting mechanism, and outputs a feedforward compensation signal.

[0020] The fractional-order active disturbance rejection control submodule receives the corrected yarn cross-sectional characteristic value and calculates the feedforward time constant using the real-time yarn travel speed and the axial distance from the detection station to the upstream drafting actuator. A discretized fractional-order extended state observer is run to track fractional-order lumped disturbances in the yarn drafting process in real time. The error feedback channel of the fractional-order extended state observer is embedded with a fractional-order calculus operator, whose order... The speed adjustment is then calculated online based on the viscoelastic characteristics of the yarn drafting mechanism within the range of 0.2-0.8. :

[0021]

[0022] in, For speed adjustment amount, The deviation of the cross-sectional characteristics after the predicted feedforward time constant. The fractional derivative of the characteristic deviation of the cross section is given by the Grunwald-Laitnikov discretization approximation. The lumped perturbation estimated by the fractional-order extended state observer. This is the feedforward compensation amount. To control the gain, , These are the proportional and differential coefficients.

[0023] Furthermore, the control module also includes an adaptive weight evolution module, which periodically acquires false alarm event records and missed alarm event records confirmed by the external re-inspection terminal, calculates the false alarm rate and missed alarm rate respectively, and constructs a non-convex loss function in the form of a weighted sum of squares of the two. The gradient descent method with a driving term and an adaptive learning rate is used to iteratively update the weight factors of each spectral channel and polarization channel used to calculate the interference confidence factor. The cumulative amount of the gradient square is updated by an exponential moving average and the decay factor is 0.9-0.99. A gradient cutoff threshold is set to prevent the single step size from being too large. The iteration continues until the decrease of the loss function for three consecutive evaluation cycles is lower than the preset threshold.

[0024] Furthermore, the control module also includes a multi-source heterogeneous fusion submodule, which receives auxiliary observations of yarn cross-sectional characteristics independently measured by non-optical principle sensors, such as capacitive yarn evenness sensors or microwave resonant mass sensors, and calculates the confidence level of optical measurements. This confidence level is obtained by multiplying four parts, including a Gaussian attenuation term for air pressure deviation, a Cauchy attenuation term for ambient light illuminance, a velocity fluctuation rate penalty term, and a polarization stability penalty term. Among them, the velocity fluctuation rate penalty term is the proportion of velocity change rate exceeding the threshold within the sliding window, the polarization stability penalty term is based on the degree of deviation of the average polarization degree from the ideal value within the window, and the confidence level of the auxiliary observation is obtained by converting its signal-to-noise ratio. Then, based on the confidence levels of both, a weighted fusion is performed on the corrected yarn cross-sectional characteristics and the auxiliary observations to give the fused yarn cross-sectional characteristics.

[0025] Furthermore, the acquisition module is equipped with an air curtain protection component at the optical window of the multispectral stroboscopic light source array and the two image sensors that are synchronously triggered in different locations. The air curtain protection component continuously sprays compressed air along the surface of the optical window to form an isolation air curtain, and the spray flow rate is not less than twice the real-time travel speed of the yarn.

[0026] Furthermore, the stroboscopic repetition frequency of the multispectral stroboscopic light source array is determined by dividing the real-time speed of the yarn by the preset sampling spatial interval along the yarn's direction of travel, rounding the resulting ratio up, and then multiplying it by a preset oversampling factor.

[0027] Furthermore, the surface fiber orientation polarization degree is obtained from the parallel polarization response image acquired by the switchable polarizer assembly. and vertical polarization response image Calculate point by point. ,in To prevent the removal of zero values; the surface fiber orientation polarization degree characterizes the degree of orderliness of the fiber orientation along the axial direction on the yarn surface. The closer the value is to 1, the more consistent the fiber orientation is, and the closer it is to 0, the more random the orientation is.

[0028] Furthermore, the reflectivity phase difference The determination is as follows: the spatial distribution sequence of reflectivity for the third band channel. Spatial distribution sequence of reflectivity in the first band channel A bandpass filter is applied along the yarn axis, with the passband range set according to the yarn twist fluctuation frequency. Then, the instantaneous phases are extracted using Hilbert transform. The two instantaneous phases are subtracted and phase unwinding is performed to obtain the final result. .

[0029] Furthermore, an online calibration module for system parameters is also provided. When a preset calibration trigger event occurs, the acquisition module is activated to measure a calibration sample bar with a known nominal size. The calibration trigger event includes system power-on, continuous operation exceeding a set time, or external re-inspection terminal reporting an excessive false alarm rate. The measured cross-sectional characteristic quantity is compared with the nominal size to obtain the deviation. Then, based on the magnitude of the deviation, the semi-transparent correction amplitude coefficient, velocity coupling adjustment coefficient, polarization enhancement coefficient, and fractional order are adjusted in segments. When the deviation is large, a large step size correction is used; when the deviation is moderate, a small step size correction is used; when the deviation is less than the tolerance value, the correction stops. The correction step size of each parameter decreases sequentially according to its influence on the measurement accuracy, and the ratio of the large step size to the small step size is preset to a fixed ratio.

[0030] Due to the adoption of the above technical solutions, the present invention has the following advantages compared with the prior art;

[0031] 1. By collecting the reflectivity differences, reflectivity phase differences, and parallel / perpendicular polarization responses of the first, second, and third bands, and performing elevation compensation, the compensation value is simultaneously correlated with the instantaneous twist angle of the yarn, the surface fiber orientation angle, and the real-time travel speed. This can correct the elevation distortion caused by internal scattering of the semi-transparent material and the high-speed movement of the yarn, and the reconstructed three-dimensional morphology truly reflects the geometric structure of the yarn.

[0032] 2. A gradient descent method with a driving term and adaptive learning rate is adopted. The weight factors of each spectral and polarization channel are periodically updated based on the false alarm rate and false negative rate fed back by the external re-inspection terminal. This allows the interference confidence factor to adapt to the changes in optical properties of different batches and materials of yarn. At the same time, through a fractional-order expansion state observer and a tension fluctuation feedforward compensator, the fractional-order order is identified online based on the viscoelastic properties of the yarn, and the fractional-order constitutive relationship of the yarn is matched. Meanwhile, the tension fluctuation caused by the change in rotation speed is predicted and compensated in advance. The accuracy of foreign fiber identification is continuously optimized, and overshoot and oscillation are effectively suppressed in the drafting control. The overall adaptive capability and robustness of the system are improved. Attached Figure Description

[0033] Figure 1 This is a flowchart of the system modules of the present invention.

[0034] Figure 2 This is a flowchart of the data acquisition module of the present invention.

[0035] Figure 3 This is a flowchart of the calculation module of the present invention.

[0036] Figure 4 This is a flowchart of the control module of the present invention.

[0037] Figure 5 This is a flowchart of the online calibration module for system parameters of the present invention. Detailed Implementation

[0038] The foregoing and other technical contents, features and effects of the present invention are described in conjunction with the appendix below. Figures 1 to 5 The detailed description of the embodiments will make this clear. All structural details mentioned in the following embodiments are based on the accompanying drawings.

[0039] Based on existing technology, in Embodiment 1, this embodiment provides an intelligent control system for light-based textile manufacturing, which includes a data acquisition module, a calculation module, and a control module.

[0040] The acquisition module includes a ring-shaped multispectral stroboscopic light source array, two asynchronously triggered image sensors, and a switchable polarizer assembly mounted in front of these two image sensors. The multispectral stroboscopic light source array projects a first center band detection beam, a second center band detection beam, and a third center band detection beam in a time-division multiplexing manner within the yarn running channel. The switchable polarizer assembly alternately acquires spatial scattering response images in parallel and perpendicular polarization directions during each exposure frame. The two asynchronously triggered image sensors synchronously acquire spatial scattering response images of the yarn surface under each detection beam and different polarization directions, providing the calculation module with a multidimensional light field data stream containing spatial coordinates, sampling time sequence, spectral labels, and polarization state labels.

[0041] After receiving the multidimensional light field data stream, the calculation module performs the following steps.

[0042] The first step involves performing stereo parallax calculation based on the multidimensional light field data stream to obtain the initial elevation distribution of the yarn profile. Stereo parallax calculation utilizes the positional differences of corresponding pixels in images acquired from different viewpoints by two image sensors. Through triangulation, the relative height of each point on the yarn surface is calculated, forming the initial elevation distribution. ,in Represents the spatial coordinates along the yarn axis. This represents the horizontal coordinate perpendicular to the yarn axis.

[0043] The second step involves extracting the reflectivity difference between the first and second band channels, extracting the phase difference between the third and first band channels, and calculating the surface fiber orientation polarization degree based on parallel and perpendicular polarization responses. The reflectivity difference, phase difference, and polarization degree are then fed into a nonlinear compression mapping function based on a yarn dynamic torsion-polarization joint model to obtain elevation compensation values ​​related to the instantaneous yarn torsion angle, surface fiber orientation angle, and real-time travel speed. These elevation compensation values ​​are then superimposed onto the initial elevation distribution to reconstruct the three-dimensional morphology of the yarn profile.

[0044] The reflectivity difference reflects the difference in reflection intensity of light of different wavelengths on the yarn surface. For semi-transparent yarns, short-wavelength (first band) light is more easily scattered by the surface fibers, while long-wavelength (second band) light penetrates deeper. Therefore, the reflectivity difference between the two is related to the fiber density and porosity distribution inside the yarn. The reflectivity phase difference is obtained by comparing the phase shift of the reflected signals of the third band and the first band. This phase difference changes periodically with the twisting of the fibers on the yarn surface, and can therefore be used to characterize the instantaneous twist angle of the fibers on the yarn surface. The surface fiber orientation polarization degree is calculated by the contrast between parallel polarization and perpendicular polarization responses. When the fibers are highly oriented along the yarn axis, the polarization degree is close to 1; when the fibers are randomly distributed, the polarization degree is close to 0.

[0045] The third step involves extracting the normalized deviation of the reflectance of each band and polarization channel relative to the yarn background reflectance benchmark, which serves as an interference confidence factor characterizing the probability of foreign fiber attachment. The yarn background reflectance benchmark is the pre-calibrated average reflectance of each channel in a pure yarn region free of foreign fibers. When foreign fibers attach, their reflectance differs from the yarn background; the larger the normalized deviation, the higher the probability of foreign fiber presence.

[0046] The fourth step involves fitting the equivalent geometric features of the yarn's main contour based on the 3D morphology. Then, the equivalent geometric features are proportionally reduced using an interference confidence factor to obtain the corrected yarn cross-sectional features. During fitting, the least squares method is used to fit the 3D morphology data into an elliptical or circular contour, extracting parameters such as cross-sectional diameter and roundness. The interference confidence factor serves as a correction coefficient; when the probability of foreign fibers is high, the equivalent geometric features are appropriately reduced to eliminate interference from foreign fibers on the cross-sectional measurement.

[0047] The control module receives the corrected yarn cross-sectional characteristic value and generates a feedforward-feedback composite control command that acts on the yarn drafting mechanism.

[0048] Example 2, based on Example 1, uses a torsion-polarization joint model for the nonlinear compression mapping function. The reconstructed three-dimensional topographic coordinates... The following intermediate variables are given.

[0049] Define normalized reflectance difference for:

[0050]

[0051] In this formula, The measured reflectivity of the first band (shortwave) channel. This represents the measured reflectance of the second band (long wave) channel. The numerator is calculated as the difference between the reflectances of the two bands, and the denominator is calculated as the sum of the reflectances of the two bands plus a very small positive number. (The value can be set to) (), to prevent the denominator from being zero. The value range is [-1, 1]. When the long-wave reflectivity is significantly higher than the short-wave reflectivity... A value close to 1 indicates strong internal scattering of the yarn, requiring significant elevation compensation; when the short-wave reflectivity is higher than the long-wave reflectivity, A negative value indicates that scattering from the yarn surface is dominant, and the compensation value decreases accordingly.

[0052] when When ≈1: This situation corresponds to That is, the long-wave reflectivity is much greater than the short-wave reflectivity.

[0053] Typical operating condition example: Dark-colored or high-density semi-transparent yarns. For example, pure cotton yarn dyed with dark dyes, or blended yarns containing a high proportion of synthetic fibers. In such yarns, short-wavelength blue light, due to its short wavelength and high energy, is more easily absorbed by the colored substances or high-density media on the yarn surface, leading to… The value is extremely low; while long-wavelength red light has stronger penetrating power, able to penetrate the surface and be scattered back to the sensor by the internal fibers of the yarn, therefore... Relatively high. At this time... The value approaching 1 indicates that the elevation compensation mainly relies on long-wavelength information. The system determines that internal scattering of the yarn is dominant, the initial elevation distribution is seriously low, and a larger elevation compensation value is required.

[0054] when When ≈0: This situation corresponds to ≈ That is, the reflectivity of the two bands is similar.

[0055] Typical example: Natural or light-colored neutral yarns. For example, undyed off-white pure cotton yarn or light-colored viscose yarn. In such yarns, the fibers have weak and flat absorption of visible light wavelengths, and the surface diffuse reflection intensity of blue and red light is similar. A value close to 0 indicates that the yarn material is uniform, and the initial elevation distribution obtained from the stereo parallax calculation is relatively accurate. The exponential term... Approaching 1, the overall elevation compensation value (1-1)≈0, no additional compensation is needed.

[0056] when When ≈-1: This situation corresponds to That is, the short-wave reflectivity is much greater than the long-wave reflectivity.

[0057] Typical operating condition example: light-colored yarns with high hairiness or rough surfaces. For example, light-colored linen yarns or low-twist viscose yarns with abundant surface hairiness and a loose structure. In such yarns, short-wavelength blue light undergoes strong multiple scattering between the loose surface fibers, and most of the blue light energy is reflected back to the sensor. Extremely high; while long-wave red light easily penetrates the fluffy surface layer into the yarn interior and is absorbed or scattered and attenuated, resulting in Relatively low. At this time... It approaches -1, although the squared terms This still causes the elevation compensation value to tend to saturate, but combined with the torsion-polarization joint factor... Due to physical constraints, the system can identify that this is a surface structure anomaly rather than internal scattering enhancement, thus avoiding over-correction of apparent elevation changes caused by hairiness and maintaining the accuracy of restoring the true geometric features of the yarn's main contour.

[0058] in, To prevent the value from being divided by zero, a typical value is taken as follows: Its function is to prevent numerical calculation overflow caused by a denominator of zero. Much smaller than the effective reflectivity signal (typically (up to the order of 1), will not affect The numerical results have a perceptible impact;

[0059] Define the torsion-polarization joint factor for:

[0060]

[0061] in, The reflectance phase difference between the third and first bands is used to characterize the instantaneous twist angle of the fibers on the yarn surface at that location. When the yarn twists, the angle of the fiber direction relative to the incident light changes, causing a shift in the phase of the reflected light. This instantaneous phase can be extracted using the Hilbert transform to obtain the reflectance. Cosine function The function is as follows: when the twist angle is 0° or 180°, the cosine value is 1, and the compensation strength is the greatest; when the twist angle is 90°, the cosine value is 0, and the compensation strength is zero. This conforms to the physical law that the surface structure changes the most and the elevation compensation demand is the strongest when the yarn twists the most violently.

[0062] The surface fiber orientation polarization degree is calculated as shown in Example 5. When the fibers are highly oriented... The factor is close to 1, and close to 0 when the fibers are disordered. Used to enhance the effect of polarization degree on compensation, among which This is the polarization enhancement factor, which is positive. When the degree of polarization is large, this factor is greater than 1, increasing the compensation intensity; when the degree of polarization is small, this factor is close to 1, and the compensation intensity mainly depends on the difference in reflectivity and the torsion angle. Using the square root form can avoid excessive amplification of the compensation value and maintain stability.

[0063] The reconstructed three-dimensional topographic coordinates for:

[0064]

[0065] in, For the initial elevation distribution, The semi-transparency correction factor is related to the yarn material (obtained by offline calibration; the value varies for different materials such as pure cotton, polyester, and blends). This refers to the real-time travel speed of the yarn. This is the speed coupling adjustment coefficient.

[0066] in the formula Square both the reflectivity difference and the torsion-polarization joint factor to ensure that the parameters within the exponent are non-negative. (Exponential term) It decreases as the squared term increases, therefore The value increases with the square term, achieving nonlinear saturation compensation: when the reflectivity difference is large and the torsion angle is appropriate, the compensation value approaches [value missing]. When the difference in reflectivity is small or the torsion angle is close to 90°, the compensation value approaches 0. Velocity coupling adjustment coefficient. This means that the faster the yarn speed, the faster the exponential decay, and the easier it is for the compensation value to saturate. This is consistent with the fact that the reflected signal from the yarn surface is less stable under high-speed motion.

[0067] in Reflecting the relative difference in reflectivity between the two bands, the value ranges from [-1, 1]. The intensity of elevation compensation is adjusted by combining the cosine of the torsion angle and polarization enhancement. It should be noted that because the yarn is a semi-transparent material (such as cotton, linen, viscose fiber, etc.), light is scattered and absorbed within the yarn, leading to variations in the initial elevation distribution obtained based on stereo parallax calculations. The initial elevation is lower than the true elevation of the yarn surface. In other words, the initial elevation reflects the equivalent position of the scattering center inside the yarn, rather than the true height of the yarn's outer contour. Therefore, the elevation compensation value of this system is non-negative and is only used to correct the initial elevation upwards to approximate the true three-dimensional shape of the yarn. For surface protrusions caused by local structures such as foreign fibers and hairiness, their elevation information has already been directly reflected in the initial elevation distribution through stereo parallax, and no additional downward correction is needed. Therefore, the compensation value is always non-negative, which is consistent with the optical imaging mechanism of semi-transparent yarns.

[0068] During continuous high-speed yarn inspection, the reflected signal acquired by the image sensor within a certain exposure time window is the integral effect of the yarn surface micro-element within that window, and the yarn travel speed... The faster the speed, the longer the equivalent scattering path length of the photons inside the yarn, and the more significant the signal attenuation. Therefore, it is necessary to dynamically adjust the nonlinear saturation speed compensation according to the speed. In the formula... As a time scalar and space function Multiplication, in its physical sense, means that the greater the speed, the faster the decay in the exponential term, and the easier it is for the compensation value to approach the equilibrium value. This simulates the physical law that the reflected signal weakens overall under high-speed motion, requiring stronger compensation. This design enables elevation compensation to adapt to the real-time running speed of the yarn, avoiding the failure of static compensation under variable speed conditions.

[0069] In continuous yarn inspection, the image sensor is constantly... Collect spatial positions on the yarn surface Difference in reflectivity and torsion-polarization joint factor At this time, the real-time speed of the yarn It is a global scalar that directly affects the equivalent scattering path length of photons inside the semi-transparent yarn within the exposure time window: the faster the speed, the longer the scattering path, and the more significant the signal attenuation.

[0070] in the formula and Multiplication, in its mathematical essence, is the uniform weighted modulation of the structural characteristics of each point in space using the velocity at the current moment. This is because all variables are defined at the same sampling time. This multiplication is dimensionally valid (both are dimensionless or have compatible dimensions). For clarity, the intermediate variable can be... = Then the compensation term becomes This form indicates that speed As an exponential decay rate factor, it controls the compensation value as a function of structural characteristics. The rate of increase and saturation varies; the greater the rate of increase, the faster the decay, and the easier it is for the compensation value to reach its upper limit. This allows for adaptive adaptation to the stronger compensation requirements under high-speed operating conditions.

[0071] In the discretization implementation, the yarn axial coordinate With sampling time satisfy Therefore, the space function , The measured values ​​already implicitly include the cumulative effect of historical speeds. The use of explicit speed modulation in this formula is a reasonable and effective engineering approximation.

[0072] Example 3: Based on Example 2, the control module includes a fractional-order active disturbance rejection control submodule and a tension fluctuation feedforward compensator.

[0073] The tension fluctuation feedforward compensator predicts the amount of yarn tension fluctuation at future moments based on the rate of change of the upstream drafting mechanism's rotational speed and the transmission delay of the yarn from the detection station to the drafting mechanism, and outputs a feedforward compensation signal. Specifically, let the rotational speed of the upstream drafting mechanism be... Its rate of change The yarn transfer distance from the inspection station to the drafting mechanism is... The real-time travel speed of the yarn is Then the transmission delay Tension fluctuation prediction value and It is directly proportional, and the proportionality coefficient is determined by the elastic modulus of the yarn. The feedforward compensator converts this predicted value into a feedforward compensation signal. Send to the controller.

[0074] The fractional-order active disturbance rejection control submodule receives the corrected yarn cross-sectional characteristic. The feedforward time constant is calculated using the real-time yarn travel speed and the axial distance from the detection station to the upstream drafting actuator. This time constant is used to predict the future. The deviation in time is used to overcome detection delay.

[0075] A discretized fractional-order extended state observer is used to track fractional-order lumped disturbances during yarn drafting in real time. The discretized recursive form of the extended state observer is as follows:

[0076] Let the sampling period be The current time is The observer state variables include: estimated cross-sectional features. Estimated rate of change of cross-sectional characteristics Estimated lumped disturbance Observation error ,in These are the measured corrected yarn cross-sectional characteristic quantities.

[0077] The observer recursive formula is:

[0078]

[0079] in, The observer gain coefficient, To control the gain, This represents the speed adjustment amount at the current moment. (Nonlinear correction function) The definition is:

[0080]

[0081] The difference between the fractional-order extended state observer and the traditional integer-order observer lies in the fact that a fractional-order calculus operator is embedded in the error feedback channel. In this embodiment, the fractional order... Online identification is performed within the range of 0.2 to 0.8. The identification method employs a recursive least squares approach based on output error: the input and output data of the yarn drafting mechanism are fitted with a fractional transfer function model, and the model is continuously updated with the goal of minimizing the model error. The value allows the model to accurately match the viscoelastic properties of the yarn. The stress-strain relationship of viscoelastic materials is usually expressed as a fractional constitutive model (fractional Kelvin-Voigt model), so using a fractional observer can obtain a more accurate estimate of the perturbation.

[0082] The controller is based on the feedforward time constant. Perturbation estimator output by fractional-order extended state observer The feedforward compensation amount output by the tension fluctuation feedforward compensator And the predicted deviation of the corrected yarn cross-sectional characteristics from the target value, resulting in a speed adjustment. :

[0083]

[0084] In the formula, The deviation of the cross-sectional characteristics after the predicted feedforward time constant, i.e. ,in For the target cross-section characteristic quantity, Predicted by extrapolating current measurements and velocity. The fractional derivative of this deviation is used, and the Grünwald-Letnikov discretization approximation is adopted. The fractional proportional-differential control is combined with disturbance estimation and feedforward compensation, which effectively overcomes the phase lag and large delay disturbance caused by yarn viscoelasticity.

[0085] In Example 4, based on Example 3, the adaptive weight evolution module periodically acquires false alarm event records and missed alarm event records confirmed by the external re-inspection terminal. The external re-inspection terminal refers to a terminal device that manually or offline re-inspects the yarn quality and records the results. A false alarm event is defined as an event where the system reports the presence of foreign fibers but re-inspection confirms the absence of foreign fibers; a missed alarm event is defined as an event where the system does not report foreign fibers but re-inspection confirms their presence.

[0086] The modules respectively calculate the false alarm rate within the current period. and underreporting rate Construct a nonconvex loss function:

[0087]

[0088] in The preset weighting coefficients, ranging from 0 to 1, are used to balance the importance of false positive and false negative rates. Larger values ​​are used when users prioritize reducing false positives, and smaller values ​​are used when they prioritize reducing false negatives. This loss function is non-convex because... and The mapping relationship between the weighting factor and the weighting factor has multiple local minima.

[0089] The gradient descent method with a driving term and an adaptive learning rate is used to iteratively update the weight factors of each spectral channel and polarization channel used to calculate the interference confidence factor. Let the weight vector be... This includes the weights for each band channel and polarization channel. The update rule is:

[0090]

[0091] in The initial learning rate is 0.01. The gradient of the loss function. The momentum coefficient is taken as 0.9. To prevent division by zero constant (take) Cumulative amount of the squared gradient. Updated using exponential moving average:

[0092]

[0093] Where the attenuation factor Use a value between 0.9 and 0.99. The exponential moving average makes the estimate of the squared gradient smoother, avoiding drastic fluctuations in the learning rate. Simultaneously, set a gradient cutoff threshold. (Take 10), when the gradient magnitude exceeds When scaling the gradient to Direction, to prevent excessively large single update steps from causing weight oscillations. Iteration continues until the decrease in the loss function over three consecutive evaluation periods is below a preset threshold (e.g., ...). ).

[0094] The multi-source heterogeneous fusion submodule receives auxiliary observations of yarn cross-sectional characteristics independently measured by non-optical sensors. These non-optical sensors can be capacitive yarn evenness sensors or microwave resonant mass sensors. Capacitive sensors obtain cross-sectional mass information by measuring the capacitance change as the yarn passes through the electrodes; microwave resonant sensors obtain density information by measuring the perturbation of the resonant cavity frequency by the yarn. Both types of sensors are unaffected by optical window contamination and changes in illumination, but their response speed is relatively slow.

[0095] Optical measurement confidence level The result is obtained by multiplying the four parts together:

[0096]

[0097] Among them, the pressure deviation from Gaussian attenuation term , This represents the real-time air pressure of the air curtain at the optical window. The preset air pressure reference threshold (e.g., 0.2 MPa) is used. The standard deviation of the air pressure measurement is given (e.g., 0.02 MPa). The confidence level decreases by a Gaussian function when the actual air pressure deviates from the reference value.

[0098] Ambient light intensity Cauchy attenuation term , Real-time ambient light illuminance This is a preset illuminance reference threshold (e.g., 100 lux). The higher the illuminance, the larger the denominator, and the lower the confidence level. The attenuation follows a Cauchy distribution (long-tailed distribution), which is suitable for scenarios where ambient light may cause severe interference.

[0099] Speed ​​volatility penalty ,in The rate of change of yarn speed. The threshold for the rate of change of velocity (e.g., 0.5 m / s²). The length of the sliding window (e.g., 1 second). This is an indicator function (1 if the condition is true, 0 otherwise). This penalty term represents the proportion of the rate of change of velocity exceeding a threshold within a statistical window; the higher the proportion, the greater the decrease in confidence, and the penalty coefficient. Take 0.3.

[0100] Polarization stability penalty term , This represents the average polarization degree of the surface fiber orientation within the current time window. When the average polarization degree is close to 1, the penalty term is close to 1; when the average polarization degree is close to 0 (fiber orientation disorder), the penalty term is [value missing]. Penalty coefficient Take 0.2.

[0101] Confidence of auxiliary observations The signal-to-noise ratio (SNR) is defined as the ratio of signal power to noise power. When the SNR is very high, the confidence level is close to 1, and when the SNR is very low, the confidence level is close to 0.

[0102] Finally, a weighted fusion of the optical measurement results and auxiliary observations is performed based on two confidence levels:

[0103]

[0104] in This refers to the corrected yarn cross-sectional characteristic quantity. To assist in observation, The fused yarn cross-sectional characteristic quantities are output to subsequent control stages.

[0105] In Example 5, based on Example 4, the online calibration module for system parameters activates the acquisition module to measure a calibration sample bar of known nominal size when a preset calibration trigger event occurs. Calibration trigger events include system power-on, continuous operation exceeding a set duration (e.g., 8 hours), or an external verification terminal reporting a false alarm rate exceeding a preset limit (e.g., 5%). The calibration sample bar is a high-precision machined metal or ceramic rod with known diameter and roundness, and its surface has diffuse reflection characteristics similar to yarn.

[0106] The measured cross-sectional characteristic quantities Compared with nominal size The deviation obtained from the comparison is denoted as . (Note the difference between this symbol and normalized reflectance) (The meaning is different; here it is only a temporary representation of deviation.) Deviation .in accordance with Size segmentation adjustment of four parameters: semi-transparency correction coefficient Speed ​​coupling adjustment coefficient Polarization enhancement coefficient and fractional order The segmentation rules are as follows:

[0107] when If the deviation is considered too large, a large step size correction is adopted: , , , The step size decreases in descending order of impact: It has the greatest impact on cross-sectional measurements and has the largest step size. Minimal impact, smallest step size.

[0108] when When the deviation is considered to be moderate, a small step size correction is adopted: the above step sizes are multiplied by 0.5 (i.e., half of the large step size).

[0109] when If the deviation is considered to be within the tolerance range, correction is stopped.

[0110] The ratio of the large step size to the small step size is preset to a fixed ratio (2:1 in this embodiment), which can be adjusted according to the actual system characteristics. After correction, the measurement is repeated until the deviation converges.

[0111] The acquisition module also incorporates an air curtain protection component at the optical windows of the multispectral stroboscopic light source array and the two asynchronously triggered image sensors. This component includes a compressed air source, nozzles, and a pressure sensor. The compressed air source continuously injects compressed airflow along the surface of the optical window through the nozzles, forming an isolation air curtain that blows away pollutants such as fly shavings and dust from the window surface. (Jet flow rate...) Set to no less than the real-time yarn travel speed twice as much as, that is This design ensures that fiber filaments carried by the high-speed moving yarn will not break through the air curtain and reach the window.

[0112] Strobe repetition frequency of multispectral strobe light source array Determined as follows: Let the real-time speed of the yarn be... The preset sampling spatial interval along the yarn travel direction is (e.g., 1mm), then the fundamental frequency is To avoid aliasing, the ratio is rounded up to obtain an integer. Then multiply by the preset oversampling factor. (Usually taken as 2), that is Oversampling increases spatial sampling density, which is beneficial for more precise reconstruction of yarn surface details.

[0113] Surface fiber orientation polarization Parallel polarization response image acquired from a switchable polarizer assembly and vertical polarization response image Point-by-point calculation:

[0114]

[0115] in To prevent small amounts from being divided by zero (take) This formula is based on the principle of polarization difference: when the fiber is oriented in a parallel direction, the parallel polarization component is strong and the perpendicular component is weak, the numerator is close to its maximum value, and the denominator is close to its maximum value. When the fiber is randomly oriented, the polarization degree is close to 1; when the fiber is randomly oriented, the two component intensities are similar, the molecular intensities are close to 0, and the polarization degree is close to 0.

[0116] Reflectivity phase difference The determination method is as follows: the spatial distribution sequence of reflectivity of the third band channel. Spatial distribution sequence of reflectivity in the first band channel A bandpass filter is applied along the yarn axis. The passband range of the bandpass filter is set according to the yarn twist fluctuation frequency, typically 0.5 revolutions / meter to 5 revolutions / meter. After filtering, the frequency components related to yarn twist are retained, while high-frequency noise and low-frequency trend terms are filtered out. Then, Hilbert transforms are performed on the two filtered sequences to obtain their respective analytical signals, and the amplitude of the analytical signals is extracted as the instantaneous phase. The phase difference is obtained by subtracting the two instantaneous phases, and then phase unwinding processing is performed (when the phase difference between adjacent points exceeds...). At that time, by adding or subtracting (Make it continuous), that is, .

[0117] In specific use, this invention, based on existing technology, involves increasing the speed of the yarn. During the process, the multispectral strobe source array is lit in time-sharing according to the first, second, and third center bands. The switchable polarizer assembly alternately switches to parallel polarization and vertical polarization direction during each frame exposure. Two image sensors that are triggered synchronously in different locations simultaneously acquire multidimensional light field data streams containing spatial coordinates, sampling time sequence, spectral labels, and polarization state labels, and transmit them to the computing module in real time.

[0118] The calculation module first performs stereo parallax calculation on the multidimensional light field data stream to obtain the initial elevation distribution of the yarn profile. Subsequently, the reflectivity difference between the first and second band channels, the reflectivity phase difference between the third and first band channels, and the surface fiber orientation polarization degree were extracted. These parameters were substituted into the torsion-polarization joint exponential decay function to calculate the elevation compensation value, which was then superimposed onto the initial elevation distribution to reconstruct the three-dimensional morphology of the yarn profile. The compensation process fully considers the scattering characteristics of the semi-transparent material, the signal attenuation caused by the high-speed movement of the yarn, and the influence of fiber twisting and orientation distribution on reflection, so that the reconstructed three-dimensional morphology can truly reflect the geometric structure of the yarn.

[0119] Based on this, the calculation module extracts the normalized deviation of the reflectance of each band and polarization channel relative to the yarn background reflectance benchmark, and obtains the interference confidence factor. This factor is used to perform proportional reduction correction on the fitted equivalent geometric features, thereby eliminating the interference of foreign fiber adhesion on the cross-section measurement and outputting the corrected yarn cross-section feature quantity.

[0120] The control module receives the corrected yarn cross-sectional characteristic value and simultaneously uses the tension fluctuation feedforward compensator to predict the tension fluctuation caused by the change in the speed of the upstream drafting mechanism, generating a feedforward compensation signal. The fractional-order active disturbance rejection control submodule operates a fractional-order extended state observer to estimate the fractional-order lumped disturbance in the yarn drafting process in real time. And control the generation of speed regulation amount The driving yarn drafting mechanism adjusts its rotational speed to stabilize the yarn cross-sectional characteristics near the target value, due to the fractional order. Based on the online identification of yarn viscoelastic properties, it can accurately match the fractional constitutive relation of the yarn, which significantly reduces overtuning and oscillations compared to the integer-order ADRC.

[0121] Meanwhile, the adaptive weight evolution module periodically updates the weight factors of each spectral channel and polarization channel using a gradient descent method with a driving quantity and adaptive learning rate based on the false alarm rate and false negative rate fed back by the external re-inspection terminal. This allows the calculation of the interference confidence factor to adapt to the changes in the optical characteristics of different batches and different materials of yarn, continuously optimizing the accuracy of foreign fiber identification. The multi-source heterogeneous fusion submodule performs weighted fusion of optical measurement results with auxiliary observations from capacitive or microwave sensors. The fusion weight is dynamically determined by the confidence of optical measurement and the confidence of auxiliary observation. The confidence of optical measurement integrates four factors: air curtain pressure, ambient light intensity, yarn speed fluctuation rate, and average polarization degree within the window. This ensures that the confidence level of optical measurement is automatically reduced under conditions such as unstable air curtain pressure, sudden changes in ambient light, drastic changes in yarn speed, or chaotic fiber orientation, thereby ensuring the reliability of the fusion output.

[0122] The online calibration module automatically initiates the measurement process for the calibration sample bar when the system is powered on, runs continuously for more than a set time, or when the false alarm rate reported by the external verification terminal exceeds the limit. It compares the measured cross-sectional characteristic quantities with the nominal dimensions and adjusts the semi-transparent correction amplitude coefficient in segments according to the magnitude of the deviation. Speed ​​coupling adjustment coefficient Polarization enhancement coefficient and fractional order The ratio between the large step size and the small step size is preset to a fixed ratio, which enables the parameters to converge quickly to the maximum value, overcoming the performance drift caused by optical device aging and window contamination after long-term system operation.

[0123] The air curtain protection component continuously sprays compressed air along the surface of the optical window, with a jet velocity no less than twice the real-time speed of the yarn. This effectively prevents fly shavings and dust from adhering to the optical window, ensuring image quality during long-term continuous operation of the acquisition module. The stroboscopic repetition frequency of the multispectral stroboscopic light source array is determined by multiplying the ratio of the real-time yarn speed to the preset sampling spatial interval by an oversampling factor, ensuring that sufficiently dense spatial sampling points are obtained at different yarn speeds, thus avoiding aliasing distortion in morphology reconstruction.

[0124] The above description is a further detailed explanation in conjunction with specific embodiments, and it should not be considered that the specific embodiments of the present invention are limited to this. For those skilled in the art to which this invention pertains and related fields, any extensions, operation methods, and data substitutions made based on the technical solution concept of this invention should fall within the protection scope of this invention.

Claims

1. A light-based intelligent control system for textile manufacturing, characterized in that, include: The acquisition module includes a multispectral stroboscopic light source array, two image sensors that are synchronously triggered in different locations, and a switchable polarizer assembly. The multispectral stroboscopic light source array projects detection beams of the first, second, and third center bands in a time-division manner within the yarn running channel. The switchable polarizer assembly alternately acquires spatial scattering response images of parallel polarization and perpendicular polarization directions during each exposure frame. The two image sensors synchronously acquire spatial scattering response images of the yarn surface under each detection beam and different polarization directions, providing the calculation module with a multidimensional light field data stream containing spatial coordinates, sampling time sequence, spectral labels, and polarization state labels. The calculation module obtains the initial elevation distribution of the yarn profile by performing stereo parallax calculation based on the multidimensional light field data stream. It extracts the reflectivity difference between the first and second band channels and the reflectivity phase difference between the third and first band channels. It also calculates the surface fiber orientation polarization degree based on the parallel polarization and perpendicular polarization responses. The reflectivity difference, phase difference, and polarization degree are input into a nonlinear compression mapping function based on the yarn dynamic torsion-polarization joint model to obtain elevation compensation values ​​related to the instantaneous torsion angle of the yarn, the surface fiber orientation angle, and the real-time travel speed. The elevation compensation values ​​are superimposed on the initial elevation distribution to reconstruct the three-dimensional morphology of the yarn profile. The normalized deviation of the reflectivity of each band and polarization channel relative to the yarn background reflectivity benchmark is extracted as an interference confidence factor characterizing the probability of foreign fiber adhesion. The equivalent geometric features of the yarn main profile are fitted according to the three-dimensional morphology. The equivalent geometric features are proportionally reduced and corrected by the interference confidence factor to obtain the corrected yarn cross-sectional feature quantities. The control module includes a fractional-order active disturbance rejection control submodule and a tension fluctuation feedforward compensator. The tension fluctuation feedforward compensator predicts the amount of yarn tension fluctuation at future moments and outputs a feedforward compensation signal based on the rate of change of the rotational speed of the upstream drafting mechanism and the transmission delay of the yarn from the detection station to the drafting mechanism. The fractional-order active disturbance rejection control submodule runs an expanded state observer embedded with a fractional-order calculus operator to track the fractional-order lumped disturbance during the yarn drafting process, and generates a control command to control the rotational speed adjustment of the yarn drafting mechanism based on the fractional-order lumped disturbance, the feedforward compensation signal, and the cross-sectional characteristic deviation after the predicted feedforward time constant.

2. The intelligent control system for optical fiber textile manufacturing according to claim 1, characterized in that... The nonlinear compression mapping function employs a torsion-polarization joint model to reconstruct the three-dimensional topographic coordinates. The following intermediate variables are given: Define normalized reflectance difference and torsion-polarization joint factor for: but: in, For the initial elevation distribution, , These are the measured reflectivities for the first and second bands, respectively. The phase difference between the reflectance of the third and first wavebands reflects the instantaneous twist angle of the fibers on the yarn surface at that position. For the surface fiber orientation polarization degree, This is a translucency correction factor related to the yarn material. This refers to the real-time speed of the yarn. This is the speed coupling adjustment coefficient. The polarization enhancement coefficient, To prevent the loss of small quantities; among which Reflecting the relative difference in reflectivity between the two bands, the value ranges from [-1, 1]. The intensity of elevation compensation is adjusted by combining torsion angle cosine and polarization enhancement.

3. The intelligent control system for optical fiber textile manufacturing according to claim 1, characterized in that, The control module includes a fractional-order active disturbance rejection control submodule and a tension fluctuation feedforward compensator. The tension fluctuation feedforward compensator predicts the amount of yarn tension fluctuation at future moments based on the rate of change of the rotational speed of the upstream drafting mechanism and the transmission delay of the yarn from the detection station to the drafting mechanism, and outputs a feedforward compensation signal. The fractional-order active disturbance rejection control submodule receives the corrected yarn cross-sectional characteristic value and calculates the feedforward time constant using the real-time yarn travel speed and the axial distance from the detection station to the upstream drafting actuator. A discretized fractional-order extended state observer is run to track fractional-order lumped disturbances in the yarn drafting process in real time. The error feedback channel of the fractional-order extended state observer is embedded with a fractional-order calculus operator, whose order... The speed adjustment is then calculated online based on the viscoelastic characteristics of the yarn drafting mechanism within the range of 0.2-0.

8. : in, For speed adjustment amount, The deviation of the cross-sectional characteristics after the predicted feedforward time constant. The fractional derivative of the characteristic deviation of the cross section is given by the Grunwald-Laitnikov discretization approximation. The lumped perturbation estimated by the fractional-order extended state observer. This is the feedforward compensation amount. To control the gain, , These are the proportional and differential coefficients.

4. The intelligent control system for optical fiber textile manufacturing according to claim 1, characterized in that, The control module also includes an adaptive weight evolution module, which periodically acquires false alarm event records and missed alarm event records confirmed by the external re-inspection terminal, calculates the false alarm rate and missed alarm rate respectively, and constructs a non-convex loss function in the form of a weighted sum of squares of the two. The gradient descent method with a driving term and an adaptive learning rate is used to iteratively update the weight factors of each spectral channel and polarization channel used to calculate the interference confidence factor. The cumulative amount of the gradient square is updated by exponential moving average and the decay factor is 0.9-0.

99. A gradient cutoff threshold is set to prevent the single step size from being too large. The iteration continues until the decrease of the loss function for three consecutive evaluation cycles is lower than the preset threshold.

5. The intelligent control system for optical fiber textiles according to claim 1, characterized in that, The control module also includes a multi-source heterogeneous fusion submodule, which receives auxiliary observations of yarn cross-sectional characteristics independently measured by non-optical sensors, such as capacitive yarn evenness sensors or microwave resonant mass sensors. It calculates the confidence level of the optical measurement, which is obtained by multiplying four parts: a Gaussian attenuation term for air pressure deviation, a Cauchy attenuation term for ambient light illuminance, a velocity fluctuation penalty term, and a polarization stability penalty term. The velocity fluctuation penalty term calculates the proportion of velocity change exceeding a threshold within a sliding window, the polarization stability penalty term is based on the degree to which the average polarization degree deviates from the ideal value within the window, and the confidence level of the auxiliary observation is calculated from its signal-to-noise ratio. Then, based on the confidence levels of both, a weighted fusion is performed on the corrected yarn cross-sectional characteristics and the auxiliary observations to provide the fused yarn cross-sectional characteristics.

6. The intelligent control system for optical fiber textile manufacturing according to claim 5, characterized in that, The acquisition module is also equipped with an air curtain protection component at the optical window of the multispectral stroboscopic light source array and the two image sensors that are synchronously triggered in different positions. The air curtain protection component continuously sprays compressed air along the surface of the optical window to form an isolation air curtain, and the spray flow rate is not less than twice the real-time travel speed of the yarn.

7. The intelligent control system for optical fiber textile manufacturing according to claim 1, characterized in that, The stroboscopic repetition frequency of the multispectral stroboscopic light source array is determined by dividing the real-time speed of the yarn by the preset sampling spatial interval along the yarn's direction of travel, rounding the resulting ratio up, and then multiplying it by the preset oversampling factor.

8. The intelligent control system for optical fiber textiles according to claim 2, characterized in that, The surface fiber orientation polarization degree is obtained from the parallel polarization response image acquired by the switchable polarizer assembly. and vertical polarization response image Calculate point by point. ,in To prevent small quantities from being reduced to zero; The surface fiber orientation polarization degree characterizes the degree of orderliness of the fiber orientation along the axial direction on the yarn surface. The closer the value is to 1, the more consistent the fiber orientation is, and the closer it is to 0, the more random the orientation is.

9. The intelligent control system for optical fiber textile manufacturing according to claim 2, characterized in that, The reflectivity phase difference The determination is as follows: the spatial distribution sequence of reflectivity for the third band channel. Spatial distribution sequence of reflectivity in the first band channel A bandpass filter is applied along the yarn axis, with the passband range set according to the yarn twist fluctuation frequency. Then, the instantaneous phases are extracted using Hilbert transform. The two instantaneous phases are subtracted and phase unwinding is performed to obtain the final result. .

10. The intelligent control system for light-based textile manufacturing according to any one of claims 1-9, characterized in that, The system also includes an online calibration module for system parameters. When a preset calibration trigger event occurs, the acquisition module is activated to measure a calibration sample bar with a known nominal size. The calibration trigger event includes system power-on, continuous operation exceeding a set time, or external verification terminal feedback that the false alarm rate exceeds the limit. The measured cross-sectional characteristic quantity is compared with the nominal size to obtain the deviation. Then, based on the magnitude of the deviation, the semi-transparent correction amplitude coefficient, velocity coupling adjustment coefficient, polarization enhancement coefficient, and fractional order are adjusted in segments. When the deviation is large, a large step size correction is used; when the deviation is moderate, a small step size correction is used; and when the deviation is less than the tolerance value, the correction stops. The correction step size of each parameter decreases sequentially according to its influence on the measurement accuracy, and the ratio of the large step size to the small step size is preset to a fixed ratio.

Images