A far infrared-based polyester filament spectral reflectance detection method

Abstract

The application provides a far-infrared-based polyester filament spectral reflectance detection method, and belongs to the technical field of polyester filament spectral reflectance detection. The method comprises the following steps: obtaining enhanced spectral data of a polyester filament sample; performing feature band extraction, calculating a correlation matrix, and obtaining a high-dimensional spectral feature vector; constructing a polyester filament spectral reflectance three-dimensional simulation model, and obtaining a spectral energy distribution diagram; performing matching degree verification, and obtaining a pre-judgment matching index; analyzing a spectral shift type, and obtaining a reflectance defect classification standard; constructing a reflection mode parameter data set, extracting a feature coincidence degree parameter, inputting the feature coincidence degree parameter into a random forest algorithm, and determining a reflectance anomaly grade; and generating a spectral reflectance detection report. The application positions an abnormal interval through three-dimensional simulation simulation, establishes a quantitative correlation between spectral anomalies and process parameters to realize tracing, generates a standardized defect classification and process optimization suggestion to form a closed-loop management and control, can be adapted to detection of multiple types of synthetic fibers, and effectively improves production quality management and control level.

Description

Technical Field

[0001] This invention belongs to the field of polyester filament spectral reflectance detection technology, specifically a method for detecting polyester filament spectral reflectance based on far-infrared radiation. Background Technology

[0002] As a widely used synthetic fiber, polyester filament's spectral reflectance characteristics directly affect the product's warmth retention, coloring effect, and stability in use. Therefore, spectral reflectance testing is a core aspect of quality control in polyester filament production.

[0003] Existing methods for detecting the spectral reflectance of polyester filament mostly rely on single-angle spectral acquisition, which is easily affected by factors such as sample morphology and environmental interference, leading to insufficient accuracy and difficulty in precisely capturing subtle spectral anomalies. Furthermore, traditional detection techniques can only achieve simple numerical measurements of reflectance, failing to establish a correlation between spectral anomalies and production process parameters, making it difficult to trace the causes of defects. This necessitates relying on experience to adjust processes, resulting in low efficiency and a lack of specificity. In addition, existing methods lack in-depth processing and simulation of spectral data, failing to intuitively present the energy distribution patterns of far-infrared light within the polyester filament, making it difficult to pinpoint the root cause of anomalies. This leads to the omission of some potential defects, affecting product quality stability. With the textile industry's continuously increasing demands for polyester filament performance, traditional detection methods can no longer meet the needs of high-precision, traceable, and highly adaptable production control. There is an urgent need for an integrated spectral reflectance detection technology that can achieve accurate detection, root cause tracing, and process guidance. Summary of the Invention

[0004] This invention provides a method for detecting the spectral reflectance of polyester filament based on far-infrared radiation, in order to overcome the deficiencies in the existing technology.

[0005] This invention provides a method for detecting the spectral reflectance of polyester filament based on far-infrared radiation, comprising: Raw far-infrared spectral data of polyester filament samples were collected and preprocessed to obtain enhanced spectral data.

[0006] Feature bands are extracted from the enhanced spectral data, and the correlation matrix between the spectral absorption coefficient and the reflection intensity is calculated to obtain a high-dimensional spectral feature vector.

[0007] A three-dimensional simulation model of the spectral reflectance of polyester filament was constructed. The model was meshed to simulate the spectral energy distribution of far-infrared light inside the polyester filament, and the coordinates of the energy peak points were extracted to obtain the spectral energy distribution map.

[0008] Based on the high-dimensional spectral feature vector, the abnormal range of spectral reflectance is located, and the energy concentration area is separated from the spectral energy distribution map. The matching degree of the pre-acquired polyester filament production process parameters is verified with the abnormal range and the energy concentration area to obtain the predicted matching index.

[0009] If the predicted matching index meets the preset matching degree threshold range, then the spectral shift type is analyzed, and the reflectivity defect classification standard is obtained by combining the spectral shift type and the production process parameter attributes.

[0010] A dataset of reflectance mode parameters is constructed, and a stability index of spectral reflectance is calculated based on the Monte Carlo method. When the stability index exceeds a preset stability threshold, the feature overlap parameter is extracted and input into the random forest algorithm to determine the reflectance anomaly level.

[0011] When the reflectance anomaly level is higher than the preset anomaly threshold, the reflectance mode parameter dataset and reflectance anomaly level are integrated to generate a spectral reflectance detection report.

[0012] The method for detecting the spectral reflectance of polyester yarn based on far-infrared radiation according to the present invention includes the following steps to obtain enhanced spectral data: Fourier transform infrared spectrometer was used to collect raw spectral data of polyester filament samples from multiple angles in the far-infrared band. The raw spectral data included raw spectral signal data of the polyester filament's reflection, absorption and transmission characteristics of far-infrared light at different wavelengths.

[0013] A polynomial fitting algorithm is used to perform baseline correction on the raw spectral data to eliminate the effects of baseline drift.

[0014] An adaptive Wiener filtering algorithm is used to filter noise from the corrected spectral data, preserving the effective spectral signal.

[0015] The filtered spectral data were normalized using the max-min normalization method to obtain enhanced spectral data.

[0016] The method for detecting the spectral reflectance of polyester filament based on far-infrared radiation according to the present invention includes the following steps for obtaining a high-dimensional spectral feature vector: The wavelet transform algorithm is used to extract the characteristic bands of the enhanced spectral data, highlighting the boundary characteristics of the characteristic spectral peaks, and thus obtaining the first spectral feature set.

[0017] The characteristic peak intensity values ​​are obtained from the first spectral feature set. When the characteristic peak intensity value is greater than the preset intensity threshold, the density clustering algorithm is used to cluster similar spectral bands to obtain the second spectral feature set.

[0018] Based on the second spectral feature set, the correlation matrix between the spectral absorption coefficient and the reflection intensity is calculated.

[0019] The eigenvalues ​​of the correlation matrix are used as weights to generate high-dimensional spectral feature vectors.

[0020] The method for detecting the spectral reflectance of polyester filament based on far-infrared radiation according to the present invention includes the following steps to obtain the spectral energy distribution map: A three-dimensional simulation model was constructed based on the diameter, fineness, and cross-sectional shape parameters of polyester filament. The model was then divided into hexahedral meshes to generate the first mesh model.

[0021] Obtain the production process parameters corresponding to polyester filament, including spinning temperature, cooling air velocity, and stretching ratio.

[0022] Based on the first grid model and production process parameters, the initial energy field is calculated using the law of light propagation, resulting in the first energy distribution set.

[0023] Based on the Lorentz filter, the energy values ​​exceeding the preset energy threshold in the first energy distribution set are smoothed to obtain the second energy distribution set.

[0024] For the second energy distribution set, the coordinates of the energy peak points are extracted, and the regions are divided according to the coordinates of the energy peak points to form a spectral energy distribution map.

[0025] According to the far-infrared-based method for detecting the spectral reflectance of polyester filament provided by the present invention, the process of obtaining the predicted matching index includes: Feature decoding is performed on the high-dimensional spectral feature vectors, and the coordinates of the midpoint of the wavelength range of the feature spectral peaks are extracted as the center coordinates of the anomaly range. The half-width at half-maximum (WHM) of the spectral peaks is calculated as the span of the anomaly range.

[0026] Based on the Sobel edge detection algorithm, the boundaries of energy concentration regions are extracted from the spectral energy distribution map to generate a dataset of concentrated regions containing coordinates and energy values.

[0027] The production process parameter attributes are obtained from the pre-established production process database. The production process parameter attributes are logically associated with the abnormal interval coordinates and the concentrated area data set to calculate the first matching degree index.

[0028] When the span of the abnormal interval exceeds the preset span threshold, the first matching degree index is corrected according to the preset index adjustment rules to obtain the predicted matching index.

[0029] According to the far-infrared spectral reflectance detection method for polyester yarn provided by the present invention, the process of obtaining the reflectance defect classification standard includes: Based on the production process parameters, the crystallinity and orientation parameters of the polyester filament are obtained.

[0030] When the predicted matching index meets the preset matching degree threshold range, the gradient boosting tree algorithm is used to determine the type of spectral shift caused by the energy value data in the abnormal interval of the energy concentration area, and the shift determination result is obtained by combining the crystallinity and orientation parameters.

[0031] A Bayesian classification algorithm is used to correlate the offset determination results with the production process parameter attributes, and the occurrence pattern of reflectivity defects under the same combination of process parameters is statistically analyzed to generate a reflectivity defect classification standard.

[0032] According to the far-infrared-based method for detecting the spectral reflectance of polyester filament provided by the present invention, the process of constructing a reflectance mode parameter dataset includes: Historical spectral reflectance pattern data are obtained from a pre-established spectral feature database. Based on the reflectance defect classification criteria, the historical spectral reflectance pattern data are classified into defects to obtain a set of classified defect type data.

[0033] Based on the defect type dataset, obtain the spectral wavelength. Based on the energy value E, the finite difference method is used to simulate the correspondence between the spectral wavelength and energy value of each type of defect, and a spectral-energy relationship dataset is obtained.

[0034] When the energy value exceeds the preset energy threshold, the data source of the spectrum-energy relationship dataset is traced, the defect origin features are extracted, and the defect origin features are classified by a hierarchical clustering algorithm to obtain a defect origin feature dataset.

[0035] For the defect origin feature dataset, a hidden Markov model is used to calculate the defect type probability of various defect features to obtain the reflection mode parameter dataset.

[0036] According to the far-infrared-based method for detecting the spectral reflectance of polyester yarn provided by the present invention, the process of determining the reflectance anomaly level includes: Based on the reflection mode parameter dataset, the core factors of the reflection mode data and energy distribution data are extracted using factor analysis to form a spectral feature dataset.

[0037] For the spectral feature dataset, multiple random sampling simulations were performed using the Monte Carlo method to calculate the spectral reflectance stability index.

[0038] By applying boundary condition constraints to the stability index, the calibration stability index is obtained.

[0039] When the calibration stability index exceeds the preset stability threshold, the mean-shift clustering algorithm is used to extract the feature overlap parameters of the reflection mode data and the energy distribution data to obtain the overlap index dataset.

[0040] Using spectral feature datasets and overlap index datasets as input, the random forest algorithm is used to classify reflectance anomaly risks and determine the reflectance anomaly level.

[0041] According to the present invention, a method for detecting the spectral reflectance of polyester filament based on far-infrared radiation is provided, wherein boundary condition constraints are applied to the stability index to obtain a calibration stability index, including: Define the boundary constraints, including the operating temperature and humidity ranges for the polyester filament.

[0042] Based on boundary constraints, a linear correction model is used to calibrate the original stability index.

[0043] According to the far-infrared-based method for detecting the spectral reflectance of polyester yarn provided by the present invention, the process of generating a spectral reflectance detection report includes: The system integrates the reflection mode parameter dataset, reflectance anomaly level, anomaly interval coordinates, energy concentration area information, and production process parameter matching results, and calculates the probability of spectral reflectance compliance.

[0044] The test report should clearly indicate the measured value of spectral reflectance, the standard value range, the basis for judging the anomaly level, and the process optimization suggestions, thus forming a complete spectral reflectance test report.

[0045] This invention provides a far-infrared-based method for detecting the spectral reflectance of polyester filament. Through multi-angle spectral acquisition and multi-step preprocessing, it effectively eliminates environmental interference and sample morphology effects, significantly improving spectral data quality and ensuring the accuracy of reflectance detection. It can capture subtle spectral anomalies and avoid missing potential defects. Using 3D simulation technology, it visually presents the energy distribution pattern of far-infrared light within the polyester filament, enabling precise location of abnormal regions and providing visual support for defect analysis. Through algorithm model construction and data correlation analysis, it establishes a quantitative relationship between spectral anomalies and production process parameters, achieving full-chain tracing from spectral anomalies to defect causes. This breaks away from the traditional experience-based process adjustment mode, improving the targeting and efficiency of process optimization. Simultaneously, this method generates standardized defect classification standards and implementable process optimization suggestions, forming a complete detection-analysis-optimization closed loop. It is adaptable to large-scale polyester filament production quality control scenarios and can be adapted to the detection of other synthetic fibers through parameter adjustments, demonstrating strong versatility. Attached Figure Description

[0046] The invention will now be further described with reference to the accompanying drawings.

[0047] Figure 1 This is a schematic flowchart of a method for detecting the spectral reflectance of polyester filament based on far-infrared radiation in this invention. Figure 2 This is a schematic diagram of the process for obtaining high-dimensional spectral feature vectors in this invention; Figure 3 This is a schematic diagram of the process for obtaining the spectral energy distribution map in this invention; Figure 4 This is a schematic diagram of the process for obtaining the predicted matching index in this invention. Detailed Implementation

[0048] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0049] like Figures 1 to 4 As shown in the figure, an embodiment of the present invention provides a method for detecting the spectral reflectance of polyester yarn based on far-infrared radiation, comprising: Raw far-infrared spectral data of polyester filament samples were collected and preprocessed to obtain enhanced spectral data. The specific implementation is as follows: A Fourier transform infrared spectrometer was selected as the core detection equipment, using a standard polystyrene film as a reference. The spectrometer's wavelength accuracy was calibrated to ±0.01μm to ensure that the measurement error in the 2.5-25μm far-infrared band met the requirements. The detector gain was set to a medium level to avoid signal saturation or loss of weak signals. The ambient temperature was stabilized at 25±1℃ and the humidity was controlled at 50±5%RH to reduce the interference of environmental factors on the spectral signal.

[0050] Example sample parameters: A sample of unblemished, unbroken polyester yarn was selected and cut to a length of 10cm. The diameter was measured multiple times using a laser diameter gauge (5 measurements, with values ​​of 0.098mm, 0.102mm, 0.100mm, 0.099mm, and 0.101mm respectively), with an average diameter of 0.1mm. The fineness was measured by gravimetric analysis (1000m length weighed 15.02g), yielding a fineness of 150D. Microscopic observation showed a circular cross-section without distortion. Differential scanning calorimetry measured a crystallinity of 45%, and X-ray diffraction measured an orientation degree of 82%. A constant tension of 5cN was applied using a tension meter to fix the sample on a dedicated spectral detection fixture, ensuring the sample was flat and free from tensile deformation to avoid spectral signal distortion caused by irregular sample shape. The far-infrared band range was set to 2.5-25μm. A multi-angle scanning strategy was adopted to scan and collect polyester filament samples from four angles: 0° (vertical incidence), 45°, 90°, and 135°. During each angle scan, the spectrometer spot diameter was adjusted to 0.5mm, the number of scans was set to 32, and three duplicate data were collected for each angle. After removing outliers, the average value was taken as the original spectral signal data for that angle. The data was integrated to obtain the original far-infrared spectral data including multi-dimensional characteristics. The data format was a wavelength-reflection intensity correspondence matrix, the wavelength step was set to 0.01μm, and the matrix dimension was 2251×4 (2251 wavelength points, 4 acquisition angles).

[0051] A sixth-order polynomial fitting algorithm is used to subtract the baseline from the original spectrum by fitting the trend curve of the spectral baseline, thus eliminating the effect of baseline drift. The correction formula is as follows: ,in to The polynomial fitting coefficients are obtained using the least squares method. The specific steps are as follows: A smooth region in the spectrum without characteristic absorption (2.5-3.0 μm, 24.0-25.0 μm) is selected as the baseline reference segment, and the wavelength and reflectance intensity data of this region are extracted. These are substituted into the polynomial equation to construct the objective function. The objective is to minimize the sum of squared residuals between the baseline fitting values ​​obtained from the polynomial calculation and the measured values ​​of the original spectral reflectance intensity acquired by the spectrometer. The fitting coefficients of each order are then obtained by solving this method.

[0052] Baseline correction is achieved by subtracting the fitted baseline function value from the original spectral data. After correction, the slow drift of the spectral baseline caused by instrument heating and environmental fluctuations can be effectively removed, making the characteristic spectral peaks clearer.

[0053] An adaptive Wiener filtering algorithm is employed, which adaptively adjusts the filter intensity based on the local variance of the spectral signal to suppress noise while preserving effective signal details. Specifically, the filter window size is 5×1 (5 points in the wavelength direction × 1 point in the angular direction), and the noise estimation threshold is 0.005. The sliding process is as follows: The four acquisition angles of 0°, 45°, 90°, and 135° are used as independent dimensions for filtering. Slide the window point by point along the wavelength dimension. The window covers the current wavelength point and two wavelength points before and after it, for a total of five consecutive wavelength points. Calculate the local mean and variance of the data within the window. If the signal variance is less than the threshold, it is determined to be a noise region and strong smoothing is performed. If the variance is greater than the threshold, it is determined to be a valid feature region and the smoothing intensity is reduced to avoid distortion of feature peaks. After sequentially traversing 2251 wavelength points and completing single-angle filtering, the sliding operation was repeated for the remaining three angles. The effect was verified using the signal-to-noise ratio (SNR) metric after filtering. The calculated SNR of the filtered spectrum increased from 28dB to 45dB, and the random noise intensity decreased to less than 3% of the original signal.

[0054] The max-min normalization method is used to eliminate intensity differences in data collected from different angles, uniformly mapping the reflection intensity to the [0,1] interval, which facilitates subsequent feature comparison and calculation. The normalization formula is as follows: ,in The reflection intensity after baseline correction and noise filtering. The minimum reflection intensity across the entire wavelength range. This represents the maximum reflection intensity across the entire spectral band. The normalized reflection intensity is calculated point-by-point to obtain the enhanced spectral data.

[0055] Feature bands are extracted from the enhanced spectral data, and the correlation matrix between the spectral absorption coefficient and the reflection intensity is calculated to obtain a high-dimensional spectral feature vector. The specific implementation is as follows: The wavelet transform algorithm is used to extract the feature bands. By decomposing the spectrum at multiple scales, the low-frequency trend and high-frequency features are separated, highlighting the effective feature peaks.

[0056] The enhanced spectral data was decomposed using the db4 wavelet basis at three levels, yielding one low-frequency component (cA3) and three high-frequency components (cD1, cD2, and cD3). The low-frequency component cA3 corresponds to the overall trend of the spectrum, while the high-frequency components correspond to the noise and detail features. Noise in the high-frequency components was removed using a thresholding method (a threshold of 0.02 was set for high-frequency components, with coefficients below the threshold set to zero), while retaining the detail information related to the characteristic peaks. The processed low-frequency and high-frequency components were then reconstructed using wavelet analysis to obtain the preliminarily selected characteristic spectra.

[0057] Based on the reconstructed spectrum, characteristic bands corresponding to low-frequency components are extracted, with a focus on highlighting the boundary characteristics of two characteristic spectral peaks: 10.5-12.5 μm (corresponding to COC stretching vibration in polyester filament) and 15.0-16.5 μm (corresponding to C=O stretching vibration in polyester filament). The starting wavelength, ending wavelength, peak position, and intensity information of each characteristic peak are determined, resulting in the first spectral feature set. The first spectral feature set is a structured parameter set, specifically including: Characteristic peak 1 (COC stretching vibration): band range 10.5-12.5μm, starting wavelength 10.5μm, ending wavelength 12.5μm, peak position 11.5μm, peak intensity, full width at half maximum (FWHM), peak area, peak shape symmetry, signal-to-noise ratio; Characteristic peak 2 (C=O stretching vibration): band range 15.0-16.5μm, starting wavelength 15.0μm, ending wavelength 16.5μm, peak position 15.75μm, peak intensity, half width at half maximum (FWHM), peak area, peak shape symmetry, signal-to-noise ratio.

[0058] The preset intensity threshold is set to 0.3. Feature peaks with an intensity value greater than 0.3 are selected from the first spectral feature set. Secondary peaks with weak intensity or easily disturbed intensity are removed, and two core feature peaks and surrounding related bands are retained.

[0059] Density clustering algorithm is used to cluster similar spectral bands. The clustering parameters need to be adjusted and optimized multiple times. The neighborhood radius is set to 0.05μm (based on wavelength step size and characteristic peak width), the minimum number of samples is set to 8 (corresponding to 8 consecutive wavelength points), and Euclidean distance is used as the distance metric to calculate the similarity of spectral intensity of different bands.

[0060] During the clustering process, all wavelength points are traversed, and the number of samples in the neighborhood of each point is marked. Points with a neighborhood sample number greater than or equal to 8 are defined as core points. Clusters are expanded around the core points, and adjacent core points and sample points with reachable density are grouped into the same cluster. Finally, the characteristic peaks are divided into two similar band clusters: the first cluster corresponds to 10.4-12.6μm (covering the first characteristic peak and surrounding bands), and the second cluster corresponds to 14.9-16.6μm (covering the second characteristic peak and surrounding bands), thus obtaining the second spectral feature set.

[0061] The second spectral feature set contains only the wavelength-reflection intensity sequences of the two high signal-to-noise ratio core feature bands of the first and second clusters, and does not include the noise segment or weak signal segment in the original spectrum.

[0062] The first spectral feature set has a dimension of 2251×4 (2251 full-band wavelength points and 4 acquisition angles). The rows represent wavelength points, the columns represent acquisition angles, and the cells represent the normalized reflection intensity of the corresponding wavelength and angle. The second spectral feature set has a dimension of 401×4 (221 wavelength points in the first cluster, 180 wavelength points in the second cluster, and 4 acquisition angles). The rows represent wavelength points within the feature bands, the columns represent acquisition angles, and the cells represent the normalized reflection intensity of the corresponding wavelength and angle.

[0063] Based on the second spectral feature set, the correlation matrix R between the spectral absorption coefficient α and the reflection intensity I is calculated. The spectral absorption coefficient α is calculated according to the Lambert-Beer law, and the relationship between α and the reflection intensity I is as follows: Where d is the thickness of the polyester filament sample. Given the incident light intensity, the absorption coefficient α for each wavelength is calculated point-by-point, forming an absorption coefficient vector and a reflection intensity vector. A correlation matrix is ​​then constructed, calculated using the following formula: The sample size n=50 (50 wavelength points are evenly selected from the first and second feature clusters as samples), and the values ​​of i and j are both in the range of 1-50, corresponding to the absorption coefficient dimension and reflection intensity dimension of the 50 sample wavelength points, respectively. The correlation coefficient between the i-th absorption coefficient dimension and the j-th reflection intensity dimension is denoted as , and its value ranges from [-1, 1]. The closer the absolute value is to 1, the stronger the correlation. Let be the absorption coefficient of the k-th sample in the i-th dimension. Let be the mean of the absorption coefficients in the i-th dimension. Let be the reflection intensity of the k-th sample in the j-th dimension. Let be the mean value of the reflection intensity in the j-th dimension.

[0064] A three-dimensional simulation model of the spectral reflectance of polyester filament was constructed. The model was meshed to simulate the spectral energy distribution of far-infrared light within the polyester filament, and the coordinates of the energy peak points were extracted to obtain the spectral energy distribution map. The specific implementation is as follows: A three-dimensional simulation model was constructed based on the actual parameters of polyester filament to ensure a high degree of consistency between the model and the actual object. The diameter of the polyester filament was measured using a laser diameter gauge, and the average value of multiple measurements was obtained as 0.1 mm. The fineness was measured by weighing, and a 1000 m length of polyester filament was weighed to calculate a fineness of 150D. The cross-sectional shape was confirmed to be circular with no obvious distortion by microscopic observation. ANSYS 2023 R1 software was used as the simulation tool. A new three-dimensional modeling project was created, and the optical module was selected to simulate the propagation, refraction, reflection, and energy distribution of light. The three-dimensional model of the polyester filament was drawn according to the measured parameters: a cylinder with a radius of 0.05 mm and a length of 10 mm was drawn with the Z-axis as the axis of the polyester filament to simulate the actual shape of the polyester filament. After the model was drawn, it was meshed. Considering the balance between simulation accuracy and computational efficiency, a hexahedral structured mesh was used, with the mesh element size set to 0.005mm × 0.005mm × 0.01mm (radial × circumferential × axial). The mesh was refined in the central region of the polyester filament cross-section (the potential energy concentration region), reducing the element size to 0.002mm × 0.002mm × 0.01mm, generating a first mesh model containing approximately 1.2 million elements and 1.5 million nodes. After meshing, the mesh quality was checked using ANSYS's built-in quality assessment tool to ensure that the mesh distortion rate was less than 5%, the aspect ratio was less than 10, and there were no negative volume elements. Only after the mesh quality met the standards could subsequent energy field simulations be performed to avoid simulation results distortion due to mesh quality issues.

[0065] The production process parameters for polyester filament were obtained by retrieving the corresponding batch's process records from the MES system in the production workshop. Core parameters included: spinning temperature 285℃, cooling air velocity 0.8 m / s, and stretching ratio 3.2 times. Auxiliary parameters such as raw material batch and cooling distance were also recorded as boundary conditions for energy field calculations. A Monte Carlo light transmission algorithm was employed as the core, combined with Lambert-Beer's law, to construct a three-dimensional energy field calculation model to simulate the refraction, reflection, scattering, and energy attenuation of far-infrared light within the polyester filament. Incident light conditions were set as follows: the incident light type was parallel far-infrared light, the incident direction was perpendicular to the polyester filament axis (Z-axis direction), the incident light intensity was normalized to 1, and the wavelength covered 2.5-25 μm, consistent with the spectral acquisition band. The production process parameters are transformed into simulation boundary conditions: spinning temperature affects the crystallinity of polyester yarn, which in turn affects optical properties. A crystallinity of 45% (calculated from spinning temperature) is substituted into the refractive index model to correct the refractive index distribution in different regions. Cooling wind speed affects the temperature gradient of polyester yarn. A corresponding convective heat transfer boundary is applied to the model surface to simulate the actual cooling effect. The stretching ratio affects the orientation of polyester yarn. An orientation of 82% corresponds to the optical anisotropy parameter, which is set to have an axial refractive index slightly higher than the radial refractive index. In this embodiment, the energy field calculation process is as follows: 10 randomly generated... 6 Each ray of light is traced along its propagation path within the 3D model of the polyester filament. For each segment of the propagation path of each ray, the intensity attenuation is calculated using the Lambert-Beer law. The energy deposition of all rays within each grid cell is statistically analyzed to obtain the far-infrared light energy value of each cell in the model, forming the first energy distribution set. The first energy distribution set includes information such as the coordinates, energy value, and intensity attenuation of each grid cell, reflecting the propagation and energy attenuation patterns of far-infrared light within the polyester filament.

[0066] Since the first energy distribution contains some high-frequency fluctuations (caused by grid discretization and boundary condition iteration), a neighborhood-weighted Lorentz filter is used for smoothing. The neighborhood-weighted Lorentz filter exhibits excellent smoothing effect on peak signals, preserving energy peaks while eliminating fluctuation interference. The filtering formula is: in This is the filtered energy value. For the 26 neighboring grid cells within the current 3×3×3 grid cell, Let be the original energy value of the nth neighboring unit. The value is the energy center value (take the energy mean of the first energy distribution set, 0.6). The half-width at half-maximum (WHM) parameter is set to 0.08. A cell-by-cell filtering calculation is performed, focusing on smoothing energy values ​​exceeding a preset energy threshold (0.5, corresponding to the effective energy region), while appropriately reducing the filtering intensity in regions below the threshold (weak energy edge regions), resulting in a second energy distribution set. The smoothing effect is verified by the energy standard deviation. The energy standard deviation values ​​are statistical results from ANSYS simulation post-processing: before filtering, the energy values ​​of several mesh elements in the entire model were extracted, and the calculated standard deviation was 0.12; after Lorentz smoothing, the energy values ​​of the entire model mesh were re-statistically calculated, and the calculated standard deviation was 0.05, indicating significant suppression of fluctuations. The coordinates of energy peak points were extracted, and peak value determination criteria were set: the energy value of a certain unit is greater than 1.2 times the energy values ​​of its 26 adjacent units, and greater than a preset peak value threshold of 0.65. Based on this, three energy peak points were selected, with coordinates of (0mm, 0mm, 2.5mm), (0mm, 0mm, 5.0mm), and (0mm, 0mm, 7.5mm), respectively, concentrated in the central region of the polyester filament cross-section (coordinate range: x∈[-0.01, 0.01]mm, y∈[-0.01, 0.01]mm, z∈[0, 10]mm). Based on these peak points, the region was divided. With the peak points as the center, continuous units with energy values ​​greater than 0.6 were divided into energy concentration areas, and the area of ​​this region was calculated to be approximately 0.001mm². 2 The spectral energy distribution map is generated in the form of a cloud map using the ANSYS post-processing module.

[0067] Based on the high-dimensional spectral feature vector, the abnormal spectral reflectance intervals are located, and the energy concentration regions are separated from the spectral energy distribution map. The pre-acquired polyester filament production process parameters are then matched with the abnormal intervals and energy concentration regions to obtain the predicted matching index. The specific implementation is as follows: Feature decoding is performed on high-dimensional spectral feature vectors to reconstruct the corresponding spectral physical information and locate the abnormal spectral reflectance intervals. A spectral feature vector inverse mapping algorithm is used to map the 4-dimensional high-dimensional spectral feature vectors (peak position, intensity, full width at half maximum (FWHM), and peak area) back to the original spectral space. The specific steps are as follows: input the 4-dimensional feature vector; extract the peak position and FWHM parameters for each feature peak; determine the wavelength interval of the feature peak with the peak position as the center and twice the FWHM as the interval span; combine the intensity and peak area parameters to reconstruct the reflectance intensity distribution within this interval; output the coordinates and span of the abnormal interval matching the original spectral space. Through the spectral feature vector inverse mapping algorithm, the wavelength intervals of each feature spectral peak are extracted: the first feature peak's wavelength interval is 10.5-12.5 μm, with the midpoint coordinates at (11.5 μm, 0.5) (the horizontal axis represents wavelength, and the vertical axis represents normalized reflectance intensity). The second feature peak's wavelength interval is 15.0-16.5 μm, with the midpoint coordinates at (15.75 μm, 0.5). The full width at half maximum (FWHM) of each characteristic peak was calculated as the span of the abnormal interval. The FWHM calculation method is as follows: find the two wavelength points corresponding to half the peak value of the characteristic peak, and the distance between the two points is the FWHM. The calculated FWHM of the first characteristic peak is 0.8 μm, and that of the second characteristic peak is 0.6 μm. To verify the rationality of the abnormal interval, the calculated interval was compared with the normal spectral interval of polyester filament in the standard spectral library. The standard interval FWHM range is 0.5-0.7 μm. Therefore, it was determined that the first characteristic peak (FWHM 0.8 μm) has an abnormal broadening, corresponding to an abnormal reflectance interval, while the second characteristic peak interval is normal. At the same time, the reflection intensity fluctuation range of the abnormal interval was recorded. The reflection intensity fluctuation of the first characteristic peak in the abnormal interval is 0.45-0.62, which exceeds the normal fluctuation range (0.50-0.58), further confirming the validity of the abnormal interval.

[0068] Based on the Sobel edge detection algorithm, the boundaries of energy concentration regions are extracted from the spectral energy distribution map to clarify the spatial range of these regions. The Sobel algorithm identifies edges by calculating the gradient values ​​of pixels. With a gradient threshold of 0.03, edge detection is performed on the spectral energy distribution map: horizontal and vertical Sobel convolution kernels are constructed and convolved on the energy distribution map to obtain horizontal and vertical gradient maps. The gradient magnitude of each pixel is calculated; points with magnitudes greater than the threshold are identified as edge points, while those less than the threshold are identified as internal or background points. Finally, edge points are connected and smoothed to remove isolated edge points, forming continuous boundary contours of energy concentration regions. This generates a dataset of concentrated regions including boundary point coordinates and corresponding energy values. This dataset contains 200 data points with energy values ​​ranging from 0.52 to 0.78, and the boundary coordinates cover the region x=0.04-0.06mm, y=0.04-0.06mm, and z=0-0.1mm, consistent with the extracted peak point concentration region. Meanwhile, outlier detection was performed on the dataset using the Grubbs criterion, removing three outlier data points that deviated from the mean by ±2 standard deviations.

[0069] Complete process parameter attributes for the corresponding batch were retrieved from the production process database. In addition to the core parameters of spinning temperature, cooling air velocity, and stretching ratio, auxiliary parameters such as raw material batch, screw speed (350 r / min), and cooling distance (50 cm) were added, resulting in four key correlation parameters selected for matching analysis. A rule base for the correlation between process parameters and spectral anomalies was established, clarifying the influence weight of each parameter on spectral characteristics. Spinning temperature directly affects the crystallinity of polyester filament, with a weight coefficient ω1=0.4. Cooling air velocity affects crystallization uniformity, with a weight coefficient ω2=0.25. Stretching ratio affects orientation, with a weight coefficient ω3=0.15. Raw material batch affects molecular chain structure, with a weight coefficient ω4=0.2. The total weight coefficient ω=ω1+ω2+ω3+ω4=1.0. Logical correlation verification was performed between the process parameter attributes and the coordinates of the anomaly interval and the data set of the concentrated area, determining whether each parameter was within the anomaly influence range: the spinning temperature of 285℃ exceeded the standard range (275-282℃), indicating a successful match. Cooling air velocity of 0.8 m / s is below the standard value (0.9-1.1 m / s), so the match is successful. The stretching ratio of 3.2 times is within the standard range (3.0-3.4 times), but considering the characteristics of the energy concentration area, it is determined to be indirectly related, so the match is successful. There are no abnormal records in the raw material batch, so the match fails. The final count of successfully matched parameters is as follows. =3, total number of parameters =4, substitute into the formula for calculating the first matching degree index The calculated value is M1 = (3 / 4) × 0.8 = 0.6. This index initially reflects the correlation between process parameters and spectral anomalies.

[0070] To improve the accuracy of the matching indicators, corrections need to be made based on the abnormal range span to eliminate the interference of abnormal span on the correlation determination. A preset span threshold S0 = 0.7 μm is calibrated, representing the upper limit of the full width at half maximum (FWHM) of the normal spectral peak of polyester filament. The acceptable limit of deviation for polyester filament is defined as the maximum permissible span. =1.0μm, corresponding to the acceptable limit deviation range for polyester filament; exceeding this range is considered a serious anomaly. In this test, the first characteristic peak's half-width at half-maximum (WHM) of 0.8μm exceeded the preset threshold S0, while the second characteristic peak's WHM of 0.6μm was within the normal range; correction was only applied to the abnormal span. Substituting into the correction formula... ,in The first matching degree index is used, and S represents the actual anomaly interval span of 0.8 μm. To preset the span threshold, To determine the maximum allowable span, the correction coefficient was calculated to be 1 - (0.8 - 0.7) / 1.0 = 0.9, and the final predicted matching index M = 0.6 × 0.9 = 0.54. To verify the rationality of this index, it was compared with similar cases in the historical database, confirming that the 0.4-0.7 range corresponds to a moderate correlation, indicating that the current reflectivity anomaly is mainly caused by deviations in process parameters.

[0071] If the predicted matching index meets the preset matching degree threshold range, then the spectral shift type is analyzed. Combining the spectral shift type and the production process parameter attributes, the reflectivity defect classification standard is obtained. The specific implementation is as follows: Based on the production process parameters, the crystallinity of this batch of polyester yarn was measured to be 45% using differential scanning calorimetry, and the orientation degree was measured to be 82% using X-ray diffraction. The crystallinity is lower than the standard value (48-55%), while the orientation degree is within the normal range (80-85%). Insufficient crystallinity is the core material factor causing the abnormal spectral characteristics. A preset matching threshold range of 0.4-0.7 was set, and the predicted matching index of 0.54 falls within this range, indicating a valid correlation between process parameters and spectral anomalies, allowing for further analysis of the spectral shift type. A gradient boosting tree algorithm was used to determine the shift type. This algorithm has strong nonlinear fitting capabilities and can accurately capture the correlation between crystallinity, orientation degree, energy value, and spectral shift. The algorithm parameters underwent multiple adjustments and optimizations: the tree depth was set to 5 to avoid overfitting; the learning rate was set to 0.1 to control the contribution weight of each tree; the minimum number of samples per leaf node was set to 20 to improve the model's generalization ability; and the number of iterations was set to 100 to ensure model convergence. Energy values, crystallinity, orientation, and process parameters of the energy concentration region were used as input features to construct a training set and a test set (ratio 7:3). The training set was based on 300 historical spectral shift cases. Through model inference, the spectral shift type was determined to be redshift, i.e., the characteristic peak shifted towards longer wavelengths. The shift was measured using the peak wavelength comparison method: the normal peak wavelength is 11.3 μm, the measured peak wavelength is 11.5 μm, and the calculated shift is 0.2 μm, which exceeds the slight shift threshold (0.1 μm). Further analysis of the cause is needed to correlate with process parameters.

[0072] A Bayesian classification algorithm was used for association analysis. The core of this approach is to calculate the posterior probability based on prior probability, quantifying the correlation between process parameter combinations and redshift defects. Based on historical production data, event A was defined as "using a specific combination of process parameters in production: spinning temperature 280-290℃, cooling air velocity 0.7-0.9 m / s, and drawing ratio 3.0-3.4 times," with a prior probability P(A) = 0.09. Event B was defined as "occurrence of a redshift defect." Statistical analysis of historical data showed that, under the condition of using process parameter combination A, the probability (likelihood probability) of a redshift defect occurring was P(B|A) = 0.75. The total probability (marginal probability) of a redshift defect occurring was calculated based on the law of total probability as P(B) = 0.42. Using Bayes' theorem P(A|B)=[P(B|A)×P(A)] / P(B), the posterior probability P(A|B)≈0.16 of the redshift defect caused by process parameter combination A was calculated under the condition that a redshift defect has already occurred. After correction based on actual detection data, the statistical occurrence rate of redshift defects under this combination reached 68%. Further analysis revealed that this occurrence rate is highly correlated with insufficient crystallinity: excessively high spinning temperature leads to insufficient molecular chain crystallization, and insufficient cooling air velocity exacerbates uneven crystallization. The combined effect of these two factors alters the absorption characteristics of polyester filament for far-infrared light, causing a spectral redshift. Based on the above analysis, a classification standard for reflectivity defects was generated, adding a new type of redshift-process parameter deviation-insufficient crystallinity, with clearly defined criteria: spectral peak shift to long wavelengths ≥0.1μm, crystallinity <48%, and corresponding spinning temperature exceeding the standard range by 2-8℃ and cooling air velocity lower than the standard value by 0.1-0.3m / s. Additionally, the defect severity classification is supplemented: an offset of 0.1-0.2μm indicates a minor defect, 0.2-0.3μm indicates a moderate defect, and >0.3μm indicates a severe defect. In this test, an offset of 0.2μm corresponds to a minor defect.

[0073] A dataset of reflectance mode parameters is constructed, and a spectral reflectance stability index is calculated based on the Monte Carlo method. When the stability index exceeds a preset stability threshold, the feature overlap parameter is extracted, and the reflectance anomaly level is determined by inputting it into a random forest algorithm. The specific implementation is as follows: Historical data on the spectral reflectance patterns of polyester filaments from the past three years were retrieved from a spectral feature database, including spectral data under different process parameters, raw material batches, and environmental conditions. Based on the previously generated reflectance defect classification criteria, a feature matching algorithm was used to filter redshift defect data: the characteristic peak positions, offsets, and crystallinity parameters of each historical data set were extracted and compared with the definition conditions in the defect classification criteria, resulting in 320 redshift defect data sets. This formed a classified defect type dataset, including 210 mild defects, 85 moderate defects, and 25 severe defects. Mild defects, the same type as those in this test case, accounted for 65.6%. For this dataset, the spectral wavelength λ (covering the 10.5-12.5μm characteristic band) and corresponding energy value E were extracted for each data set. The finite difference method was used to simulate the correspondence between the two. This method can accurately calculate the rate of change of energy value with wavelength, reflecting the changing law of spectral characteristics. The core formula is... where the wavelength step size =0.02μm, based on the wavelength accuracy setting of the spectral acquisition to ensure calculation accuracy. The spectral-energy relationship dataset was obtained by point-by-point calculation. Through data statistical analysis, it was found that in the redshift defect scenario, the energy value decreased by an average of 0.03 for every 0.1μm increase in wavelength, and the rate of decrease accelerated with the increase of offset. When the offset exceeded 0.2μm, the rate of decrease in energy value increased to 0.05 / 0.1μm.

[0074] A preset energy threshold of 0.5 was set, which is the lower limit of the effective energy range. Energy data below this value was considered invalid interference. Data source tracing was performed on the spectral-energy relationship dataset to track the origin characteristics of redshift defects. The source tracing process adopted a reverse reasoning method, which reversed the correlation between spectral anomaly data and parameters of the entire production process to extract key origin features: process parameter anomalies included spinning temperature 5-10℃ higher, cooling wind speed 0.1-0.2m / s lower than the standard value, and stretching ratio fluctuation ±0.2 times. Raw material performance deviations included chip moisture content exceeding the standard by 0.05-0.1% and uneven molecular weight distribution of chips. Environmental factors included humidity in the spinning workshop exceeding the 45-55%RH range. A hierarchical clustering algorithm was used to classify the origin features and construct a cluster tree: based on feature similarity, a similarity threshold of 0.7 was set. Features related to process parameter anomalies were grouped into one category, features related to raw material performance deviations were grouped into another category, and environmental factors were temporarily merged into the raw material performance deviation category because the sample proportion was less than 5%. The final results yielded two feature sets: abnormal process parameters and deviations in raw material performance. The abnormal process parameters accounted for 75% of the set, while the deviations in raw material performance accounted for 25%, clearly indicating that abnormal process parameters are the primary cause of redshift defects. To verify the accuracy of the clustering results, the silhouette coefficient method was used for evaluation. A silhouette coefficient of 0.82 indicates good clustering performance and clear feature classification.

[0075] For a dataset of defect origin features, a Hidden Markov Model (HMM) is used to calculate the probability of redshift defects caused by various defect features. This model can effectively handle probability prediction of sequence data and adapt to the temporal characteristics of parameter changes in the production process. Model construction includes three steps: state definition, observation sequence definition, and parameter training. State definitions include abnormal process parameters, raw material performance deviations, and two types of hidden states. Observation sequences include observable features such as spectral shift, energy value change rate, and crystallinity. Model parameters include initial state probability, state transition probability, and observation probability. A time step of T=10 is set, corresponding to 10 key nodes in the production process (slice drying, slice melting, spinning, cooling and forming, pre-stretching, heat setting, drawing and texturing, winding and forming, balancing, and finished product inspection). The forward probability α(T) and backward probability β(T) for each time step are calculated: the forward probability represents the probability of being in state T from the initial state to step t and generating the first t observation sequences. The backward probability represents the probability of being in state T from step t and generating the remaining observation sequences. Substituting these into the probability calculation formula... The probabilities of occurrence for two types of hidden states were calculated: the probability of redshift defects caused by abnormal process parameters was 0.72, and the probability of defects caused by raw material performance deviations was 0.28. By integrating hidden state probabilities, origin feature data, spectral-energy relationship data, and process parameter information, a reflection mode parameter dataset was constructed. This dataset includes 12 feature dimensions, providing comprehensive input data for subsequent stability analysis and anomaly level determination.

[0076] Based on the reflection mode parameter dataset, factor analysis was used to extract core factors, reducing data dimensionality while retaining key information. The dataset was then standardized to eliminate differences in the magnitude of different feature dimensions; the standardization formula is as follows: Where μ is the characteristic mean and σ is the characteristic standard deviation. A feature matrix is ​​constructed by screening highly correlated features through correlation analysis. Principal component analysis is used to extract core factors. Based on the principle that the eigenvalue is greater than 1, two core factors are identified: a process parameter influence factor (variance contribution rate of 62%), including parameters such as spinning temperature, cooling wind speed, and stretching ratio; and an energy distribution factor (variance contribution rate of 28%), including energy value, energy change rate, peak coordinates, etc., with a cumulative variance contribution rate of 90%, which can fully characterize the original data information, forming a spectral feature dataset. Based on this dataset, the Monte Carlo method is used to calculate the spectral reflectance stability index S. This method simulates the impact of uncertainties on the results through a large number of random samplings, improving the reliability of the index. The sampling number m = 1000 is set, and each sampling randomly selects 80% of the data from the spectral feature dataset as a sample, calculating the corresponding reflectance value R. The mean reflectance is obtained through statistical analysis. Substitute into the stability index calculation formula The calculated sample standard deviation is 0.018, and the stability index S = 1 - (0.018 / 0.82) ≈ 0.78. The closer this index is to 1, the more stable the spectral reflectance is.

[0077] Since the temperature and humidity of the testing environment have a slight impact on spectral reflectance, boundary conditions need to be applied to the stability index to obtain the calibration stability index S'. The standard environmental conditions for polyester filament spectral testing are defined as follows: standard temperature T0 = 25℃, standard humidity H0 = 50%RH, operating ambient temperature range [25℃, 80℃], humidity range [30%, 70%], and results outside these ranges are invalid. In this test, the actual ambient temperature T = 25℃ and humidity H = 50%RH were under standard conditions, but calibration calculations were still performed to ensure universality. Temperature correction factor k. T =0.002 / ℃, meaning that for every 1℃ deviation of the temperature from the standard value, the stability index changes by 0.002. Humidity correction factor k H =0.001 / %RH, meaning that for every 1% RH deviation from the standard humidity, the stability index changes by 0.001. Substituting this into the calibration formula... The calculated value is S' = 0.78 × [1 + 0.002 × (25 - 25) + 0.001 × (50 - 50)] = 0.78. If the actual environment deviates from the standard conditions, calibration according to this formula is required to ensure the accuracy of the stability index. A preset stability threshold of 0.7 is set. The calibration stability index of 0.78 exceeds the threshold, indicating that although there is an anomaly in the spectral reflectance, the stability is good, and the feature overlap parameter can be further extracted. Using the mean-shift clustering algorithm, with the spectral feature dataset and energy distribution data as input, the feature overlap parameters of the reflectance mode data and energy distribution data are extracted, including three dimensions: spatial overlap rate, intensity consistency, and wavelength matching degree. The calculated spatial overlap rate is 82%, directional consistency is 88%, and wavelength matching degree is 79%, forming an overlap index dataset. The spectral feature dataset and overlap index dataset were input into a random forest algorithm. The number of decision trees was set to 50, the tree depth to 6, and the minimum number of sample splits to 10. Through model training and inference, and combined with a pre-defined three-level anomaly classification standard, the reflectance anomaly level was determined: mild anomaly (stability index ≥ 0.8, redshift offset 0.1-0.15 μm, measured reflectance 0.80-0.85), moderate anomaly (stability index 0.7-0.8, redshift offset 0.15-0.25 μm, measured reflectance 0.75-0.80), and severe anomaly (stability index < 0.7, redshift offset > 0.25 μm, measured reflectance < 0.75). In this test, the stability index (0.78), redshift offset (0.2 μm), and measured reflectance (0.78) all fell within the moderate anomaly range. Therefore, it was ultimately classified as a moderate anomaly, and the corresponding defect handling priority was medium.

[0078] When the reflectance anomaly level exceeds the preset anomaly threshold, the reflectance mode parameter dataset and the reflectance anomaly level are integrated to generate a spectral reflectance detection report. The specific implementation is as follows: The system integrates all detection data, including reflectance mode parameter datasets, reflectance anomaly levels, anomaly interval coordinates (10.5-12.5μm, half-width at half-maximum 0.8μm), and energy concentration region information (x=0.04-0.06mm, y=0.04-0.06mm, z=0-0.1mm, area 0.001mm). 2 Based on the matching results of production process parameters (three parameters showing correlation anomalies), a data support system for the inspection report is constructed. The probability P of spectral reflectance compliance is calculated using a multi-defect type joint probability calculation method, the formula is as follows: The defect types were identified as k=2: redshift defects caused by abnormal process parameters (probability P1=0.72) and redshift defects caused by raw material performance deviations (probability P2=0.28). These two types of defects are independent. Substituting these values ​​into the formula, we get P=(1-0.72)×(1-0.28)=0.28×0.72=0.2016, approximately 20.2%, indicating a high risk of non-compliance in the spectral reflectance of the current batch of polyester filament, requiring targeted process optimization. The calculation process for the measured reflectance value was also supplemented: based on enhanced spectral data, the average reflectance intensity in the 10.5-12.5μm band was taken, and combined with the calibration coefficient, the measured reflectance was calculated to be 0.78. The standard value range was set at 0.85-0.95 based on industry standards and historical qualified data. The measured value is lower than the lower limit of the standard, further verifying the anomaly judgment results.

[0079] According to the standardized test report format, systematically sort out the test results and analysis conclusions, and clarify the core content of each module. Mark the basic test information at the beginning of the report, including sample number, test date, test equipment model, environmental conditions, etc., to ensure traceability. In the core test results section, clearly mark the measured value of spectral reflectance as 0.78 (standard value range 0.85 - 0.95), abnormal level (medium abnormality), and pass probability of 20.2%. Attach enhanced spectral diagrams and spectral energy distribution diagrams as evidence. In the determination basis section, elaborate: the redshift offset is 0.2 μm, exceeding the slight offset threshold of 0.1 μm. The stability index is 0.78. Although it exceeds the preset threshold of 0.7, the measured reflectance is lower than the standard value, and the comprehensive determination is medium abnormality. The process parameter matching degree is 0.54, and it is confirmed that the abnormality is directly related to the too high spinning temperature and insufficient cooling air speed. In the process optimization suggestion section, combined with the test results and historical optimization cases, put forward targeted measures: lower the spinning temperature from 285°C to 280°C, and adopt a segmented temperature control strategy (282°C in the front section, 280°C in the middle section, 278°C in the rear section) to improve the crystallization uniformity. Increase the cooling air speed from 0.8 m / s to 0.9 m / s, adjust the cooling air angle to 45°, and enhance the cooling effect. At the same time, increase the drying time of the chips and control the moisture content below 0.03% to avoid interference from raw material factors. It is recommended to re-sample and test after optimization, compare the change in reflectance. If the measured value increases above 0.85 and the redshift offset ≤ 0.1 μm, it is determined to be qualified.

[0080] In summary, this embodiment provides a method for detecting the spectral reflectance of polyester filaments based on far-infrared. Through multi-angle spectral acquisition and multi-step preprocessing, it effectively eliminates environmental interference and the influence of sample morphology, greatly improves the quality of spectral data, ensures the accuracy of reflectance detection, can capture subtle spectral characteristic abnormalities, and avoids omission of potential defects. With the help of three-dimensional simulation technology, it intuitively presents the energy distribution law of far-infrared light inside the polyester filaments, realizes the precise positioning of abnormal intervals, and provides visual support for defect analysis. Through algorithm model construction and data correlation analysis, it establishes a quantitative relationship between spectral abnormalities and production process parameters, realizes the full-chain traceability from spectral abnormalities to defect causes, breaks the traditional empirical process adjustment mode, and improves the pertinence and efficiency of process optimization. At the same time, this method generates a standardized defect classification standard and implementable process optimization suggestions, forms a complete detection - analysis - optimization closed loop, adapts to the quality control scenario of large-scale polyester filament production, and can be adapted to the detection of other synthetic fibers through parameter adjustment, with strong versatility.

[0081] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0082] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting the spectral reflectance of polyester filament based on far-infrared radiation, characterized in that, include: Raw far-infrared spectral data of polyester filament samples were collected and preprocessed to obtain enhanced spectral data; Feature bands are extracted from the enhanced spectral data, and the correlation matrix between the spectral absorption coefficient and the reflection intensity is calculated to obtain a high-dimensional spectral feature vector. A three-dimensional simulation model of the spectral reflectance of polyester filament was constructed. The model was meshed to simulate the spectral energy distribution of far-infrared light inside the polyester filament, and the coordinates of the energy peak points were extracted to obtain the spectral energy distribution map. Based on the high-dimensional spectral feature vector, the abnormal spectral reflectance interval is located, and the energy concentration region is separated from the spectral energy distribution map. The matching degree of the pre-acquired polyester filament production process parameters is verified with the abnormal interval and the energy concentration region to obtain the predicted matching index. The process of obtaining the predictive matching index includes: The high-dimensional spectral feature vector is decoded to extract the midpoint coordinates of the wavelength range of the feature spectral peak as the center coordinates of the anomaly range, and the half-width at half-maximum of the spectral peak is calculated as the span of the anomaly range. Based on the Sobel edge detection algorithm, the boundary of the energy concentration region is extracted from the spectral energy distribution map to generate a concentration region data set containing coordinates and energy values; The production process parameter attributes are obtained from the pre-established production process database. The production process parameter attributes are logically associated with the abnormal interval coordinates and the concentrated area data set to calculate the first matching degree index. When the span of the abnormal interval exceeds the preset span threshold, the first matching degree index is corrected according to the preset index adjustment rule to obtain the predicted matching index. If the predicted matching index meets the preset matching degree threshold range, then the spectral shift type is analyzed, and the reflectivity defect classification standard is obtained by combining the spectral shift type and the production process parameter attribute. A dataset of reflectance mode parameters is constructed, and a spectral reflectance stability index is calculated based on the Monte Carlo method. When the stability index exceeds a preset stability threshold, the feature overlap parameter is extracted and input into a random forest algorithm to determine the reflectance anomaly level. The process of determining the level of reflectivity anomaly includes: Based on the reflection mode parameter dataset, the core factors of the reflection mode data and energy distribution data are extracted using factor analysis to form a spectral feature dataset. For the aforementioned spectral feature dataset, multiple random sampling simulations were performed using the Monte Carlo method to calculate the spectral reflectance stability index. By applying boundary condition constraints to the stability index, a calibration stability index is obtained; When the calibration stability index exceeds the preset stability threshold, the mean drift clustering algorithm is used to extract the feature overlap parameters of the reflection mode data and the energy distribution data to obtain the overlap index dataset. Using the spectral feature dataset and the overlap index dataset as input, the random forest algorithm is used to classify the risk of reflectivity anomalies and determine the reflectivity anomaly level. When the reflectance anomaly level is higher than a preset anomaly threshold, the reflectance mode parameter dataset and the reflectance anomaly level are integrated to generate a spectral reflectance detection report.

2. The method for detecting the spectral reflectance of polyester filament based on far-infrared radiation according to claim 1, characterized in that, The process of obtaining enhanced spectral data includes: Using a Fourier transform infrared spectrometer, raw spectral data of polyester filament samples were collected from multiple angles within the far-infrared band. The raw spectral data included raw spectral signal data of the polyester filament's reflection, absorption, and transmission characteristics of far-infrared light at different wavelengths. A polynomial fitting algorithm is used to perform baseline correction on the raw spectral data to eliminate the effects of baseline drift. An adaptive Wiener filtering algorithm is used to filter noise from the corrected spectral data while retaining the effective spectral signal. The filtered spectral data were normalized using the max-min normalization method to obtain enhanced spectral data.

3. The method for detecting the spectral reflectance of polyester filament based on far-infrared radiation according to claim 1, characterized in that, The process of obtaining high-dimensional spectral eigenvectors includes: The enhanced spectral data is subjected to wavelet transform algorithm to extract feature bands, highlighting the boundary characteristics of feature spectral peaks, and a first spectral feature set is obtained; The characteristic peak intensity value is obtained from the first spectral feature set. When the characteristic peak intensity value is greater than the preset intensity threshold, the density clustering algorithm is used to cluster similar spectral bands to obtain the second spectral feature set. Based on the second spectral feature set, the correlation matrix between the spectral absorption coefficient and the reflection intensity is calculated; The high-dimensional spectral feature vector is generated by using the eigenvalues ​​of the correlation matrix as weights.

4. The method for detecting the spectral reflectance of polyester filament based on far-infrared radiation according to claim 1, characterized in that, The process of obtaining the spectral energy distribution map includes: A three-dimensional simulation model was constructed based on the diameter, fineness, and cross-sectional shape parameters of polyester filament. The model was then divided into hexahedral meshes to generate the first mesh model. Obtain the production process parameters corresponding to polyester filament, including spinning temperature, cooling air velocity, and stretching ratio; Based on the first grid model and the production process parameters, the initial energy field is calculated using the law of light propagation to obtain the first energy distribution set. Based on the Lorentz filter, the energy values ​​in the first energy distribution set that exceed a preset energy threshold are smoothed to obtain the second energy distribution set. For the second energy distribution set, the coordinates of the energy peak points are extracted, and the regions are divided according to the coordinates of the energy peak points to form the spectral energy distribution map.

5. The method for detecting the spectral reflectance of polyester filament based on far-infrared radiation according to claim 1, characterized in that, The process of obtaining the classification criteria for reflectivity defects includes: Based on the aforementioned production process parameter attributes, the crystallinity and orientation parameters of the polyester filament are obtained; When the predicted matching index meets the preset matching degree threshold range, the crystallinity and orientation parameters are combined, and the gradient boosting tree algorithm is used to determine the type of spectral shift caused by the energy value data of the energy concentration region in the abnormal interval, and the shift determination result is obtained. A Bayesian classification algorithm is used to correlate the offset determination results with the production process parameter attributes, and the occurrence pattern of reflectivity defects under the same combination of process parameters in the offset type is statistically analyzed to generate the reflectivity defect classification standard.

6. The method for detecting the spectral reflectance of polyester filament based on far-infrared radiation according to claim 1, characterized in that, The process of constructing the reflection mode parameter dataset includes: Historical spectral reflectance pattern data are obtained from a pre-established spectral feature database. Based on the reflectance defect classification criteria, the historical spectral reflectance pattern data are classified into defects to obtain a set of classified defect type data. Based on the defect type data set, obtain the spectral wavelength. The spectral wavelength and energy value E are used to simulate the correspondence between the spectral wavelength and energy value for each type of defect using the finite difference method, thus obtaining the spectral-energy relationship dataset. When the energy value exceeds a preset energy threshold, the data source of the spectrum-energy relationship dataset is traced, defect origin features are extracted, and the defect origin features are classified using a hierarchical clustering algorithm to obtain a defect origin feature dataset. For the defect origin feature dataset, a hidden Markov model is used to calculate the defect type probability of various defect features to obtain the reflection mode parameter dataset.

7. The method for detecting the spectral reflectance of polyester filament based on far-infrared radiation according to claim 1, characterized in that, The process of applying boundary condition constraints to the stability index to obtain the calibrated stability index includes: Define the boundary constraints, including the operating temperature and humidity ranges for the polyester filament; Based on boundary constraints, a linear correction model is used to calibrate the original stability index.

8. The method for detecting the spectral reflectance of polyester filament based on far-infrared radiation according to claim 1, characterized in that, The process of generating a spectral reflectance test report includes: The reflection mode parameter dataset, reflectance anomaly level, anomaly interval coordinates, energy concentration area information, and production process parameter matching results are integrated, and the probability of spectral reflectance qualification is calculated. The test report should clearly indicate the measured value of spectral reflectance, the standard value range, the basis for judging the anomaly level, and the process optimization suggestions, thus forming a complete spectral reflectance test report.

Images