A method for real-time analysis of low-temperature plasma pyrolysis thermodynamic processes

Abstract

This application provides a method for real-time analysis of the thermodynamic process of low-temperature plasma pyrolysis, comprising: using a hyperspectral imaging device to acquire emission spectral information within the pyrolysis cavity during the low-temperature plasma pyrolysis of carbon fiber precursor, obtaining a spatial image of the pyrolysis cavity containing emission spectral information, which is used to characterize the spectral features of the carbon fiber molecular chain breakage process; by monitoring the grayscale image of the local region of the vortex circulation, calculating the curvature features of the vortex morphology, identifying the geometric shape and change law of the vortex region, and determining the region with curvature feature similarity higher than the similarity threshold as a vortex region with self-organization characteristics, wherein the vortex region represents the degree of rearrangement and graphitization of the carbon fiber molecular chain.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a method for real-time analysis of low-temperature plasma pyrolysis thermodynamic processes. Background Technology

[0002] During the low-temperature plasma pyrolysis of carbon fiber precursor fibers, self-organized structures, such as vortices and ripples, appear in specific regions of the plasma boundary layer and pyrolysis cavity. These characteristics are closely related to the breakage, rearrangement, and graphitization of carbon fiber molecular chains. The formation mechanism of these self-organized structures is still unclear, and their impact on the carbon fiber pyrolysis process and the evolution of carbon fiber molecular chain structure requires further investigation. The current technical challenges are twofold: First, the emergence of self-organized structures is closely related to parameters such as plasma temperature, pressure, and electric field strength. These parameters not only affect the plasma state but also directly influence the active groups on the carbon fiber surface. However, these parameters change dynamically during pyrolysis. Real-time monitoring and control of these parameters are crucial to manage the formation of self-organized structures and accurately identify and locate unstable vortex circulation within the cavity. Second, the frequency of spectral characteristic data directly affects the accuracy of dominant frequency determination. Accurately extracting the dominant frequency of the vortex circulation and using it in conjunction with the center position information to obtain the coefficient of variation and grayscale image that reveals the structural evolution distribution pattern, thereby characterizing the degree of breakage and orientation state of the carbon fiber molecular chains, is a critical technical problem that needs to be solved. When a grayscale image exhibits a vortex shape with self-organized characteristics, how can we accurately calculate its curvature characteristics to quantify the complexity of the morphological structure, and accurately collect its ripple wavelength and circulation velocity data? Combined with the charge distribution characteristics of the active groups on the carbon fiber surface, we can study the formation mechanism of these self-organized structures and their impact on the carbon fiber pyrolysis process to achieve real-time analysis of the thermodynamic process of low-temperature plasma pyrolysis of carbon fibers. Summary of the Invention

[0003] This invention provides a method for real-time analysis of low-temperature plasma pyrolysis thermodynamic processes, mainly including:

[0004] Using a hyperspectral imaging device, emission spectral information in the pyrolysis chamber during the low-temperature plasma pyrolysis of carbon fiber precursor was acquired, and a spatial image of the pyrolysis chamber containing emission spectral information was obtained to characterize the spectral features of the carbon fiber molecular chain breakage process.

[0005] The spectral intensity values ​​of the emission spectrum information of the spatial image of the fracture cavity are obtained, including the characteristic spectral lines generated when the carbon fiber molecular chains break. If the spectral intensity value of the spectral characteristic band exceeds the preset threshold, it is confirmed that an unstable vortex circulation has appeared inside the fracture cavity and the vortex circulation is located. The vortex circulation center position information is recorded and integrated into the high-intensity spectral dataset.

[0006] Fourier transform is performed on the local region image to obtain the spectral image. The spatial location is grouped according to the frequency peak and the spectral features are obtained. The local region includes the vortex circulation center and the fracture region of the active group on the surface of the carbon fiber precursor. When the amplitude of a certain spatial frequency exceeds the amplitude threshold, it is determined as the dominant frequency of the unstable vortex circulation.

[0007] Based on the dominant frequency as the time sampling reference, the center position of the vortex circulation is mapped to the spatial coordinate system, and the coefficient of variation is used as the weight to perform a weighted average of the gray values ​​of pixels in the high-intensity spectral dataset to obtain the gray-scale image and spatial position image of the local region of the vortex circulation, which can be used to characterize the degree of breakage and orientation state of carbon fiber molecular chains.

[0008] By monitoring grayscale images of local vortex circulation regions, the curvature characteristics of vortex morphology are calculated, the geometric shape and variation patterns of vortex regions are identified, and regions with curvature feature similarity higher than the similarity threshold are identified as vortex regions with self-organizing characteristics. Vortex regions represent the degree of rearrangement and graphitization of carbon fiber molecular chains.

[0009] We acquire the wall charge accumulation data of the vortex region and the charge distribution characteristics of the active groups on the carbon fiber surface. We then combine the high-intensity spectral dataset to build a machine learning model to predict the correlation between charge distribution and self-organized structure. When the rate of change of wall charge distribution exceeds the rate threshold, it is identified as a nearby region that is prone to forming self-organized structure. We then integrate the relevant data into an information chain of the vortex region.

[0010] Data on the arc discharge frequency and vortex center magnetic field strength of the pyrolysis cavity are obtained and correlated with the information chain of the vortex region. By measuring the changes in vortex behavior during arc discharge, the correlation between vortex structure evolution, wall charge distribution and pyrolysis thermodynamic parameters is established. The processes of carbon fiber molecular chain breakage, rearrangement and graphitization are analyzed, forming a complete analytical chain.

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

[0012] This invention discloses a method for real-time analysis of the thermodynamic process of low-temperature plasma pyrolysis. The method utilizes a hyperspectral imaging device to acquire emission spectral information within the pyrolysis chamber, identifying and locating unstable vortex circulations by analyzing the intensity values ​​of characteristic spectral bands. Fourier transforms are performed on local regions of the vortex circulations to determine the dominant frequencies, and the vortex center position is mapped based on these frequencies. By monitoring the curvature characteristics and wall charge distribution of the vortex region, regions with self-organized features are identified, and the correlation between charge distribution and self-organized structures is predicted. Combining arc discharge frequency and magnetic field strength data, a correlation is established between vortex structure evolution, wall charge distribution, and pyrolysis thermodynamic parameters, enabling a comprehensive analysis of carbon fiber molecular chain breakage, rearrangement, and graphitization processes, providing a basis for optimizing the pyrolysis process. Attached Figure Description

[0013] Figure 1 This is a flowchart of a method for real-time analysis of a low-temperature plasma pyrolysis thermodynamic process according to the present invention.

[0014] Figure 2 This is a schematic diagram of a method for real-time analysis of a low-temperature plasma pyrolysis thermodynamic process according to the present invention. Detailed Implementation

[0015] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0016] like Figure 1-2 The method for real-time analysis of a low-temperature plasma pyrolysis thermodynamic process in this embodiment may specifically include:

[0017] Step S101: Using a hyperspectral imaging device, emission spectral information of the pyrolysis chamber during the low-temperature plasma pyrolysis of carbon fiber precursor is acquired to obtain a spatial image of the pyrolysis chamber containing emission spectral information, which is used to characterize the spectral features of the carbon fiber molecular chain breaking process.

[0018] A hyperspectral imager is used to acquire spectral data within the carbon fiber precursor pyrolysis chamber. Fourier transform is used to perform frequency domain decomposition on the spectral data to obtain a first spectral characteristic frequency distribution map. Wavelet transform is then performed on the first spectral characteristic frequency distribution map to remove background noise, resulting in a second spectral characteristic frequency distribution map. Spectral peak positions and intensity data are extracted from the second spectral characteristic frequency distribution map. Spectral absorption is used to measure the second spectral characteristic frequency distribution map to obtain gas component concentration data within the chamber. Data processing generates a second gas state dataset. A deep neural network model is established for the second gas state dataset. The spectral peak positions and intensity data are input into the deep neural network model to obtain predicted gas component concentrations. The location of carbon fiber molecular chain breakage is determined based on a pre-established standard carbon fiber molecular fragment spectral database.

[0019] For example, a three-dimensional spatial spectral database is established based on the emission spectral intensity distribution pattern within the carbon fiber precursor pyrolysis cavity. A hyperspectral imager is used to acquire spectral data from multiple angles within the cavity. Fourier transform is used to perform frequency domain decomposition of the spectral data to obtain a first spectral characteristic frequency distribution map. Spectral signal preprocessing is performed on the first spectral characteristic frequency distribution map, and wavelet transform is used to remove background noise to obtain a second spectral characteristic frequency distribution map. Spectral peak positions and intensity data are extracted from the second spectral characteristic frequency distribution map. The relationship between gas density and plasma state within the pyrolysis cavity is monitored in real time using two-photon absorption spectroscopy, recording the ratio of excited-state particles to ground-state particles to establish a first gas state dataset. A gas flow numerical model is established based on the first gas state dataset, setting the cavity boundary temperature field distribution as the boundary condition. Spectral absorption spectroscopy is used to measure the gas component concentration within the cavity to obtain a second gas state dataset. Data fusion is performed between the second gas state dataset and the second spectral characteristic frequency distribution map. A deep neural network is used to establish a model corresponding to the gas component concentration and spectral characteristics. The model input is spectral peak position and intensity data, and the output is the predicted gas component concentration. A pre-established standard carbon fiber molecular fragment spectral database was used to compare and verify the predicted values ​​of gas component concentrations. A similarity score was calculated using a spectral matching algorithm to determine the location of carbon fiber molecular chain breakage. During the monitoring of the carbon fiber precursor filament pyrolysis process, a hyperspectral imaging device acquired emission spectral data within the cavity through multi-angle sampling. The sampling angles included 0 degrees, 45 degrees, 90 degrees, 135 degrees, and 180 degrees, acquiring spectral data at 100 wavelengths per angle, forming a 500×100 spectral data matrix. After Fourier transform, the frequency domain characteristic distribution was obtained. In the spectral data matrix, the wavelength range was 200-800 nm, the wavelength interval was 6 nm, and the spectral intensity values ​​ranged from 0 to 4095 digital quantization units. The acquired spectral data underwent wavelet transform denoising, using the db4 wavelet basis function. The original spectral data was decomposed into five levels with a soft threshold of 15. The reconstructed denoised spectral data showed major characteristic peaks at 308 nm, 336 nm, and 391 nm, corresponding to the characteristic emission lines of OH radicals, NH radicals, and N2+ ions, respectively. Two-photon absorption spectroscopy was used to monitor the gas state within the pyrolysis chamber, with a laser wavelength of 532 nm and a pulse energy density of 2.5 J / cm². 2 The intensity of the absorbed signal was recorded using a photomultiplier tube, and the excited-state particle number density was calculated to be 3.2 × 10⁻⁶. 16 / cm 3 The ground-state particle number density is 8.5 × 10⁻⁶. 18 / cm 3The ratio is 0.0038. In the numerical model of gas flow, the cavity boundary temperature is set to 1200K, the pressure is standard atmospheric pressure, and the Navier-Stokes equations are used to describe the gas flow, where the gas viscosity coefficient μ is 1.8 × 10⁻⁶. -5 With a thermal conductivity λ of 0.025 W / (m·K), the temperature and velocity field distributions within the cavity were obtained using the finite volume method. A deep neural network model employs a 5-layer fully connected network structure. The input layer contains 100 nodes representing spectral peak positions and intensities, the hidden layers have 64, 32, and 16 nodes respectively, and the output layer has 10 nodes corresponding to different gas component concentrations. The ReLU function is used as the activation function, the mean squared error function as the loss function, and the learning rate is set to 0.001. The model converged after 10,000 iterations. A standard carbon fiber molecular fragment spectral database was used, containing characteristic spectral data for 20 typical molecular fragments, including fragments resulting from the breakage of CC, CH, and CN bonds. The spectral matching algorithm uses cosine similarity calculation, with a similarity threshold of 0.85. When the similarity between the measured spectrum and the standard spectrum exceeds the threshold, it is identified as a molecular fragment of that type. The location and extent of carbon fiber molecular chain breakage can be determined by the type and concentration distribution of molecular fragments.

[0020] Step S102: Obtain the spectral intensity values ​​of the spectral characteristic bands of the emission spectrum information of the spatial image of the fracture cavity, including the characteristic spectral lines generated when the carbon fiber molecular chains break. If the spectral intensity value of the spectral characteristic band exceeds a preset threshold, it is confirmed that an unstable vortex circulation has appeared inside the fracture cavity and the vortex circulation is located. The vortex circulation center position information is recorded and integrated into the high-intensity spectral dataset.

[0021] A multi-angle spectral acquisition device is used to acquire emission spectral data of carbon fiber molecular chain fracture. The spectral data includes the position of characteristic spectral lines and spectral intensity values. The spectral intensity values ​​are then assessed to determine if they exceed a preset threshold. If the spectral intensity values ​​exceed the preset threshold, a vortex field distribution map is obtained using particle image velocimetry. A region growing algorithm is used to extract the vortex circulation core region from the vortex field distribution map, and a centroid positioning algorithm is used to calculate the vortex feature vector. The vortex feature vector includes the vortex center coordinates, vortex peak value, and vortex core region area. A temporal correlation is established between the vortex feature vector and the spectral data, and the least squares method is used to fit the vortex center motion trajectory curve. The vortex center motion trajectory curve records the vortex circulation evolution process.

[0022] For example, based on the emission spectrum information generated by the breakage of carbon fiber molecular chains within the fracture cavity, spectral characteristic band data are collected at 0°, 45°, 90°, and 135°. The positions and intensity values ​​of characteristic spectral lines are extracted using peak-valley identification to obtain the first spectral dataset. A threshold judgment is applied to the spectral intensity values ​​in the first spectral dataset, setting the threshold at 80% of the standard intensity of the characteristic spectral lines. If the spectral intensity value exceeds the threshold, a particle image velocimetry method is used to obtain the velocity distribution map of the flow field within the fracture cavity, and the first vorticity field distribution map is calculated using the curl operator. Gaussian filtering is applied to the first vorticity field distribution map to reduce noise, resulting in a second vorticity field distribution map. A region growing algorithm is used to extract regions with vorticity values ​​greater than twice the average vorticity as the vortex circulation core region. A centroid localization algorithm is used to calculate the vortex center coordinates for the vortex circulation core region, constructing a vortex feature vector containing the vortex center coordinates, peak vorticity, and vortex core region area. Based on the temporal correlation between the vortex feature vector and the first spectral dataset, a light intensity correction algorithm is used to compensate for the spectral intensity of the vortex region, generating a high-intensity spectral dataset. To capture the vortex circulation center location information in the high-intensity spectral dataset, the least squares method was used to fit the vortex center trajectory curve, recording the vortex circulation evolution process. During spectral acquisition within the carbon fiber pyrolysis chamber, four fiber optic probes at different angles were used to simultaneously acquire emission spectra. Each probe had a 2mm aperture, a sampling frequency of 100Hz, a spectral range covering 200-800nm, and a spectral resolution of 0.1nm. Feature spectral lines were extracted using a peak-valley identification algorithm, primarily including three characteristic spectral lines: CH (431.4nm), CN (388.3nm), and C2 (516.5nm). When the intensity of a characteristic spectral line exceeded 80% of the standard spectral line intensity threshold, a particle image velocimetry system was activated. Alumina tracer particles with a diameter of 0.5μm were used, with a pulsed laser wavelength of 532nm, a pulse interval of 100μs, a camera resolution of 2048×2048 pixels, and a spatial resolution of 0.05mm / pixel to measure the flow velocity field. The spatial gradient of the velocity field is calculated using the central difference method, and then the curl field distribution is solved. The curl calculation formula is as follows: Where u and v are the velocity components in the x and y directions, respectively. The calculated curl field is then denoised using a Gaussian filter with a kernel size of 5×5 and a standard deviation σ = 1.5. In the filtered curl field, the vorticity value is greater than twice the average vorticity, approximately 150 s. -1 The region is designated as the vortex core region. The vortex core region is typically elliptical in shape, with a major axis of approximately 4 mm, a minor axis of approximately 2 mm, and an area of ​​approximately 6.28 mm². 2The vortex feature vector includes parameters such as the vortex center coordinates (x, y), the peak vorticity ωmax, and the core area S. The vortex center coordinates are calculated using the region centroid formulas xc = ∑(xi·ωi) / ∑ωi and yc = ∑(yi·ωi) / ∑ωi, where ωi is the vorticity value at the i-th point. The vortex region's influence on spectral acquisition is mainly manifested as light intensity attenuation; the attenuation coefficient is proportional to the vortex intensity, with typical values ​​between 0.85 and 0.95. When performing least-squares fitting on the vortex center trajectory, a second-order polynomial model y = ax is used. 2 +bx+c, where a typical coefficient value is a = -0.05mm -1 b = 0.8, c = 2.5 mm. The root mean square error of the fitted curve is usually less than 0.2 mm, indicating that the vortex motion has good regularity. The vortex center moves within a range of 10-30 mm in the x-direction and 5-15 mm in the y-direction, with a period of approximately 0.1 s. The evolution of the vortex circulation shows an initial increase in intensity followed by a decrease, with the maximum vortex volume reaching 300 s⁻¹ and a duration of approximately 0.2 s.

[0023] Step S103: Perform Fourier transform on the local region image to obtain a spectral image. Group the spatial locations according to the frequency peaks and obtain spectral features. The local region includes the vortex circulation center and the fracture region of the active groups on the surface of the carbon fiber precursor. When the amplitude of a certain spatial frequency exceeds the amplitude threshold, it is determined as the dominant frequency of the unstable vortex circulation.

[0024] Based on the location of the vortex circulation center, a sliding window is used to divide the local region image into blocks, and a first frequency domain image is obtained through two-dimensional fast Fourier transform. A Gaussian filter is applied to the first frequency domain image for noise suppression, and a region growing clustering algorithm is used to cluster regions where the frequency difference between adjacent locations is less than a difference threshold, resulting in a second frequency domain image. Based on the second frequency domain image, a random forest algorithm is used to classify the frequency domain features at different spatial locations, and feature vectors are extracted from the frequency domain features to obtain a frequency domain feature dataset. Double threshold detection is performed on the frequency amplitudes in the frequency domain feature dataset, and a maximum entropy algorithm is used to sort the frequency domain features to obtain the dominant frequency components of the vortex circulation.

[0025] For example, based on the coordinates of the fracture area of ​​the active groups on the surface of the carbon fiber precursor, a 32×32 pixel sliding window is used to divide the local image into blocks with a sliding step size of 16 pixels. A two-dimensional fast Fourier transform is used to perform a frequency domain transformation on each image block to obtain the first frequency domain image. A 5×5 pixel Gaussian filter is used to suppress noise in the first frequency domain image, with a standard deviation of 1.2. A peak detection algorithm is used to extract the spatial frequency peaks in the local area. Regions with frequency differences of less than 10% between adjacent locations are clustered using region growing to generate the second frequency domain image. Based on the statistical frequency amplitude distribution of the second frequency domain image, a random forest algorithm is used to classify the frequency domain features at different spatial locations. Feature vectors are extracted from the frequency domain features, including frequency peaks, amplitudes, bandwidth, symmetry, and skewness parameters, to obtain a frequency domain feature dataset. For the frequency amplitudes in the frequency domain feature dataset, a double-threshold detection is performed with a low threshold of 3 times the mean background amplitude and a high threshold of 5 times the mean amplitude, to obtain the third frequency domain image. The frequency domain features were sorted using the maximum entropy algorithm based on the entropy value of the third frequency domain image. The entropy value was calculated using the Shannon entropy formula to determine the dominant frequency component of the vortex circulation. Time-frequency analysis was then performed on the dominant frequency component, using continuous wavelet transform to calculate the frequency variation over time and record the evolution of the dominant frequency. In the vortex circulation analysis of the carbon fiber precursor pyrolysis process, extracting frequency domain features from local image regions is crucial. When using a 32×32 pixel sliding window for image segmentation, a sliding step size of 16 pixels ensured a 50% overlap between adjacent blocks, which is beneficial for capturing continuously changing frequency domain features. The spectrum obtained after two-dimensional fast Fourier transform of each image block had a frequency resolution equal to the sampling frequency divided by the window size; for a 10kHz sampling frequency, the frequency resolution was 312.5Hz. In the noise suppression process, the standard deviation of the 5×5 Gaussian filter, 1.2, was an optimized parameter value that effectively suppressed high-frequency noise while maintaining the main spectral features. For Gaussian white noise with an amplitude of 100, the filtered noise amplitude was reduced to below 15. Frequency peak detection employs the local maximum method, setting a minimum peak-to-valley ratio of 3:1 and ensuring an interval of no less than 5 frequency points between adjacent peaks to avoid detecting false peaks. During region growing clustering, a 10% frequency difference threshold is used to reflect the spatial continuity of the vortex circulation frequency domain characteristics. A typical vortex circulation region has an area of ​​approximately 1000 pixels and contains 3-5 main frequency components. In the frequency domain feature vector, the frequency peaks reflect the vortex rotation frequency, typically within the range of 200-500Hz. Bandwidth characterizes the fluctuation of vortex intensity, while symmetry and skewness describe the statistical characteristics of the spectral shape. In dual-threshold detection, the low threshold is set to 3 times the average background amplitude (approximately 45 units), and the high threshold is set to 5 times (approximately 75 units), based on the optimal segmentation ratio obtained through experimental statistics.When sorting frequency domain features using the maximum entropy algorithm, the entropy value is calculated using the formula H = -Σ(pi × log₂pi), where pi is the normalized frequency amplitude. The entropy value of the dominant frequency is typically between 4.5 and 5.5, significantly higher than that of the secondary frequency components. The continuous wavelet transform uses Morlet wavelets with a scale parameter range of 1-64, corresponding to a frequency range of 50-1000 Hz. Time-frequency analysis results show that the dominant frequency of the vortex circulation exhibits an upward trend in the early stages of formation, increasing from 200 Hz to 350 Hz, lasting approximately 50 ms, after which it tends to stabilize, with frequency fluctuations controlled within ±20 Hz. The slope of the frequency evolution curve reflects the formation speed of the vortex circulation, while the amplitude of fluctuations in the stable segment indicates the stability of the vortex structure.

[0026] Step S104: Based on the dominant frequency as the time sampling reference, the center position of the vortex circulation is mapped to the spatial coordinate system, and the gray value of the pixel in the high-intensity spectral dataset is weighted and averaged using the coefficient of variation as the weight to obtain the gray value image and spatial position image of the local region of the vortex circulation, which is used to characterize the degree of breakage and orientation state of the carbon fiber molecular chain.

[0027] Based on the location of the vortex circulation center, time-series resampling is performed using cubic spline interpolation. The first spatial location image is obtained by mapping to a spatial rectangular coordinate system using a rotation and translation matrix. For the high-intensity spectral dataset, a sliding window is used to calculate the statistical parameters of pixel grayscale values ​​in local areas. The first weighted image is obtained by the ratio of standard deviation to mean. The high-intensity spectral dataset is then subjected to pixel-level weighted averaging based on the first weighted image. The weighted grayscale value distribution is then clustered using a random forest algorithm to obtain a second grayscale image. The second grayscale image and the first spatial location image are then fused at the feature level. A spatial feature vector containing location coordinates, grayscale values, and gradient directions is extracted using a deep neural network.

[0028] For example, the sampling time interval is set according to the dominant frequency of the vortex circulation, and the reciprocal of the period is used as the reference time unit. The center position of the vortex circulation is resampled in time series using cubic spline interpolation. A rotation and translation matrix is ​​constructed to map the center position to a spatial rectangular coordinate system, resulting in the first spatial position image. For pixels in the high-intensity spectral dataset, a 16×16 pixel sliding window is used to calculate the statistical parameters of the pixel grayscale values ​​within a local region. The window sliding step size is 8 pixels. The coefficient of variation is calculated using the ratio of the standard deviation to the mean, generating the first weighted image. A pixel-level weighted average is performed on the high-intensity spectral dataset based on the first weighted image. The weighting coefficients are normalized to the 0-1 range. A random forest algorithm is used to perform cluster analysis on the weighted grayscale value distribution, resulting in the second grayscale image. The gradient distribution of grayscale values ​​is calculated for the second grayscale image. Edge features are extracted using the Sobel operator, and a gradient direction histogram is established, resulting in the third grayscale image. Feature-level fusion is performed between the third grayscale image and the first spatial position image. A deep neural network is used to extract spatial feature vectors, containing three components: position coordinates, grayscale value, and gradient direction. A system of quadratic curve equations was constructed based on the spatial feature vectors. The spatial orientation parameters of the carbon fiber molecular chains were fitted using the least squares method to calculate the curvature values ​​and inflection point positions, thus determining the degree of molecular chain breakage. During carbon fiber pyrolysis, the dominant frequency of the vortex circulation is typically in the range of 300-500 Hz. Taking 400 Hz as an example, the corresponding baseline sampling time interval is 2.5 ms. When resampling the vortex center position using cubic spline interpolation, four control points were selected to construct a cubic polynomial. The interpolation node interval was 1 / 4 of the baseline time interval, i.e., 0.625 ms, thus obtaining a smooth center trajectory curve. The rotation and translation matrix adopted a 4×4 homogeneous coordinate transformation matrix, containing the rotation angle θ and the translation vector (tx, ty, tz). In the processing of the high-intensity spectral dataset, a 16×16 pixel sliding window corresponds to an actual spatial size of approximately 0.8 mm × 0.8 mm, with each window containing 256 pixels. The coefficient of variation was calculated as the ratio of the standard deviation σ to the mean μ, i.e., CV = σ / μ. For images with grayscale values ​​ranging from 0 to 255, the typical coefficient of variation is between 0.1 and 0.5. A larger coefficient of variation indicates significant fluctuations in grayscale values ​​within that region. When weighting the grayscale values, the weight coefficient *w* is normalized using the sigmoid function: *w* = 1 / (1 + e^(-CV)), limiting the weight values ​​to the range of 0-1. The weighted grayscale value *g'* is calculated as *g'* = Σ(wi×gi) / Σwi*, where *gi* is the original grayscale value. Edge detection is performed using the Sobel operator on the weighted result. The horizontal operator is [-1,-2,-1; 0,0,0; 1,2,1], and the vertical operator is [-1,0,1; -2,0,2; -1,0,1], calculating the gradient magnitude and direction.When calculating the gradient orientation histogram, the 360-degree direction is divided into 18 intervals, each 20 degrees. The number of gradient orientations falling into each interval is counted and weighted. Feature vector fusion uses a three-layer deep neural network. The input layer has 5 nodes, containing x and y coordinates, grayscale values, gradient magnitude, and direction. The hidden layers have 16 and 8 nodes respectively, and the output layer has 3 nodes. The spatial orientation curve uses a quadratic function y = ax. 2 The curve is fitted using the formula +bx+c, where coefficient 'a' reflects the curvature of the curve, 'b' represents the overall tilt trend, and 'c' is the longitudinal offset. Locations with a curvature value greater than 0.5 mm⁻¹ are identified as potential breakage points. In practical applications, the typical curvature value of carbon fiber precursor is in the range of 0.2-0.8 mm⁻¹, while the curvature value in the fracture region often exceeds 0.6 mm⁻¹, and the curvature value exhibits abrupt changes at the breakage point.

[0029] Step S105: By monitoring the grayscale image of the local area of ​​the vortex circulation, the curvature characteristics of the vortex shape are calculated, the geometric shape and change pattern of the vortex region are identified, and the region with curvature characteristic similarity higher than the similarity threshold is identified as the vortex region with self-organizing characteristics. The vortex region represents the degree of rearrangement and graphitization of carbon fiber molecular chains.

[0030] For the grayscale image of the vortex circulation, the Sobel operator is used to extract the vortex edge contour point sequence. Based on the vortex edge contour point sequence, a first curvature feature image is obtained by calculating a cubic spline curve. Based on the first curvature feature image, the random forest algorithm is used to extract the shape features of the vortex region, including roundness, eccentricity, and area ratio, to obtain a second curvature feature image. Based on the second curvature feature image, the feature similarity of the vortex region is calculated using cosine distance. If the feature similarity is higher than a preset threshold, a region growing method is used to obtain a third curvature feature image. Based on the third curvature feature image, the vortex region morphological parameters are calculated, including region area, perimeter, and equivalent diameter. The interlayer spacing parameter is obtained through a deep neural network to determine the graphitization conversion level.

[0031] For example, based on the grayscale image of a local region of the vortex circulation, the Sobel operator is used to extract the vortex edge contour. A sequence of edge contour points is fitted using a cubic spline curve, and the tangent vectors and normal vectors of adjacent points on the curve are calculated. The curvature distribution of the vortex shape is calculated using the discrete curvature formula k = (x'y”-y'x”) / ((x'^2+y'^2)^(3 / 2)), resulting in a first curvature feature image. Morphological processing is then performed on the first curvature feature image. Opening operations are performed using circular structuring elements with a radius of 3 pixels, and closing operations are used to fill the gaps in the edge contour. A random forest algorithm is used to extract shape features such as roundness, eccentricity, and area ratio of the vortex region, generating a second curvature feature image. A feature vector is constructed based on the second curvature feature image, containing the mean, standard deviation, maximum, and minimum curvature values. The feature similarity between adjacent vortex regions is calculated using cosine distance, resulting in a first similarity matrix. The first similarity matrix is ​​normalized. If the similarity value is higher than a preset threshold of 0.85, the vortex regions are merged using a region growing method. Regions with gray value differences less than the standard deviation are set as growth conditions to obtain the third curvature feature image. Based on the third curvature feature image, the morphological parameters of the self-organized vortex regions are calculated, including region area, perimeter, and equivalent diameter. A deep neural network is used to extract the spatial distribution features of the region's gray values ​​to generate the second feature matrix. The interlayer spacing parameter d = λ / (2sinθ) of the carbon fiber molecular chains is calculated based on the second feature matrix, where λ is the incident light wavelength and θ is the diffraction angle. The graphitization conversion level is determined by the deviation between the interlayer spacing and the standard graphite lattice spacing. In the carbon fiber pyrolysis process, the morphological feature analysis of the vortex circulation begins with the extraction of the edge contours. When using the Sobel operator for edge detection, the difference operators in the horizontal and vertical directions are [-1,-2,-1; 0,0,0; 1,2,1] and [-1,0,1;-2,0,2;-1,0,1], respectively. For a 256-level grayscale image, the typical edge response intensity threshold is set to 50. The extracted edge point sequence is fitted using a cubic spline curve, with each control point spaced 4 pixels apart to ensure the smoothness of the curve. Curvature calculation is expressed parametrically: for a point P(x(t),y(t)) on the curve, its curvature k = (x'y”-y'x”) / ((x'^2+y'^2)^(3 / 2)). Taking the vortex core region as an example, the curvature value usually fluctuates between 0.2 and 0.8, with the curvature peak appearing at the sharp bend of the vortex circulation. In morphological processing, a circular structuring element with a radius of 3 pixels corresponds to an actual size of approximately 150 micrometers, and the noise spot area eliminated by the opening operation is less than 28 square pixels. Shape characteristics include roundness, 4π × area / perimeter 2 The ideal circle has a roundness of 1, while the actual roundness of the vortex region is between 0.7 and 0.9. The eccentricity reflects the degree of vortex ellipticity, and the calculation formula is e = √(1 - b). 2 / a 2The similarity is calculated using cosine distance. For feature vectors vi and vj, the similarity is s = vi·vj / (|vi|×|vj|). The feature vectors include curvature statistics with a mean of 0.4-0.6, a standard deviation of 0.1-0.2, kurtosis of 2.5-4.0, and skewness of ±0.5. A threshold of 0.85 is set to ensure that the merged vortex regions have similar morphological features. During region growth, seed points are selected from local curvature maxima. Growth conditions include a grayscale value difference less than the current region's standard deviation, 15-25 grayscale levels, and a curvature change rate of less than 30%. The input feature dimension of the deep neural network is 128, including a 64-dimensional normalized grayscale histogram and a 64-dimensional morphological feature. In the calculation of the interlayer spacing of carbon fiber molecular chains, the incident light wavelength λ = 0.154 nm, copper target X-rays, and the diffraction angle θ varying from 22° to 26° correspond to a decrease in interlayer spacing d from 0.395 nm to 0.335 nm. The interlayer spacing of standard graphite is 0.3354 nm. The deviation of the interlayer spacing of the actual sample from the standard value reflects the degree of graphitization; a deviation of less than 0.002 nm indicates a high degree of graphitization.

[0032] Step S106: Obtain the wall charge accumulation data of the vortex region and the charge distribution characteristics of the active groups on the carbon fiber surface. Combine the high-intensity spectral dataset to build a machine learning model to predict the correlation between charge distribution and self-organized structure. When the rate of change of wall charge distribution exceeds the rate threshold, it is determined to be a nearby region that is prone to forming self-organized structure. The relevant data are then integrated into an information chain of the vortex region.

[0033] The Laplace operator is used to calculate the spatial distribution gradient of the charge obtained by the wall charge detector in the vortex region, and a first charge distribution feature map is obtained through a Poisson equation solver. Based on the electric field intensity data in the first charge distribution feature map, a Gaussian smoothing filter is used to obtain a second charge distribution feature map. The second charge distribution feature map is aligned with the high-intensity spectral dataset, and a random forest algorithm is used to establish a mapping relationship between charge distribution and spectral features to obtain a third charge distribution feature map. Based on the charge density, accumulation rate, and spectral feature data of the third charge distribution feature map, a prediction model is constructed using a deep neural network. If the accumulation rate exceeds a rate threshold, it is marked as a self-organized neighboring region.

[0034] For example, based on the charge density data collected by the wall charge detector in the vortex region, a five-point Laplace operator is used to calculate the spatial distribution gradient of the charge. The spatial distribution of wall charge density and electric field intensity is then calculated using a Poisson equation solver, resulting in a first charge distribution feature map. For the electric field intensity data in the first charge distribution feature map, Gaussian smoothing filtering is used to remove noise, and the kernel size is set to 5×5, resulting in a second charge distribution feature map. Data alignment is performed between the second charge distribution feature map and the high-intensity spectral dataset. A random forest algorithm is used to establish a mapping relationship between charge distribution and spectral features. The feature vector includes electric field intensity, charge density, spectral peak position, and peak intensity, generating a third charge distribution feature map. The wall charge change rate dρ / dt is calculated for the third charge distribution feature map. A sliding window of 32 time points with a step size of 8 time points is set, and the central difference method is used to calculate the charge accumulation rate, resulting in a first rate distribution map. Charge accumulation features were extracted from the first rate distribution map, and a prediction model was constructed using a deep neural network. The input layer contained three sets of data: charge density, accumulation rate, and spectral features. The number of hidden layer nodes was 64-32-16. When the accumulation rate exceeded a preset threshold, it was marked as a self-organized neighborhood region. A feature data table was constructed for the self-organized neighborhood region, containing four data blocks: spatial coordinates, charge distribution, spectral features, and accumulation rate. A unified information chain structure was generated using a data merger. During the carbon fiber pyrolysis process, the wall charge distribution was calculated using a five-point Laplacian operator. For discrete points with a grid spacing h = 0.1 mm, the typical potential value φ was distributed between -500 V and 500 V, and the calculated charge density ρ ranged from -2 × 10⁻⁶. -6 Up to 2×10 -6 C / m 3 When solving the Poisson equation, the dielectric constant ε0 is taken as 8.85 × 10⁻⁶. -12 The boundary condition is set to wall potential φ = 0V, with a current of F / m. Gaussian smoothing filtering using a 5×5 kernel matrix and a standard deviation σ = 1.2 is applied to reduce noise in the electric field strength data. Before processing, the electric field strength fluctuation range is 4×10⁻⁶. 5 ±2×10 4 V / m, the fluctuation after filtering is reduced to ±5×10 3 The electric field density (V / m) maintains the main characteristics of the electric field distribution. A random forest algorithm constructs 100 decision trees, with feature vectors containing electric field strength, charge density, and spectral peak positions: CH is 431.4 nm, CN is 388.3 nm, and C2 is 516.5 nm, along with their corresponding peak intensities. The charge accumulation rate is calculated using a 32-point sliding window, corresponding to an actual time span of 80 ms, with a window step size of 8 points and 20 ms. The central difference calculation formula is dρ / dt=(ρi+1-ρi-1) / (2Δt), where Δt is the sampling time interval of 2.5 ms. Typical charge accumulation rates are within ±5×10⁻⁶.-4 C / (m 3 The cumulative rate varies within the range of ·s, when it exceeds 1×10 -3 C / (m 3 When ·s), it is marked as a self-organized neighborhood region. The input layer of the deep neural network model contains 72 nodes, with 16 nodes each for charge density and accumulation rate, and 40 nodes for spectral features, containing intensity values ​​at 20 wavelengths. The hidden layer uses the ReLU activation function, with a dropout rate of 0.3 and an initial learning rate of 0.001. The prediction model is trained using the mean squared error loss function, and after 10,000 iterations, the validation set accuracy reaches 92%. In the structure design of the feature data table, the spatial coordinates record the position of the vortex center (x,y,z) with an accuracy of 0.1mm; the charge distribution data includes charge density ρ, electric field intensity E, and accumulation rate dρ / dt; the spectral features record the wavelength position λ, peak intensity I, and full width at half maximum (FWHM) Δλ of the feature peaks. The information chain is stored in binary format, with each data block header containing a 4-byte identifier and 4-byte length information to ensure data integrity and traceability. The typical size of the information chain structure is 4KB, containing the key parameters required to characterize the self-organizing process.

[0035] Step S107: Obtain the arc discharge frequency and vortex center magnetic field strength data of the pyrolysis cavity, associate them with the vortex region information chain, and establish the correlation between vortex structure evolution, wall charge distribution and pyrolysis thermodynamic parameters by measuring the changes in vortex behavior in arc discharge. Analyze the breaking, rearrangement and graphitization process of carbon fiber molecular chains to form a complete analysis chain.

[0036] The discharge frequency data acquired by the arc detector and the magnetic field strength distribution data measured by the Hall sensor array are used to obtain a first spectral feature map through fast Fourier transform. Based on the frequency domain components in the first spectral feature map, a Butterworth low-pass filter is used to process the data to obtain a second spectral feature map. The second spectral feature map and the magnetic field strength distribution data are correlated in time, and the vortex motion features are extracted using a random forest algorithm to obtain a first evolution feature map. A feature matrix is ​​constructed based on the first evolution feature map and the wall charge distribution data. A deep neural network is used to establish a mapping relationship of thermodynamic parameters, and the fracture location is determined by the bond angle change rate.

[0037] For example, discharge frequency data is obtained from arc detectors arranged in a ring array within the fracture cavity. A 4×4 Hall sensor array with a spacing of 5 mm is used to measure the magnetic field strength distribution in the vortex center region. The frequency domain components of the arc discharge are calculated using a 1024-point Fast Fourier Transform, and a first spectral feature map is generated by combining this with the vortex region information chain. For the frequency domain components in the first spectral feature map, a Butterworth low-pass filter is used for noise reduction, with the cutoff frequency set to 1 / 4 of the sampling frequency, resulting in a second spectral feature map. The second spectral feature map is temporally correlated with the magnetic field strength data, and a random forest algorithm is used to extract vortex motion features, including the vortex center trajectory, angular velocity, and magnetic field gradient, generating a first evolutionary feature map. Vortex structural parameters are extracted from the first evolutionary feature map, and principal component analysis is used to reduce the dimensionality of the parameters. Feature vectors with a cumulative contribution rate exceeding 85% are selected as principal components, resulting in a second evolutionary feature map. A feature matrix is ​​constructed based on the second evolutionary feature map and wall charge distribution data. A deep neural network is used to establish a mapping relationship between thermodynamic parameters, with the input layer containing three sets of data: temperature field, pressure field, and electric field strength. The stress distribution of molecular chains was calculated based on the mapping relationship of thermodynamic parameters, the fracture location was determined by the bond angle change rate, and the degree of graphitization was quantified using the lattice interlayer spacing calculation method. A complete link table was established based on the molecular chain structure evolution data, including four data blocks: vortex characteristics, field strength distribution, thermodynamic parameters, and structural characteristics. A hierarchical storage structure was used to record the correspondence between each link. In the monitoring of carbon fiber pyrolysis, arc discharge frequency and magnetic field strength are key parameters. A 4×4 Hall sensor array covered a 20×20mm measurement area, with each sensor having a sensitivity of 50mV / mT and a measurement range of 0-100mT. The arc discharge frequency sampling rate was set to 10kHz, and a 1024-point FFT was used to obtain the 0-5kHz spectrum distribution, with the main discharge frequency components concentrated in the 500-2000Hz range. A Butterworth low-pass filter with a fourth-order design and a cutoff frequency of 2.5kHz was used to suppress high-frequency noise through the -3dB amplitude-frequency characteristic at the cutoff frequency. The signal-to-noise ratio of the filtered signal was improved from the original 15dB to 28dB, while maintaining the main characteristics of the discharge frequency. In the vortex motion feature extraction, the center trajectory is located by the point of maximum magnetic field strength, and the angular velocity is calculated from the position change between adjacent time points, with typical values ​​in the range of 200-600 rad / s. The maximum magnetic field gradient reaches 5 mT / mm. When principal component analysis is used to reduce the dimensionality of the vortex structure parameters, the original features contain 15 parameters: position coordinates (x, y, z), velocity components (vx, vy, vz), acceleration components (ax, ay, az), magnetic field strength (Bx, By, Bz), and its gradient. The cumulative contribution of the first four principal components reached 87%. The first principal component, approximately 45%, mainly reflects the spatial location characteristics of the vortex; the second principal component, approximately 23%, characterizes the motion state; and the third and fourth principal components, approximately 12% and 7%, respectively, correspond to the magnetic field distribution characteristics. Thermodynamic parameter mapping employed a five-layer neural network. The input layer included a temperature field of 1000-2500 K, a pressure field of 0.1-1 MPa, and an electric field strength of 1-5 × 10⁻⁵ K. 5 Three sets of data (V / m) were used, each with 20 sampling points. The number of hidden layer nodes were 128, 64, 32, and 16, respectively, and the ReLU activation function was used. The molecular chain breakage criterion was based on the C / C bond angle change rate; a breakage location was determined when the bond angle change exceeded 15° / ps. The degree of graphitization was characterized by the interlayer spacing of the crystal planes. The standard value for complete graphitization was 0.3354 nm, and the actual measured value varied in the range of 0.3354-0.344 nm. The complete link table adopted a hierarchical storage structure. The first layer stored 512 bytes of vortex features, including position, velocity, and acceleration data; the second layer recorded 1024 bytes of field strength distribution, including electromagnetic field strength and gradient; the third layer stored 2048 bytes of thermodynamic parameters, including temperature, pressure, and energy data; and the fourth layer stored 1024 bytes of structural features, including bond angle, bond length, and interlayer spacing information.

[0038] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for real-time analysis of low-temperature plasma pyrolysis thermodynamic processes, characterized in that, The method includes: using a hyperspectral imaging device to acquire emission spectral information within the pyrolysis cavity during the low-temperature plasma pyrolysis of carbon fiber precursor, obtaining a spatial image of the pyrolysis cavity containing emission spectral information, used to characterize the spectral features during the carbon fiber molecular chain breakage process; acquiring the spectral intensity values ​​of the spectral characteristic bands of the emission spectral information of the pyrolysis cavity spatial image, including characteristic spectral lines generated during carbon fiber molecular chain breakage; if the spectral intensity value of the spectral characteristic band exceeds a preset threshold, then an unstable vortex circulation is confirmed to exist inside the pyrolysis cavity, the vortex circulation is located, and the vortex circulation center position information is recorded and integrated into a high-intensity spectral dataset; performing a Fourier transform on the local region image to obtain a spectral image, and characterizing the spatial region based on the frequency peak value. Location grouping and spectral feature acquisition are performed. Local regions include the vortex circulation center and the fractured areas of active groups on the carbon fiber precursor surface. When the amplitude of a certain spatial frequency exceeds the amplitude threshold, it is identified as the dominant frequency of the unstable vortex circulation. Based on the dominant frequency as the time sampling reference, the position of the vortex circulation center is mapped to a spatial coordinate system. Using the coefficient of variation as a weight, the gray values ​​of pixels in the high-intensity spectral dataset are weighted and averaged to obtain grayscale images and spatial location images of the local regions of the vortex circulation, which are used to characterize the degree of fracture and orientation state of carbon fiber molecular chains. By monitoring the grayscale images of the local regions of the vortex circulation, the curvature features of the vortex morphology are calculated to identify the geometric shape and variation patterns of the vortex region. Regions with curvature feature similarity higher than [a certain value] are grouped accordingly. The similarity threshold region is defined as a vortex region with self-organizing characteristics. The vortex region represents the degree of rearrangement and graphitization of carbon fiber molecular chains. The process includes: extracting a sequence of vortex edge contour points from the grayscale image of the vortex circulation using the Sobel operator; calculating a first curvature feature image based on the vortex edge contour point sequence using cubic spline curves; extracting the shape features of the vortex region using a random forest algorithm based on the first curvature feature image, the shape features including roundness, eccentricity, and area ratio, to obtain a second curvature feature image; calculating the feature similarity of the vortex region based on the second curvature feature image using cosine distance; if the feature similarity is higher than a preset threshold, a third curvature feature image is obtained using a region growing method; and so on. The curvature feature image is used to calculate the morphological parameters of the vortex region, including the region area, perimeter, and equivalent diameter. The interlayer spacing parameter is obtained through a deep neural network to determine the graphitization conversion level. The wall charge accumulation data of the vortex region and the charge distribution characteristics of the active groups on the carbon fiber surface are obtained. A machine learning model is constructed by combining the high-intensity spectral dataset to predict the correlation between charge distribution and self-organized structure. When the rate of change of wall charge distribution exceeds the rate threshold, it is determined to be a nearby region that is prone to forming self-organized structure. The relevant data are integrated into the information chain of the vortex region, including: using the Laplace operator to calculate the charge spatial distribution gradient obtained by the wall charge detector of the vortex region, and obtaining the first charge distribution feature map through the Poisson equation solver.Based on the electric field intensity data in the first charge distribution feature map, a Gaussian smoothing filter is used to obtain a second charge distribution feature map. The second charge distribution feature map is then aligned with a high-intensity spectral dataset, and a random forest algorithm is used to establish a mapping relationship between charge distribution and spectral features to obtain a third charge distribution feature map. Based on the charge density, accumulation rate, and spectral feature data of the third charge distribution feature map, a deep neural network is used to construct a prediction model. If the accumulation rate exceeds a rate threshold, it is marked as a self-organized neighboring region. The arc discharge frequency and vortex center magnetic field intensity data of the fracture cavity are acquired and correlated with the vortex region information chain. By measuring the changes in vortex behavior during arc discharge, the evolution of the vortex structure, wall charge distribution, and fracture thermodynamic parameters are established. The correlation between the data and the magnetic field strength distribution data was analyzed to determine the fracture, rearrangement, and graphitization processes of carbon fiber molecular chains, forming a complete analytical chain. This includes: receiving discharge frequency data acquired by an arc detector and magnetic field strength distribution data measured by a Hall sensor array, and obtaining a first spectral feature map through fast Fourier transform; processing the frequency domain components in the first spectral feature map using a Butterworth low-pass filter to obtain a second spectral feature map; performing time-series correlation between the second spectral feature map and the magnetic field strength distribution data, and extracting vortex motion features using a random forest algorithm to obtain a first evolutionary feature map; constructing a feature matrix based on the first evolutionary feature map and wall charge distribution data, establishing a thermodynamic parameter mapping relationship using a deep neural network, and determining the fracture location through the bond angle change rate.

2. The method according to claim 1, characterized in that, The method of using a hyperspectral imaging device to acquire emission spectral information within the pyrolysis cavity during the low-temperature plasma pyrolysis of carbon fiber precursor, and obtaining a spatial image of the pyrolysis cavity containing emission spectral information, used to characterize the spectral features during the carbon fiber molecular chain breakage process, includes: acquiring spectral data within the pyrolysis cavity of carbon fiber precursor using a hyperspectral imager, and performing frequency domain decomposition on the spectral data through Fourier transform to obtain a first spectral feature frequency distribution map. Based on the first spectral characteristic frequency distribution map, wavelet transform is performed to remove background noise, resulting in a second spectral characteristic frequency distribution map. Spectral peak positions and intensity data are extracted from the second spectral characteristic frequency distribution map. The second spectral characteristic frequency distribution map is measured using spectral absorption to obtain gas component concentration data within the cavity. A second gas state dataset is generated through data processing. A deep neural network model is established for the second gas state dataset. The spectral peak positions and intensity data are input into the deep neural network model to obtain predicted gas component concentrations. The location of carbon fiber molecular chain breakage is determined based on a pre-established standard carbon fiber molecular fragment spectral database.

3. The method according to claim 1, characterized in that, The acquisition of emission spectrum information of the fracture cavity spatial image includes the spectral intensity values ​​of the spectral characteristic bands, including the characteristic spectral lines generated when the carbon fiber molecular chains break. If the spectral intensity value of the spectral characteristic band exceeds a preset threshold, it is confirmed that an unstable vortex circulation has appeared inside the fracture cavity, and the vortex circulation is located and the vortex circulation center position information is recorded and integrated into a high-intensity spectral dataset. This includes: acquiring emission spectrum data of carbon fiber molecular chain breakage using a multi-angle spectral acquisition device, the spectral data including the position of characteristic spectral lines and spectral intensity values; determining whether the spectral intensity value exceeds a preset threshold, and if the spectral intensity value exceeds the preset threshold, acquiring the vortex field distribution map using particle image velocimetry; extracting the vortex circulation core region using a region growing algorithm for the vortex field distribution map, and calculating the vortex feature vector using a centroid positioning algorithm, the vortex feature vector including the vortex center coordinates, vortex peak value, and vortex core region area; and performing a time-series correlation between the vortex feature vector and the spectral data, fitting the vortex center motion trajectory curve using the least squares method, the vortex center motion trajectory curve recording the vortex circulation evolution process.

4. The method according to claim 1, characterized in that, The process involves performing a Fourier transform on a local region image to obtain a spectral image, grouping spatial locations based on frequency peaks, and acquiring spectral features. The local region includes the vortex circulation center and areas where the active groups on the surface of the carbon fiber precursor are broken. When the amplitude of a certain spatial frequency exceeds an amplitude threshold, it is determined as the dominant frequency of the unstable vortex circulation. This includes: dividing the local region image into blocks using a sliding window based on the vortex circulation center location; obtaining a first frequency domain image through a two-dimensional fast Fourier transform; applying a Gaussian filter to suppress noise in the first frequency domain image; clustering regions where the frequency difference between adjacent locations is less than a difference threshold using a region growing clustering algorithm to obtain a second frequency domain image; classifying the frequency domain features at different spatial locations using a random forest algorithm based on the second frequency domain image; extracting feature vectors from the frequency domain features to obtain a frequency domain feature dataset; performing dual-threshold detection on the frequency amplitudes in the frequency domain feature dataset; and sorting the frequency domain features using a maximum entropy algorithm to obtain the dominant frequency components of the vortex circulation.

5. The method according to claim 1, characterized in that, The method, based on the dominant frequency as the time sampling benchmark, maps the center position of the vortex circulation to a spatial coordinate system and uses the coefficient of variation as weight to perform a weighted average of the pixel gray values ​​in the high-intensity spectral dataset, obtaining gray-scale images and spatial position images of the local region of the vortex circulation, used to characterize the degree of breakage and orientation state of carbon fiber molecular chains. This includes: resampling the time series using cubic spline interpolation based on the center position of the vortex circulation, mapping it to a spatial rectangular coordinate system using a rotation and translation matrix to obtain a first spatial position image; calculating the statistical parameters of pixel gray values ​​in the local region using a sliding window for the high-intensity spectral dataset, obtaining a first weighted image by the ratio of standard deviation to mean; performing pixel-level weighted averaging on the high-intensity spectral dataset based on the first weighted image, and performing cluster analysis on the weighted gray value distribution using a random forest algorithm to obtain a second gray-scale image; and performing feature-level fusion on the second gray-scale image and the first spatial position image, extracting spatial feature vectors containing position coordinates, gray values, and gradient directions using a deep neural network.

Images