Intelligent control method for textile finished product production line

Abstract

The invention relates to the field of textile production, and discloses an intelligent control method for a textile finished product production line, which comprises the following steps: acquiring a textile surface image through a camera array to obtain textile surface image data; performing fiber arrangement density analysis on the textile surface image data to obtain fiber arrangement density difference data; according to the yarn gap distribution characteristic extraction method and device, environmental factors are monitored and compensated in real time through the temperature and humidity correlation deformation correction coefficient, it is ensured that under different environmental conditions, fabric tension is kept stable, the product quality and the production stability are further improved, and the yarn gap distribution characteristic data are obtained. An equipment cooperative control model constructed by using a depth deterministic strategy gradient algorithm realizes real-time speed adjustment and tension compensation of the textile machinery group through edge computing nodes, and the collaboration and automation level of the production process are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of textile production technology, specifically to an intelligent control method for textile finished product production lines. Background Technology

[0002] In traditional textile production, fabric quality, production efficiency, and the coordinated control of textile machinery typically rely on manual adjustments and empirical rules. This approach is inherently uncertain and easily affected by environmental factors such as temperature and humidity changes, leading to unstable fabric tension and consequently impacting production efficiency and product quality. Furthermore, traditional textile production lines lack precise quality monitoring and real-time control methods, making it difficult to accurately assess and adjust crucial parameters such as surface quality, yarn gaps, and fabric tension. Therefore, a textile production line control method based on intelligent control technology is needed, capable of real-time data acquisition and dynamic adjustments to cope with various external environmental changes and production status fluctuations.

[0003] Currently, some methods have attempted to introduce intelligent monitoring and control into textile production processes, such as using machine vision to detect fabric defects and employing control algorithms for tension compensation. However, these methods mostly focus on single control factors and struggle to comprehensively consider multiple variables in textile production, such as fiber arrangement, yarn gap distribution, ambient temperature and humidity, and production cycle time. Furthermore, existing technologies suffer from low system integration and lack in-depth application of coordinated equipment control, making precise dynamic adjustments impossible. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent control method for textile finished product production lines, in order to solve the problems of low system integration, lack of in-depth application of equipment collaborative control, and inability to achieve precise dynamic adjustment in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an intelligent control method for a textile finished product production line, comprising: Textile surface images are acquired using a camera array to obtain textile surface image data; fiber arrangement density analysis is performed on the textile surface image data to obtain fiber arrangement density difference data; yarn gap distribution features are extracted based on the fiber arrangement density difference data to obtain yarn gap distribution feature data. Based on the yarn gap distribution characteristic data, the warp and weft yarn interlacing points are dynamically tracked to obtain the warp and weft yarn interlacing point motion trajectory data; based on the warp and weft yarn interlacing point motion trajectory data, the fabric tension consumption is evaluated to obtain the fabric tension consumption evaluation data; combined with the fabric tension consumption evaluation data, the production cycle adaptability analysis is performed to obtain the initial production cycle adaptability data. Obtain the workshop environment temperature and humidity data; according to the workshop environment temperature and humidity data, the material deformation coefficient is corrected to the fabric tension consumption evaluation data, and the temperature and humidity related deformation correction coefficient is obtained; based on the temperature and humidity related deformation correction coefficient, the initial production rhythm adaptive data is calculated, and the dynamic rhythm compensation data is obtained; Through the deep deterministic policy gradient algorithm, the dynamic rhythm compensation data is used for equipment collaborative control model construction, and the equipment collaborative control model is obtained; the equipment collaborative control model is deployed to the edge computing node, and the model output parameter is used for real-time speed regulation and tension compensation of the textile machinery group, and the intelligent control of the whole process of textile production is executed.

[0006] Further, the camera array is used for textile surface image acquisition, and fiber arrangement density difference data is obtained, including: The camera array is used for synchronous acquisition of textile surface image, and original textile surface image data is obtained; the original textile surface image data is subjected to self-adaptive filtering and denoising treatment, and denoised textile surface image data is obtained; based on the denoised textile surface image data, fiber direction feature extraction is carried out, and fiber direction feature data is obtained; According to the fiber direction feature data, local area density clustering analysis is carried out, and fiber density clustering center data is obtained; combined with the fiber density clustering center data, the density difference heat map is constructed, and the fiber density difference heat map data is obtained; the abnormal density area marking data is obtained by using the dynamic threshold segmentation algorithm to mark the fiber density difference heat map data; Obtain the camera array calibration parameters; according to the camera array calibration parameters, the three-dimensional space coordinate mapping of the abnormal density area marking data is carried out, and the three-dimensional abnormal density area coordinate data is obtained; based on the three-dimensional abnormal density area coordinate data, the fiber arrangement looseness calculation is carried out, and the regional fiber arrangement looseness data is obtained; The support vector machine algorithm is used for density level classification of regional fiber arrangement looseness data, and fiber density level classification data is obtained; combined with the fiber density level classification data, the full image density difference is counted, and the fiber arrangement density difference data is generated.

[0007] Further, according to the fiber arrangement density difference data, yarn gap distribution feature extraction is carried out, and yarn gap distribution feature data is obtained, including: The fiber arrangement density difference data is subjected to Gaussian smoothing pretreatment, and the denoised density difference data is obtained; based on the denoised density difference data, a multi-scale edge detection operator is constructed, and yarn boundary preliminary positioning data is obtained; the gap region extraction is carried out on the yarn boundary preliminary positioning data by using the adaptive threshold segmentation algorithm, and the yarn gap candidate region data is obtained; Obtaining laser scanner point cloud data; performing three-dimensional space verification on the yarn gap candidate region data according to the laser scanner point cloud data to obtain three-dimensional gap verification data; performing morphological closing operation processing on the three-dimensional gap verification data to obtain optimized yarn gap contour data.

[0008] Further, yarn gap distribution feature extraction is performed according to the fiber arrangement density difference data to obtain yarn gap distribution feature data, and the yarn gap distribution feature data further comprises: Based on the optimized yarn gap contour data, connected region labeling is performed to obtain independent yarn gap unit data; minimum circumscribed rectangle fitting is performed on the independent yarn gap unit data to obtain gap geometric parameter data; distribution mode recognition is performed on the gap geometric parameter data through a density clustering algorithm to obtain yarn gap distribution mode data; A generative adversarial network is introduced to perform data enhancement processing on the yarn gap distribution mode data to obtain enhanced gap distribution data; statistical feature extraction is performed based on the enhanced gap distribution data to generate a multi-dimensional feature vector.

[0009] Further, warp and weft yarn interlacing point dynamic tracking is performed based on the yarn gap distribution feature data to obtain warp and weft yarn interlacing point motion trajectory data, which comprises: Multi-scale morphological opening operation processing is performed on the yarn gap distribution feature data to eliminate small gap noise to obtain optimized yarn gap data; a warp and weft yarn crossing region detection operator is constructed based on the optimized yarn gap data, and candidate interlacing points are located through direction gradient histogram feature matching to obtain initial interlacing point coordinate data; A camera array is introduced to capture continuous frame textile surface images, and cross-frame optical flow tracking is performed in combination with the initial interlacing point coordinate data to obtain interlacing point sub-pixel level motion vector data; Kalman filtering algorithm is used to perform trajectory smoothing processing on the motion vector data to generate preliminary interlacing point motion trajectory data; Laser scanner point cloud data is obtained, and the preliminary interlacing point motion trajectory data is mapped to a three-dimensional space coordinate system; the Z-axis deviation of the two-dimensional trajectory is corrected through a point cloud registration algorithm to obtain three-dimensional interlacing point motion trajectory data; an interlacing point motion velocity field model is constructed based on the three-dimensional trajectory data to generate warp and weft yarn interlacing point motion state label data.

[0010] Further, fabric tension consumption evaluation is performed according to the warp and weft yarn interlacing point motion trajectory data to obtain fabric tension consumption evaluation data, which comprises: Kinematic feature extraction is performed on the warp and weft yarn interlacing point motion trajectory data to calculate the interlacing point instantaneous speed, acceleration and direction change frequency to generate interlacing point dynamic behavior feature data; an interlacing point motion energy model is constructed based on the dynamic behavior feature data to quantify the change amount of kinetic energy per unit time to obtain interlacing point kinetic energy fluctuation data; Introduce the yarn elastic modulus database, combine the interlacing point kinetic energy fluctuation data to carry out tension kinetic energy mapping analysis, identify the abnormal tension fluctuation area through dynamic threshold comparison, and generate initial fabric tension consumption label data; adopt the isolated forest algorithm to detect the abnormal points of the initial label data, and obtain the optimized fabric tension consumption label data; Obtain the real-time running parameters of the textile machinery, correlate and analyze the optimized fabric tension consumption label data and the mechanical parameters, and construct a tension consumption process parameter correlation matrix; extract the dominant influencing factors in the correlation matrix through principal component analysis, and generate tension consumption dominant factor data; Based on the tension consumption dominant factor data, a long short-term memory network tension prediction model is constructed, and historical tension data is input to train the model parameters; the trained LSTM model is deployed to the edge computing node, the tension consumption value at the future time is predicted through real-time interlacing point trajectory data, and fabric tension consumption evaluation data is generated.

[0011] Further, the production rhythm adaptability analysis is carried out combined with the fabric tension consumption evaluation data, and the initial production rhythm adaptation data is obtained, including: The fabric tension consumption evaluation data is subjected to time sequence feature decomposition, the tension fluctuation period, peak value distribution density and energy attenuation rate are extracted, and tension dynamic behavior feature data is generated; based on the dynamic behavior feature data, a production rhythm tension response surface model is constructed, the tension consumption trend under different rhythm settings is quantified, and rhythm tension correlation mapping data is obtained; Introduce the real-time running parameters of the loom, fuse the rhythm tension correlation mapping data and the mechanical parameters in multiple dimensions, construct a production rhythm adaptability evaluation matrix; through fuzzy C-means clustering algorithm, the rhythm adaptability of the evaluation matrix is graded, the optimal rhythm interval with low tension consumption is identified, and initial production rhythm adaptation label data is generated; Obtain the rhythm adjustment records in the historical production data, and construct a reinforcement learning environment model combined with the initial production rhythm adaptation label data; adopt the deep Q network algorithm to simulate the tension consumption feedback under different rhythm adjustment strategies, optimize the rhythm selection strategy through the reward function, and generate dynamic rhythm adjustment strategy data; Based on the dynamic rhythm adjustment strategy data, a production rhythm tension consumption prediction model is constructed, and the optimal production rhythm is predicted by inputting the fabric tension data; the predicted rhythm parameters are transmitted to the loom control system, the rhythm self-adaptive adjustment is executed, the initial production rhythm adaptation data is generated and fed back to the central control console.

[0012] Further, obtain the workshop environment temperature and humidity data, obtain the temperature and humidity correlation deformation correction coefficient, including: An array of temperature and humidity sensors is deployed in key areas of the workshop to collect multi-dimensional environmental data in real time through a wireless transmission protocol. The original temperature and humidity data is denoised and temporally and spatially aligned through Kalman filtering to obtain calibrated workshop environmental temperature and humidity data. A database of fiber material deformation characteristics is constructed to store parameters such as yarn elastic modulus, thermal expansion coefficient, and moisture absorption expansion rate under different temperature and humidity conditions. Based on the calibrated environmental temperature and humidity data, the database is queried to obtain the material deformation reference value corresponding to the current environment through a bilinear interpolation algorithm, and temperature and humidity material deformation mapping data is generated. The temperature and humidity material deformation mapping data is fused and analyzed with the fabric tension consumption evaluation data, and a temperature and humidity tension consumption coupling model is established using partial least squares regression to quantify the contribution of environmental factors to tension fluctuations and generate initial deformation correction coefficients. The initial correction coefficients are simulated under extreme conditions using a generative adversarial network to generate virtual deformation data in high-temperature and high-humidity, low-temperature and low-humidity edge scenarios. The coupling model parameters are optimized through transfer learning to improve the adaptability of the correction coefficients to unconventional environments, and temperature and humidity-related deformation correction coefficients are obtained and transmitted to the dynamic beat compensation module.

[0013] Furthermore, based on the temperature and humidity-related deformation correction coefficients, dynamic beat compensation calculations are performed on the initial production beat adaptation data to obtain dynamic beat compensation data, including: A temperature and humidity beat compensation mapping model is constructed to fuse the temperature and humidity-related deformation correction coefficients with the initial production beat adaptation data, extract key environmental beat-related features through an attention mechanism, and generate compensation weight matrix data. Environmental monitoring stream data is introduced, combined with the compensation weight matrix data to perform dynamic weight adjustment, and a sliding window algorithm is used to capture temperature and humidity mutation signals to generate instantaneous beat compensation increment data. A digital twin simulation platform is constructed to input the instantaneous beat compensation increment data into a virtual weaving environment model, simulate the changes in fabric tension after compensation through finite element analysis, and generate simulation optimization suggestion data. Based on the simulation optimization suggestion data, a deep reinforcement learning compensation strategy model is constructed, with the deviation between the actual production beat and the target beat as the reward signal. The compensation parameters are optimized through a policy gradient algorithm to generate dynamic beat compensation data and synchronize it to the loom control system.

[0014] Furthermore, a device collaborative control model is constructed through a deep deterministic policy gradient algorithm on the dynamic beat compensation data to obtain a device collaborative control model, including: A multi-device collaborative simulation environment is constructed to integrate dynamic beat compensation data and device state parameters, and a virtual control scenario is generated through digital twin technology. Based on the simulation environment, a continuous action space is defined to convert the device collaborative control target into a reinforcement learning reward function. A deep deterministic policy gradient algorithm framework is designed, an Actor network outputs device control instructions, and a Critic network evaluates action values; dynamic beat compensation data is introduced as network input features, and through an attention mechanism, multiple device state information is fused to generate initial collaborative control policy data; In the simulation environment, DDPG policy iteration is performed, an experience replay mechanism is used to store historical control trajectories, and a target network is used to soften policy updates; a noise exploration mechanism is introduced to add orthogonal perturbations in the action output, expand the policy search space, and generate optimized collaborative control policy data; The trained DDPG model is deployed to an edge computing node to receive dynamic beat compensation data and device state feedback; through model output parameter adjustment, multiple device control instructions are executed to perform full-process device collaborative control in textile production, and a device collaborative control model is generated and packaged as a cloud platform callable service.

[0015] Compared with the prior art, the beneficial effects of the present application are: (1) The camera array and laser scanner can accurately identify the fiber arrangement density difference and timely detect abnormal areas by collecting and analyzing the textile surface image, and through dynamic adjustment of the yarn gap distribution characteristics, high-quality production of the fabric can be realized; (2) By combining fabric tension consumption evaluation data and production beat adaptability analysis, the production beat can be intelligently optimized to avoid unnecessary tension fluctuations, improve production efficiency, and reduce the rate of defective products; (3) By using the temperature and humidity correlation deformation correction coefficient, the environmental factors are monitored and compensated in real time to ensure that the fabric tension remains stable under different environmental conditions, further improving product quality and production stability; (4) The device collaborative control model constructed by the deep deterministic policy gradient algorithm (DDPG) realizes real-time speed adjustment and tension compensation of the textile machinery group through the edge computing node, effectively improving the collaboration and automation level of the production process. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 The flowchart of the method of the present application. DETAILED DESCRIPTION

[0017] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0018] Please refer to Figure 1The application provides a technical solution: an intelligent control method for a textile finished product production line, comprising: Textile surface image data is obtained by camera array textile surface image acquisition, and fiber arrangement density analysis is performed on the textile surface image data to obtain fiber arrangement density difference data; yarn gap distribution feature extraction is performed according to the fiber arrangement density difference data to obtain yarn gap distribution feature data; The camera array is composed of a system of multiple high-resolution cameras, and multi-dimensional image data of the textile surface is obtained by synchronous shooting; the fiber arrangement density analysis uses image processing technology to process the images collected by the camera array, quantifies the number and distribution uniformity of fibers in a unit area; the yarn gap distribution feature extraction is based on the fiber arrangement density data, and the gap area, shape and distribution law between the yarns are calculated by morphological operation to identify the gap area between the yarns; Based on the yarn gap distribution feature data, the warp and weft yarn interlacing point dynamic tracking is performed to obtain the warp and weft yarn interlacing point motion trajectory data; the fabric tension consumption evaluation is performed according to the warp and weft yarn interlacing point motion trajectory data to obtain the fabric tension consumption evaluation data; the production rhythm adaptability analysis is performed in combination with the fabric tension consumption evaluation data to obtain the initial production rhythm adaptation data; The warp and weft yarn interlacing point dynamic tracking is to use a target tracking algorithm to monitor the motion trajectory of the warp and weft yarn interlacing point in real time; the fabric tension consumption evaluation is to combine the interlacing point motion trajectory and calculate the tension loss of the fabric in the production process due to stretching, friction and other factors through a mechanical model; the production rhythm adaptability analysis is to predict the efficiency bottleneck under the current production speed according to the tension consumption data through a linear regression or neural network model to generate an initial rhythm adjustment scheme, so as to ensure that the equipment operation matches the process requirements; The workshop environment temperature and humidity data are obtained; the material deformation coefficient correction is performed on the fabric tension consumption evaluation data according to the workshop environment temperature and humidity data to obtain the temperature and humidity related deformation correction coefficient; the dynamic rhythm compensation calculation is performed on the initial production rhythm adaptation data based on the temperature and humidity related deformation correction coefficient to obtain the dynamic rhythm compensation data; The temperature and humidity related deformation correction coefficient is based on the environment temperature and humidity data, and combines the moisture absorption mechanics relationship model in textile material science to calculate the deformation compensation value of the material caused by temperature and humidity changes; the dynamic rhythm compensation calculation is to use the correction coefficient to adjust the initial rhythm data in real time, and a dynamic rhythm instruction is generated through a PID control algorithm or fuzzy logic to ensure that the production rhythm adapts to the environmental changes; The device collaborative control model is constructed by using a deep deterministic policy gradient algorithm on the dynamic rhythm compensation data to obtain the device collaborative control model; the device collaborative control model is deployed to an edge computing node, and the textile machinery group is adjusted in real time in speed and tension compensation through the model output parameters to execute the intelligent control of the whole process of textile production. wherein the deep deterministic policy gradient (DDPG) algorithm is a reinforcement learning algorithm that combines the Actor-Critic architecture with deep neural networks, outputs continuous actions such as device speed adjustment values through deterministic policy, and stabilizes the training process using experience replay and target network technology; the edge computing node is a computing device deployed near the production line, which has data preprocessing, local decision-making such as fault warning, protocol conversion (Modbus to MQTT), and security protection functions; the model output parameters include device speed adjustment amount, tension compensation value, etc., which are directly applied to textile machinery such as frequency converters and tension controllers through edge nodes to form a closed-loop control circuit, ensuring stable fabric quality.

[0019] It should be noted that during operation, through the whole-process closed loop of "visual inspection-data analysis-dynamic control-environment adaptation", the textile production is fully intelligentized from quality detection to device control, effectively improving product consistency, production efficiency and environmental adaptability, while reducing energy consumption and cost.

[0020] In an embodiment, textile surface image acquisition is performed through a camera array to obtain fiber arrangement density difference data, including: Synchronous acquisition of textile surface images is performed through a camera array to obtain original textile surface image data; adaptive filtering and denoising processing is performed on the original textile surface image data to obtain denoised textile surface image data; fiber direction feature extraction is performed based on the denoised textile surface image data to obtain fiber direction feature data; Local region density clustering analysis is performed according to the fiber direction feature data to obtain fiber density clustering center data; density difference heat map construction is performed in combination with the fiber density clustering center data to obtain fiber density difference heat map data; abnormal density region marking is performed on the fiber density difference heat map data through a dynamic threshold segmentation algorithm to obtain abnormal density region marking data; Fiber density clustering center calculation quantifies fiber distribution uniformity through local density clustering, and the formula is as follows: wherein, : the coordinate of the i-th density clustering center, unit: millimeter; : the coordinate of the i-th density clustering center, unit: millimeter; : the pixel set in the i-th clustering region; : the pixel set in the i-th clustering region; : the pixel set in the i-th clustering region; ​Obtain camera array calibration parameters; perform three-dimensional spatial coordinate mapping on the abnormal density region marker data according to the camera array calibration parameters to obtain three-dimensional abnormal density region coordinate data; calculate fiber arrangement looseness based on the three-dimensional abnormal density region coordinate data to obtain regional fiber arrangement looseness data; The fiber density classification data is obtained by classifying the regional fiber arrangement looseness data using the support vector machine algorithm. Then, the density difference statistics of the whole map are performed by combining the fiber density classification data to generate fiber arrangement density difference data.

[0021] With this design, this embodiment acquires images of the textile surface through a camera array and combines multiple technologies such as adaptive filtering and noise reduction, fiber orientation feature extraction, and local region density clustering analysis to generate fiber arrangement density difference data. This can comprehensively describe the fiber arrangement of the textile and classify the density level through the support vector machine algorithm, which can help with quality control in the production process, identify surface defects or uneven structures of textiles, and optimize the production process and the quality of the final product.

[0022] In one embodiment, yarn gap distribution features are extracted based on fiber arrangement density difference data to obtain yarn gap distribution feature data, including: Gaussian smoothing preprocessing is performed on fiber arrangement density difference data to obtain denoised density difference data; a multi-scale edge detection operator is constructed based on the denoised density difference data to obtain preliminary yarn boundary positioning data; gap regions are extracted from the preliminary yarn boundary positioning data using an adaptive threshold segmentation algorithm to obtain yarn gap candidate region data; Acquire point cloud data from a laser scanner; perform 3D spatial verification on candidate yarn gap regions based on the laser scanner point cloud data to obtain 3D gap verification data; combine the 3D gap verification data with morphological closing operations to obtain optimized yarn gap contour data.

[0023] This design involves multi-step processing of fiber arrangement density difference data to extract yarn gap distribution characteristics. Through Gaussian smoothing, morphological operations, and laser scanner verification, an optimized yarn gap profile is finally obtained. This process provides basic data for precise control of yarns during production, effectively identifying gaps between different yarns, thereby helping to improve the weaving process and enhance the quality and stability of the fabric.

[0024] In one embodiment, yarn gap distribution features are extracted based on fiber arrangement density difference data to obtain yarn gap distribution feature data, and the method further includes: Based on the optimized yarn gap profile data, a connected region marking is performed to obtain independent yarn gap unit data; a minimum bounding rectangle fitting is performed on the independent yarn gap unit data to obtain gap geometric parameter data; a distribution mode recognition is performed on the gap geometric parameter data through a density clustering algorithm to obtain yarn gap distribution mode data; A generative adversarial network is introduced to perform data enhancement processing on the yarn gap distribution mode data to obtain enhanced gap distribution data; statistical feature extraction is performed based on the enhanced gap distribution data to generate a multi-dimensional feature vector.

[0025] In this way, by performing geometric parameter fitting and distribution mode recognition on the yarn gap distribution feature data, the embodiment can reveal the structural characteristics of the yarn gap, and further improve the diversity of the data and the prediction ability of the model by combining the generative adversarial network for data enhancement. This method can be widely applied to quality monitoring and optimization in the fabric production process, ensuring the uniformity of the gap distribution of the fabric, reducing defects and unqualified products in the production process, and at the same time, the distribution mode data provides strong support for subsequent intelligent optimization and automated production, which helps to improve production efficiency and reduce costs.

[0026] In an embodiment, based on the yarn gap distribution feature data, a dynamic tracking of warp and weft yarn intersection points is performed to obtain warp and weft yarn intersection point motion trajectory data, including: A multi-scale morphological opening operation is performed on the yarn gap distribution feature data to eliminate small gap noise to obtain optimized yarn gap data; a warp and weft yarn intersection region detection operator is constructed based on the optimized yarn gap data, and a candidate intersection point is located through a histogram of oriented gradients feature matching to obtain initial intersection point coordinate data; A camera array is introduced to capture continuous frame textile surface images, and cross-frame optical flow tracking is performed in combination with the initial intersection point coordinate data to obtain intersection point sub-pixel level motion vector data; a Kalman filter algorithm is used to perform trajectory smoothing processing on the motion vector data to generate preliminary intersection point motion trajectory data; Laser scanner point cloud data is acquired, and the preliminary intersection point motion trajectory data is mapped to a three-dimensional coordinate system, and a point cloud registration algorithm is used to correct the Z-axis deviation of the two-dimensional trajectory to obtain three-dimensional intersection point motion trajectory data; a motion velocity field model of the intersection point is constructed based on the three-dimensional trajectory data to generate warp and weft yarn intersection point motion state label data.

[0027] Thus designed, the embodiment utilizes continuous frame images and Kalman filtering algorithm to track the dynamic motion trajectory of the warp and weft interlacing points, combines with laser scanning point cloud data to correct deviations, and finally generates accurate interlacing point motion trajectory data, which can monitor the interlacing state of the yarn in the fabric in real time, find unstable factors or potential faults in the interlacing process, provide data support for fabric quality monitoring and production adjustment, and accurate motion trajectory analysis can help optimize the loom setting, improve the fabric quality, avoid damage, and thus improve the production efficiency and product consistency.

[0028] In an embodiment, the fabric tension consumption is evaluated according to the motion trajectory data of the warp and weft interlacing points, and the fabric tension consumption evaluation data is obtained, including: The kinematic characteristics of the warp and weft interlacing point motion trajectory data are extracted, the instantaneous speed, acceleration and direction change frequency of the interlacing point are calculated, and the interlacing point dynamic behavior characteristic data is generated; based on the dynamic behavior characteristic data, an interlacing point motion energy model is constructed, the change amount of kinetic energy in unit time is quantified, and the interlacing point kinetic energy fluctuation data is obtained; Interlacing point kinetic energy calculation: , Wherein, : the kinetic energy of the i-th interlacing point, unit: Joule, : the unit mass of the yarn, unit: kilogram, calculated from the yarn linear density and length, : the instantaneous speed of the interlacing point, unit: meter / second; The yarn elastic modulus database is introduced, and the tension kinetic energy mapping analysis is performed combined with the interlacing point kinetic energy fluctuation data, the abnormal tension fluctuation area is identified through dynamic threshold comparison, and the initial fabric tension consumption label data is generated; the isolated forest algorithm is used to detect the abnormal points of the initial label data, and the optimized fabric tension consumption label data is obtained; The real-time running parameters of the textile machinery are obtained, the optimized fabric tension consumption label data is correlated and analyzed with the mechanical parameters, and a tension consumption process parameter correlation matrix is constructed; the dominant influencing factors in the correlation matrix are extracted through principal component analysis, and the tension consumption dominant factor data is generated; Based on the tension consumption dominant factor data, a long short-term memory network tension prediction model is constructed, and historical tension data is input to train the model parameters; the trained LSTM model is deployed to the edge computing node, the tension consumption value at the future time is predicted through the real-time interlacing point trajectory data, and the fabric tension consumption evaluation data is generated.

[0029] ​In this way, by kinematic feature extraction, energy model construction and tension consumption mapping analysis, the tension consumption of the fabric can be accurately evaluated, the tension anomaly can be detected by combining the isolated forest algorithm, and the correlation analysis can be performed with the real-time parameters of the loom, thereby providing accurate basis for tension control in the production process. By constructing an LSTM model to predict future tension consumption, the predictability in the production process is improved, and the operation of the loom is finely adjusted, the fabric quality is optimized, and the production loss is reduced, thereby laying a foundation for the implementation of intelligent manufacturing.

[0030] In an embodiment, production beat adaptability analysis is performed in combination with the fabric tension consumption evaluation data to obtain initial production beat adaptation data, including: The fabric tension consumption evaluation data is subjected to time sequence feature decomposition to extract tension fluctuation period, peak value distribution density and energy decay rate, and generate tension dynamic behavior feature data. A production beat tension response surface model is constructed based on the dynamic behavior feature data to quantify the tension consumption trend under different beat settings, and beat tension correlation mapping data is obtained. Real-time operation parameters of the loom are introduced, the beat tension correlation mapping data is fused with the mechanical parameters in multiple dimensions, and a production beat adaptability evaluation matrix is constructed. The evaluation matrix is subjected to beat adaptability classification by a fuzzy C-means clustering algorithm to identify an optimal beat interval with low tension consumption, and initial production beat adaptation label data is generated. Beat adjustment records in historical production data are obtained, and a reinforcement learning environment model is constructed in combination with the initial production beat adaptation label data. A deep Q network algorithm is used to simulate tension consumption feedback under different beat adjustment strategies, and a beat selection strategy is optimized by a reward function to generate dynamic beat adjustment strategy data. A production beat tension consumption prediction model is constructed based on the dynamic beat adjustment strategy data, fabric tension data is input to predict the optimal production beat, the predicted beat parameters are transmitted to the loom control system to perform beat self-adaptive adjustment, and initial production beat adaptation data is generated and fed back to the central control console.

[0031] In this way, by time sequence feature decomposition of the fabric tension consumption evaluation data, in combination with the loom parameters, the influence of beat setting on tension fluctuation is quantified, thereby providing strong data support for production beat optimization. By introducing a deep reinforcement learning algorithm, a feedback mechanism under different beat strategies is simulated, the beat selection is gradually optimized, the tension consumption is reduced, the production efficiency is improved, and the dynamic beat adjustment strategy data generated can realize self-adaptive adjustment of the production beat, further optimize the operation efficiency of the loom control system, reduce energy consumption, and improve the production automation level.

[0032] In an embodiment, workshop environment temperature and humidity data are obtained to obtain a temperature and humidity correlation deformation correction coefficient, including: An array of temperature and humidity sensors is deployed in the key area of the workshop to collect multi-dimensional environmental data in real time through a wireless transmission protocol; the original temperature and humidity data are denoised and temporally and spatially aligned through Kalman filtering to obtain calibrated temperature and humidity data of the workshop environment; A database of deformation characteristics of fiber materials is constructed to store parameters such as elastic modulus, thermal expansion coefficient, and moisture absorption expansion rate of yarns under different temperature and humidity conditions; based on the calibrated environmental temperature and humidity data, the database is queried to obtain the material deformation reference value corresponding to the current environment through a bilinear interpolation algorithm to generate temperature and humidity material deformation mapping data; Bilinear interpolation deformation coefficient calculation: ; : temperature and humidity related deformation correction coefficient, unitless, : ambient temperature, unit: ℃, : ambient relative humidity, unit: %, , : reference temperature and humidity, 25℃, 65%, , , : fitting coefficient, calibrated through the material deformation database; The temperature and humidity material deformation mapping data are fused and analyzed with the fabric tension consumption evaluation data, a temperature and humidity tension consumption coupling model is established using a partial least squares regression to quantify the contribution of environmental factors to tension fluctuations to generate an initial deformation correction coefficient; A generative adversarial network is introduced to simulate extreme working conditions for the initial correction coefficient to generate virtual deformation data under high temperature and high humidity, low temperature and low humidity edge scenarios; the coupling model parameters are optimized through transfer learning to improve the adaptability of the correction coefficient to unconventional environments to obtain a temperature and humidity related deformation correction coefficient and transmit it to the dynamic beat compensation module.

[0033] In this way, the array of temperature and humidity sensors collects workshop environmental data and generates a temperature and humidity related deformation correction coefficient in combination with the influence of temperature and humidity on the deformation characteristics of yarns, which, through fusion analysis with fabric tension consumption evaluation data, can accurately adjust the production process and improve production stability under different climate conditions, providing reliable support for the environmental adaptability of fabric production.

[0034] In an embodiment, the initial production beat adaptation data is dynamically beat compensated based on the temperature and humidity related deformation correction coefficient to obtain dynamic beat compensation data, including: ​A temperature and humidity beat compensation mapping model is constructed, the temperature and humidity correlation deformation correction coefficient is fused with the initial production beat adaptation data, key environmental beat correlation features are extracted through an attention mechanism, and a compensation weight matrix data is generated; Attention mechanism weight update: , wherein, : attention weight matrix, unitless, : learnable parameter matrix, unitless , : environmental and beat feature vectors, unitless, : feature dimensions, such as temperature, humidity, and beat speed; Environmental monitoring flow data is introduced, dynamic weight adjustment is performed in combination with the compensation weight matrix data, a sliding window algorithm is used to capture temperature and humidity mutation signals, and instantaneous beat compensation increment data is generated; A digital twin simulation platform is constructed, the instantaneous beat compensation increment data is input into a virtual weaving environment model, the fabric tension change after compensation is simulated through finite element analysis, and simulation optimization suggestion data is generated; Based on the simulation optimization suggestion data, a deep reinforcement learning compensation strategy model is constructed, the deviation between the actual production beat and the target beat is used as a reward signal, the compensation parameters are optimized through a policy gradient algorithm, dynamic beat compensation data is generated and synchronized to the loom control system.

[0035] In this way, the initial production beat adaptation data is compensated and calculated using the temperature and humidity correlation deformation correction coefficient, key features are extracted through an attention mechanism, and the fabric tension change after compensation is simulated through a sliding window algorithm and a digital twin simulation platform, providing precise data support for beat adjustment in the production process. Not only can the influence of temperature and humidity changes on production beats be compensated in real time, but also the compensation strategy can be optimized through deep reinforcement learning to improve the adaptive adjustment capability in the production process, further improving weaving quality and production efficiency.

[0036] In an embodiment, a device collaborative control model is constructed through a deep deterministic policy gradient algorithm on the dynamic beat compensation data, and the device collaborative control model is obtained, including: A multi-device collaborative simulation environment is constructed, integrating dynamic beat compensation data and device state parameters, and generating a virtual control scenario through digital twin technology; based on the simulation environment, a continuous action space is defined, and the device collaborative control target is converted into a reinforcement learning reward function; A deep deterministic policy gradient algorithm framework is designed, an actor network outputs device control instructions, and a critic network evaluates action values; dynamic beat compensation data is introduced as network input features, and multiple device state information is fused through an attention mechanism to generate initial collaborative control policy data; DDPG policy iteration is performed in a simulation environment, an experience replay mechanism is used to store historical control trajectories, and a target network is used to soften policy updates; a noise exploration mechanism is introduced to add orthogonal perturbations in action outputs, expand the policy search space, and generate optimized collaborative control policy data. The trained DDPG model is deployed to an edge computing node, receives dynamic beat compensation data and device state feedback, adjusts multiple device control instructions through model output parameters, performs full-process device collaborative control in textile production, generates a device collaborative control model, and encapsulates it as a cloud platform callable service.

[0037] In this way, a device collaborative control model is constructed through a deep deterministic policy gradient algorithm (DDPG), and dynamic beat compensation data and device states are optimized collaboratively to improve the automation and intelligence level of fabric production. Through digital twin technology and a simulation environment, different device collaborative working states can be accurately simulated to provide global optimization strategies for device control in the production process.

[0038] The embodiments of the application are described in detail above in combination with the drawings, but the application is not limited thereto, and various changes can be made within the knowledge of those skilled in the art without departing from the spirit of the application.

Claims

1. A method for intelligent control of a textile manufactured product line, characterized in that, The method comprises the following steps: acquiring textile surface image data by capturing a textile surface image through a camera array; analyzing fiber arrangement density based on the textile surface image data to obtain fiber arrangement density difference data; and extracting yarn gap distribution characteristics based on the fiber arrangement density difference data to obtain yarn gap distribution characteristic data; tracking the interlacing points of warp and weft yarns dynamically based on the yarn gap distribution characteristic data to obtain interlacing point trajectory data of the warp and weft yarns; evaluating fabric tension consumption based on the interlacing point trajectory data of the warp and weft yarns to obtain fabric tension consumption evaluation data; and analyzing production rhythm adaptability in combination with the fabric tension consumption evaluation data to obtain initial production rhythm adaptation data; acquiring workshop environment temperature and humidity data; correcting the material deformation coefficient based on the fabric tension consumption evaluation data according to the workshop environment temperature and humidity data to obtain a temperature and humidity related deformation correction coefficient; performing dynamic rhythm compensation calculation based on the initial production rhythm adaptation data based on the temperature and humidity related deformation correction coefficient to obtain dynamic rhythm compensation data; constructing a device collaborative control model based on the dynamic rhythm compensation data through a deep deterministic policy gradient algorithm to obtain the device collaborative control model; and deploying the device collaborative control model to an edge computing node to adjust the speed and compensate the tension of a textile machinery group in real time through model output parameters to perform intelligent control of the whole process of textile production.

2. A method for intelligent control of a textile manufactured product production line as claimed in claim 1, characterized in that, The method comprises the following steps: synchronously acquiring original textile surface image data through a camera array; performing adaptive filtering and denoising on the original textile surface image data to obtain denoised textile surface image data; and extracting fiber direction features based on the denoised textile surface image data to obtain fiber direction feature data; performing local region density clustering analysis based on the fiber direction feature data to obtain fiber density clustering center data; constructing a density difference heat map in combination with the fiber density clustering center data to obtain fiber density difference heat map data; and marking abnormal density regions through a dynamic threshold segmentation algorithm to obtain abnormal density region marking data; acquiring camera array calibration parameters; mapping the abnormal density region marking data to three-dimensional space coordinates based on the camera array calibration parameters to obtain three-dimensional abnormal density region coordinate data; and calculating fiber arrangement looseness based on the three-dimensional abnormal density region coordinate data to obtain regional fiber arrangement looseness data; performing density level classification on the regional fiber arrangement looseness data through a support vector machine algorithm to obtain fiber density level classification data; and performing full-image density difference statistics in combination with the fiber density level classification data to generate fiber arrangement density difference data.

3. A method of intelligent control for a textile manufactured product production line as claimed in claim 2, wherein, The method comprises the following steps: extracting yarn gap distribution characteristics based on the fiber arrangement density difference data to obtain yarn gap distribution characteristic data, including: The fiber arrangement density difference data is subjected to Gaussian smoothing preprocessing to obtain denoised density difference data; a multi-scale edge detection operator is constructed based on the denoised density difference data to obtain yarn boundary preliminary positioning data; a self-adaptive threshold segmentation algorithm is used to extract gap regions from the yarn boundary preliminary positioning data to obtain yarn gap candidate region data; Point cloud data of a laser scanner is acquired; the yarn gap candidate region data is subjected to three-dimensional space verification based on the point cloud data of the laser scanner to obtain three-dimensional gap verification data; morphological closing operation processing is performed in combination with the three-dimensional gap verification data to obtain optimized yarn gap contour data.

4. A method of intelligent control for a textile manufactured product production line as claimed in claim 3, wherein, Yarn gap distribution feature data is extracted from the fiber arrangement density difference data, and further includes Connected region labeling is performed based on the optimized yarn gap contour data to obtain independent yarn gap unit data; Minimum circumscribed rectangle fitting is performed on the independent yarn gap unit data to obtain gap geometric parameter data; Distribution pattern data of the yarn gap is obtained by performing distribution pattern recognition on the gap geometric parameter data through a density clustering algorithm; Data enhancement processing is performed on the yarn gap distribution pattern data by introducing a generative adversarial network to obtain enhanced gap distribution data; Statistical feature extraction is performed based on the enhanced gap distribution data to generate a multi-dimensional feature vector.

5. A method of intelligent control for a textile manufactured product production line as claimed in claim 4, wherein, Based on the yarn gap distribution feature data, warp and weft yarn interlacing point dynamic tracking is performed to obtain warp and weft yarn interlacing point motion trajectory data, including: Multi-scale morphological opening operation processing is performed on the yarn gap distribution feature data to eliminate small gap noise and obtain optimized yarn gap data; a warp and weft yarn intersection region detection operator is constructed based on the optimized yarn gap data, and candidate interlacing points are located by direction gradient histogram feature matching to obtain initial interlacing point coordinate data; A camera array is introduced to capture continuous frame textile surface images, and cross-frame optical flow tracking is performed in combination with the initial interlacing point coordinate data to obtain interlacing point sub-pixel level motion vector data; Kalman filtering algorithm is used to perform trajectory smoothing processing on the motion vector data to generate preliminary interlacing point motion trajectory data; Point cloud data of a laser scanner is acquired, and the preliminary interlacing point motion trajectory data is mapped to a three-dimensional space coordinate system; the Z-axis deviation of the two-dimensional trajectory is corrected by a point cloud registration algorithm to obtain three-dimensional interlacing point motion trajectory data; an interlacing point motion velocity field model is constructed based on the three-dimensional trajectory data to generate warp and weft yarn interlacing point motion state label data.

6. A method of intelligent control for a textile manufactured product production line as claimed in claim 5, wherein, Based on the warp and weft yarn interlacing point motion trajectory data, fabric tension consumption evaluation is performed to obtain fabric tension consumption evaluation data, including: Kinematic feature extraction is performed on the warp and weft yarn interlacing point motion trajectory data to calculate the interlacing point instantaneous speed, acceleration and direction change frequency to generate interlacing point dynamic behavior feature data; an interlacing point motion energy model is constructed based on the dynamic behavior feature data to quantify the change amount of kinetic energy per unit time to obtain interlacing point kinetic energy fluctuation data; A yarn elastic modulus database is introduced, combined with the interlacing point kinetic energy fluctuation data to perform tension kinetic energy mapping analysis, and an abnormal tension fluctuation area is identified through dynamic threshold comparison to generate initial fabric tension consumption label data; an isolation forest algorithm is used to detect abnormal points in the initial label data to obtain optimized fabric tension consumption label data; Real-time operation parameters of the textile machine are acquired, and the optimized fabric tension consumption label data is correlated and analyzed with the machine parameters to construct a tension consumption process parameter correlation matrix; dominant influencing factors in the correlation matrix are extracted through principal component analysis to generate tension consumption dominant factor data; A long short-term memory network tension prediction model is constructed based on the tension consumption dominant factor data, and historical tension data are input to train model parameters; the trained LSTM model is deployed to an edge computing node to predict tension consumption values at future time points through real-time interlacing point trajectory data to generate fabric tension consumption evaluation data.

7. A method of intelligent control for a textile manufactured product production line as claimed in claim 6, wherein, Production rhythm adaptability analysis is performed in combination with the fabric tension consumption evaluation data to obtain initial production rhythm adaptation data, including: Fabric tension consumption evaluation data are subjected to time series feature decomposition to extract tension fluctuation period, peak value distribution density and energy decay rate to generate tension dynamic behavior feature data; a production rhythm tension response surface model is constructed based on the dynamic behavior feature data to quantify tension consumption trends under different rhythm settings to obtain rhythm tension correlation mapping data; Real-time operation parameters of the loom are introduced, and the rhythm tension correlation mapping data are fused with the machine parameters in multiple dimensions to construct a production rhythm adaptability evaluation matrix; the evaluation matrix is subjected to rhythm adaptability classification through a fuzzy C-means clustering algorithm to identify an optimal rhythm interval with low tension consumption to generate initial production rhythm adaptation label data; Rhythm adjustment records in historical production data are acquired, and an enhanced learning environment model is constructed in combination with the initial production rhythm adaptation label data; a deep Q network algorithm is used to simulate tension consumption feedback under different rhythm adjustment strategies, and a rhythm selection strategy is optimized through a reward function to generate dynamic rhythm adjustment strategy data; A production rhythm tension consumption prediction model is constructed based on the dynamic rhythm adjustment strategy data, and fabric tension data are input to predict optimal production rhythms; predicted rhythm parameters are transmitted to a loom control system to perform rhythm self-adaptive adjustment to generate initial production rhythm adaptation data and feedback to a central control console.

8. A method of intelligent control for a textile manufactured product production line as claimed in claim 7, characterized in that, Workshop environment temperature and humidity data are acquired to obtain temperature and humidity correlation deformation correction coefficients, including: Temperature and humidity sensor arrays are deployed at key areas in the workshop to collect multi-dimensional environment data in real time through a wireless transmission protocol; original temperature and humidity data are subjected to Kalman filter denoising and spatio-temporal alignment processing to obtain calibrated workshop environment temperature and humidity data; A fiber material deformation characteristic database is constructed to store parameters such as yarn elastic modulus, thermal expansion coefficient and moisture absorption expansion rate under different temperature and humidity conditions; the calibrated environment temperature and humidity data are used to query the database, and material deformation reference values corresponding to the current environment are acquired through a bilinear interpolation algorithm to generate temperature and humidity material deformation mapping data; The temperature and humidity material deformation mapping data and the fabric tension consumption evaluation data are fused and analyzed, a temperature and humidity tension consumption coupling model is established by using a partial least squares regression, a contribution degree of environmental factors to tension fluctuation is quantified, and initial deformation correction coefficients are generated; An initial correction coefficient is simulated in an extreme working condition by introducing a generative adversarial network, virtual deformation data in a high temperature and high humidity, low temperature and low humidity edge scene are generated, coupling model parameters are optimized by using a transfer learning, adaptability of the correction coefficient to a non-conventional environment is improved, temperature and humidity related deformation correction coefficients are obtained, and are transmitted to a dynamic beat compensation module.

9. A method of intelligent control for a textile manufactured product production line as claimed in claim 8, wherein, Based on the temperature and humidity related deformation correction coefficients, initial production beat adaptation data are dynamically compensated to obtain dynamic beat compensation data, including: A temperature and humidity beat compensation mapping model is constructed, temperature and humidity related deformation correction coefficients and initial production beat adaptation data are fused in features, key environmental beat related features are extracted by using an attention mechanism, and compensation weight matrix data are generated; Environmental monitoring flow data are introduced, dynamic weight adjustment is performed in combination with the compensation weight matrix data, a sliding window algorithm is used to capture temperature and humidity mutation signals, and instantaneous beat compensation increment data are generated; A digital twin simulation platform is constructed, the instantaneous beat compensation increment data are input into a virtual weaving environment model, fabric tension changes after compensation are simulated by using a finite element analysis, and simulation optimization suggestion data are generated; Based on the simulation optimization suggestion data, a deep reinforcement learning compensation strategy model is constructed, a deviation between actual production beat and target beat is taken as a reward signal, compensation parameters are optimized by using a policy gradient algorithm, dynamic beat compensation data are generated, and are synchronized to a loom control system.

10. A method of intelligent control for a textile manufactured product production line as claimed in claim 9, wherein, A device collaborative control model is constructed by using a deep deterministic policy gradient algorithm on the dynamic beat compensation data, the device collaborative control model is obtained, including: A multi-device collaborative simulation environment is constructed, dynamic beat compensation data and device state parameters are integrated, a virtual control scene is generated by using digital twin technology; a continuous action space is defined based on the simulation environment, a device collaborative control target is converted into a reinforcement learning reward function; A deep deterministic policy gradient algorithm framework is designed, an Actor network outputs device control instructions, a Critic network evaluates action values; dynamic beat compensation data are introduced as network input features, multi-device state information is fused by using an attention mechanism, and initial collaborative control strategy data are generated; DDPG policy iteration is performed in the simulation environment, an experience replay mechanism is used to store historical control trajectories, a target network is used to soften policy updates; a noise exploration mechanism is introduced, orthogonal disturbances are added in action outputs, a policy search space is expanded, and optimized collaborative control strategy data are generated; The trained DDPG model is deployed to an edge computing node, receives dynamic beat compensation data and device state feedback; multi-device control instructions are adjusted by using model output parameters, device collaborative control in a whole process of textile production is performed, a device collaborative control model is generated, and is packaged as a cloud platform callable service.

Images