Vortex spinning broken yarn detection method and system based on image features

Abstract

The invention relates to a vortex spinning broken yarn detection method and system based on image features, and belongs to the technical field of intelligent control, and the detection method comprises the steps: collecting a dual-channel original image sequence of vortex spinning yarns; preprocessing the dual-channel original image sequence to generate an enhanced image, and extracting a yarn center line coordinate and a hairiness distribution matrix in the enhanced image; based on the yarn center line coordinates and the hairiness distribution matrix, spatial-temporal feature extraction is carried out, and feature vectors and a time sequence feature matrix composed of continuous multi-frame feature vectors are generated; according to the feature vector, the time sequence feature matrix and parameters in a preset knowledge base, calculating a broken yarn risk index and generating a control instruction; spinning equipment control operation corresponding to the control instruction is executed, and a fault data packet in the execution process is collected; and updating parameters in a preset knowledge base according to the fault data packet and measurement data fed back by the external calibration device. The accuracy and the automation level of vortex spinning broken yarn detection are improved.

Description

Technical Field

[0001] This application relates to the field of intelligent control technology, and in particular to a method and system for detecting yarn breakage in eddy current spinning based on image features. Background Technology

[0002] Vortex spinning is an important modern spinning process. Its working principle utilizes a high-speed rotating airflow to generate vortices, causing fibers to spiral and entangle under the influence of the airflow. Finally, a twisting device binds the loose fiber aggregates into a yarn with a certain strength and structural stability. Compared to traditional ring spinning and rotor spinning, vortex spinning offers significant advantages such as simpler processes, lower energy consumption, a wider range of suitable fibers, and higher production efficiency. It has been widely used in the processing of chemical fibers, cotton-linen blends, and functional fiber products.

[0003] However, with the widespread application of eddy current spinning technology, its high speed and high output have led to traditional mechanical sensing-based yarn breakage detection methods gradually revealing problems such as slow response, high false alarm rates, and high maintenance costs. Especially under complex operating conditions, such as changes in ambient light, frequent switching of yarn types, and state drift caused by long-term equipment operation, existing detection methods often struggle to accurately capture early subtle defect characteristics, thus failing to take timely and effective intervention measures. This ultimately results in batch yarn breakage accidents, affecting not only the quality of finished products but also potentially triggering a chain of shutdowns and increasing energy consumption and maintenance burdens. Summary of the Invention

[0004] To accurately identify yarn breakage risks and respond promptly, this application provides a method and system for detecting yarn breakage in eddy current spinning based on image features.

[0005] Firstly, this application provides a method for detecting yarn breakage in eddy current spinning based on image features, employing the following technical solution:

[0006] A method for detecting yarn breakage in eddy current spinning based on image features, the method comprising:

[0007] Acquire dual-channel raw image sequences of eddy current spinning yarn, including synchronously acquired infrared and blue light channel images;

[0008] The dual-channel original image sequence is preprocessed to generate an enhanced image, and the yarn centerline coordinates and hair distribution matrix are extracted from the enhanced image.

[0009] Spatiotemporal features are extracted based on the yarn centerline coordinates and the hair distribution matrix to generate feature vectors and a temporal feature matrix composed of feature vectors from multiple consecutive frames.

[0010] Based on the feature vector, the time-series feature matrix, and the parameters in the preset knowledge base, calculate the yarn breakage risk index and generate control instructions;

[0011] Execute the spinning equipment control operation corresponding to the control command, and collect fault data packets during the execution process;

[0012] The parameters in the preset knowledge base are updated based on the fault data packet and the measurement data fed back by the external calibration equipment.

[0013] By adopting the above technical solution, a highly integrated visual sensing and intelligent reasoning platform has been constructed. By fully utilizing the complementary advantages of visible light and near-infrared dual-spectral imaging, it can efficiently identify subtle defects without contacting the target and take proactive control measures, fundamentally changing the traditional passive management model that relies primarily on post-construction maintenance. This technical solution not only improves product quality stability and yield rate but also provides strong support for the digital transformation and upgrading of the textile industry under the background of intelligent manufacturing.

[0014] Optionally, before the step of acquiring the dual-channel raw image sequence of the eddy current spinning yarn, the method further includes: acquiring real-time process parameters of the spinning equipment and ambient light intensity sensor data in real time; wherein, the real-time process parameters include spinning speed, nozzle air pressure, and ambient temperature and humidity; calculating a dynamic frame rate adjustment coefficient based on the real-time process parameters to generate a target image acquisition frame rate; controlling the image sensor to acquire the dual-channel raw image sequence based on the target image acquisition frame rate; and calculating the pulse intensity compensation value of the dual-channel light source based on the difference between the ambient light intensity sensor data and the preset reference illumination.

[0015] The pulse width of the LED array of the dual-channel light source is adjusted based on the pulse intensity compensation value.

[0016] By adopting the above technical solution, a three-in-one control system integrating multi-parameter sensing, adaptive adjustment, and intelligent compensation was constructed, achieving precise control and optimization of the eddy current spinning yarn image acquisition process. This technical solution not only adapts to production needs under different process conditions but also effectively overcomes the impact of environmental interference factors on imaging quality, providing a reliable data foundation for subsequent yarn quality analysis and defect detection.

[0017] Optionally, the steps of preprocessing the dual-channel original image sequence to generate an enhanced image and extracting the yarn centerline coordinates and hair distribution matrix from the enhanced image include: parsing the dual-channel original image sequence into an infrared light channel image and a blue light channel image; performing edge enhancement processing on the infrared light channel image and contrast enhancement processing on the blue light channel image to obtain a processed dual-channel image; weighting and fusing the processed dual-channel images based on a preset scaling factor to generate a fused image; performing adaptive filtering processing on the fused image to output an enhanced image; loading historical yarn trajectories, predicting the yarn trajectory of the current frame based on the Kalman filter algorithm, and locating the yarn centerline coordinates in the enhanced image; extracting the pixel set outside the main yarn region in the enhanced image, separating hair pixels using a threshold segmentation method, and mapping the coordinates of the hair pixels to a hair distribution matrix.

[0018] By adopting the above technical solution, not only is the complementary information advantage of infrared and blue light channel images fully utilized, but key technologies such as adaptive filtering, trajectory prediction, and intelligent segmentation are also introduced, significantly improving the accuracy and efficiency of yarn image processing. Especially in high-speed continuous inspection scenarios, this technical solution can reduce computational overhead while ensuring image quality, providing a reliable data foundation and technical support for subsequent yarn quality control, defect detection, and intelligent manufacturing.

[0019] Optionally, the step of extracting spatiotemporal features based on the yarn centerline coordinates and the hair distribution matrix to generate feature vectors and a temporal feature matrix composed of feature vectors from multiple consecutive frames includes: receiving a sequence of yarn centerline coordinates and a hair distribution matrix; and calculating the yarn diameter and hair density based on the current frame enhanced image and the yarn centerline coordinates.

[0020] A series of continuous multi-frame yarn centerline coordinate sequences are constructed and a Fast Fourier Transform is performed to extract the amplitude of the main spectral components, thus obtaining the vibration frequency characteristics. The tension fluctuation value is calculated by combining the standard deviation of the continuous multi-frame yarn diameter data sequence and a preset yarn elastic modulus. The yarn diameter, hairiness density, vibration frequency, and tension fluctuation value are combined into a feature vector. The continuous feature vectors are then integrated according to a preset fixed time window to generate a time-series feature matrix. By adopting the above technical solution, multi-dimensional and multi-level feature extraction of eddy current spinning yarn states is achieved, which not only improves the accuracy and robustness of yarn breakage detection but also possesses good engineering adaptability and expansion potential.

[0021] Optionally, the step of calculating the yarn breakage risk index and generating control instructions based on the feature vector, the time-series feature matrix, and parameters in the preset knowledge base includes: receiving the feature vector and the time-series feature matrix at the current moment; calling the parameter threshold and feature weight coefficient matching the yarn type in the preset knowledge base; calculating the yarn breakage risk index by weighting each component in the feature vector according to the parameter threshold and feature weight coefficient; generating graded control instructions based on the yarn breakage risk index and tension fluctuation value; and recording the time-series feature matrix and the corresponding graded control instructions in real time, merging them as decision data to update the learning sample set of the preset knowledge base.

[0022] By adopting the above technical solution, feature vectors and time series data are effectively integrated. Relying on a differentiated parameter matching mechanism supported by a knowledge base, a risk index calculation model enables early warning of potential yarn breakage. Furthermore, through multi-level control command issuance strategies and online learning feedback loops, the system's responsiveness and handling rationality in the face of emergencies are significantly improved, while also providing valuable empirical resources for long-term operation and maintenance management.

[0023] Optionally, the step of updating the parameters in the preset knowledge base based on the fault data packet and the measurement data fed back by the external calibration device includes: receiving the fault data packet fed back by the execution control unit, which includes the abnormal feature vector when the yarn breakage occurs and real-time environmental parameters; calling the measured data of yarn physical quantities provided by the external calibration device; calculating the parameter correction amount and marking the actual yarn breakage state based on the deviation between the abnormal feature vector in the fault data packet and the measured data, and obtaining the false alarm / missed alarm statistical analysis results; iteratively updating the threshold parameters in the preset knowledge base based on the parameter correction amount; and adjusting the feature weight coefficients in the preset knowledge base based on the false alarm / missed alarm statistical analysis results.

[0024] By adopting the above technical solution, dynamic updating and intelligent optimization of the knowledge base parameters in the yarn breakage detection system are achieved, significantly improving the system's adaptability and detection accuracy. This technical solution is not only applicable to yarn breakage detection scenarios in the textile industry, but can also be extended to other industrial detection fields requiring high real-time performance and accuracy, demonstrating broad application prospects and technological promotion value.

[0025] Optionally, after executing the spinning equipment control operation corresponding to the control command and collecting the fault data packet during the execution process, the method further includes: parsing the abnormal feature vector and real-time environmental parameters contained in the fault data packet; combining the abnormal feature vector and real-time environmental parameters into diagnostic input data, and performing similarity matching with a preset historical fault case library; when the detection shows that the mutation rate of the hair density feature value relative to the historical mean exceeds a preset first threshold and the vibration frequency offset exceeds a preset second threshold, the fiber raw material defect detection process is initiated; when the detection shows that the tension fluctuation feature value continuously exceeds a preset safety threshold and the change rate of the yarn diameter feature value is lower than a preset change rate threshold, the mechanical fault diagnosis process is initiated. By adopting the above technical solution, a multi-level progressive intelligent diagnostic architecture is constructed, realizing full-process coverage from macroscopic state perception to microscopic root cause tracing. This technical solution uses a historical knowledge base to carry out rapid matching and screening based on similarity, and then activates more targeted special detection modules according to different types of fault feature manifestations, ultimately achieving the dual prevention and control goals of raw material defects and mechanical equipment hazards.

[0026] Optional, the specific steps of the fiber raw material defect detection process include:

[0027] Call the externally connected fiber fineness detection device to obtain the fineness distribution data of the current yarn raw material batch;

[0028] If the dispersion of the fineness distribution data exceeds the preset raw material defect threshold, a raw material batch replacement alarm command is output to the spinning equipment operation and maintenance system.

[0029] By adopting the above technical solution, the system outputs raw material batch replacement alarm instructions to the spinning equipment operation and maintenance system. This timely replacement of problematic raw material batches prevents the production of substandard products, embodying the concepts of predictive maintenance and quality control in modern intelligent manufacturing.

[0030] Optionally, the mechanical fault diagnosis process includes: collecting real-time spectrum data from the vibration sensor of the spinning equipment; if there are abnormal frequency components in the real-time spectrum data that have a correlation with the fundamental harmonic of the spindle exceeding a preset overlap, outputting a spindle bearing maintenance command to the spinning equipment operation and maintenance system.

[0031] By adopting the above technical solution, when the fault judgment conditions are met, the system outputs a spindle bearing maintenance command to the spinning equipment operation and maintenance system, realizing automated closed-loop management from fault detection to maintenance decision-making. This maintenance strategy falls under the category of predictive maintenance, which has significant advantages over traditional periodic maintenance or post-fault repair, and can significantly reduce equipment downtime, reduce maintenance costs, and increase equipment lifespan.

[0032] Secondly, this application provides an image feature-based eddy current spinning yarn breakage detection system, which adopts the following technical solution:

[0033] A vortex spinning yarn breakage detection system based on image features, the detection system comprising:

[0034] The image acquisition module is used to acquire dual-channel raw image sequences of eddy current spinning yarn, including synchronously acquired infrared light channel images and blue light channel images;

[0035] The image processing module is used to preprocess the dual-channel original image sequence to generate an enhanced image, and extract the yarn centerline coordinates and hair distribution matrix from the enhanced image;

[0036] The feature extraction module is used to extract spatiotemporal features based on the yarn centerline coordinates and the hair distribution matrix, and generate feature vectors and a temporal feature matrix composed of feature vectors from multiple consecutive frames.

[0037] The yarn breakage risk control module is used to calculate the yarn breakage risk index and generate control instructions based on the feature vector, the time-series feature matrix and the parameters in the preset knowledge base.

[0038] The execution feedback module is used to execute the spinning equipment control operation corresponding to the control command and collect fault data packets during the execution process;

[0039] The parameter update module is used to update the parameters in the preset knowledge base based on the fault data packet and the measurement data fed back by the external calibration equipment.

[0040] In summary, this application includes at least one of the following beneficial technical effects: it realizes intelligent real-time monitoring and control of yarn breakage risk in eddy spinning. The system adopts dual-channel image acquisition and advanced image processing technology, which can accurately extract key features such as yarn centerline coordinates and hair distribution. Through spatiotemporal feature analysis, it generates a yarn breakage risk index, realizing accurate risk warning and automatic equipment control. Simultaneously, it possesses self-learning capabilities, dynamically optimizing detection parameters based on actual operating data, significantly improving the accuracy and automation level of eddy spinning yarn breakage detection. Attached Figure Description

[0041] Figure 1 This is a schematic diagram of the first process of a vortex spinning yarn breakage detection method according to one embodiment of this application.

[0042] Figure 2 This is a schematic diagram of the second process of the eddy current spinning yarn breakage detection method according to one embodiment of this application.

[0043] Figure 3 This is a schematic diagram of the third process of the eddy current spinning yarn breakage detection method according to one embodiment of this application.

[0044] Figure 4 This is a schematic diagram of the fourth process of the eddy current spinning yarn breakage detection method according to one embodiment of this application.

[0045] Figure 5 This is a schematic diagram of the fifth process of the eddy current spinning yarn breakage detection method according to one embodiment of this application.

[0046] Figure 6 This is a schematic diagram of the sixth process of the eddy current spinning yarn breakage detection method according to one embodiment of this application.

[0047] Figure 7 This is a schematic diagram of the seventh process of the eddy current spinning yarn breakage detection method according to one embodiment of this application.

[0048] Figure 8 This is a schematic diagram of the eighth process of the eddy current spinning yarn breakage detection method according to one embodiment of this application.

[0049] Figure 9 This is a schematic diagram of the ninth process of the eddy current spinning yarn breakage detection method according to one embodiment of this application. Detailed Implementation

[0050] To make the purpose, technical solution, and advantages of this application clearer, the following description is provided in conjunction with the appendix. Figure 1-9 The present application will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the application.

[0051] This application discloses a method for detecting yarn breakage in eddy current spinning based on image features.

[0052] Reference Figure 1 A method for detecting yarn breakage in eddy current spinning based on image features, the method comprising:

[0053] Step S101: Acquire the dual-channel raw image sequence of the eddy current spinning yarn, including the infrared light channel image and the blue light channel image acquired simultaneously;

[0054] The core purpose of this step is to enhance image contrast by utilizing the differences in the reflective properties of different wavelengths of light on the material surface, thereby improving the accuracy of subsequent image recognition and analysis.

[0055] Specifically, infrared light (wavelength 850nm) has strong penetrating power and can effectively reflect the internal structural information of yarn; while blue light (wavelength 460nm) focuses more on capturing surface texture details.

[0056] In this embodiment, a high-speed CMOS sensor, combined with a narrow pulse beam (30-50 μs) emitted by a ring array light source, can achieve high-quality image capture in an extremely short time, avoiding errors caused by motion blur and ensuring temporal consistency between the two channels. This dual-channel image acquisition method not only improves the image signal-to-noise ratio but also enhances the identifiability of the target object, laying the foundation for subsequent complex image analysis.

[0057] Step S102: Preprocess the dual-channel original image sequence to generate an enhanced image, and extract the yarn centerline coordinates and hair distribution matrix from the enhanced image;

[0058] The preprocessing step begins with weighted fusion, where data from the infrared and blue light channels are superimposed pixel-wise according to specific weights. This ensures that the final synthesized image retains both the deep structural features of the infrared image and the rich edge and texture information of the blue light channel image. Following this, an adaptive median filtering algorithm is implemented to remove image noise while preserving edge sharpness as much as possible, preventing distortion or breakage of the target contour due to random interference.

[0059] Building upon this foundation, the system further introduces a yarn localization strategy based on prior trajectory prediction. This involves establishing a mathematical model using the yarn movement trends in historical frames to calculate the possible location range at the next moment, thereby narrowing the search window and significantly reducing computational complexity. The final step is the feather region segmentation operation. This typically employs methods such as Otsu's thresholding, region growing, or deep learning semantic segmentation to separate the fine fibrous protrusions, the so-called "feathers," from the background. The results are represented as a binary matrix for subsequent quantitative statistical use.

[0060] Step S103: Based on the yarn centerline coordinates and the hair distribution matrix, perform spatiotemporal feature extraction to generate feature vectors and a temporal feature matrix composed of feature vectors from multiple consecutive frames; wherein, the feature vectors include spatial domain features (yarn diameter and hair density) and temporal domain features (vibration frequency and tension fluctuation).

[0061] Specifically, spatial domain characteristics refer to the geometric properties of the yarn itself in a static state, such as diameter and cross-sectional shape changes. To accurately measure these indicators, the system first uses classic edge detection operators such as Canny and Sobel to determine the yarn boundary positions, and then estimates its average radius or maximum circumscribed circle diameter. As for the hairiness density, it is obtained by normalizing the proportion of non-zero elements in the aforementioned hairiness distribution matrix, reflecting the number of freely floating fibers per unit area.

[0062] On the other hand, time-domain features focus on the dynamic behavior of the yarn in continuous motion, such as the frequency of shaking and the amplitude of periodic offset. Extracting these characteristics relies on tracking and recording the coordinates of the yarn center point in multiple frames of images, and then performing Fast Fourier Transform (FFT) processing to convert the time-domain signal to the frequency domain, thereby accurately identifying the dominant frequency component corresponding to the dominant vibration mode. Furthermore, tension fluctuation parameters are obtained by multiplying the variation in yarn diameter between consecutive frames by its material elastic modulus, indirectly reflecting whether the yarn is subjected to stable and uniform stress.

[0063] Step S104: Based on the feature vector, time-series feature matrix, and parameters in the pre-set knowledge base, calculate the yarn breakage risk index and generate control instructions. The system pre-establishes a standard threshold system for different types of yarn and importance ranking rules for various features (i.e., feature weights), which constitute the pre-set knowledge base. During actual operation, all feature values ​​of the current sample are read in real time and compared one by one with the reference standards in the pre-set knowledge base. A weighted summation formula is then used to derive a probability score between 0 and 1, i.e., the yarn breakage risk index. The higher the index, the greater the probability of yarn breakage. It should be noted that this judgment mechanism does not simply set a fixed threshold to trigger an alarm, but rather considers the differences in normal fluctuation ranges under different operating conditions. Therefore, it allows for flexible adjustment of the judgment conditions according to different yarn varieties, demonstrating good versatility and robustness.

[0064] Step S105: Execute the control operation of the spinning equipment corresponding to the control command, and collect the fault data packet during the execution process; among them, once the system determines that there is a high risk of yarn breakage (such as the index exceeding 0.7), it will issue a deceleration command to notify the frequency converter drive to reduce the spindle speed to a safe level; if the situation is urgent (such as the index exceeding 0.9 and accompanied by significant tension abnormality), the emergency stop procedure will be initiated immediately, the air supply valve will be closed and the negative pressure suction mechanism will be activated to quickly remove residual fibers to prevent entanglement and spread.

[0065] At the same time, all relevant event snapshots, environmental variable records, and abnormal feature vectors at the moment of occurrence will be packaged and saved for future retrospective analysis, forming a valuable on-site fault sample resource.

[0066] Step S106: Update the parameters in the preset knowledge base based on the fault data packet and the measurement data fed back by the external calibration equipment.

[0067] It is understandable that as production batches change, raw materials are replaced, or even seasonal temperature and humidity changes occur, the original judgment criteria in the pre-built knowledge base may gradually become ineffective, which requires regular correction and optimization.

[0068] To this end, the system has a standardized API interface that can be connected to precision instruments such as laser diameter gauges to obtain more accurate real measurement values ​​as a reference benchmark for correcting basic threshold parameters such as yarn diameter.

[0069] Furthermore, if frequent false positives are detected within a certain period (the consecutive false positive rate exceeds the set limit), a retraining process will be automatically triggered to realign the relative importance weights of various features, making the classifier more realistic. More importantly, the complete logs generated from each successful interception of a potential yarn breakage incident will be included in a new training set, continuously expanding and improving the knowledge base of the underlying machine learning model, enabling it to self-evolve.

[0070] The above implementation constructs a highly integrated visual sensing and intelligent reasoning platform. By fully utilizing the complementary advantages of visible light and near-infrared dual-spectral images, it can efficiently identify subtle defect signs without contacting the target and take proactive control measures, fundamentally changing the traditional passive management model that relies primarily on post-construction maintenance. This technical solution not only improves product quality stability and yield rate but also provides strong support for the digital transformation and upgrading of the textile industry under the background of intelligent manufacturing.

[0071] Reference Figure 2 As a further implementation of the eddy current spinning yarn breakage detection method, before the step of acquiring the dual-channel original image sequence of the eddy current spinning yarn, the method further includes:

[0072] Step S201: Real-time process parameters of the spinning equipment and ambient light intensity sensor data are acquired; wherein, the real-time process parameters include spinning speed, nozzle air pressure and ambient temperature and humidity.

[0073] Among these parameters, spinning speed, as a core parameter, determines the speed and dwell time of the yarn within the imaging area, directly affecting the image acquisition time window; nozzle air pressure influences the morphological stability and surface hair distribution characteristics of the yarn; and ambient temperature and humidity indirectly affect the yarn surface condition by influencing the physical properties of the fiber material (such as moisture regain and electrostatic effects). Meanwhile, ambient light intensity sensor data provides illumination baseline information for the current working environment, offering a quantitative basis for subsequent light source compensation.

[0074] Step S202: Calculate the dynamic frame rate adjustment coefficient based on real-time process parameters to generate the target image acquisition frame rate;

[0075] Among them, by establishing a direct proportional function relationship between the spinning speed and the image acquisition frame rate, it is ensured that an ideal image sampling density can be obtained under different production speed conditions. Specifically, when the spinning speed increases, the distance that the yarn passes through the imaging area per unit time increases. If the frame rate is kept fixed, the displacement of the yarn within the time interval between adjacent image frames increases, which may lead to motion blur or sparse sampling points, affecting the accuracy of defect detection. Therefore, by increasing the image acquisition frame rate to compensate for the increase in the yarn movement speed, a constant spatial sampling resolution can be maintained.

[0076] Specifically, a calculation formula for the target image acquisition frame rate is established: F = ⌈k(V) × V × φ(T, H)⌉. In the above formula, V is the spinning speed (m / min), k is the resolution coefficient, T is the temperature, and H is the humidity.

[0077] In the embodiments of this application, φ(T, H) is the temperature-humidity coupling coefficient: φ = 1 + 0.05∣T - 25∣ + 0.03max(H - 70%, 0). k(V) is a piecewise function of speed. When V ≤ 300, k = 0.8; when 300 < V ≤ 600, k = 1.2; when V > 600, k = 0.6log 10 (V). That is, a fixed coefficient is adopted when V ≤ 300V to avoid frame rate fluctuations under low-speed working conditions; when V > 600, logarithmic attenuation is introduced to suppress the image smear effect caused by high-speed movement.

[0078] It can be understood that this dynamic adjustment mechanism is based on kinematic principles and digital sampling theory, and follows the basic requirements of the Nyquist sampling theorem to ensure that the image sequence can completely capture the detailed change characteristics of the yarn surface. In terms of technical implementation, this adjustment algorithm usually adopts a proportional-integral-derivative (PID) control strategy or a fuzzy logic control method, and realizes precise control of the image acquisition device by real-time monitoring the change of the spinning speed and calculating the corresponding frame rate adjustment amount.

[0079] Step S203, based on the target image acquisition frame rate, control the image sensor to collect a dual-channel raw image sequence;

[0080] Step S204, calculate the pulse intensity compensation value of the dual-channel light source according to the difference between the ambient light intensity sensor data and the preset reference illumination;

[0081] Among them, this compensation mechanism is based on optical measurement principles and automatic exposure control theory, and realizes dynamic optimization of the imaging illumination conditions by establishing an inverse compensation model between the ambient light intensity and the light source pulse intensity.

[0082] Specifically, when ambient light intensity increases, the impact of background illumination on the image sensor increases, potentially leading to decreased image contrast and compressed dynamic range. Conversely, when ambient light intensity decreases, relatively strong background noise may mask subtle defects on the yarn surface. By adjusting the pulse intensity of the dual-channel light source to compensate for changes in ambient light, a constant signal-to-noise ratio and contrast level can be maintained.

[0083] For example, the specific formula for calculating the pulse intensity compensation value is as follows:

[0084] ;

[0085] In the above formula, I c To compensate for the pulse intensity, I0 is the reference intensity, and E env E represents ambient light intensity. ref The preset reference illumination intensity is β, which is the attenuation factor (default value 0.7).

[0086] Step S205: Adjust the pulse width of the LED array of the dual-channel light source based on the pulse intensity compensation value.

[0087] In the embodiments of this application, the dual-channel light source design typically employs LED array technology, configuring light source components of different wavelengths (such as red and blue light) respectively. By independently controlling the pulse width and peak power of each channel, optimal illumination effects are achieved for yarns of different materials and colors. This compensation algorithm typically integrates a photometric calculation model, considering the spectral response characteristics of the light source, the quantum efficiency curve of the image sensor, and the optical reflection characteristics of the yarn material. A real-time feedback adjustment mechanism ensures the stability and consistency of imaging quality.

[0088] In the above embodiments, a three-in-one control system integrating multi-parameter sensing, adaptive adjustment, and intelligent compensation is constructed, achieving precise control and optimization of the eddy current spinning yarn image acquisition process. This technical solution can not only adapt to the production needs under different process conditions, but also effectively overcome the influence of environmental interference factors on imaging quality, providing a reliable data foundation for subsequent yarn quality analysis and defect detection.

[0089] Reference Figure 3 As one implementation of step S102, the steps of preprocessing the dual-channel original image sequence to generate an enhanced image and extracting the yarn centerline coordinates and hair distribution matrix from the enhanced image include:

[0090] Step S301: The dual-channel original image sequence is parsed into infrared light channel images and blue light channel images;

[0091] Among them, the dual-channel image acquisition method can enrich the image information dimension from the perspective of physical characteristics. Infrared light is sensitive to the thermal radiation response of materials and is suitable for capturing the temperature distribution and internal structural differences of yarn surfaces; while blue light is more sensitive to surface reflection characteristics and helps to highlight the yarn outline and texture details.

[0092] Step S302: Perform edge enhancement processing on the infrared light channel image and contrast enhancement processing on the blue light channel image to obtain the processed dual-channel image.

[0093] Specifically, infrared images typically have strong penetrating power and can reflect changes in the internal structure of yarn, but edge information is blurred; while blue light channel images have higher contrast and clearer edge performance, but are easily affected by ambient lighting. Therefore, before weighted fusion, edge enhancement processing (such as Laplacian operator, Canny edge detection, etc.) is performed on the infrared image to improve its edge sharpness; at the same time, contrast enhancement processing (such as histogram equalization, CLAHE, etc.) is performed on the blue light channel image to enhance the representation of local details.

[0094] Step S303: The processed dual-channel images are weighted and fused based on a preset scaling factor to generate a fused image;

[0095] The process involves linearly weighting and fusing the processing results of the two channels according to a preset scaling factor to obtain a fused image. This process essentially uses a weighted average to ensure that the fused image retains both the deep structural information of the infrared image and the apparent features of the blue light channel image, thus providing more discriminative input data for subsequent processing.

[0096] Step S304: Perform adaptive filtering on the fused image to output an enhanced image;

[0097] Because the yarn moves at high speed during production, images are easily affected by factors such as impulse noise and background texture interference, making it difficult for traditional fixed-parameter filters to adapt to dynamically changing image features. Therefore, this embodiment employs an adaptive filtering mechanism to dynamically adjust the filter window size based on the real-time speed of the yarn.

[0098] Specifically, when the yarn movement speed is high, the filtering window is appropriately enlarged to improve the noise reduction effect; conversely, the window is reduced to retain more detailed information. Based on this, a median filtering algorithm is used to effectively suppress impulse noise (such as salt-and-pepper noise). This algorithm effectively removes isolated noise points without destroying edge information by sorting the pixel values ​​in the neighborhood and taking the median. Simultaneously, Gaussian filtering is used to smooth the image background area, reducing the interference of complex background textures on subsequent analysis, while preserving the original gradient features of the yarn edges to avoid excessive blurring. The final enhanced image output has clearer target contours, lower noise levels, and stronger feature discernibility.

[0099] Step S305: Load historical yarn trajectories, predict the yarn trajectory of the current frame based on the Kalman filter algorithm, and locate the coordinates of the yarn centerline in the enhanced image;

[0100] Because yarn exhibits continuity and regularity in its movement during weaving, its trajectory changes show a certain temporal correlation. Therefore, the system loads historical data on the center positions of the yarn from the most recent N frames and uses a Kalman filter algorithm to predict the yarn trajectory of the current frame.

[0101] Specifically, Kalman filtering is a recursive estimation algorithm that can optimally estimate the system state through a state transition model and an observation model, maintaining high prediction accuracy even in the presence of measurement noise. This prediction mechanism allows for the delineation of a small region of interest (ROI) in the current frame image, significantly narrowing the search range for subsequent edge detection.

[0102] Within this predicted area, sub-pixel-level edge detection algorithms (such as Sobel+ interpolation, Zernike moments, etc.) can be used to accurately locate the yarn edges, and the coordinate array of the yarn centerline can be obtained through curve fitting or skeleton extraction methods. This step not only improves the detection speed but also enhances the system's robustness to yarn movement jitter and image shift.

[0103] Step S306: Extract the set of pixels outside the main yarn region in the enhanced image, separate the feather pixels using the threshold segmentation method, and map the coordinates of the feather pixels into a feather distribution matrix.

[0104] Among them, hair, as a protruding fiber bundle on the surface of the yarn, directly affects the yarn quality and fabric performance.

[0105] In this step, the first step is to extract the pixel set outside the main yarn region from the enhanced image, typically achieved through masking or region growing algorithms. Then, a thresholding method is used to separate the feather pixels from the background. This threshold is dynamically set, with the formula T = μ + kσ, where μ represents the mean grayscale value of the background region, σ is the standard deviation, and k is a pre-calibrated coefficient based on the yarn material characteristics. This dynamic thresholding mechanism can adapt to different lighting conditions and variations in material reflectivity, avoiding mis-segmentation problems caused by a fixed threshold.

[0106] After segmentation, the two-dimensional coordinates of the feather pixels are mapped to a binary matrix with the same size as the original image, where 1 represents a feather pixel and 0 represents a non-feather region. This matrix is ​​the feather distribution matrix, which has good numerical stability and computability, facilitating subsequent quantitative analysis, defect identification, or quality assessment.

[0107] The above implementation not only fully leverages the complementary information advantages of infrared and blue light channel images but also introduces key technologies such as adaptive filtering, trajectory prediction, and intelligent segmentation, significantly improving the accuracy and efficiency of yarn image processing. Especially in high-speed continuous inspection scenarios, this technical solution can reduce computational overhead while ensuring image quality, providing a reliable data foundation and technical support for subsequent yarn quality control, defect detection, and intelligent manufacturing.

[0108] Reference Figure 4 As one implementation of step S103, the steps of extracting spatiotemporal features based on the yarn centerline coordinates and the hair distribution matrix to generate feature vectors and a temporal feature matrix composed of feature vectors from multiple consecutive frames include:

[0109] Step S401: Receive the yarn centerline coordinate sequence and the hair distribution matrix;

[0110] In the field of industrial visual inspection, the yarn centerline typically refers to a curve that runs through the yarn body, obtained by edge recognition or skeletonization of the yarn region in an enhanced image. It represents the spatial direction and positional information of the yarn. This coordinate sequence can be two-dimensional pixel coordinates (x, y) recorded point by point in consecutive frames, reflecting the dynamic trajectory of the yarn in image space.

[0111] The yarn hair distribution matrix is ​​a two-dimensional grayscale or binary image data structure used to describe the distribution of hair on the yarn surface. Non-zero elements represent the presence and intensity of hair, while zero elements represent hairless regions. This matrix not only reflects the microstructural characteristics of the yarn surface but also indirectly reflects the uniformity and density of fiber arrangement.

[0112] Step S402: Calculate the yarn diameter and hair density based on the current frame enhanced image and the yarn centerline coordinates;

[0113] In this process, image enhancement usually refers to a high-quality image after preprocessing such as filtering, sharpening, and contrast adjustment, which aims to improve edge clarity and noise suppression capabilities, thereby improving the accuracy of subsequent measurements.

[0114] The specific steps for calculating the yarn diameter include: selecting several sampling points along the yarn centerline and extracting a grayscale profile along the normal direction at each point; then determining the positions of the two edges using gradient extremum detection algorithms (such as the Sobel operator or Canny edge detection), calculating the pixel distance between the two edges, and converting it into a diameter value in physical units using the pixel equivalent coefficient obtained from camera calibration. This method avoids the errors caused by the traditional circular assumption and is particularly suitable for yarns with irregular cross-sections or local deformations.

[0115] As for the fuzz density, it is normalized by counting the number of non-zero pixels within the fuzz distribution matrix and dividing by the total area of ​​the yarn detection region. A material correction coefficient is then introduced to compensate for the differences in the impact of different fiber materials on image response, ultimately outputting a standardized fuzz density index. This index not only reflects the number of fuzzes but also implicitly includes yarn surface roughness and fiber adhesion status, helping to assess the stability of yarn quality.

[0116] Step S403: Construct a continuous multi-frame yarn centerline coordinate sequence and perform a fast Fourier transform to extract the amplitude of the main spectral component and obtain the vibration frequency characteristics;

[0117] In particular, due to the influence of airflow disturbance and tension fluctuation on the yarn during the vortex spinning process, the yarn's motion trajectory on the transmission path will exhibit certain periodic vibration characteristics.

[0118] To capture this subtle yet crucial motion pattern, the system needs to acquire a sequence of N consecutive frames of the yarn centerline Y-axis coordinates, treating it as a one-dimensional time-series signal. A Fast Fourier Transform (FFT) is then performed on this sequence to map the time-domain signal to the frequency domain, extracting the main spectral components and their corresponding amplitudes. These frequency components are often closely related to equipment operating frequencies, airflow disturbance frequencies, etc., and abnormal frequency bands may exhibit energy concentration in the precursory stage of yarn breakage. Therefore, extracting vibration frequency characteristics essentially transforms the spatial displacement of the yarn into energy distribution characteristics, thereby aiding in the determination of whether a potential yarn breakage trend exists.

[0119] Step S404: Calculate the tension fluctuation value by combining the standard deviation of the continuous multi-frame yarn diameter data sequence and the preset yarn elastic modulus;

[0120] In actual production, changes in yarn tension often lead to corresponding fluctuations in its diameter, and this relationship can be approximated using Hooke's Law. Specifically, the system first acquires a sequence of yarn diameters for K consecutive frames within a time window, then calculates its standard deviation to measure the amplitude of diameter fluctuations; subsequently, it multiplies this by a preset elastic modulus parameter (which can be pre-entered into a database based on the yarn material) to derive the corresponding tension fluctuation value. This process essentially couples image measurement results with physical parameters, transforming the originally visually significant diameter change into a mechanical indicator with engineering implications.

[0121] It should be noted that the elastic modulus, as an inherent property of the material, is not limited to a specific value in this application embodiment. Instead, it is used as a configurable parameter to support the application expansion of various yarn types, thereby enhancing the versatility and adaptability of the system.

[0122] Step S405: Combine yarn diameter, hair density, vibration frequency, and tension fluctuation value into a feature vector;

[0123] This combination is not a simple numerical patchwork, but a fusion strategy based on the potential correlations and complementarities between various features. For example, there is a causal relationship between yarn diameter and tension fluctuations, while hair density may affect the distribution of vibration frequency, and vibration frequency may be one of the manifestations of tension instability.

[0124] Therefore, unifying these features from different dimensions (spatial geometry, surface structure, dynamic behavior, and mechanical response) into a high-dimensional vector not only enhances the richness of feature representation but also provides more comprehensive data support for subsequent classifiers or decision models. This feature vector essentially provides a multi-faceted characterization of the yarn's current state, more accurately reflecting whether it is in normal operation or nearing the breakage threshold.

[0125] Step S406: Integrate continuous feature vectors according to a preset fixed time window to generate a time-series feature matrix.

[0126] Since yarn breakage events often occur gradually and cumulatively, relying solely on single-frame features is insufficient to effectively capture their evolution. Therefore, the system employs a sliding time window mechanism, arranging feature vectors from multiple consecutive time points in chronological order to construct a two-dimensional matrix structure where rows represent time points and columns represent feature dimensions. This matrix form not only preserves the changing trends of each feature over time but also facilitates further analysis using temporal modeling tools such as Recurrent Neural Networks (RNNs) and Long Short-Term Memory Networks (LSTMs).

[0127] In addition, the fixed-length time window design helps control memory usage and improve real-time processing efficiency, making it particularly suitable for applications in high-speed spinning production lines.

[0128] The above implementation method realizes the extraction of multi-dimensional and multi-level features of eddy current spinning yarn state, which not only improves the accuracy and robustness of yarn breakage detection, but also has good engineering adaptability and expansion potential.

[0129] Reference Figure 5 As one implementation of step S104, the step of calculating the yarn breakage risk index and generating control instructions based on the feature vector, the time-series feature matrix, and parameters in the preset knowledge base includes:

[0130] Step S501: Receive the feature vector and time series feature matrix at the current time.

[0131] The feature vector is a multidimensional numerical set composed of multiple key physical indicators, typically including parameters such as yarn diameter, hairiness density, vibration frequency, and tension fluctuation. These parameters represent quality information across multiple dimensions, including yarn morphological stability, surface roughness variation trends, mechanical motion disturbance intensity, and the response consistency of the tension system. The time-series feature matrix further introduces dynamic evolution characteristics from a time-series perspective. It not only reflects a snapshot of the state at a single point in time but also includes the trajectory and trend patterns of various parameters over a period of time, such as the historical curve of tension fluctuations and the temporal distribution of periodic oscillation amplitudes.

[0132] Understandably, this two-tiered data organization allows the system to overlay dynamic perception capabilities on top of static assessments, thereby enabling more accurate risk identification of potential yarn breakage events.

[0133] Step S502: Call the parameter threshold and feature weight coefficient that match the yarn type from the preset knowledge base;

[0134] Among them, the pre-built knowledge base is an expert experience model library that has been trained and encapsulated with a large amount of historical data. Internally, it establishes differentiated standard threshold tables and sensitive factor weight mapping relationships based on different types (such as cotton type, polyester filament, blended yarn, etc.) and specifications (such as fineness grade Ne or Tex unit).

[0135] Specifically, each yarn category corresponds to a set of exclusive benchmark reference values ​​(i.e., thresholds). For example, for a specific yarn count, its maximum allowable hairiness density, optimal tension range, and reasonable upper limit of vibration frequency are all clearly defined. The feature weight coefficients, on the other hand, quantify the relative importance of different features in their likelihood of causing yarn breakage. Some factors may be more likely to cause breakage than others, thus requiring higher weighting. Through this stage of knowledge retrieval and parameter configuration, the subsequent algorithm processing is ensured to have a high degree of customization capability and effectively avoids the problem of increased misjudgment rate due to insufficient generalization of general rules.

[0136] Step S503: Calculate the yarn breakage risk index by weighting each component in the feature vector according to the parameter threshold and feature weight coefficient;

[0137] The calculation formula is as follows:

[0138] ;

[0139] In the above formula, Risk is the yarn breakage risk index, w i V represents the feature weight coefficients. i T represents the actual feature value of the i-th feature. i V is the parameter threshold corresponding to the i-th feature. i With T i The closer the ratio is to 1, the more stable the system is; conversely, a smaller ratio indicates an abnormal tendency.

[0140] Specifically, each original observation is compared with the standard threshold of its category to determine the degree of deviation, and then weighted accordingly to synthesize the final risk score. Using absolute values ​​ensures that all deviations are considered uniformly regardless of their direction.

[0141] It should be noted that the feature weight coefficient w i It is not fixed but can be adaptively adjusted according to external environmental variables, which greatly enhances the self-optimization potential of the entire evaluation framework in the face of complex operating conditions and disturbances.

[0142] Step S504: Generate graded control instructions based on the yarn breakage risk index and tension fluctuation value;

[0143] This mechanism transforms abstract probabilistic predictions into concrete action commands, enabling downstream equipment to take timely and effective intervention measures. The hierarchical control mechanism fully considers the safety margin requirements of industrial operations, avoiding abrupt action problems that may occur under traditional binary judgment methods.

[0144] For example, when the risk index is within the first threshold range, a normal operation command is output; when the risk index exceeds the first threshold but is lower than the second threshold, a pre-deceleration command is generated; when the risk index exceeds the second threshold and the tension fluctuation value exceeds the preset safety threshold, an emergency stop command is generated.

[0145] Understandably, when the calculated risk score is low, the existing work cycle will continue to operate by default. If the score exceeds the first-level warning line but has not yet reached the emergency shutdown condition, a warning signal will be issued and a pre-deceleration program will be started to reduce the speed gradient in the stretching zone, providing a buffer space for subsequent observation windows. Only when the risk index exceeds the higher-level safety boundary and a violent oscillation phenomenon (exceeding the set multiple threshold) is detected simultaneously will the forced shutdown mechanism be triggered.

[0146] It should be noted that the reason for adopting a dual verification logic instead of immediately braking based solely on the risk score is that some transient disturbances may temporarily raise local parameter readings but will not actually develop into substantial failures. Therefore, an additional layer of cross-verification from the tension sensing channel is necessary to make a final judgment, thereby improving the reliability and noise resistance of the overall early warning system.

[0147] Step S505: Record the time-series feature matrix and the corresponding hierarchical control instructions in real time, and merge them as decision data to update the learning sample set of the preset knowledge base.

[0148] Whenever a valid control response event occurs, the system automatically captures the complete feature input set (including feature vectors and time series matrices), the weight settings, the control output command and its triggering cause chain, and other contextual background information, and packages and uploads it to the central database for permanent storage.

[0149] In the above implementation, feature vectors and time series data are effectively fused, and a differentiated parameter matching mechanism supported by a knowledge base is used to achieve early warning capabilities for potential yarn breakage through a risk index calculation model. Furthermore, through multi-level control command issuance strategies and online learning feedback loops, the system's responsiveness and handling rationality in the face of emergencies are significantly improved, while also providing valuable empirical resources for long-term operation and maintenance management.

[0150] Reference Figure 6 As one implementation of step S106, the step of updating the parameters in the preset knowledge base based on the fault data packet and the measurement data fed back by the external calibration device includes:

[0151] Step S601: Receive the fault data packet fed back by the execution control unit, which includes the abnormal feature vector when the yarn breakage occurs and real-time environmental parameters;

[0152] The fault data packet is an information carrier automatically encapsulated and uploaded by the system after detecting an abnormal state (such as an emergency stop). Its content includes not only abnormal feature vectors extracted by the image processing module, such as key indicators like yarn diameter, hairiness density, vibration frequency, and tension fluctuations, but also environmental parameters at the time, such as sensor readings for spinning speed, temperature, and humidity. This information collectively constitutes the set of criteria for determining whether a yarn breakage event has actually occurred.

[0153] Understandably, environmental parameters provide context and help identify factors that may cause misjudgments under certain conditions. For example, tension sensors are susceptible to interference and false alarms under high temperature and high humidity conditions.

[0154] Step S602: Call the measured data of yarn physical quantities provided by the external calibration equipment;

[0155] Because image recognition or indirect sensing methods may be subject to systematic errors or noise interference, relying solely on internal data cannot fully guarantee accurate judgments. Therefore, the system connects to external calibration devices such as laser diameter gauges and high-frequency tension sensors via industrial bus interfaces (e.g., Profinet, EtherCAT, etc.) to obtain more precise data on physical quantities such as yarn diameter and instantaneous tension fluctuations. This design not only enhances the authority and reliability of the data source but also provides a benchmark reference for subsequent deviation calculations.

[0156] It should be noted that, in order to ensure the time consistency between data from different sources, it is also necessary to perform timestamp alignment processing on the data streams from different devices, that is, to accurately match the image feature acquisition time with the sampling time of the external sensor, so as to avoid the deviation calculation distortion caused by time sequence misalignment.

[0157] Step S603: Based on the deviation between the abnormal feature vector in the fault data packet and the measured data, calculate the parameter correction amount and mark the actual yarn breakage state to obtain the statistical analysis results of false alarm / missed alarm.

[0158] For example, taking yarn diameter as an example, the system compares the yarn diameter value obtained from image recognition with the measured diameter provided by the laser diameter gauge, calculates the difference between the two, and uses this as the basis for diameter threshold correction. This deviation not only reflects the error level of the current detection system, but also reveals whether the threshold set in the preset knowledge base is reasonable.

[0159] Furthermore, by labeling actual yarn breakage states—positive samples indicating system-triggered and actual yarn breakage, and negative samples indicating false alarms or missed alarms—a labeled training sample set can be established, providing a foundation for subsequent model optimization. Sample labeling is essentially a labeling process in supervised learning, a typical binary classification problem-solving method that helps the system distinguish between valid and invalid alarms, thus providing a basis for parameter adjustment decisions. In addition, this deviation calculation process can be extended to other feature dimensions, such as tension fluctuations and vibration frequencies, forming a multi-dimensional deviation matrix, providing comprehensive support for evaluating system performance.

[0160] Step S604: Iteratively update the threshold parameters in the preset knowledge base based on the parameter correction amount;

[0161] Traditional threshold settings are often based on experience or static rules, making them difficult to adapt to complex and ever-changing production environments. This application employs an exponentially weighted moving average algorithm to update key threshold parameters, such as the diameter threshold. EWMA is a commonly used time series smoothing method that assigns higher weights to recent observations while gradually decaying the weights on historical data, thus maintaining stability while rapidly responding to new changes.

[0162] For example, the new diameter threshold parameter T_new = α × T_old + (1 − α) × median of the measured diameter, where α is a smoothing factor and T_old is the original threshold parameter, determining the fusion ratio of the old and new data. When multiple negative samples (i.e., false alarms) occur consecutively, the system also triggers a threshold adjustment mechanism. For instance, if the false alarm rate triggered by tension fluctuation features exceeds 15% in 30 consecutive false alarms, its judgment threshold is increased by 10% to reduce the probability of future false alarms. This adaptive threshold adjustment mechanism not only enhances the robustness of the system but also effectively addresses long-term influencing factors such as equipment aging and environmental drift, thereby achieving the intelligent evolution of the knowledge base.

[0163] Step S605: Based on the statistical analysis results of false positives / false negatives, adjust the feature weight coefficients in the preset knowledge base.

[0164] The system identifies which features are more likely to cause incorrect judgments by statistically analyzing the contribution rate of each feature in false alarm events over the past 24 hours, and dynamically reduces their weight accordingly. For example, if a feature appears frequently in false alarms, it indicates that its discrimination ability is weak or easily interfered with, and its influence in the final decision should be appropriately reduced.

[0165] The specific adjustment formula is: w i =w i ×(1−0.2×false alarm rate), where w i w represents the original weights. iThe adjusted weights are represented by a penalty coefficient of 0.2. This mechanism is essentially a feature importance re-evaluation mechanism based on false alarm feedback, which can effectively suppress the interference of noisy features on the overall judgment.

[0166] Furthermore, when the number of newly added positive samples reaches a certain threshold (e.g., 50 groups), the system will initiate the retraining process of the convolutional neural network classifier. Through incremental learning, it continuously optimizes the model structure and parameters, thereby improving the accuracy and generalization ability of yarn breakage detection. This conditional triggering mechanism avoids the waste of resources caused by frequent retraining, while ensuring that the model can maintain good performance when facing new scenarios.

[0167] The above embodiments enable dynamic updating and intelligent optimization of the knowledge base parameters in the yarn breakage detection system, significantly improving the system's adaptability and detection accuracy. This technical solution is not only applicable to yarn breakage detection scenarios in the textile industry, but can also be extended to other industrial detection fields requiring high real-time performance and accuracy, demonstrating broad application prospects and technological promotion value.

[0168] Reference Figure 7 As a further implementation of the eddy current spinning yarn breakage detection method, after the steps of executing the spinning equipment control operation corresponding to the control command and collecting fault data packets during the execution process, the method further includes:

[0169] Step S701: parse the abnormal feature vector and real-time environmental parameters contained in the fault data packet;

[0170] The fault data package includes, but is not limited to, the density of yarn surface hairs, the frequency spectrum characteristics of mechanical vibrations generated during equipment operation, the instantaneous change curve of yarn tension, and yarn geometric parameters. These parameters exhibit relatively stable statistical regularities under normal production conditions, but once raw material quality deteriorates, mechanical parts wear out, or process conditions deviate from the optimal range, the corresponding characteristic indicators will show obvious deviation trends.

[0171] Meanwhile, the synchronous acquisition of environmental parameters is crucial for improving diagnostic accuracy. Among these, temperature (T), humidity (H), and spinning speed (V) constitute the fundamental boundary conditions affecting fiber behavior and equipment dynamic response. Temperature changes directly affect the fiber's plastic deformation capacity and coefficient of friction, thereby altering the fiber's aggregation state in the eddy current field. Humidity fluctuations relate to the distribution of static charge on the fiber surface and the magnitude of the adhesion forces between fibers, exerting a subtle but significant effect on yarn uniformity. Spinning speed, as a key process control parameter, determines the number of tensile stress cycles and the intensity of cumulative fatigue effects borne by the fiber per unit time.

[0172] Step S702: Combine the abnormal feature vector with the real-time environmental parameters to form diagnostic input data, and perform similarity matching with the preset historical fault case library;

[0173] Among them, the historical failure case library is a collection of expert knowledge formed by systematically summarizing and organizing typical failure events accumulated in long-term production practice. Each case contains a complete failure description template, covering multiple dimensions of information elements such as failure type identification, cause classification, range of associated characteristic parameters, and corresponding handling strategy suggestions.

[0174] In this process, the diagnostic input data is not simply a list of raw measurement values, but a standardized expression after standardized preprocessing, ensuring that physical quantities of different dimensions and orders of magnitude can participate in comparison calculations within a unified framework.

[0175] To quantify the similarity between new and old samples, similarity measurement models are usually established using Euclidean distance, cosine angle, Mahalanobis distance, or more complex kernel function mapping, taking into account the differences in the contribution weights of each feature dimension to the final classification result.

[0176] It should be noted that the introduction of environmental parameters greatly enhances matching accuracy because it effectively eliminates false positives caused by factors such as seasonal changes and regional climate variations, thereby improving the robustness and generalization ability of diagnostic conclusions. When the candidate case with the highest matching score is selected, not only can the best response plan for the current situation be obtained, but it can also correct and improve the rule entries of the knowledge base itself, forming a virtuous cycle mechanism of self-evolution.

[0177] Step S703: When the mutation rate of the hair density characteristic value relative to the historical average exceeds the preset first threshold and the vibration frequency offset exceeds the preset second threshold, the fiber raw material defect detection process is started.

[0178] Among these, fiber hairiness density is an extremely sensitive apparent quality indicator, its value directly reflecting the cohesion and surface smoothness of the fibers during twisting. Under normal circumstances, hairiness is randomly distributed along the yarn axis, and its density value remains within a relatively narrow and controllable range. However, if the raw materials have problems such as excessive short fiber content, poor fiber length uniformity, or uneven oiling, a large number of free ends will form at the exit of the drafting zone. These insufficiently covered fiber bundles are very easy to detach from the main structure under external disturbance, forming protruding hairiness. This trend often has the characteristics of suddenness and rapid spread. If it cannot be identified and the propagation path blocked in time, it will quickly spread to downstream processes, causing batch waste.

[0179] At the same time, the vibration frequency shift phenomenon reveals the potential changes in the dynamic characteristics of the equipment. In particular, the abnormal phenomena such as resonance peak drift and the enhancement of harmonic components near high-speed rotating parts are likely caused by the imbalance of inertial torque caused by unbalanced load.

[0180] In this context, the coupling relationship between two seemingly independent monitoring signals suggests a common triggering factor: the additional disturbance caused by substandard raw material performance. Therefore, setting dual triggering conditions not only improves the selectivity of the early warning mechanism and avoids the risk of false alarms caused by occasional noise interference, but also demonstrates the advantages of multi-sensor fusion diagnostics.

[0181] Step S704: When the tension fluctuation characteristic value is detected to continuously exceed the preset safety threshold and the rate of change of the yarn diameter characteristic value is lower than the preset rate of change threshold, the mechanical fault diagnosis process is initiated.

[0182] Specifically, in theory, an ideal tension control system should be able to automatically compensate for the effects of various internal and external disturbances such as raw material variations, fluctuations in workshop temperature and humidity, and spindle speed adjustments, and always maintain the output tension in a small oscillation around the predetermined target value.

[0183] However, in reality, due to wear and tear on the transmission mechanism, lag in the control system response, and drift in feedback loop parameters, significant tension fluctuations or even loss of control frequently occur. Of particular concern is that simple tension anomalies are not always accompanied by a significant change in yarn diameter. This is because modern precision drafting systems possess strong adaptive adjustment capabilities, which can, to some extent, offset the negative impact of tension disturbances by adjusting the draft ratio. However, this passive compensation mechanism has its limits; severe tension impacts exceeding the tolerance range can still cause serious consequences such as fiber breakage, sliver knotting, and yarn breakage and splashing.

[0184] Therefore, this application innovatively proposes to combine the duration and amplitude of tension fluctuations as the basis for judgment, and supplement it with the yarn diameter change rate as an auxiliary constraint, so as to accurately locate mechanical faults that are in the nascent stage and have not yet fully developed.

[0185] Understandably, tension fluctuations exceeding the safety threshold continuously indicate abnormal yarn tension, potentially caused by mechanical component wear, transmission system malfunctions, or sensor abnormalities. Conversely, a yarn diameter change rate below the preset threshold suggests a relatively stable yarn forming process, ruling out tension anomalies caused by yarn quality fluctuations. This combination of "abnormal tension but stable diameter" strongly points to a mechanical equipment malfunction rather than a process parameter issue. Therefore, initiating a mechanical fault diagnosis process can accurately pinpoint the root cause, avoid misdiagnosis and missed detections, and improve the accuracy of fault diagnosis and maintenance efficiency.

[0186] It should be noted that steps S703 and S704 are process steps executed separately according to different situations, so there is no distinction in the order of execution; they can be executed when the corresponding conditions are met.

[0187] In the above embodiments, a multi-layered, progressive intelligent diagnostic architecture is constructed, achieving full-process coverage from macroscopic state perception to microscopic root cause tracing. This technical solution utilizes a historical knowledge base to conduct rapid matching and screening based on similarity, and then activates more targeted specialized detection modules according to the characteristics of different types of faults, ultimately achieving the dual prevention and control goals of raw material defects and potential mechanical equipment hazards.

[0188] Reference Figure 8 As a further implementation of the eddy current spinning yarn breakage detection method, the specific steps of the fiber raw material defect detection process include:

[0189] Step S801: Call the externally connected fiber fineness detection device to obtain the fineness distribution data of the current yarn raw material batch;

[0190] Among them, the statistical distribution information of fiber diameter or cross-sectional area can be obtained through laser diffraction, capacitance sensing or other precision measurement methods. This fineness distribution data essentially reflects the geometric dimensional uniformity characteristics of fiber raw materials and is a key physical indicator for evaluating the quality of raw materials.

[0191] Step S802: Determine whether the dispersion of the fineness distribution data exceeds the preset raw material defect threshold; if yes, proceed to step S803; if no, do not perform any operation.

[0192] Step S803: Output the raw material batch replacement alarm command to the spinning equipment operation and maintenance system.

[0193] Specifically, when the dispersion of the fineness distribution data exceeds the preset raw material defect threshold, it means that there is a significant unevenness in the fineness of the batch of raw materials. This unevenness may lead to problems such as increased breakage rate, uneven yarn strength, and uneven yarn evenness in the subsequent spinning process.

[0194] In the above implementation, the system outputs a raw material batch replacement alarm command to the spinning equipment operation and maintenance system. By promptly replacing the problematic raw material batches, defective products are avoided, reflecting the concepts of predictive maintenance and quality control in modern intelligent manufacturing.

[0195] Reference Figure 9 As a further implementation of the eddy current spinning yarn breakage detection method, the mechanical fault diagnosis process includes:

[0196] Step S901: Collect real-time spectrum data from the vibration sensor of the spinning equipment;

[0197] Vibration sensors typically employ piezoelectric, capacitive, or magnetoelectric sensors, which can convert the mechanical vibration of mechanical equipment into electrical signals. Through digital signal processing techniques such as Fast Fourier Transform (FFT), the time-domain signals are converted into frequency-domain signals, thereby obtaining frequency characteristic information of the equipment's operating status.

[0198] Step S902: Determine whether there are abnormal frequency components in the real-time spectrum data whose correlation with the spindle fundamental frequency harmonics exceeds a preset overlap degree; if yes, proceed to step S903; if no, do not perform any operation.

[0199] Step S903: Output the spindle bearing maintenance command to the spinning equipment operation and maintenance system.

[0200] As the core rotating component of spinning equipment, the spindle generates a specific fundamental frequency signal and its integer multiples of harmonic components during normal operation. These frequency components constitute the characteristic frequency spectrum of the equipment's normal operation. When mechanical faults such as wear, poor lubrication, or fatigue damage occur in the spindle bearings, specific fault frequency components will be generated in the vibration signal. These fault frequencies often have a certain correlation with the spindle's rotational speed frequency and its harmonics.

[0201] In this embodiment, by calculating the correlation coefficient between abnormal frequency components in the spectral data and the spindle's fundamental harmonics, and comparing it with a preset overlap threshold, early signs of spindle bearing failure can be effectively identified. This fault diagnosis method based on spectral correlation analysis has high sensitivity and accuracy, enabling the timely detection of potential problems before serious equipment failures occur.

[0202] In the above implementation, when the fault determination conditions are met, the system outputs a spindle bearing maintenance command to the spinning equipment operation and maintenance system, realizing automated closed-loop management from fault detection to maintenance decision-making. This maintenance strategy falls under the category of predictive maintenance, which has significant advantages over traditional periodic maintenance or post-fault repair, significantly reducing equipment downtime, maintenance costs, and extending equipment lifespan.

[0203] This application establishes a comprehensive quality assurance mechanism from raw materials to production equipment by constructing a dual monitoring system of raw material quality inspection and equipment fault diagnosis. The fiber raw material defect detection process, through statistical analysis of the dispersion of fineness distribution data, enables early identification and warning of raw material quality problems, effectively preventing quality accidents caused by the use of inferior raw materials in production. The mechanical fault diagnosis process utilizes vibration spectrum analysis technology to achieve precise location and predictive maintenance of key bearing component faults through in-depth analysis of the correlation of spindle fundamental frequency harmonics. The two processes work together to ensure both the stability of input raw material quality and the good operating condition of production equipment, thereby significantly improving the product quality consistency, production efficiency, and equipment reliability of the entire spinning production line, reflecting the technological development trend of modern textile industry towards intelligent and digital transformation and upgrading.

[0204] This application also discloses an image feature-based eddy current spinning yarn breakage detection system.

[0205] A vortex spinning yarn breakage detection system based on image features, the detection system comprising:

[0206] The image acquisition module is used to acquire dual-channel raw image sequences of eddy current spinning yarn, including synchronously acquired infrared light channel images and blue light channel images;

[0207] The image processing module is used to preprocess the dual-channel original image sequence, generate an enhanced image, and extract the yarn centerline coordinates and hair distribution matrix from the enhanced image;

[0208] The feature extraction module is used to extract spatiotemporal features based on the yarn centerline coordinates and the hair distribution matrix, and generate feature vectors and a temporal feature matrix composed of feature vectors from multiple consecutive frames.

[0209] The yarn breakage risk control module is used to calculate the yarn breakage risk index and generate control instructions based on the feature vector, the time-series feature matrix and the parameters in the preset knowledge base.

[0210] The execution feedback module is used to execute the control operations of the spinning equipment corresponding to the control commands, and to collect fault data packets during the execution process;

[0211] The parameter update module is used to update the parameters in the preset knowledge base based on the fault data packet and the measurement data fed back by the external calibration equipment.

[0212] The image feature-based eddy current spinning yarn breakage detection system of this application embodiment can implement any of the above-mentioned eddy current spinning yarn breakage detection methods, and the specific working process of each module in the eddy current spinning yarn breakage detection system can refer to the corresponding process in the above-mentioned method embodiments.

[0213] In the several embodiments provided in this application, it should be understood that the provided methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for example, the division of a certain module is merely a logical functional division, and in actual implementation there may be other division methods, such as multiple modules can be combined or integrated into another system, or some features can be ignored or not executed.

[0214] In this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0215] Although this application has been described herein in conjunction with various embodiments, those skilled in the art, by reviewing the accompanying drawings, disclosure, and appended claims, will understand and implement other variations of the disclosed embodiments in carrying out the claimed application. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude a plurality. A single processor or other unit can implement several functions listed in the claims. While different dependent claims may recite certain measures, this does not mean that these measures cannot be combined to produce a good effect.

[0216] The above are all preferred embodiments of this application and are not intended to limit the scope of protection of this application. Any feature disclosed in this specification (including the abstract and drawings) may be replaced by other equivalent or similar features unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is only one example of a series of equivalent or similar features.

Claims

1. A method of detecting a broken yarn in a vortex spinning process based on image features, characterized by, The detection method comprises: Collecting a double-channel original image sequence of the vortex spun yarn, including synchronously acquired infrared light channel images and blue light channel images; Pretreating the double-channel original image sequence to generate an enhanced image, and extracting yarn center line coordinates and a hair distribution matrix in the enhanced image; Extracting spatio-temporal features based on the yarn center line coordinates and the hair distribution matrix to generate a feature vector and a time sequence feature matrix composed of continuous multiple frames of feature vectors; According to the feature vector, the time sequence feature matrix, and parameters in a preset knowledge base, calculating a yarn breakage risk index and generating a control instruction; Performing a spinning equipment control operation corresponding to the control instruction, and collecting fault data packets in the execution process; According to the fault data packets and measurement data fed back by an external calibration device, updating the parameters in the preset knowledge base.

2. A yarn breakage detection method based on image features of a vortex spinning according to claim 1, characterized in that, Before the step of collecting the double-channel original image sequence of the vortex spun yarn, further comprising: Real-time acquisition of real-time process parameters of the spinning equipment and ambient light intensity sensor data; wherein the real-time process parameters include spinning speed, nozzle air pressure, and ambient temperature and humidity; Calculating a dynamic frame rate adjustment coefficient based on the real-time process parameters to generate a target image acquisition frame rate; Controlling the image sensor to collect the double-channel original image sequence based on the target image acquisition frame rate; According to a difference between the ambient light intensity sensor data and a preset reference light, calculating a pulse intensity compensation value of the double-channel light source; Adjusting the pulse width of the LED array of the double-channel light source based on the pulse intensity compensation value.

3. The yarn breakage detection method based on image features of a vortex spinning according to claim 1, characterized in that, The step of pretreating the double-channel original image sequence to generate an enhanced image, and extracting yarn center line coordinates and a hair distribution matrix in the enhanced image comprises: Parsing the double-channel original image sequence into infrared light channel images and blue light channel images; Performing edge enhancement processing on the infrared light channel images and contrast enhancement processing on the blue light channel images to obtain processed double-channel images; Based on a preset proportion coefficient, performing weighted fusion on the processed double-channel images to generate a fusion image; Performing adaptive filtering processing on the fusion image to output an enhanced image; Loading a historical yarn trajectory, predicting a current frame of yarn trajectory based on a Kalman filtering algorithm, and locating yarn center line coordinates in the enhanced image; Extracting a pixel set outside a yarn main body region in the enhanced image, separating hair pixel points through a threshold segmentation method, and mapping coordinates of the hair pixel points into a hair distribution matrix.

4. A yarn breakage detection method based on image features of a vortex spinning according to claim 3, characterized in that, The step of extracting spatio-temporal features based on the yarn center line coordinates and the hair distribution matrix to generate a feature vector and a time sequence feature matrix composed of continuous multiple frames of feature vectors comprises: Receiving a yarn center line coordinate sequence and a hair distribution matrix; Based on a current frame of enhanced image and yarn center line coordinates, calculating yarn diameter and hair density; Constructing a continuous multiple frames of yarn center line coordinate sequence and performing fast Fourier transform to extract main spectral component amplitudes to obtain vibration frequency features; Combining a standard deviation of a continuous multiple frames of yarn diameter data sequence and a preset yarn elastic modulus to calculate a tension fluctuation value; The yarn diameter, hairiness density, vibration frequency and tension fluctuation value are combined as a feature vector; The continuous feature vectors are integrated according to a preset fixed time window to generate a time sequence feature matrix.

5. A yarn breakage detection method based on image features of a vortex spinning according to claim 4, characterized in that, The step of calculating a yarn breakage risk index and generating a control instruction according to the feature vector, the time sequence feature matrix and parameters in a preset knowledge base comprises: receiving the feature vector and the time sequence feature matrix at the current moment; calling the parameter threshold value and the feature weight coefficient matched with the yarn type in the preset knowledge base; weighting and calculating the yarn breakage risk index according to each component in the feature vector according to the parameter threshold value and the feature weight coefficient; generating a hierarchical control instruction based on the yarn breakage risk index and the tension fluctuation value; real-time recording the time sequence feature matrix and the corresponding hierarchical control instruction, and merging as a learning sample set to update the preset knowledge base.

6. The yarn breakage detection method based on image features of a vortex spinning according to claim 1, characterized in that, The step of updating the parameters in the preset knowledge base according to the fault data packet and the measurement data fed back by the external calibration equipment comprises: receiving the fault data packet fed back by the execution control unit, including the abnormal feature vector and the real-time environmental parameter when the yarn breakage occurs; calling the measured data of the physical quantity of the yarn provided by the external calibration equipment; calculating the parameter correction amount and marking the actual yarn breakage state according to the deviation between the abnormal feature vector in the fault data packet and the measured data, to obtain a false alarm / miss alarm statistical analysis result; iteratively updating the threshold parameters in the preset knowledge base based on the parameter correction amount; adjusting the feature weight coefficient in the preset knowledge base based on the false alarm / miss alarm statistical analysis result.

7. A method of detecting a broken yarn in a vortex spinning process based on image features according to any one of claims 1 to 6, characterized in that, After the step of executing the spinning equipment control operation corresponding to the control instruction and collecting the fault data packet in the execution process, further comprising: analyzing the abnormal feature vector and the real-time environmental parameter included in the fault data packet; combining the abnormal feature vector and the real-time environmental parameter as diagnostic input data, and performing similarity matching with a preset historical fault case library; when it is detected that the mutation rate of the hairiness density feature value relative to the historical average value exceeds a preset first threshold value, and the vibration frequency offset exceeds a preset second threshold value, starting a fiber raw material defect detection process; when it is detected that the tension fluctuation feature value continuously exceeds a preset safety threshold value and the change rate of the yarn diameter feature value is lower than a preset change rate threshold value, starting a mechanical fault diagnosis process.

8. A yarn breakage detection method based on image features of a vortex spinning according to claim 7, characterized in that, The specific steps of the fiber raw material defect detection process comprise: calling an externally connected fiber fineness detection device to obtain fineness distribution data of the current yarn raw material batch; if the dispersion degree of the fineness distribution data exceeds a preset raw material defect threshold value, outputting a raw material batch replacement alarm instruction to a spinning equipment operation and maintenance system.

9. The yarn breakage detection method based on image features of a vortex spinning according to claim 7, characterized in that, The mechanical fault diagnosis process comprises: collecting real-time frequency spectrum data of a spinning equipment vibration sensor; if there is an abnormal frequency component related to the spindle base frequency harmonic in the real-time frequency spectrum data, which exceeds a preset coincidence degree, outputting a spindle bearing maintenance instruction to a spinning equipment operation and maintenance system.

10. An image feature based vortex spinning broken end detection system, characterized in that, The detection system comprises: an image acquisition module, configured to acquire a double-channel original image sequence of the vortex spun yarn, including an infrared light channel image and a blue light channel image acquired synchronously; An image processing module is configured to preprocess the double-channel raw image sequence, generate an enhanced image, and extract yarn centerline coordinates and a hair distribution matrix in the enhanced image; A feature extraction module is configured to perform spatiotemporal feature extraction based on the yarn centerline coordinates and the hair distribution matrix, generate a feature vector and a time-series feature matrix composed of continuous multiple frames of feature vectors; A yarn breakage risk control module is configured to calculate a yarn breakage risk index and generate a control instruction according to the feature vector, the time-series feature matrix and parameters in a preset knowledge base; An execution feedback module is configured to perform a spinning equipment control operation corresponding to the control instruction, and collect fault data packets in the execution process; A parameter updating module is configured to update the parameters in the preset knowledge base according to the fault data packets and measurement data fed back by an external calibration device.

Images