Image processing-based carbon powder particle distribution real-time analysis method

Abstract

The invention provides a carbon powder particle distribution real-time analysis method based on image processing, and the method comprises the steps: applying an adjusted gray threshold parameter to the processing and analysis of a carbon powder particle image, and carrying out the edge extraction, geometric feature extraction and particle size distribution analysis of carbon powder particles again, the adjusted analysis result is used as a basis for regulating and controlling a carbon fiber recovery process; establishing a quantitative relationship between the particle size distribution of the carbon powder particles and the centrifugal rotation rate of the solution by adopting a mathematical statistical method according to a particle size distribution analysis result, and analyzing the influence of the centrifugal rotation rate on particle aggregation and particle size distribution; the carbon powder particle distribution in the carbon fiber centrifugation process is analyzed and monitored in real time, the centrifugation process parameters are dynamically regulated and controlled according to the real-time analysis result, and the online monitored data are used for optimizing the carbon fiber recycling process.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a real-time analysis method for carbon powder particle distribution based on image processing. Background Art

[0002] During the carbon fiber recycling process, centrifugal separation of carbon powder particles is a very crucial step, which directly affects the quality and performance of the recycled carbon fiber. In the real-time analysis of carbon powder particle distribution, the particle agglomeration phenomenon caused by the centrifugation process of carbon fiber poses a challenge to image processing algorithms. When the centrifugal rotation rate of the solution changes, the morphology of the carbon powder particles also changes, which requires the image processing algorithm to be able to adjust the image gray threshold for particle edge recognition in real time. However, the adjustment of the image gray threshold often affects the result of the overall particle size distribution analysis. To deeply understand this problem, it is necessary to explore the internal relationship between the adjustment of the image gray threshold and the result of the particle size distribution analysis. Specifically, when the centrifugal rotation rate increases, the agglomeration effect between carbon powder particles enhances, resulting in blurred particle edges. At this time, if the image gray threshold is not adjusted in time or appropriately, the agglomerated particles may be recognized as single large particles, thus underestimating the number of small particles and overestimating the number of large particles, ultimately leading to the deviation of the particle size distribution curve. On the contrary, if the centrifugal rotation rate decreases, the degree of particle agglomeration weakens, and the particle edges become clear. At this time, if the image gray threshold is adjusted improperly, a single particle may be recognized as multiple small particles, thus overestimating the number of small particles and underestimating the number of large particles, resulting in another deviation of the particle size distribution curve. Therefore, there is a complex non-linear relationship between the adjustment of the image gray threshold and the result of the particle size distribution analysis, and it is necessary to establish a quantitative model through a large number of experiments and theoretical analyses to guide the optimization and improvement of the image processing algorithm. Summary of the Invention

[0003] The present invention provides a real-time analysis method for carbon powder particle distribution based on image processing, mainly including:

[0004] Preprocess the image of carbon powder particles during the carbon fiber centrifugation process to obtain the preprocessed carbon powder particle image. According to the preprocessed carbon powder particle image, use the region growing algorithm, set the gray threshold parameter for carbon powder particle edge recognition, extract the edge of the carbon powder particles, and obtain the edge contour information of the carbon powder particles;

[0005] According to the edge contour information of the carbon powder particles, use the morphological processing method to eliminate the noise in the edge contour of the carbon powder particles, including breakpoints and burrs, to obtain the repaired edge contour of the carbon powder particles. By extracting the geometric feature of the repaired edge contour of the carbon powder particles, obtain the geometric feature parameters of the carbon powder particles in the carbon fiber, including size, shape, and fiber orientation;

[0006] According to the geometric characteristic parameters of the carbon powder particles of carbon fiber, a method based on cluster analysis is adopted to classify and identify the carbon powder particles, and carbon powder particle categories with different morphologies and sizes are obtained. By statistically analyzing the quantity and distribution of carbon powder particles in different categories, the particle size distribution information of the carbon powder particles is obtained, and the quality attributes of the carbon fiber are correlated;

[0007] During the centrifugation process, the centrifugal rotation rate of the carbon fiber solution is obtained in real time. According to the change of the centrifugal rotation rate, the gray threshold parameter is dynamically adjusted to adapt to the change of the particle morphology. When the centrifugal rotation rate increases, the gray threshold is increased; when the centrifugal rotation rate decreases, the gray threshold is decreased;

[0008] The adjusted gray threshold parameter is applied to the processing and analysis of the carbon powder particle image. The edge extraction, geometric feature extraction and particle size distribution analysis of the carbon powder particles are re-performed, and the adjusted analysis result is used as the basis for regulating the carbon fiber recovery process;

[0009] According to the particle size distribution analysis result, a mathematical statistics method is adopted to establish a quantitative relationship between the particle size distribution of the carbon powder particles and the centrifugal rotation rate of the solution, and the influence of the centrifugal rotation rate on particle agglomeration and particle size distribution is analyzed;

[0010] The distribution of carbon powder particles during the carbon fiber centrifugation process is analyzed and monitored in real time. According to the real-time analysis result, the centrifugation process parameters are dynamically regulated, and the online monitoring data is used to optimize the carbon fiber recovery and utilization process.

[0011] The technical solution provided by the embodiment of the present invention may include the following beneficial effects:

[0012] The present invention discloses a real-time analysis method for carbon powder particle distribution based on image processing. This method obtains the geometric features of carbon powder particles through image processing technology, and uses cluster analysis to classify and identify the particles to realize the statistical analysis of particle size distribution. The innovation lies in dynamically adjusting the gray threshold parameter of image processing according to the real-time change of the centrifugal rotation rate to adapt to the change of particle morphology. By establishing a quantitative relationship model between particle size distribution and centrifugal rate, the influence of process parameters on particle characteristics is reflected. The present invention realizes the online monitoring and analysis of carbon powder particle distribution, provides a basis for dynamically regulating centrifugation process parameters, helps to optimize the carbon fiber recovery and utilization process, and improves the quality of carbon fiber. This method combines image analysis, data modeling and process control, providing a new technical approach for the intelligent monitoring and optimization of the carbon fiber production process. Brief Description of the Drawings

[0013] Figure 1 It is a flowchart of a real-time analysis method for carbon powder particle distribution based on image processing of the present invention. Detailed Embodiment

[0014] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0015] As Figure 1 , a real-time analysis method for carbon powder particle distribution based on image processing in this embodiment may specifically include:

[0016] S101. Preprocess the image of carbon powder particles during the centrifugation of carbon fiber to obtain the preprocessed carbon powder particle image. According to the preprocessed carbon powder particle image, use the region growing algorithm to set the gray threshold parameter for carbon powder particle edge recognition, and extract the edges of the carbon powder particles to obtain the edge contour information of the carbon powder particles.

[0017] Obtain the original image of carbon powder particles collected by the carbon fiber centrifuge, perform denoising processing on the original image through adaptive Gaussian filtering with a set kernel size to obtain a preprocessed image; perform edge detection on the preprocessed image, obtain edge candidate points by setting a gradient threshold, and use the region growing algorithm with an eight-neighborhood search method to segment and label the preprocessed image to obtain a segmented and labeled image; perform morphological erosion operation on the segmented and labeled image to remove noise points, and at the same time perform morphological dilation operation to complement the edge information to obtain an optimized image; calculate the gray difference between adjacent particles according to the gray value distribution of carbon powder particles in the optimized image. If the gray difference is less than a preset threshold, it is determined as an adhesion area, and the adhesion area is segmented to obtain the edge contour information of the carbon powder particles.

[0018] Specifically, the original images of carbon powder particles are collected from a carbon fiber centrifuge. The images are denoised by adaptive Gaussian filtering with a kernel size of 5×5. According to the distribution of the image gray histogram, the contrast enhancement range is set between 0.2 and 0.8, and the gray image of the carbon powder particles is smoothed to obtain a preprocessed image. The Sobel operator is used to detect the edges of the preprocessed image to calculate the image gradient values. A gradient threshold is set to obtain edge candidate points as the initial point set for region growing. According to the gray threshold range of the carbon powder particle edges, a region growing algorithm using an eight-neighborhood search method is used to dynamically segment and label the carbon powder particle regions. The carbon powder particle images within the segmented and labeled regions are subjected to morphological erosion operations using a circular structuring element with a radius of 3 pixels to remove noise point interference, and then morphological dilation operations using the same structuring element are performed to complement the edge information, obtaining an optimized particle region image. According to the gray value distribution of the carbon powder particles in the optimized particle region image, the gray difference between adjacent carbon powder particles is calculated. When the difference is less than a preset threshold, it is determined as an adhesion region, and the watershed algorithm is used to segment the adhered carbon powder particle region. The edges of the segmented carbon powder particles are smoothed using a cubic spline interpolation algorithm. The smoothing parameters are set according to the contour curvature change, and the edge contour information of the carbon powder particles is extracted to obtain the complete edge contour data of the carbon powder particles. When collecting the carbon powder particle images during the carbon fiber centrifugation process, a high-resolution industrial camera is used. The image resolution is set to 2048×1536 pixels, and the sampling frequency is 30 frames per second. The original images of the carbon powder particles obtained have noise due to the vibration during the centrifugation process and environmental light interference. The images are denoised by setting an adaptive Gaussian filtering kernel of 5×5. The standard deviation of the Gaussian filter kernel is adaptively adjusted according to the gray variance of the local image region. The standard deviation is smaller in the carbon powder particle edge region to keep the edges clear, and larger in the smooth region to fully remove noise. When analyzing the gray histogram of the images, it is found that the gray values of the carbon powder particle images are mainly distributed between 35 and 220. To enhance the image contrast, the piecewise linear stretching method is used to map the gray values, and the pixel points in the gray interval between 0.2 and 0.8 are linearly stretched, so that the contrast between the carbon powder particles and the background is improved. After the image preprocessing, the edge features of the carbon powder particles are more obvious, facilitating subsequent edge detection and segmentation. When using the Sobel operator to detect the edges of the preprocessed image, the image gradients are calculated in the horizontal and vertical directions respectively. When the gradient value of a certain pixel point is greater than 25, it is marked as an edge candidate point. After obtaining the edge candidate points, these points are used as seed points for region growing, and the search direction uses the eight-neighborhood method, that is, the 8 adjacent pixel points around the current pixel point are judged. When the gray difference between the adjacent pixel point and the current pixel point is less than the preset threshold of 8, the pixel point is added to the current region.During the morphological processing, a circular structuring element with a radius of 3 pixels is selected for erosion and dilation operations. The erosion operation can remove small noise points, and the circular structuring element can maintain the smoothness of the edges of toner particles. The subsequent dilation operation then complements the edge missing caused by erosion, making the contour of the toner particles more complete. For the segmentation of adhered toner particles, the adhesion situation is judged by calculating the gray difference between adjacent regions. When the average gray difference between two adjacent regions is less than the threshold value of 12, these two regions are considered as adhered particles. When using the watershed algorithm, the local minimum points of the particle region are used as the initial marker points, and the region is gradually expanded through the gradient descent method until the boundaries of adjacent regions meet to form a watershed line. In the edge contour extraction step, the contour of the segmented toner particles is smoothed using the cubic spline interpolation algorithm, and the interval of the interpolation points is set to 4 pixels. By adjusting the curvature factor, the contour curve is kept smooth without losing detailed features, and finally, the complete edge contour data of the toner particles is obtained.

[0019] S102. According to the edge contour information of the toner particles, a morphological processing method is adopted to eliminate the noise in the edge contour of the toner particles, including breakpoints and burrs, to obtain the repaired edge contour of the toner particles. By extracting the geometric feature of the repaired edge contour of the toner particles, the geometric feature parameters of the toner particles in the carbon fiber are obtained, including size, shape, and fiber orientation.

[0020] The morphological opening and closing operations are performed on the edge contour of the toner particles using a structuring element to obtain a complete edge contour image; boundary tracking coding is performed according to the complete edge contour image, and abnormal points are removed by setting the included angle threshold of adjacent three points to obtain smooth edge contour data; for the smooth edge contour data, the projection distance from the contour points to the eigenvector is fitted to obtain the major axis and minor axis of the toner particle contour; the carbon fiber orientation angle is calculated according to the major axis and minor axis of the toner particle contour, the ratio of the inertia principal axis moment to the area is calculated using the area moment method, and the roundness parameter is obtained by the ratio of the square of the contour perimeter to the area.

[0021] Specifically, based on the edge contour data of toner particles, morphological opening operation is performed using a circular structuring element with a radius of 5 pixels to eliminate the burrs in the edge contour. Then, morphological closing operation is performed using the same structuring element to connect and repair the breakpoints in the obtained edge contour where the interval is less than the diameter of the structuring element, resulting in a complete edge contour image. 8-connected boundary tracking is performed on the complete edge contour image, and Freeman chain code is used to encode and store the contour boundary points. By setting the angle formed by adjacent three points to be less than 45 degrees as the curvature anomaly determination threshold, the abnormal points in the contour are removed to obtain smooth edge contour data. For the contour point coordinates in the smooth edge contour data, the covariance matrix eigenvalue decomposition method is used to calculate the principal direction eigenvector, and the projection distance from the contour points to the eigenvector is fitted by the least squares method to obtain the major axis and minor axis of the toner particle contour. According to the spatial projection relationship between the major axis and the minor axis, the carbon fiber orientation angle is calculated, and the area moment method is used to calculate the ratio of the inertia principal axis moment to the area of the toner particle to obtain the aspect ratio parameter of the particle. The curvature change of the toner particle contour is statistically analyzed, the roundness parameter is obtained by calculating the ratio of the square of the contour perimeter to the area, and the regularity parameter is calculated according to the distance variance from the upsampled points on the contour to the contour center to obtain a complete set of geometric feature parameters of the toner particle. When processing the edge contour of toner particles, the selection of the structuring element for morphological operations is crucial for noise elimination and feature preservation. The opening operation is performed using a circular structuring element with a radius of 5 pixels, where the erosion operation removes the burr protrusions less than 5 pixels on the edge contour, and the dilation operation maintains the basic shape features of the contour. For breakpoint repair, the dilation operation in the closing operation connects the contour breakpoints within 10 pixels, and the subsequent erosion operation maintains the thickness consistency of the repaired contour. During the boundary tracking process, the 8-connected Freeman chain code starts from the contour starting point and encodes and records the contour point direction according to the directions from 0 to 7. Each direction number represents the direction from the current point to the next boundary point. When processing a typical toner particle contour, the length of the obtained chain code sequence is usually between 200 and 300. By analyzing the direction change formed by adjacent three encoded points, the curvature abnormal points are judged. If the angle between the three points is less than 45 degrees, it indicates that the curvature change is severe at this point and needs to be smoothed. In the principal direction analysis, a 2×2 covariance matrix is constructed for the contour point set, and the matrix elements contain the second-order statistics of the contour point coordinates. By calculating the eigenvalues and eigenvectors of the covariance matrix, the eigenvector corresponding to the larger eigenvalue is the principal direction of the contour. For a typical toner particle, its major axis is usually between 20 and 50 pixels, and the minor axis is between 10 and 25 pixels. The ratio of the major axis to the minor axis reflects the elongation degree of the particle. When calculating the carbon fiber orientation, the angle between the principal direction eigenvector and the horizontal direction is the orientation angle, and the value range is between 0 and 180 degrees. In the area moment method, by calculating the second moment of the contour points relative to the centroid, the inertia principal axis direction is obtained, and this direction is consistent with the principal direction obtained by eigenvalue decomposition, verifying the reliability of the orientation calculation.For regular toner particles, their aspect ratio parameter is usually between 1.5 and 3.0. In the shape feature extraction step, the roundness parameter is obtained by calculating the ratio of the square of the contour perimeter to the area. The closer this value is to 4π, the closer the particle shape is to a circle. The regularity parameter is characterized by calculating the coefficient of variation of the distances from the contour sampling points to the center. For regular particles with smooth edges, its coefficient of variation is usually less than 0.15. In practical applications, a typical toner particle image contains 50 to 200 particles, and the overall distribution characteristics can be obtained through batch processing. When the contour roundness parameter is greater than 14 and the regularity parameter is less than 0.1, it indicates that the particle has good morphological characteristics.

[0022] S103. According to the geometric feature parameters of the toner particles of carbon fiber, a method based on cluster analysis is adopted to classify and identify the toner particles, obtain the toner particle categories with different morphologies and sizes, and obtain the particle size distribution information of the toner particles through statistical analysis of the quantity and distribution of toner particles in different categories, and correlate with the quality attributes of carbon fiber.

[0023] Receive the size, shape and fiber orientation angle data of the toner particles to construct a feature matrix, classify the feature matrix using a density-based clustering algorithm, and obtain the morphology classification result by calculating the product of the local density value and the distance of the sample points in the feature space; perform standardization processing on the feature matrix according to the morphology classification result, and extract the principal component vectors with the feature contribution rate greater than the contribution rate threshold using the principal component decomposition method to obtain the dimensionality-reduced feature data; calculate the particle number frequency within the particle size interval according to the dimensionality-reduced feature data, and perform kernel density estimation on the frequency distribution data using a Gaussian kernel function to obtain the particle size distribution probability density function; perform moment analysis on the probability density function to calculate the distribution parameters, and use the recursive least squares method to iteratively calculate the parameters of the elastic modulus prediction equation to obtain the prediction result of the quality attributes of carbon fiber.

[0024] Specifically, a feature matrix is constructed based on the size, shape, and fiber orientation angle data of toner particles. The density-based clustering algorithm is used to classify the feature matrix. The clustering center points are determined by calculating the product of the local density value and distance of sample points in the feature space, and the morphological classification result of toner particles is obtained. The classified toner particle feature matrix is subjected to zero-mean normalization. A covariance matrix is constructed according to the particle category labels, and the principal component vectors with feature contribution rates greater than a preset threshold are extracted using the principal component decomposition method to obtain the reduced-dimensional toner particle feature data. The particle number frequency in each particle size interval is calculated based on the reduced-dimensional feature data, and the Gaussian kernel function is used for kernel density estimation of the frequency distribution data. The continuous particle size distribution probability density function is obtained through bandwidth parameter optimization. Moment analysis is performed on the particle size distribution probability density function, and the mean, variance, and skewness parameters of the distribution function are calculated. A polynomial regression equation is used to fit the carbon fiber strength value to obtain the strength prediction model. According to the coefficient matrix of the carbon fiber strength prediction model, an elastic modulus prediction equation is constructed, and the equation parameters are iteratively calculated by the recursive least squares method to obtain the prediction result of the quality attributes of carbon fiber. In the process of constructing the toner particle feature matrix, each particle sample contains multiple feature dimensions. Among them, the particle size is represented by the equivalent diameter and is usually distributed in the range of 5 to 50 microns. The shape features include roundness and aspect ratio parameters, and the fiber orientation angle is distributed in the range of 0 to 180 degrees. The density-based clustering algorithm determines the clustering center by calculating the local density value of each sample point. When the number of samples within a 10-micron range around a point exceeds 20, the point is marked as a high-density point and thus determined as the clustering center. When normalizing the feature matrix, different normalization methods are used for feature parameters with different dimensions. For example, the maximum-minimum normalization is used for particle size to map the values to the range of 0 to 1, while the z-score normalization is used for shape features to make their mean 0 and variance 1. In the principal component decomposition process, it is found through calculating the eigenvalues that the cumulative contribution rate of the first 3 principal components reaches 85%, so these 3 principal components are selected as the reduced-dimensional feature vectors. When constructing the particle size distribution, the particle size is divided into 12 intervals with a width of 5 microns each, and the obtained frequency distribution shows irregular fluctuations. The Gaussian kernel function is used for kernel density estimation, and the bandwidth parameter of the kernel function is optimized to 1.2 microns through the cross-validation method, so that the smoothed probability density curve not only maintains the distribution characteristics but also avoids over-smoothing. For the carbon fiber strength prediction, it is found through analysis that there is an obvious negative correlation between the mean value of the particle size distribution and the strength. When the average particle diameter increases from 15 microns to 25 microns, the carbon fiber strength decreases from 4.2 GPa to 3.8 GPa. At the same time, the variance of the particle size distribution reflects the uniformity of particle size. For every 0.5 increase in variance, the strength decreases by approximately 0.15 GPa. Based on these correlation features, a third-order polynomial regression equation is constructed. In the elastic modulus prediction, it is found through iterative calculation by the recursive least squares method that when the number of iterations reaches 50, the prediction error converges to a stable value.Model validation shows that when the average size of carbon powder particles is around 20 microns and the variance of particle size distribution is less than 0.8, the predicted elastic modulus is in the range of 230 to 250 GPa, and the deviation from the measured value is controlled within 5%. The influence of distribution skewness on the modulus is relatively small. When the skewness varies in the range of -0.5 to 0.5, the change in modulus does not exceed 10 GPa. The prediction method based on the particle size distribution characteristics establishes a quantitative correlation between the microscopic structure and macroscopic properties of carbon fibers.

[0025] S104. During centrifugation, the centrifugal rotation speed of the carbon fiber solution is obtained in real time. According to the change of the centrifugal rotation speed, the gray threshold parameter is dynamically adjusted to adapt to the change of particle morphology. When the centrifugal rotation speed increases, the gray threshold is increased; when the centrifugal rotation speed decreases, the gray threshold is decreased.

[0026] Obtain the real-time rotation speed signal from the carbon fiber centrifuge, perform signal smoothing processing on the rotation speed signal using a Gaussian filter to obtain the filtered rotation speed data; calculate the acceleration value per unit time according to the filtered rotation speed data, and perform arithmetic processing on the acceleration value using the proportional-integral method to obtain the reference gray threshold value; perform a clamping operation on the reference gray threshold value using the hyperbolic tangent function, and limit the operation result through the preset upper threshold and lower threshold to obtain the dynamic gray threshold; perform block processing on the carbon fiber solution image, perform region segmentation on the image block according to the dynamic gray threshold, and perform an optimization operation on the segmentation result using the iterative threshold method to obtain the segmented image.

[0027] Specifically, real-time rotational speed data is collected from the carbon fiber centrifuge. An angular rotation rate signal is obtained through a rotational speed sensor every preset sampling time. A Gaussian filter is used to smooth the rotational speed signal, and the rate change is calculated based on the rate difference between two adjacent sampling moments. The centrifugal acceleration value per unit time is calculated based on the filtered rotational speed data. A proportional-integral algorithm with an integral time constant equal to the sampling period and a proportionality coefficient equal to a preset reference value is used to calculate the gray threshold adjustment amount, and the numerically updated reference gray threshold value is obtained. The gray threshold adjustment range is set through pre-calibrated process parameters. A hyperbolic tangent function is used to clamp the threshold values outside the range. The adjusted gray threshold is limited by the upper and lower thresholds to obtain the dynamic gray threshold value. The real-time image obtained by the carbon fiber solution image acquisition device is divided into blocks, and the gray mean of each image block is calculated. The image blocks are regionally segmented based on the dynamic gray threshold to obtain the initial segmented image. An iterative threshold method is used to optimize the initial segmented image. The iteration termination condition is judged by calculating the difference in gray means between two consecutive iterations, and the segmentation threshold is dynamically updated to obtain the final segmentation result adapted to the rotational speed change. During the carbon fiber centrifugation process, the acquisition frequency of the rotational speed signal is set to 100 times per second, and real-time rotational speed data is obtained through a Hall sensor installed on the main shaft of the centrifuge. The original rotational speed signal often contains high-frequency noise and is smoothed using a Gaussian filter with a window width of 5. The filtered rotational speed curve shows the typical change characteristics of the centrifugation process gradually accelerating from 0 revolutions per minute to 12,000 revolutions per minute. When calculating the centrifugal acceleration, the velocity difference between adjacent sampling points is divided by the sampling time interval. When the centrifuge is in the acceleration stage, the acceleration value remains at about 800 revolutions per minute per second, while in the steady operation stage, the acceleration fluctuation value is controlled within 50 revolutions per minute per second. In the proportional-integral algorithm, the proportionality coefficient is set to 0.5, and the integral time constant is set to the sampling period of 0.01 seconds, enabling the gray threshold to quickly respond to the rotational speed change. During the gray threshold adjustment process, the reference threshold is set to 128. Considering the imaging characteristics of the carbon fiber solution, the threshold adjustment range is set between 50% and 150% of the reference value, i.e., 64 to 192. A hyperbolic tangent function is used for clamping, and the gain coefficient of the function is set to 0.5 to ensure the smoothness of the threshold adjustment and avoid step changes. When performing image segmentation, the original 1920×1080 pixel image is divided into image blocks of size 128×128 pixels, and the overlap rate between adjacent image blocks is set to 25%. For each image block, its gray mean is calculated and compared with the dynamic threshold. When the gray mean of the image block is higher than the dynamic threshold, the block is marked as the carbon fiber solution area. During the iterative threshold optimization process, the maximum number of iterations is set to 10, and the iteration is judged to converge when the difference in gray means between two consecutive iterations is less than 2.In practical applications, during the process of increasing the centrifuge speed from 6000 revolutions per minute to 9000 revolutions per minute, the gray threshold gradually increases from 120 to 165, and the corresponding image segmentation results show that the boundary clarity of the carbon fiber solution area is significantly improved. For areas with uneven local gray levels, by calculating the gray difference between adjacent image blocks and setting the difference threshold to 20, a transition area is inserted between the two image blocks when the threshold is exceeded, achieving a smooth transition of image segmentation.

[0028] S105. Apply the adjusted gray threshold parameter to the processing and analysis of the toner particle image, re-perform edge extraction, geometric feature extraction, and particle size distribution analysis of the toner particles, and use the adjusted analysis results as the basis for regulating the carbon fiber recovery process.

[0029] Perform edge detection on the toner particle image, adopt an adaptive edge tracking algorithm based on gradient to extract the toner particle contour, repair the contour breakpoints by calculating the edge point curvature value to obtain a complete edge contour image; for the complete edge contour image, use a recursive method to mark the independent particle regions, calculate the area and perimeter of the particle regions by integral projection method, calculate the major axis ratio of the particles according to the minimum circumscribed rectangle to obtain geometric feature data; perform normalization processing on the geometric feature data, classify the particle features, and obtain the particle size distribution curve by calculating the particle number distribution in different size intervals; for the particle size distribution curve, establish a mapping relationship with the carbon fiber strength and modulus by the least square method, and calculate the temperature and pressure parameter values of the recovery process by third-order polynomial fitting to obtain a process parameter sequence.

[0030] Specifically, edge detection is performed on the toner particle image according to the dynamically updated grayscale threshold. The adaptive edge tracking algorithm based on gradient is used to extract the toner particle contour. The continuity determination threshold is set by calculating the edge point curvature value, and the contour breakpoints are repaired to obtain a complete toner particle edge contour image. The toner particles in the complete edge contour image are marked with eight-connected regions, numbered recursively for each independent particle region, and the area and perimeter of the particle region are calculated by the integral projection method. The major axis ratio is calculated according to the minimum circumscribed rectangle to obtain the geometric feature data of the toner particles. The geometric feature data is normalized by the maximum-minimum method, and the density-based clustering method is used to classify the particle features. The clustering radius is set as the sample average distance, and the particle number distribution in different size intervals is calculated to obtain the particle size distribution curve. The peak position and distribution width parameters are extracted from the particle size distribution curve, and the mapping relationship with the carbon fiber strength and modulus is established by the least squares method. The temperature and pressure parameter values of the recycling process are calculated by third-order polynomial fitting. The calculated process parameters are constrained, and the process parameters are quantified by setting the change step of temperature and pressure. The recycling process parameter sequence is gradually updated according to the parameter change direction to obtain the regulation data of the carbon fiber recycling process. During the toner particle image processing, the edge detection uses the gradient-based adaptive algorithm, and the edge points are determined by calculating the grayscale gradient value of the image. When the image gradient value is greater than the dynamic threshold, the point is marked as an edge candidate point. For typical toner particle images, the dynamic threshold varies between 80 and 150. During the edge tracking process, the curvature value of adjacent edge points is calculated, and when the curvature is greater than 0.8, it is determined as the breakpoint position, and cubic spline interpolation is used to repair the breakpoint. In the toner particle region marking stage, the eight-connected region growing method is used to segment the particles. The image is scanned from the upper left corner. When an unmarked edge point is encountered, it is used as a seed point to start region growing, and the label number starts to increase from 1. Regions with an area less than 25 square pixels are determined as noise points and removed. When calculating the particle area by the integral projection method, the projection curves in the horizontal and vertical directions reflect the spatial distribution characteristics of the particles, and the number of peaks of the projection curve corresponds to the aggregation degree of the particles. In the normalization of the feature data, for the area feature, the largest and smallest area particles are selected as the normalization reference and linearly mapped to the interval of 0 to 1. For the major axis ratio feature, considering the shape characteristics of the toner particles, the value is limited between 1 and 5. In the density-based clustering analysis, the clustering radius is set as 0.15 times the sample average distance. When the number of sample points in a certain region exceeds 5% of the total sample size, the region is identified as a clustering center. When establishing the mapping relationship between the particle size distribution and the carbon fiber performance, the characteristic parameters of the particle size distribution curve are extracted, including the curve peak position, half-peak width, and skewness coefficient.Verified by a large amount of experimental data, when the peak position of the particle size distribution curve is around 15 microns and the full width at half maximum is less than 8 microns, the corresponding tensile strength of carbon fiber is usually higher than 3.8 GPa. A mapping relationship is established by fitting with a third-order polynomial, and the goodness of fit R-squared value reaches 0.92, indicating that the model has good predictive ability. During the process parameter regulation, the change step of the temperature parameter is set to 5 degrees Celsius, and the change step of the pressure parameter is 0.2 MPa. When the peak of the particle size distribution shifts towards the small size direction, the temperature parameter gradually decreases and the pressure parameter gradually increases. By real-time monitoring the change of particle size distribution, the process parameters are dynamically optimized to keep the recycling quality of carbon fiber in the best state. In practical applications, the temperature parameter is adjusted in the range of 250 to 320 degrees Celsius, and the pressure parameter varies in the range of 5 to 8 MPa, realizing the precise control of the carbon fiber recycling process.

[0031] S106. According to the analysis results of particle size distribution, use mathematical statistics methods to establish a quantitative relationship between the particle size distribution of carbon powder particles and the centrifugal rotation rate of the solution, and analyze the influence of the centrifugal rotation rate on particle agglomeration and particle size distribution.

[0032] Perform moment quantization calculation according to the particle size distribution curve of carbon powder particles, and obtain the mean parameter, standard deviation parameter, skewness parameter and kurtosis parameter of the distribution curve through integral operation; sample the centrifugal rotation rate data with a sliding window of a preset time period, and obtain the rotational speed change characteristic data by calculating the first-order difference value and the second-order difference value of the speed within the sliding window; construct a sample matrix according to the mean parameter, standard deviation parameter, skewness parameter and kurtosis parameter of the distribution curve and the rotational speed change characteristic data, and establish a support vector regression model through a radial basis kernel function to obtain a non-linear correlation equation; perform singular value decomposition on the eigenvectors of the non-linear correlation equation, and if the cumulative contribution rate of the eigenvalues reaches the cumulative contribution rate threshold, use a polynomial function to fit the data points in the feature space to obtain a particle size distribution change equation.

[0033] Specifically, according to the particle size distribution curve of toner particles, the moment quantization method is used to calculate the first to fourth moments of the distribution curve. The mean, standard deviation, skewness, and kurtosis parameters of the distribution curve are obtained through integral operations to quantitatively describe the particle size distribution characteristics and obtain the characteristic data representing the distribution characteristics. The centrifugal rotation speed data is sampled at fixed time intervals, and a sliding window with a length of a preset time period is used to calculate the speed change trend. By calculating the first and second differences of the speed within the window, the acceleration and deceleration characteristics of the centrifugal process are described, and the characteristic data representing the rotational speed change is obtained. A sample matrix is constructed based on the particle size distribution characteristic data and the rotational speed change characteristic data. The support vector regression method with a radial basis kernel function is used to establish a mapping relationship, and the kernel function parameters are optimized through cross-validation to obtain the non-linear correlation equation between the particle size distribution and the rotational speed. The coefficient matrix of the non-linear correlation equation is subjected to singular value decomposition. The eigenvectors are sorted by calculating the eigenvalue magnitudes, and the cumulative contribution rate method is used to determine the number of main features, resulting in a reduced-dimensional feature space. A quantitative relationship equation between the particle size distribution and the rotational speed is established in the reduced-dimensional feature space. The data points in the feature space are fitted by a polynomial function, and the feature coefficients are normalized to obtain a mathematical expression representing the variation law of the particle size distribution with the rotational speed. In the moment quantization process of the particle size distribution curve, the first moment reflects the average level of particle size, usually distributed in the range of 15 to 35 microns. The second moment represents the degree of dispersion of the distribution, and the standard deviation is between 3 and 8 microns. The third moment describes the asymmetry of the distribution. When the skewness is positive, it indicates that the distribution shifts towards the large-size direction, and the typical value is between 0.2 and 0.8. The fourth moment reflects the sharpness of the distribution, and the kurtosis value is usually between 2.5 and 4.0. The time series analysis of the centrifugal rotation speed uses a sliding time window with a length of 5 seconds, and the window moves 1 second each time for sampling. In the acceleration stage, the first difference value remains at about 800 revolutions per minute per second, showing relatively stable acceleration characteristics. The second difference reflects the change rate of acceleration, with a large value in the starting stage and dropping below 50 revolutions per minute per second squared after stable operation. When establishing the non-linear mapping relationship, the kernel width parameter of the radial basis kernel function is optimized by the grid search method, with the search range from 0.1 to 10 and a step size of 0.1. The cross-validation uses the 5-fold method. When the kernel width is 2.5, the root mean square error on the validation set is minimized, reaching 0.15. The penalty factor of the support vector regression is set to 100, and the epsilon tube width is 0.1, ensuring the generalization ability of the model. The eigenvalues obtained from the singular value decomposition are sorted from largest to smallest, and the ratio of the first three eigenvalues is approximately 4.5:2.8:1.0, with a cumulative contribution rate reaching 92%. This indicates that the relationship between the particle size distribution and the rotational speed can be described by three main characteristic patterns. The first eigenvector corresponds to the change trend of the average particle size with the rotational speed, showing an obvious negative correlation. The second eigenvector reflects the change law of the distribution width. When the rotational speed exceeds 8000 revolutions per minute, the distribution width begins to decrease significantly.In the dimensionality-reduced feature space, a third-order polynomial function is used to fit the data points, and the coefficients are normalized to be in the range of -1 to 1. In practical applications, when the rotational speed increases from 3000 revolutions per minute to 9000 revolutions per minute, the average particle size decreases from 35 microns to 18 microns, the standard deviation of the distribution decreases from 7.5 microns to 4.2 microns, and the skewness decreases from 0.75 to 0.35, indicating that the particle agglomeration phenomenon under high-speed centrifugation conditions is effectively inhibited. The data points in the feature space are evenly distributed along the polynomial curve, and the sum of squared residuals is less than 0.08, verifying that the established quantitative relationship model has good fitting accuracy.

[0034] S107. Perform real-time analysis and monitoring on the carbon powder particle distribution during the centrifugation of carbon fibers. According to the real-time analysis results, dynamically adjust the centrifugation process parameters, and use the online monitoring data to optimize the carbon fiber recycling process.

[0035] Obtain the carbon powder particle distribution image of the carbon fiber centrifuge, perform segmentation processing on the carbon powder particle distribution image to obtain particle distribution detection data; perform normalization preprocessing according to the particle distribution detection data, and perform dynamic prediction on the normalized preprocessed data through a recurrent neural network with long short-term memory units to obtain a distribution trend evaluation result; construct a deviation function according to the distribution trend evaluation result, and use a proportional-integral controller to calculate the deviation function to obtain a parameter correction amount; convert the parameter correction amount into a control signal using a communication protocol and transmit it to the centrifuge control unit through an industrial bus to obtain the carbon fiber centrifugation process parameters.

[0036] Specifically, real-time operation data is collected from the carbon fiber centrifuge. A high-speed image acquisition device is used to obtain the carbon powder particle distribution images within a fixed sampling period. Median filtering is employed to remove image noise. The region growing algorithm is used to segment the carbon powder particle regions, obtaining the particle distribution detection data. The particle distribution detection data is preprocessed by normalization. A recurrent neural network containing double-layer long short-term memory units is used to construct a time series predictor. By calculating the root mean square error between the predicted value and the measured value, the dynamic evaluation of the carbon powder particle distribution trend is carried out. A deviation function is constructed based on the distribution trend evaluation result. A proportional-integral controller with adjustable proportional coefficient and integral time is used to calculate the parameter correction amount. The control output is limited by setting the upper and lower limits of the correction amount, obtaining the adjustment instructions for the centrifugal speed and time. The adjustment instructions are quantized and encoded, and the instructions are converted into control signals using a communication protocol and transmitted to the centrifuge control unit through an industrial bus to adjust the carbon fiber centrifugation process parameters in real time. Feedback data is collected according to the adjusted process parameters. A state observer is used to estimate the dynamic characteristics of the process. By calculating the state error, a feedback gain matrix is designed to optimize the carbon fiber recovery process parameters online. During the real-time monitoring of the carbon fiber centrifugation process, a high-speed camera device is used for image acquisition. The sampling period is set to 0.1 second, and the image resolution is 1920×1080 pixels. Median filtering with a 5×5 window is used for the collected image sequence to remove noise points, and the signal-to-noise ratio of the filtered image is increased to over 35 dB. The region growing algorithm uses the pixel points with image gray values greater than the threshold as seed points, and the particle regions are segmented and marked by 8-neighborhood expansion. The recurrent neural network adopts a double-layer structure, with each layer containing 128 long short-term memory units. The input layer receives the particle distribution data at 10 consecutive time points. The network training uses the stochastic gradient descent method, and the learning rate is set to 0.01. The training stops when the root mean square error on the validation set is less than 0.05. The predictor predicts the distribution trend for the next 3 time points, and the average relative error of the prediction results is controlled within 8%. In the design of the proportional-integral controller, the initial value of the proportional coefficient is set to 2.5, and the integral time constant is set to 1.5 seconds. The control parameters are dynamically adjusted according to the system response characteristics through an online parameter tuning method. The control dead zone is set to plus or minus 2% of the centrifugal speed. The upper limit of the correction amount is 10% of the current speed, and the lower limit is -10% of the current speed to avoid overly drastic parameter adjustment. The industrial bus uses the real-time Ethernet protocol, with a communication cycle of 10 milliseconds and a data frame length of 32 bytes. The control instructions use a 4-byte encoding format, where the upper 2 bytes represent the speed correction amount and the lower 2 bytes represent the time correction amount. The cyclic redundancy check is used to ensure the accuracy of data transmission, and the data frame is automatically retransmitted when a communication error occurs. The state observer uses the Kalman filter algorithm, and the observation vector includes three components: centrifugal speed, motor torque, and power. The diagonal elements of the process noise covariance matrix are set to 0.01, and the diagonal elements of the measurement noise covariance matrix are set to 0.05.The feedback gain matrix is obtained by solving the Lyapunov equation, and the real parts of the eigenvalues are all less than -0.5, ensuring the stability of the closed-loop system. In practical applications, when the centrifugal speed increases from 5000 revolutions per minute to 8000 revolutions per minute, the average particle size of the toner particles decreases from 30 microns to 20 microns, and the distribution standard deviation decreases from 6 microns to 4 microns. The dynamic response time of the control system is less than 2 seconds, and the overshoot is controlled within 5%. During the feedback control process, the fluctuation range of the process parameters is controlled within plus or minus 3% of the set value, reflecting the good dynamic tracking performance and anti-interference ability of the control system.

[0037] As described above, this is only the specific implementation manner of this specification. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the systems, modules, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be repeated here. It should be understood that the protection scope of this specification is not limited thereto. Any person skilled in the art within the technical scope disclosed in this specification can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should all be covered within the protection scope of this specification.

Claims

1. A real-time analysis method for toner particle distribution based on image processing, characterized in that, The method includes: Preprocessing the image of carbon powder particles during the centrifugation of carbon fiber to obtain the preprocessed image of carbon powder particles. According to the preprocessed image of carbon powder particles, using the region growing algorithm, setting the gray threshold parameter for carbon powder particle edge recognition, and extracting the edge of carbon powder particles to obtain the edge contour information of carbon powder particles; According to the edge contour information of carbon powder particles, using the morphological processing method to eliminate the noise in the edge contour of carbon powder particles, including breakpoints and burrs, to obtain the repaired edge contour of carbon powder particles. By extracting the geometric features of the repaired edge contour of carbon powder particles, obtaining the geometric feature parameters of carbon powder particles in carbon fiber, including size, shape, and fiber orientation; According to the geometric feature parameters of carbon powder particles in carbon fiber, using the method based on cluster analysis to classify and identify carbon powder particles, obtaining carbon powder particle categories with different morphologies and sizes. By statistically analyzing the quantity and distribution of different categories of carbon powder particles, obtaining the particle size distribution information of carbon powder particles and correlating with the quality attributes of carbon fiber; During centrifugation, the centrifugal rotation rate of the carbon fiber solution is obtained in real time. According to the change of the centrifugal rotation rate, the gray threshold parameter is dynamically adjusted to adapt to the change of particle morphology. When the centrifugal rotation rate increases, the gray threshold is increased; when the centrifugal rotation rate decreases, the gray threshold is decreased; Applying the adjusted gray threshold parameter to the processing and analysis of the carbon powder particle image, re-performing the edge extraction, geometric feature extraction, and particle size distribution analysis of carbon powder particles, and using the adjusted analysis result as the basis for regulating the carbon fiber recycling process; According to the particle size distribution analysis result, using the mathematical statistics method to establish the quantitative relationship between the carbon powder particle size distribution and the centrifugal rotation rate of the solution, and analyzing the influence of the centrifugal rotation rate on particle agglomeration and particle size distribution; Performing real-time analysis and monitoring on the carbon powder particle distribution during the centrifugation of carbon fiber. According to the real-time analysis result, dynamically regulating the centrifugation process parameters, and using the online monitoring data to optimize the carbon fiber recycling process; 2. The method according to claim 1, wherein The preprocessing of the image of carbon powder particles during the centrifugation of carbon fiber to obtain the preprocessed image of carbon powder particles. According to the preprocessed image of carbon powder particles, using the region growing algorithm, setting the gray threshold parameter for carbon powder particle edge recognition, and extracting the edge of carbon powder particles to obtain the edge contour information of carbon powder particles includes: Obtaining the original image of carbon powder particles collected by the carbon fiber centrifuge, and performing denoising processing on the original image through adaptive Gaussian filtering with a set kernel function size to obtain the preprocessed image; Performing edge detection on the preprocessed image, obtaining edge candidate points by setting the gradient threshold, and performing segmentation and marking on the preprocessed image using the region growing algorithm with an eight-neighborhood search method to obtain the segmentation and marking image; Performing morphological erosion operation on the segmentation and marking image to remove noise points, and at the same time performing morphological dilation operation to complement the edge information to obtain the optimized image; Calculate the gray - scale difference between adjacent toner particles according to the gray - scale value distribution of toner particles in the optimized image. If the gray - scale difference is less than the preset threshold, it is determined as an adhesion area, and the adhesion area is segmented to obtain the edge contour information of the toner particles.

3. The method according to claim 1, characterized in that, According to the edge contour information of the toner particles, a morphological processing method is used to eliminate the noise in the edge contour of the toner particles, including breakpoints and burrs, to obtain the repaired edge contour of the toner particles. By extracting the geometric feature parameters of the toner particles in the carbon fiber through the repaired edge contour of the toner particles, including size, shape, and fiber orientation, it includes: Perform morphological opening and closing operations on the edge contour of the toner particles using a structural element to obtain a complete edge contour image; Perform boundary tracking coding based on the complete edge contour image, and eliminate abnormal points by setting the adjacent three - point angle threshold to obtain smooth edge contour data; For the smooth edge contour data, fit the projection distance from the contour points to the eigenvector to obtain the major axis and minor axis of the toner particle contour; Calculate the carbon fiber orientation angle according to the major axis and minor axis of the toner particle contour, calculate the ratio of the moment of inertia of the principal axis to the area using the area - moment method, and obtain the roundness parameter through the ratio of the square of the contour perimeter to the area.

4. The method according to claim 1, wherein According to the geometric feature parameters of the toner particles of the carbon fiber, a method based on cluster analysis is used to classify and identify the toner particles to obtain the categories of toner particles with different morphologies and sizes. By statistically analyzing the quantity and distribution of different categories of toner particles, the particle size distribution information of the toner particles is obtained, and the quality attributes of the carbon fiber are associated, including: Receive the data of the size, shape, and fiber orientation angle of the toner particles to construct a feature matrix, classify the feature matrix using a density - based clustering algorithm, and obtain the morphology classification result by calculating the product of the local density value and distance of the sample points in the feature space; Perform standardization processing on the feature matrix according to the morphology classification result, and extract the principal component vectors with the feature contribution rate greater than the contribution rate threshold using the principal component decomposition method to obtain the dimensionality - reduced feature data; Calculate the particle number frequency within the particle size interval according to the dimensionality - reduced feature data, and perform kernel density estimation on the frequency distribution data using a Gaussian kernel function to obtain the particle size distribution probability density function; Perform moment analysis on the probability density function to calculate the distribution parameters, and use the recursive least - squares method to iteratively calculate the parameters of the elastic modulus prediction equation to obtain the prediction result of the carbon fiber quality attributes.

5. The method according to claim 1, characterized in that, During the centrifugation process, the centrifugal rotation speed of the carbon fiber solution is obtained in real - time. According to the change of the centrifugal rotation speed, the gray - scale threshold parameter is dynamically adjusted to adapt to the change of the particle morphology. When the centrifugal rotation speed increases, the gray - scale threshold is increased; when the centrifugal rotation speed decreases, the gray - scale threshold is decreased, including: Obtain the real - time rotation speed signal from the carbon fiber centrifuge, and perform signal smoothing processing on the rotation speed signal using a Gaussian filter to obtain the filtered rotation speed data; Calculate the acceleration value per unit time according to the filtered rotation speed data, and perform arithmetic processing on the acceleration value using the proportional - integral method to obtain the reference gray - scale threshold value; Perform a clamping operation on the reference grayscale threshold value using the hyperbolic tangent function, and limit the operation result by the preset upper threshold and lower threshold to obtain a dynamic grayscale threshold; Perform a block processing on the carbon fiber solution image, perform a region segmentation on the image block according to the dynamic grayscale threshold, and perform an optimization operation on the segmentation result using the iterative threshold method to obtain a segmented image.

6. The method according to claim 1, characterized in that Apply the adjusted grayscale threshold parameter to the processing and analysis of the toner particle image, re-perform the edge extraction, geometric feature extraction and particle size distribution analysis of the toner particles, and use the adjusted analysis result as the basis for regulating the carbon fiber recycling process, including: Perform edge detection on the toner particle image, use an adaptive edge tracking algorithm based on gradient to extract the toner particle contour, and repair the contour breakpoints by calculating the edge point curvature value to obtain a complete edge contour image; For the complete edge contour image, use a recursive method to mark the independent particle regions, calculate the area and perimeter of the particle regions by the integral projection method, and calculate the major axis ratio of the particles according to the minimum circumscribed rectangle to obtain geometric feature data; Perform normalization processing on the geometric feature data, classify the particle features, and calculate the particle number distribution in different size intervals to obtain a particle size distribution curve; For the particle size distribution curve, establish a mapping relationship with the carbon fiber strength and modulus using the least squares method, and calculate the temperature and pressure parameter values of the recycling process by third-order polynomial fitting to obtain a process parameter sequence.

7. The method according to claim 1, wherein According to the particle size distribution analysis result, use mathematical statistics methods to establish a quantitative relationship between the toner particle size distribution and the centrifugal rotation rate of the solution, and analyze the influence of the centrifugal rotation rate on particle agglomeration and particle size distribution, including: Perform moment quantization calculation on the particle size distribution curve of the toner particles, and obtain the mean parameter, standard deviation parameter, skewness parameter and kurtosis parameter of the distribution curve through integral operation; Sample the centrifugal rotation rate data using a sliding window with a preset time period, and obtain the rotational speed change characteristic data by calculating the first-order difference value and the second-order difference value of the speed within the sliding window; Construct a sample matrix according to the mean parameter, standard deviation parameter, skewness parameter and kurtosis parameter of the distribution curve and the rotational speed change characteristic data, and establish a support vector regression model using the radial basis kernel function to obtain a non-linear correlation equation; Perform singular value decomposition on the eigenvector of the non-linear correlation equation. If the cumulative contribution rate of the eigenvalues reaches the cumulative contribution rate threshold, then use a polynomial function to fit the data points in the feature space to obtain a particle size distribution change equation.

8. The method according to claim 1, characterized in that Perform real-time analysis and monitoring on the toner particle distribution during the carbon fiber centrifugation process. According to the real-time analysis result, dynamically regulate the centrifugation process parameters, and use the online monitoring data to optimize the carbon fiber recycling process, including: Obtain the toner particle distribution image of the carbon fiber centrifuge, and perform segmentation processing on the toner particle distribution image to obtain particle distribution detection data; Perform normalization preprocessing based on the detected particle distribution data, and perform dynamic prediction on the data after normalization preprocessing through a recurrent neural network with long short-term memory units to obtain a distribution trend evaluation result; Construct a deviation function according to the distribution trend evaluation result, and use a proportional-integral controller to calculate the deviation function to obtain a parameter correction amount; Convert the parameter correction amount into a control signal using a communication protocol, and transmit it to the centrifuge control unit through an industrial bus to obtain carbon fiber centrifugation process parameters.

Images