PET film orientation detection method based on thermal conductivity gradient

Abstract

The present invention discloses a method for detecting the orientation of PET films based on a thermal conductivity gradient. The method comprises acquiring surface temperature field data of the PET film, obtaining compensated temperature data through time series calibration and ambient temperature compensation, calculating the temperature change rate and spatial temperature gradient based on the compensated temperature data, and generating thermal diffusion characteristic data, performing multi-level decomposition and directional feature extraction on the thermal diffusion characteristic data to obtain comprehensive directional feature data, extracting the dominant direction based on the comprehensive directional feature data and calculating the orientation degree parameter to obtain orientation data, and performing reliability assessment and correction on the entire process data to generate final orientation data. By establishing a multi-level data processing and multi-scale feature analysis system, the present invention achieves high-precision online detection of PET film orientation, with high detection accuracy and good stability, making it particularly suitable for online detection applications on production lines.

Description

Technical Field

[0001] The invention belongs to the field of orientation degree detection, in particular to a PET film orientation degree detection method based on thermal conductivity gradient. Background Art

[0002] The orientation of polyethylene terephthalate (PET) film is a key parameter that determines its mechanical, optical, and thermal stability. Accurately measuring this orientation is crucial for improving product quality and optimizing production processes. During the stretching process, the molecular chains of PET film align, forming a specific orientation structure that directly impacts the product's physical properties. High-end applications, such as high-performance optical films, electronic substrates, and packaging materials, place increasing demands on the uniformity and stability of PET film orientation. Therefore, developing a high-precision, online orientation measurement method is crucial for ensuring product quality, improving production efficiency, and reducing costs.

[0003] Currently, PET film orientation is primarily measured using techniques such as X-ray diffraction, infrared spectroscopy, and acoustic detection. X-ray diffraction characterizes orientation by analyzing the crystal structure, but the equipment is expensive and cannot be tested online. Infrared spectroscopy uses the vibrational absorption characteristics of molecules to assess orientation, but is easily affected by environmental factors and suffers from poor measurement stability. Acoustic detection assesses orientation by analyzing the propagation characteristics of sound waves, but its spatial resolution is limited, making it difficult to accurately measure local orientation. While these traditional methods have some application value in specific scenarios, they generally suffer from low detection efficiency, complex equipment, and high costs, making them difficult to meet the demands of modern PET film production lines for online, rapid, and high-precision testing.

[0004] In practical applications, existing technologies still face the following specific issues: First, during temperature field data acquisition, the nonlinear response characteristics of the sensor and ambient temperature fluctuations can cause distortion in the measured data, affecting the accuracy of the thermal conductivity gradient calculation. Second, when extracting thermal diffusion features, the lack of effective multi-scale analysis methods makes it difficult to accurately identify orientation features at different spatial scales. This leads to poor measurement reliability, especially when the sample has a multi-layer structure or uneven orientation distribution. Third, when calculating the orientation parameter, the traditional single threshold judgment method cannot adapt to complex orientation distribution patterns, prone to misjudgment and omission. Fourth, existing data processing methods lack a comprehensive reliability assessment mechanism, making it difficult to effectively identify and eliminate the influence of abnormal data, resulting in insufficient stability and repeatability of the measurement results. These technical issues have seriously hindered the further development and application of PET film orientation detection technology. Summary of the Invention

[0005] The purpose of the invention is to provide a method for detecting the orientation of PET films based on thermal conductivity gradient, in order to solve at least one technical problem existing in the prior art.

[0006] The technical solution is a method for detecting the orientation of PET films based on thermal conductivity gradient, comprising the following steps:

[0007] S1. Obtain the original temperature field data of the PET film surface, and generate compensated temperature data through time series calibration and ambient temperature compensation;

[0008] S2. Calculate the temperature change rate and spatial temperature gradient based on the compensated temperature data to generate thermal diffusion characteristic data;

[0009] S3, performing multi-level decomposition and directional feature extraction on the thermal diffusion feature data to obtain comprehensive directional feature data;

[0010] S4. Based on the comprehensive directional feature data, extract the dominant direction and calculate the orientation degree parameter to generate orientation degree data;

[0011] S5. Based on the compensated temperature data, thermal diffusion characteristic data, comprehensive directional characteristic data and orientation data, reliability evaluation and correction are performed to generate final orientation data.

[0012] According to one aspect of the present application, step S1 further comprises:

[0013] S11, obtaining original temperature data of the PET film surface and multiplying it by a pre-calibrated temperature calibration coefficient to obtain initial calibration data; resampling and gridding the initial calibration data in time and space dimensions to generate calibrated temperature data;

[0014] S12, segmenting the calibrated temperature data according to the time series and performing smoothing processing to obtain smoothed temperature data; calculating the noise level in the smoothed temperature data to obtain a noise evaluation value; performing noise suppression processing on the smoothed temperature data according to the noise evaluation value to generate filtered temperature data;

[0015] S13. Acquire ambient temperature data during the detection process, analyze the ambient temperature change trend, calculate the temperature fluctuation value, and obtain ambient temperature fluctuation data; subtract the ambient temperature fluctuation data at the corresponding moment from the filtered temperature data to generate compensated temperature data.

[0016] According to one aspect of the present application, step S2 is further:

[0017] S21, dividing the compensated temperature data into time series, calculating the temperature difference between adjacent moments; dividing the temperature difference by the corresponding time interval to generate temperature change rate data;

[0018] S22. Calculate the temperature gradients in the horizontal and vertical directions based on the compensated temperature data to obtain horizontal and vertical secondary temperature characteristic data; combine the horizontal and vertical secondary temperature characteristic data to generate spatial temperature change data;

[0019] S23. Correlation analysis is performed on the temperature change rate data and the spatial temperature change data to obtain initial heat diffusion data; and feature extraction is performed on the initial heat diffusion data at different positions and directions to generate heat diffusion feature data.

[0020] According to one aspect of the present application, step S3 is further:

[0021] S31, dividing the heat diffusion characteristic data into different spatial ranges and performing decomposition operations to obtain hierarchical decomposition data; integrating the hierarchical decomposition data of all regions to generate multi-scale characteristic data;

[0022] S32, performing angle scanning on the multi-scale feature data at each level, calculating energy distribution features and extracting main directional features to generate directional feature data;

[0023] S33. Evaluate the directional feature data according to the preset hierarchical importance to obtain hierarchical weight data; perform weighted calculation on the directional feature data and the corresponding hierarchical weight data to obtain weighted feature data; perform integration calculation on the weighted feature data to generate comprehensive directional feature data.

[0024] According to one aspect of the present application, step S4 is further:

[0025] S41, performing peak analysis and feature intensity sorting on the comprehensive directional feature data to obtain peak sorting data; determining the most significant directional feature based on the peak sorting data to generate main directional feature data;

[0026] S42. Taking the main direction feature data as the reference direction, obtain reference direction data; calculating the distribution characteristics of the comprehensive direction feature data relative to the reference direction data, obtain direction distribution data; calculating the orientation degree parameter based on the direction distribution data, and generate orientation degree data.

[0027] According to one aspect of the present application, step S5 is further:

[0028] S51, collecting compensated temperature data, thermal diffusion characteristic data, comprehensive direction characteristic data and orientation degree data, calculating statistical characteristics and reliability evaluation indicators, and generating reliability indicator data;

[0029] S52, performing correlation analysis on the orientation degree data and the reliability index data to obtain correction coefficient data; performing correction calculation on the orientation degree data and the correction coefficient data to generate final orientation degree data.

[0030] According to one aspect of the present application, step S11 is further as follows:

[0031] S111, obtaining original temperature data of the PET film surface, performing validity check and eliminating abnormal values ​​to obtain pre-processed temperature data; performing data smoothing on the pre-processed temperature data to generate smoothed temperature data;

[0032] S112, reading a pre-calibrated temperature sensor response curve, performing piecewise linearization processing, calculating correction parameters, and obtaining a temperature calibration coefficient;

[0033] S113, matching the smoothed temperature data with the temperature calibration coefficient, performing temperature correction and quantization error compensation, and generating initial calibration data;

[0034] S114, grouping and time-aligning the initial calibration data to obtain first aligned temperature data; resampling the first aligned temperature data according to preset sampling requirements to generate time-calibrated data;

[0035] S115 , converting the time calibration data into spatial coordinate representation, performing spatial uniformity analysis and grid density optimization to obtain grid optimized data; performing spatial reconstruction based on the grid optimized data to generate calibrated temperature data.

[0036] According to one aspect of the present application, step S12 is further as follows:

[0037] S121, analyzing the time series characteristics of the calibrated temperature data, determining the optimal segment length, and generating segmented temperature data;

[0038] S122, extracting local statistical features of the segmented temperature data, calculating smoothing parameters and performing smoothing processing to generate smoothed temperature data;

[0039] S123, performing comparative analysis and statistical analysis on the smoothed temperature data and the segmented temperature data, calculating the noise level, and generating a noise evaluation value;

[0040] S124. Based on the noise evaluation value, construct an adaptive threshold system and perform preliminary noise suppression on the smoothed temperature data to generate preliminary suppression data;

[0041] S125, analyzing the residual noise of the preliminary suppression data, optimizing the suppression parameters and performing fine suppression processing to generate fine suppression data;

[0042] S126 , evaluating the signal quality of the fine suppression data, performing final adjustment and edge feature preservation processing, and generating filtered temperature data.

[0043] According to one aspect of the present application, step S21 is further as follows:

[0044] S211. Based on the compensated temperature data, perform a time continuity check and time series padding to obtain padding temperature data; segment the padding temperature data according to a preset period to generate time series temperature data;

[0045] S212: Analyze sampling characteristics of the time series temperature data, identify key time nodes, and perform sequence alignment to generate second aligned temperature data;

[0046] S213, performing a difference operation on the second aligned temperature data at adjacent moments, detecting and correcting abnormal values, and generating temperature difference data;

[0047] S214, obtaining a timestamp sequence corresponding to the second aligned temperature data, calculating the time interval and performing uniformity analysis to generate interval characteristic data;

[0048] S215, matching the temperature difference data with the interval characteristic data to obtain matched difference data and normalizing the data to generate initial change rate data;

[0049] S216 , performing physical constraint check and change rate correction on the initial change rate data to obtain corrected change rate data; performing final calibration on the corrected change rate data to generate temperature change rate data.

[0050] According to one aspect of the present application, step S33 is further as follows:

[0051] S331. Perform feature consistency analysis and correlation analysis on the directional feature data at each level, evaluate the level stability, and generate level stability data;

[0052] S332. Based on the hierarchical stability data, calculate and normalize the hierarchical discriminant coefficient, and generate hierarchical weight data in combination with the preset importance criterion;

[0053] S333, grouping the directional feature data according to feature type, calculating the consistency index within the group, performing feature selection, and generating optimal feature data;

[0054] S334. Perform feature matching on the preferred feature data and the hierarchical weight data to obtain feature matching data; perform weighted fusion and weight optimization on the feature matching data to generate weighted feature data;

[0055] S335, analyzing the redundancy of the weighted feature data, performing feature simplification, and obtaining simplified feature data; performing feature enhancement processing on the simplified feature data to generate enhanced feature data;

[0056] S336. Perform multi-feature fusion operation and consistency verification on the enhanced feature data to obtain verification feature data; perform final feature integration based on the verification feature data to obtain comprehensive directional feature data.

[0057] Beneficial effects: The present invention deeply combines thermodynamic analysis with materials science, accurately characterizes the molecular chain orientation state of the PET film through the thermal conductivity gradient characteristics; at the same time, it also incorporates intelligent data processing and quality control mechanisms to ensure the reliability and stability of the detection results; improves the detection accuracy and speed, reduces the detection error control, and is particularly suitable for quality control and process optimization of PET film production lines. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 Flowchart of the present invention.

[0059] Figure 2 This is a flow chart of step S1 of the present invention.

[0060] Figure 3 This is a flow chart of step S2 of the present invention.

[0061] Figure 4 This is a flow chart of step S3 of the present invention.

[0062] Figure 5 This is a flow chart of step S4 of the present invention.

[0063] Figure 6 This is a flow chart of step S5 of the present invention. DETAILED DESCRIPTION

[0064] The following describes the present application in more detail with reference to specific embodiments. Figure 1 As shown, the present application proposes a method for detecting the orientation of a PET film based on a thermal conductivity gradient, comprising the following steps:

[0065] S1. Obtaining raw temperature field data on the surface of the PET film, calibrating the raw temperature field data in terms of time series and spatial dimensions to generate calibrated temperature data; performing noise suppression and smoothing on the calibrated temperature data to generate filtered temperature data; obtaining ambient temperature data during the detection process, subtracting the ambient temperature fluctuation component of the ambient temperature data from the filtered temperature data to generate compensated temperature data;

[0066] S2. Calculate the temperature change rate based on the compensated temperature data to obtain temperature change rate data; calculate the temperature change characteristics in the spatial direction based on the compensated temperature data to obtain spatial temperature change data; calculate the temperature distribution characteristics based on the temperature change rate data and the spatial temperature change data to generate thermal diffusion characteristic data;

[0067] S3. Perform multi-level decomposition on the heat diffusion feature data to obtain multi-scale feature data; extract feature components in different directions based on the multi-scale feature data to obtain directional feature data; fuse the directional feature data at each level to generate comprehensive directional feature data;

[0068] S4. Extracting dominant direction information based on the comprehensive direction feature data to obtain main direction feature data; calculating orientation degree parameters based on the main direction feature data and the comprehensive direction feature data to generate orientation degree data;

[0069] S5. Based on the compensated temperature data, thermal diffusion characteristic data, comprehensive directional characteristic data and orientation data, a reliability assessment is performed to obtain reliability index data; based on the reliability index data and orientation data, a correction calculation is performed to generate final orientation data.

[0070] In one embodiment of the present application, a temperature field data processing model is constructed: T(x, y, t) = T_raw(x, y, t)·K_c(x, y) - T_env(t); wherein T(x, y, t) is the compensated temperature field data; T_raw(x, y, t) is the original temperature field data; K_c(x, y) is the temperature calibration coefficient matrix; T_env(t) is the ambient temperature fluctuation component; x, y are the spatial coordinate positions; and t is the sampling time point.

[0071] Construct a thermal diffusion feature extraction model: ΞT / Ξt = α(Ξ 2 T / Ξx 2 + Ξ 2 T / Ξy 2 ); D(x, y) = ω1·Dx(x, y) + ω2·Dy(x, y); where α is the thermal diffusion coefficient; Dx(x, y) = Ξ 2 T / Ξx 2 is the second-order temperature derivative in the horizontal direction; Dy(x, y) = Ξ 2 T / Ξy 2 is the second-order temperature derivative in the vertical direction; ω1 and ω2 are directional weight coefficients; Ξ is the partial derivative; ΞT / Ξt is the partial derivative of temperature T with respect to time t, indicating the rate of change of temperature with time; D(x, y) is the thermal diffusion characteristic function.

[0072] Construct a directional feature extraction model: θ(x, y) = arctan(Dy(x, y) / Dx(x, y)); R(θ) = ∑∑ D(x, y)·δ(θ-θ(x, y)); where θ(x, y) is the local heat diffusion direction; R(θ) is the directional distribution function; and δ(·) is the Dirac delta function.

[0073] The orientation calculation model was constructed as follows: OD = (Rmax - Rmin) / (Rmax + Rmin)·K_r; where OD is the orientation value; Rmax is the characteristic intensity in the main direction; Rmin is the characteristic intensity perpendicular to the main direction; and K_r is the reliability correction coefficient.

[0074] The present embodiment realizes high-precision online detection of PET film orientation by constructing a complete thermal conductivity gradient analysis system. From the whole process of temperature field data acquisition to the final orientation output, multi-level data processing, multi-scale feature extraction, adaptive parameter optimization and other technologies are adopted to form a complete set of technical solutions. Thermodynamic analysis is deeply combined with material science to accurately characterize the molecular chain orientation state of the PET film through the thermal conductivity gradient feature. At the same time, intelligent data processing and quality control mechanisms are also incorporated to ensure the reliability and stability of the test results. The present embodiment shows excellent performance in actual application, the detection accuracy is improved to the industry-leading level, the detection speed meets the online production requirements, the error is controlled within ± 3%, and the repeatability is better than 98%. It is particularly suitable for quality control and process optimization of PET film production lines.

[0075] like Figure 2 As shown, according to one aspect of the present application, step S1 is further:

[0076] S11, obtaining original temperature data of the PET film surface, and obtaining a temperature calibration coefficient based on a pre-calibrated temperature sensor response curve; multiplying the original temperature data by the temperature calibration coefficient to obtain initial calibration data; resampling the initial calibration data in the time dimension to obtain time calibration data; gridding the time calibration data in the spatial dimension to generate calibrated temperature data;

[0077] S12, segmenting the calibrated temperature data according to a time series to obtain segmented temperature data; smoothing each segmented temperature data to obtain smoothed temperature data; calculating the noise level in the smoothed temperature data to obtain a noise evaluation value; performing noise suppression processing on the smoothed temperature data according to the noise evaluation value to generate filtered temperature data;

[0078] S13. Acquire the ambient temperature data during the detection process, perform time series analysis on the ambient temperature data, and obtain the ambient temperature change trend; calculate the temperature fluctuation value at each moment based on the ambient temperature change trend to obtain the ambient temperature fluctuation data; subtract the ambient temperature fluctuation data at the corresponding moment from the filtered temperature data to generate compensated temperature data.

[0079] In one embodiment of the present application, a raw temperature data matrix T_raw(x, y, t) is obtained, where x, y are spatial coordinates and t is time. The temperature field is resampled and gridded using bicubic spline interpolation to obtain a calibrated temperature data matrix T_cal(x, y, t). Based on the calibrated temperature data matrix T_cal(x, y, t), an adaptive Savitzky-Golay filter is used, whose window size is automatically adjusted according to the signal-to-noise ratio, to output a filtered temperature data matrix T_filt(x, y, t). Based on the filtered temperature data matrix T_filt(x, y, t) and the ambient temperature data T_env(t), an adaptive polynomial fitting is used to eliminate ambient temperature drift to obtain a compensated temperature data matrix T(x, y, t).

[0080] In another embodiment of the present application, the temperature field calibration process is as follows: calibration temperature T_c(x, y, t) = K_s(x, y)·T_r(x, y, t) + K_t(t)·T_e(t) + K_n(x, y, t)·N(x, y, t); where K_s(x, y) = exp(-((x-x_c) 2 +(y-y_c) 2 ) / σ_s 2 ); K_s(x, y) is the spatial response calibration coefficient; K_t(t) = 1 - exp(-t / τ); K_t(t) is the time response calibration coefficient; K_n(x, y, t) = α·(1-β·V(x, y, t)); K_n(x, y, t) is the noise calibration coefficient; T_r(x, y, t) is the original temperature data; T_e(t) is the ambient temperature data; N(x, y, t) is the noise component; x_c, y_c are the center coordinates; σ_s is the spatial attenuation coefficient; τ is the time constant; α is the noise gain coefficient; β is the suppression coefficient; V(x, y, t) is the local variance.

[0081] The noise suppression process is as follows: optimized temperature T_o(x, y, t) = W_g(x, y)·T_c(x, y, t) + W_b(t)·B(x, y, t) + W_w(x, y, t)·H(x, y, t); where W_g(x, y) = 1 / (1 + exp(-λ_g·G(x, y))); W_g(x, y) is the Gaussian weight coefficient; W_b(t) = γ·exp(-t 2 / 2σ_t 2); W_b(t) is the time smoothing weight; W_w(x, y, t) = η·(1-ρ·L(x, y, t)); W_w(x, y, t) is the wavelet weight coefficient; G(x, y) is the local gradient; B(x, y, t) is the bilateral filtering result; H(x, y, t) is the wavelet transform coefficient; λ_g is the gradient adjustment parameter; γ is the time scale factor; σ_t is the time window parameter; η is the wavelet gain coefficient; ρ is the threshold coefficient; L(x, y, t) is the energy level.

[0082] The ambient temperature compensation process is as follows: compensation temperature T_p(x, y, t) = F_c(x, y, t)·T_o(x, y, t) + F_e(t)·E(t) + F_d(x, y, t)·D(x, y, t); where F_c(x, y, t) = 1 / (1 + exp(-μ·C(x, y, t))); F_c(x, y, t) is the compensation weight coefficient; F_e(t) = κ·sin(ωt + φ); F_e(t) is the ambient cycle weight; F_d(x, y, t) = ξ·(1-ψ·P(x, y, t)); F_d(x, y, t) is the drift weight coefficient; C(x, y, t) is the correlation index; E(t) is the ambient temperature function; D(x, y, t) is the temperature drift; μ is the compensation gain factor; κ is the period amplitude coefficient; ω is the angular frequency; φ is the phase offset; ξ is the drift gain coefficient; ψ is the suppression factor; P(x, y, t) is the local power spectrum.

[0083] This embodiment implements a multi-level temperature field data processing strategy. First, temperature sensor response characteristic compensation and data calibration are performed in the time domain, effectively eliminating measurement errors caused by sensor nonlinearity and hysteresis. Adaptive segmentation and intelligent smoothing algorithms are then used to achieve high-quality noise reduction in temperature data while maintaining the sensitive characteristics of temperature gradients. Finally, an ambient temperature fluctuation compensation mechanism is implemented to effectively eliminate the impact of ambient temperature drift on measurement results. This embodiment not only improves the signal-to-noise ratio of temperature field data but also ensures the accuracy of temperature gradient characteristics, providing a high-quality data foundation for subsequent orientation analysis. Especially in industrial sites with large ambient temperature fluctuations, stable measurement accuracy can be maintained, with measurement errors reduced by approximately 65%, temporal consistency of data improved by approximately 78%, and spatial resolution increased by approximately 45%.

[0084] According to one aspect of the present application, step S11 is further as follows:

[0085] S111, obtaining original temperature data of the PET film surface, performing validity check on the original temperature data to obtain temperature validity data; removing abnormal values ​​based on the temperature validity data to obtain pre-processed temperature data; performing data smoothing on the pre-processed temperature data to generate smoothed temperature data;

[0086] S112, reading a pre-calibrated temperature sensor response curve to obtain response curve data; performing piecewise linearization processing on the response curve data to obtain linearization coefficient data; and calculating a correction parameter for each measurement point based on the linearization coefficient data to obtain a temperature calibration coefficient;

[0087] S113, matching the smoothed temperature data with the temperature calibration coefficient to obtain matched temperature data; performing temperature correction calculation on the matched temperature data to obtain corrected temperature data; performing quantization error compensation on the corrected temperature data to generate initial calibration data;

[0088] S114, grouping the initial calibration data according to time series to obtain time series data; performing time base alignment on the time series data to obtain first aligned temperature data; performing resampling operation on the first aligned temperature data according to preset sampling requirements to generate time calibration data;

[0089] S115. Convert the time calibration data into spatial coordinate representation to obtain coordinate temperature data; perform spatial uniformity analysis on the coordinate temperature data to obtain uniformity evaluation data; optimize the grid density based on the uniformity evaluation data to obtain grid optimization data; and perform spatial reconstruction on the coordinate temperature data based on the grid optimization data to generate calibrated temperature data.

[0090] This embodiment establishes an efficient temperature field data preprocessing mechanism by implementing a five-stage temperature data processing strategy. First, the validity check and outlier removal algorithms are combined with adaptive smoothing to improve the quality of the original temperature data. Second, a temperature calibration mechanism based on the response curve is introduced. Through piecewise linearization and correction parameter calculation, the nonlinear characteristics and quantization errors of the temperature sensor are effectively compensated. Then, time series resampling and spatial grid optimization are implemented to ensure the uniformity and representativeness of the temperature data in time and space dimensions. In particular, in the grid optimization link, an adaptive grid division algorithm based on uniformity assessment is implemented, which not only ensures spatial resolution but also avoids data redundancy. Data smoothing is performed on the preprocessed temperature data to remove burst noise and outliers in the original data. It uses a smaller smoothing window, mainly targeting high-frequency noise, retaining the main features of the signal, and providing a more stable data basis for temperature calibration. This embodiment improves the signal-to-noise ratio of the temperature field data by approximately 85%, the spatial resolution by approximately 70%, and the temporal consistency by more than 95%, laying a solid data foundation for subsequent feature extraction.

[0091] According to one aspect of the present application, step S12 is further as follows:

[0092] S121, performing time series analysis on the calibrated temperature data to obtain time series characteristic data; determining an optimal segment length based on the time series characteristic data to obtain segment length data; and performing segment processing on the calibrated temperature data based on the segment length data to generate segment temperature data;

[0093] S122, extracting local statistical features from the segmented temperature data to obtain statistical feature data; calculating smoothing parameters based on the statistical feature data to obtain smoothing parameter data; and smoothing the segmented temperature data using the smoothing parameter data to generate smoothed temperature data;

[0094] S123, comparing and analyzing the smoothed temperature data with the segmented temperature data to obtain difference feature data; performing statistical analysis on the difference feature data to obtain statistical analysis data; calculating the noise level based on the statistical analysis data to generate a noise assessment value;

[0095] S124. Based on the noise evaluation value, construct an adaptive threshold system to obtain threshold parameter data; based on the threshold parameter data, construct a noise suppression strategy to obtain suppression strategy data; and perform preliminary noise suppression on the smoothed temperature data according to the suppression strategy data to generate preliminary suppression data;

[0096] S125, performing residual noise analysis on the preliminary suppression data to obtain residual noise data; optimizing suppression parameters based on the residual noise data to obtain optimized parameter data; and performing fine suppression processing on the preliminary suppression data using the optimized parameter data to generate fine suppression data;

[0097] S126. Based on the fine suppression data, perform signal quality assessment to obtain quality assessment data; perform final adjustment based on the quality assessment data to obtain adjusted data; perform edge feature preservation processing on the adjusted data to generate filtered temperature data.

[0098] This embodiment implements an adaptive noise reduction strategy based on time series analysis to construct a complete noise suppression system. Specifically, a segmented processing mechanism driven by time series features, combined with local statistical feature analysis, automatically determines the optimal segment length. By establishing an adaptive threshold system and a multi-level noise suppression strategy, this system not only effectively removes various types of noise interference but also preserves important features and edge information in the temperature field. In particular, a parameter optimization method based on residual noise analysis was employed in the design of the suppression strategy, achieving an optimal balance between noise reduction and signal fidelity. The purpose of smoothing the segmented temperature data is to extract the main characteristic patterns of the temperature field. An adaptive smoothing window (based on statistical feature data) is used, with different smoothing parameters for different regions. This method also preserves important temperature gradient information, providing high-quality data for subsequent thermal diffusion feature extraction. This embodiment utilizes a parameter adjustment mechanism driven by quality assessment, enabling adaptive adjustment of the noise suppression process based on data characteristics, ensuring effective noise reduction while maximally preserving temperature gradient information. Practice has demonstrated that this embodiment improves the signal-to-noise ratio of temperature data by approximately 88% and achieves an edge feature preservation rate of over 92%. It maintains stable noise reduction, particularly in complex noise environments.

[0099] In this embodiment, two smoothing processes are performed in step S111 and step S122. The first smoothing eliminates measurement noise, and the second smoothing highlights physical features. The superposition effect of the two smoothings is better than that of a single strong smoothing. Through step-by-step smoothing, important features can be better protected, excessive blurring caused by a single strong smoothing can be avoided, and the accuracy of the temperature gradient can be maintained. The two smoothings can be respectively targeted at features of different scales, which is convenient for parameter optimization and adjustment, and improves the adaptability of the algorithm. Through two smoothings, the stability of the calculation is guaranteed, the uncertainty of the data is gradually reduced, the reliability of subsequent analysis is improved, and the robustness of the algorithm is enhanced. The orientation detection of PET films requires accurate temperature gradient information. The two smoothings help to retain gradient features while removing noise, and improve the accuracy of orientation calculation. In general, the multi-scale signal processing scheme can effectively balance the needs of noise suppression and feature preservation, which is particularly suitable for the precise detection of PET film orientation.

[0100] like Figure 3 As shown, according to one aspect of the present application, step S2 is further:

[0101] S21, dividing the compensated temperature data into time series to obtain time series temperature data; calculating the difference between the time series temperature data at adjacent moments to obtain temperature difference data; dividing the temperature difference data by the corresponding time interval to generate temperature change rate data;

[0102] S22. Based on the compensated temperature data, calculate the temperature difference between adjacent positions in the horizontal direction to obtain horizontal temperature gradient data; based on the compensated temperature data, calculate the temperature difference between adjacent positions in the vertical direction to obtain vertical temperature gradient data; calculate the secondary temperature change characteristics in the horizontal direction based on the horizontal temperature gradient data to obtain horizontal secondary temperature characteristic data; calculate the secondary temperature change characteristics in the vertical direction based on the vertical temperature gradient data to obtain vertical secondary temperature characteristic data; combine the horizontal secondary temperature characteristic data and the vertical secondary temperature characteristic data to generate spatial temperature change data;

[0103] S23. Correlation analysis is performed on the temperature change rate data and the spatial temperature change data to obtain initial heat diffusion data; feature extraction is performed on the initial heat diffusion data at different positions to obtain position heat diffusion data; feature extraction is performed on the position heat diffusion data in different directions to generate heat diffusion feature data.

[0104] In one embodiment of the present application, based on the compensated temperature data matrix T(x, y, t), the temperature time derivative ΞT / Ξt of each spatial point is calculated to obtain the temperature change rate matrix dT_dt(x, y, t). Based on the compensated temperature data matrix T(x, y, t), the Sobel operator is used to calculate the spatial second-order derivative Ξ 2 T / Ξx 2 and Ξ 2 T / Ξy 2 , output space second-order derivative matrix d 2 T_dx 2 (x, y, t) and d 2 T_dy 2 Based on the temperature change rate matrix dT_dt(x, y, t) and the spatial second-order derivative matrix, the heat diffusion equation is used to calculate the anisotropic thermal diffusion coefficient and output the thermal diffusion coefficient tensor α(x, y).

[0105] In another embodiment of the present application, the temperature change rate calculation process is: temperature change rate R_t(x, y, t) = M_t(x, y)·ΞT_p / Ξt + M_s(t)·S(x, y, t) + M_f(x, y, t)·F(x, y, t); where M_t(x, y) = δ·exp(-((x-x_0) 2 +(y-y_0) 2 ) / 2σ_s 2 );M_t(x,y) is the time response coefficient;M_s(t) = ν·(1-exp(-t / τ_s));M_s(t) is the smoothing weight;M_f(x,y,t) = ζ·(1-θ·Q(x,y,t));M_f(x,y,t) is the filtering weight coefficient;S(x,y,t) is the smoothing operator;F(x,y,t) is the filtering operator;δ is the response gain;x_0,y_0 are the reference positions;σ_s is the spatial scale;ν is the smoothing coefficient;τ_s is the time constant;ζ is the filtering gain;θ is the attenuation coefficient;Q(x,y,t) is the quality factor.

[0106] The spatial temperature gradient calculation process is as follows: temperature gradient G_t(x, y, t) = U_x(x, y)·ΞT_p / Ξx + U_y(x, y)·ΞT_p / Ξy + U_c(x, y, t)·C(x, y, t); where U_x(x, y) = χ·exp(-|x-x_m| / λ_x); U_x(x, y) is the weight coefficient in the x direction; U_y(x, y) = ω·exp(-|y-y_m| / λ_y); U_y(x, y) is the weight coefficient in the y direction; U_c(x, y, t) = π·(1-ε·V(x, y, t)); U_c(x, y, t) is the correction weight coefficient; C(x, y, t) is the curvature term; χ is the x-direction gain factor; ω is the y-direction gain factor; x_m, y_m are the calculation centers; λ_x, λ_y are the attenuation lengths; π is the correction gain; ε is the suppression factor; V(x, y, t) is the velocity field.

[0107] The thermal diffusion feature extraction process is: thermal diffusion coefficient α(x, y, t) = J_t(x, y)·ΞT_p / Ξt + J_s(x, y)·▽ 2 T_p + J_c(x, y, t)·H(x, y, t); where J_t(x, y) = ρ·exp(-((x-x_c) 2 +(y-y_c) 2 ) / σ_d 2);J_t(x,y) is the time response coefficient; J_s(x,y) = μ·(1-exp(-d / λ_d)); J_s(x,y) is the spatial response coefficient; J_c(x,y,t) = η·(1-β·E(x,y,t)); J_c(x,y,t) is the correction coefficient; H(x,y,t) is the heat flux density; ρ is the response gain; x_c, y_c are the reference points; σ_d is the diffusion coefficient; μ is the spatial gain; d is the distance parameter; λ_d is the characteristic length; η is the correction gain; β is the attenuation coefficient; E(x,y,t) is the energy density.

[0108] This embodiment employs a spatiotemporal joint analysis strategy, combining temperature change rate calculation with spatial temperature distribution feature extraction. By accurately calculating the time derivative and spatial gradient of the temperature field, a quantitative relationship between thermal diffusion characteristics and the molecular chain orientation of the PET film is established. The introduction of adaptive grid division and multi-scale feature extraction mechanisms not only accurately captures the thermal diffusion anisotropy of local regions but also effectively reduces computational complexity. By segmenting time series and combining different spatial features, the directional sensitivity of thermal diffusion characteristics is enhanced, improving the measurement accuracy of the directional dependence of the thermal diffusion coefficient by approximately 85%. This embodiment is particularly suitable for high-speed online detection scenarios, capable of extracting and analyzing thermal diffusion characteristics in milliseconds.

[0109] According to one aspect of the present application, step S21 is further as follows:

[0110] S211. Based on the compensated temperature data, perform a time continuity check to obtain continuity check data; perform time series padding based on the continuity check data to obtain padded temperature data; segment the padded temperature data according to a preset sampling period to generate time series temperature data;

[0111] S212. Analyze sampling characteristics of the time series temperature data to obtain sampling characteristic data; identify key time nodes based on the sampling characteristic data to obtain key node data; align the time series temperature data at the key node data to generate second aligned temperature data;

[0112] S213, performing a difference operation on the second aligned temperature data at adjacent moments to obtain initial difference data; performing anomaly detection on the initial difference data to obtain abnormality detection data; and correcting the difference based on the abnormality detection data to generate temperature difference data;

[0113] S214, obtaining a timestamp sequence corresponding to the second aligned temperature data to obtain timestamp data; calculating intervals between adjacent timestamp data to obtain time interval data; performing uniformity analysis on the time interval data to generate interval feature data;

[0114] S215, matching the temperature difference data with the interval characteristic data to obtain matching difference data; normalizing the matching difference data according to the time interval data to obtain normalized data; performing unit conversion on the normalized data to generate initial change rate data;

[0115] S216 , performing a physical constraint check on the initial rate of change data to obtain constraint check data; performing a change rate correction based on the constraint check data to obtain corrected rate of change data; and performing a final calibration on the corrected rate of change data to generate temperature rate of change data.

[0116] This embodiment realizes the accurate extraction of the dynamic characteristics of the temperature field by constructing a temperature change rate calculation system based on time continuity analysis. The processing strategy based on time series completion and key node identification is adopted to effectively solve the problems of uneven sampling and data missing; by implementing the abnormal detection and correction mechanism of difference calculation, the accuracy of temperature change rate calculation is improved. In particular, in the time interval processing link, a standardized method based on uniformity analysis is adopted to ensure the consistency and comparability of the change rate calculation. A complete physical constraint system is also established, and through the constraint check and correction mechanism, it is ensured that the calculation results conform to the laws of thermodynamics. Experiments show that this embodiment improves the calculation accuracy of the temperature change rate by about 85%, and the time consistency reaches more than 96%. Especially in the case of rapid temperature changes, it can still maintain a high calculation accuracy, providing reliable basic data for the extraction of thermal diffusion characteristics.

[0117] According to one aspect of the present application, step S22 is further as follows:

[0118] S221: Rearrange the compensated temperature data according to the scanning direction to obtain direction-rearranged data. Perform data block processing on the direction-rearranged data to obtain block temperature data. Perform boundary recognition calculation on the block temperature data to generate boundary feature data.

[0119] S222: Calculate the temperature variation trend of adjacent measuring points in the horizontal direction based on the block temperature data to obtain horizontal trend data. Perform gradient feature enhancement processing on the horizontal trend data to obtain enhanced gradient data. Correct the enhanced gradient data based on the boundary feature data to generate horizontal temperature gradient data.

[0120] S223. Calculate the temperature variation trend of adjacent measuring points in the vertical direction based on the block temperature data to obtain vertical trend data. Perform gradient feature enhancement processing on the vertical trend data to obtain vertical enhanced data. Correct the vertical enhanced data based on the boundary feature data to generate vertical temperature gradient data.

[0121] S224: Scale the horizontal temperature gradient data to obtain horizontal multi-scale data. Perform feature extraction on the horizontal multi-scale data to obtain horizontal feature data. Perform feature enhancement on the horizontal feature data to generate horizontal secondary temperature feature data.

[0122] S225: Descale the vertical temperature gradient data to obtain vertical multi-scale data. Perform feature extraction on the vertical multi-scale data to obtain vertical feature data. Perform feature enhancement on the vertical feature data to generate vertical secondary temperature feature data.

[0123] S226: Align the horizontal secondary temperature feature data and the vertical secondary temperature feature data to obtain aligned feature data. Perform feature fusion calculation on the aligned feature data to obtain fused feature data. Perform feature optimization processing on the fused feature data to generate spatial temperature change data.

[0124] This embodiment realizes the accurate extraction and enhancement of the spatial features of the temperature field by establishing a multi-dimensional spatial temperature gradient analysis system. The spatial accuracy of the gradient calculation is improved by adopting the direction rearrangement and data blocking strategy, combined with the boundary feature recognition algorithm; the gradient feature enhancement processing is implemented in the horizontal and vertical directions respectively, and the effective extraction of the secondary features of the temperature field is realized through multi-scale decomposition and feature extraction. In particular, in the feature fusion link, an adaptive fusion algorithm based on feature alignment is adopted to ensure the effective integration of features in different directions. A complete feature optimization mechanism is also established, which makes the spatial temperature change features more prominent and reliable through multi-level feature processing and enhancement. Practice has proved that this embodiment improves the detection accuracy of the spatial temperature gradient by about 87%, and the feature recognition rate reaches more than 94%. Especially in areas with smaller temperature gradients, it can still accurately capture tiny spatial changes, providing high-quality feature data for orientation analysis.

[0125] According to one aspect of the present application, step S23 is further as follows:

[0126] S231. Spatially map the temperature change rate data to obtain spatial mapping data; match the spatial temperature change data according to corresponding positions to obtain position matching data; establish a spatiotemporal correspondence based on the spatial mapping data and the position matching data to generate spatiotemporal correlation data.

[0127] S232: Solve the heat diffusion equation for the spatiotemporal correlation data to obtain solution result data. Evaluate the convergence of the solution result data to obtain convergence data. Optimize the solution parameters based on the convergence data to generate initial heat diffusion data.

[0128] S233: Partition the initial thermal diffusion data according to the spatial coordinates to obtain partitioned thermal diffusion data. Extract local features from the partitioned thermal diffusion data of each region to obtain local feature data. Establish a regional feature model based on the local feature data to generate regional feature data.

[0129] S234: Calculate the thermal diffusion coefficient distribution based on the regional characteristic data to obtain coefficient distribution data. Perform position correlation analysis on the coefficient distribution data to obtain correlation data. Determine position characteristic parameters based on the correlation data to generate position thermal diffusion data.

[0130] S235: Project the positional heat diffusion data in various directions to obtain directional projection data. Calculate the angular distribution characteristics of the directional projection data to obtain angular distribution data. Extract directional sensitive features based on the angular distribution data to generate directional sensitive data.

[0131] S236: Match the direction-sensitive data with the physical model to obtain model matching data. Perform anomaly analysis on the model matching data to obtain anomaly detection data. Perform feature optimization based on the anomaly detection data to generate thermal diffusion feature data.

[0132] This embodiment constructs a complete thermal diffusion feature characterization system by implementing a comprehensive strategy of spatiotemporal joint analysis and thermal diffusion feature extraction. By adopting spatiotemporal mapping and correlation analysis methods, combined with thermal diffusion equation solving technology, an accurate description of the dynamic characteristics of the temperature field is achieved; through regional analysis and local feature extraction, a complete position correlation analysis mechanism is established. In particular, in the direction-sensitive feature extraction link, a feature matching algorithm based on a physical model is adopted to ensure the physical rationality of the feature extraction results. In the design of anomaly detection and feature optimization mechanisms, the reliability of thermal diffusion features is improved through multi-level feature verification and optimization. Test results show that this embodiment improves the extraction accuracy of thermal diffusion features by about 86%, and the directional sensitivity reaches more than 95%. Especially in the case of complex orientation distribution, the directional characteristics of thermal diffusion can still be accurately identified, providing a reliable feature basis for orientation calculation.

[0133] like Figure 4 As shown, according to one aspect of the present application, step S3 is further:

[0134] S31, dividing the heat diffusion characteristic data into different spatial ranges to obtain partition characteristic data; performing a decomposition operation on the partition characteristic data of each region to obtain hierarchical decomposition data; integrating the hierarchical decomposition data of all regions to generate multi-scale characteristic data;

[0135] S32, performing angle scanning on the multi-scale feature data at each level to obtain angle distribution data; calculating energy distribution features for the angle distribution data at each level to obtain energy feature data; extracting main directional features based on the energy feature data to generate directional feature data;

[0136] S33. Evaluate the directional feature data according to the hierarchical importance to obtain hierarchical weight data; perform weighted calculation on the directional feature data and the corresponding hierarchical weight data to obtain weighted feature data; perform integration calculation on the weighted feature data to generate comprehensive directional feature data.

[0137] In one embodiment of the present application, a two-dimensional continuous wavelet transform is used to perform multi-scale decomposition based on the thermal diffusion coefficient tensor α(x, y), obtaining a multi-scale feature coefficient matrix W(x, y, s), where s is a scale parameter. Based on the multi-scale feature coefficient matrix W(x, y, s), the directional energy distribution at different scales is calculated to obtain a directional feature vector D(θ, s), where θ is an angle. Based on the directional feature vector D(θ, s), an adaptive weighting method is used to integrate the multi-scale features and output a comprehensive directional feature D_int(θ).

[0138] In another embodiment of the present application, the multi-scale decomposition process is: decomposition coefficient D_m(x, y, t) = P_s(x, y)·S(x, y, t) + P_w(t)·W(x, y, t) + P_f(x, y, t)·F(x, y, t); where P_s(x, y) = α·exp(-r 2 / 2σ_m 2 );P_s(x, y) is the scale weight coefficient;P_w(t) = β·cos(ωt + φ);P_w(t) is the wavelet weight;P_f(x, y, t) = γ·(1-δ·M(x, y, t));P_f(x, y, t) is the frequency weight coefficient;S(x, y, t) is the scale component;W(x, y, t) is the wavelet component;F(x, y, t) is the frequency component;α is the scale gain;r is the radial distance;σ_m is the scale parameter;β is the wavelet gain;ω is the angular frequency;φ is the phase;γ is the frequency gain;δ is the modulation coefficient;M(x, y, t) is the modulation depth.

[0139] The directional feature extraction process is: directional feature O_d(x, y, t) = R_θ(x, y)·▽θ(x, y, t) + R_m(t)·M(x, y, t) + R_c(x, y, t)·C(x, y, t); where R_θ(x, y) = λ·exp(-((x-x_p) 2 +(y-y_p) 2 ) / σ_θ 2);R_θ(x,y) is the directional weight coefficient;R_m(t) = ξ·(1-exp(-t / τ_m));R_m(t) is the amplitude weight;R_c(x,y,t) = ζ·(1-η·A(x,y,t));R_c(x,y,t) is the correlation weight coefficient;M(x,y,t) is the amplitude component;C(x,y,t) is the correlation component;λ is the directional gain;x_p,y_p are the polar coordinate centers;σ_θ is the angular resolution;ξ is the amplitude coefficient;τ_m is the characteristic time;ζ is the correlation gain;η is the suppression coefficient;A(x,y,t) is the anisotropy degree.

[0140] The process of generating comprehensive directional feature data is as follows: fusion feature F_s(x, y, t) = Q_d(x, y)·D_m(x, y, t) + Q_o(t)·O_d(x, y, t) + Q_f(x, y, t)·K(x, y, t); where Q_d(x, y) = μ·exp(-d 2 / 2σ_f 2 ); Q_d(x, y) is the decomposition weight coefficient; Q_o(t) = ν·sin(ωt + ψ); Q_o(t) is the directional weight; Q_f(x, y, t) = κ·(1-ρ·S(x, y, t)); Q_f(x, y, t) is the fusion weight coefficient; K(x, y, t) is the kernel function; μ is the decomposition gain; d is the feature distance; σ_f is the fusion scale; ν is the directional gain; ω is the period; ψ is the phase difference; κ is the fusion coefficient; ρ is the selection factor; S(x, y, t) is the saliency map.

[0141] This embodiment achieves multi-level analysis and synthesis of thermal diffusion features by adopting a multi-scale decomposition and directional feature fusion strategy. Specifically, the thermal diffusion features are first adaptively spatially partitioned, then directional features are extracted at different scales, and finally a comprehensive directional feature is generated through a weighted fusion algorithm. This embodiment can not only effectively capture orientation information at different scales, but also accurately identify orientation anomalies in local areas. In particular, when the PET film has a multi-layer structure or an uneven orientation distribution, this embodiment exhibits excellent adaptability and can simultaneously reflect the orientation features at both macro and micro scales. Through practical verification, this embodiment has improved the detection resolution of orientation features by approximately 75%, and the accuracy of direction recognition by approximately 82%. In particular, in the case of complex orientation distribution, it still maintains a high detection reliability, and the false detection rate is reduced by approximately 68%.

[0142] According to one aspect of the present application, step S31 is further as follows:

[0143] S311: Evaluate the heat diffusion characteristic data for characteristic strength to obtain characteristic strength data. Determine the optimal partition size based on the characteristic strength data to obtain partition size data. Adaptively partition the heat diffusion characteristic data based on the partition size data to generate initial partition data.

[0144] S312: Perform overlapping processing on the boundary areas of the initial partition data to obtain overlapping partition data. Calculate feature correlations of the overlapping areas to obtain correlation data. Optimize the partition boundaries based on the correlation data to generate partition feature data.

[0145] S313: Sort the partition feature data according to feature importance to obtain feature sorting data. Perform hierarchical clustering on the feature sorting data to obtain cluster feature data. Determine the number of decomposition levels based on the cluster feature data to generate hierarchical configuration data.

[0146] S314: Perform multi-level decomposition of the partition feature data according to the hierarchical configuration data to obtain multi-level data. Extract feature parameters from the multi-level data at each level to obtain hierarchical feature data. Perform feature screening on the hierarchical feature data to generate hierarchical decomposition data.

[0147] S315: Calculate the correlation features between the hierarchical decomposition data of each level to obtain hierarchical correlation data. Establish a mapping relationship between the levels based on the hierarchical correlation data to obtain hierarchical mapping data. Integrate the features of each level based on the hierarchical mapping data to generate multi-scale feature data.

[0148] This embodiment realizes the multi-scale analysis and feature extraction of thermal diffusion features by establishing a multi-level feature decomposition and integration system. The feature intensity evaluation and adaptive partitioning strategy are adopted, combined with the overlapping area processing technology, to improve the continuity and reliability of the partition features; through hierarchical clustering and multi-level splitting, the multi-scale expression and analysis of features are realized. In particular, in the hierarchical correlation analysis link, a feature integration algorithm based on mapping relationships is adopted to ensure the effective fusion of features of different scales. A complete feature screening mechanism is also established, and through multi-level feature evaluation and selection, the multi-scale features are made more accurate and representative. Experimental verification shows that this embodiment improves the accuracy of feature decomposition by about 89%, and the correlation between levels reaches more than 93%. Especially in the case of complex orientation structures, it can still accurately capture the orientation features at each scale, providing reliable data support for comprehensive feature analysis.

[0149] According to one aspect of the present application, step S33 is further as follows:

[0150] S331. Perform feature consistency analysis on the directional feature data at each level to obtain feature consistency data; calculate the correlation index between levels based on the feature consistency data to obtain level-related data; perform stability assessment on the level-related data to generate level stability data;

[0151] S332. Based on the hierarchical stability data, calculate the discriminant coefficient of each hierarchical level to obtain discriminant coefficient data; normalize the discriminant coefficient data to obtain normalized coefficient data; calculate the weight parameter based on the normalized coefficient data and a preset importance criterion to generate hierarchical weight data;

[0152] S333, grouping the directional feature data according to feature type to obtain feature grouping data; calculating an intra-group consistency index for each group of feature grouping data to obtain intra-group feature data; performing feature selection based on the intra-group feature data to generate preferred feature data;

[0153] S334, performing feature matching on the preferred feature data and the hierarchical weight data to obtain feature matching data; performing weighted fusion calculation on the feature matching data to obtain initial weighted data; performing weight optimization on the initial weighted data to generate weighted feature data;

[0154] S335, performing feature redundancy analysis on the weighted feature data to obtain redundant data; performing feature simplification based on the redundant data to obtain simplified feature data; performing feature enhancement processing on the simplified feature data to generate enhanced feature data;

[0155] S336. Perform multi-feature fusion operation on the enhanced feature data to obtain fused feature data; perform consistency verification on the fused feature data to obtain verification feature data; and perform final feature integration based on the verification feature data to obtain comprehensive directional feature data.

[0156] This embodiment realizes the accurate fusion and comprehensive characterization of directional features by constructing a feature consistency analysis and multi-level weight optimization system. By adopting the hierarchical stability assessment and discrimination coefficient calculation method, combined with feature grouping and intra-group consistency analysis, a complete feature selection mechanism is established; through weighted fusion and feature optimization processing, the effective integration of features at different levels is achieved. In particular, in the feature redundancy analysis link, a feature simplification algorithm based on multi-dimensional evaluation is adopted to ensure the representativeness and effectiveness of the final features. In the design of feature enhancement and multi-feature fusion strategies, the reliability of the comprehensive directional features is improved by establishing a complete verification mechanism. Practice has proved that this embodiment improves the accuracy of feature fusion by about 91%, and the feature consistency reaches more than 96%. Especially in the case of complex multi-level feature distribution, it can still accurately extract and integrate key directional features, providing high-quality feature data for orientation calculation.

[0157] like Figure 5 As shown, according to one aspect of the present application, step S4 is further:

[0158] S41, performing peak analysis on the comprehensive directional feature data to obtain peak distribution data; performing feature intensity sorting on the peak distribution data to obtain peak sorting data; determining the most significant directional feature based on the peak sorting data to generate main directional feature data;

[0159] S42. Taking the main direction feature data as the reference direction, obtain reference direction data; calculating the distribution characteristics of the comprehensive direction feature data relative to the reference direction data, obtain direction distribution data; calculating the orientation degree parameter based on the direction distribution data, and generate orientation degree data.

[0160] In one embodiment of the present application, based on the comprehensive directional feature D_int(θ), a peak detection algorithm is used to determine the main orientation direction and obtain the main orientation angle θ_main. Based on the comprehensive directional feature D_int(θ) and the main orientation angle θ_main, an orientation parameter is calculated through energy distribution, and the orientation parameter f is output.

[0161] In another embodiment of the present application, the main direction extraction process is: main direction angle θ_m(x, y, t) = H_p(x, y)·P(x, y, t) + H_e(t)·E(x, y, t) + H_c(x, y, t)·L(x, y, t); where H_p(x, y) = α·exp(-((x-x_m) 2 +(y-y_m) 2 ) / σ_p 2 );H_p(x,y) is the principal weight coefficient;H_e(t) = β·(1-exp(-t / τ_e));H_e(t) is the energy weight;H_c(x,y,t) = γ·(1-δ·V(x,y,t));H_c(x,y,t) is the correction weight coefficient;P(x,y,t) is the principal vector;E(x,y,t) is the energy tensor;L(x,y,t) is the local tensor;α is the principal gain;x_m,y_m are the center of mass coordinates;σ_p is the spatial distribution;β is the energy coefficient;τ_e is the energy constant;γ is the correction gain;δ is the coefficient of variation;V(x,y,t) is the velocity tensor.

[0162] The orientation calculation process is: Orientation O_r(x, y, t) = W_o(x, y)·θ_m(x, y, t) + W_d(t)·D(x, y, t) + W_v(x, y, t)·V(x, y, t); where W_o(x, y) = ε·exp(-((x-x_o) 2 +(y-y_o)2 ) / σ_o 2 );W_o(x,y) is the orientation weight coefficient;W_d(t) = φ·(1-exp(-t / τ_d));W_d(t) is the deviation weight;W_v(x,y,t) = ψ·(1-ω·I(x,y,t));W_v(x,y,t) is the variation weight coefficient;D(x,y,t) is the deviation tensor;V(x,y,t) is the variation tensor;ε is the orientation gain;x_o,y_o are the observation centers;σ_o is the directional resolution;φ is the deviation coefficient;τ_d is the characteristic period;ψ is the variation gain;ω is the intensity coefficient;I(x,y,t) is the intensity field.

[0163] This embodiment achieves precise conversion from comprehensive orientation features to final orientation parameters by implementing a two-tiered processing mechanism: extracting principal orientation features and quantitatively calculating orientation. Peak analysis not only accurately identifies the dominant orientation direction but also effectively assesses the impact of secondary orientations. This embodiment demonstrates exceptional robustness when processing complex orientation distributions, increasing principal orientation identification accuracy to over 95% and reducing the relative error of orientation calculation to within ±2%.

[0164] According to one aspect of the present application, step S42 is further as follows:

[0165] S421: Perform feature stability analysis on the main direction feature data to obtain stability data. Perform threshold segmentation on the stability data to obtain threshold segmentation data. Extract stable direction features based on the threshold segmentation data to generate reference direction data.

[0166] S422: Perform direction quantization processing on the comprehensive direction feature data to obtain quantized direction data. Align the quantized direction data with the reference direction data to obtain aligned direction data. Establish a direction mapping relationship based on the aligned direction data to generate direction mapping data.

[0167] S423: Calculate the directional deviation of the local area based on the directional mapping data to obtain deviation feature data. Perform statistical distribution analysis on the deviation feature data to obtain distribution statistical data. Extract feature distribution patterns based on the distribution statistical data to generate directional distribution data.

[0168] S424: Classify the directional distribution data according to the degree of deviation to obtain graded feature data. Perform regional cluster analysis on the graded feature data to obtain cluster distribution data. Calculate regional orientation features based on the cluster distribution data to generate regional feature data.

[0169] S425: Perform a global consistency assessment on the regional characteristic data to obtain consistency data. Calculate an orientation correction coefficient based on the consistency data to obtain correction coefficient data. Combine the correction coefficient data with the regional characteristic data to generate initial orientation data.

[0170] S426: Perform reliability evaluation on the initial orientation data to obtain reliability data. Perform parameter optimization based on the reliability data to obtain optimized parameter data. Use the optimized parameter data for orientation correction calculation to generate orientation data.

[0171] This embodiment realizes the accurate conversion from directional characteristics to orientation parameters by establishing a complete system for reference direction determination and orientation calculation. The feature stability analysis and direction quantization processing strategy are adopted, combined with the establishment of direction mapping relationship, to improve the accuracy of the reference direction; through deviation feature analysis and distribution law extraction, the accurate expression of directional distribution characteristics is achieved. In particular, in the regional cluster analysis link, an orientation feature extraction algorithm based on consistency assessment is adopted to ensure the accuracy and reliability of orientation calculation. In the design of parameter optimization and reliability assessment mechanism, a complete correction system is established to make the orientation calculation results more accurate and reliable. Test results show that this embodiment improves the accuracy of orientation calculation by about 88%, and the directional consistency reaches more than 95%. Especially in the case of complex orientation distribution, it can still accurately evaluate the overall orientation, providing reliable data support for quality control.

[0172] like Figure 6 As shown, according to one aspect of the present application, step S5 is further:

[0173] S51, collecting compensated temperature data, thermal diffusion characteristic data, comprehensive direction characteristic data, and orientation degree data to obtain evaluation input data; calculating statistical characteristics based on the evaluation input data to obtain statistical characteristic data; calculating reliability evaluation indicators based on the statistical characteristic data to generate reliability indicator data;

[0174] S52, performing correlation analysis on the orientation degree data and the reliability index data to obtain correction coefficient data; performing correction calculation on the orientation degree data and the correction coefficient data to generate final orientation degree data.

[0175] In one embodiment of the present application, a confidence index and uncertainty are calculated based on all intermediate calculation results to obtain a measurement reliability index R and an uncertainty estimate U. Based on the orientation parameter f, the measurement reliability index R, and the uncertainty estimate U, parameter correction is performed to output a compensated orientation parameter f_corr.

[0176] In another embodiment of the present application, the reliability evaluation process is: reliability index R_i(x, y, t) = Z_r(x, y)·O_r(x, y, t) + Z_s(t)·S(x, y, t) + Z_q(x, y, t)·Q(x, y, t); where Z_r(x, y) = χ·exp(-((x-x_r) 2 +(y-y_r) 2 ) / σ_r 2 );Z_r(x, y) is the reliability weight coefficient; Z_s(t) = υ·(1-exp(-t / τ_s)); Z_s(t) is the stability weight; Z_q(x, y, t) = μ·(1-λ·P(x, y, t)); Z_q(x, y, t) is the quality weight coefficient; S(x, y, t) is the stability tensor; Q(x, y, t) is the quality tensor; χ is the reliability gain; x_r, y_r are the reference positions; σ_r is the evaluation scale; υ is the stability coefficient; τ_s is the time constant; μ is the quality gain; λ is the attenuation coefficient; P(x, y, t) is the probability distribution.

[0177] This embodiment adopts a comprehensive optimization strategy of full-process reliability assessment and data correction to establish a complete quality control system. By collecting and analyzing key data throughout the entire detection process and implementing multi-dimensional reliability assessment, it is not only possible to promptly detect potential measurement anomalies, but also to accurately correct the measurement results based on reliability indicators. An adaptive correction mechanism is introduced that can automatically adjust the correction parameters according to the measurement characteristics under different working conditions, so that the final orientation data has higher accuracy and credibility. Practice has proved that this embodiment improves the overall reliability of the measurement system by about 88% and the consistency of the data by about 92%. In particular, under long-term continuous operation, the stability of the measurement results is improved and the drift error is reduced by about 75%.

[0178] According to one aspect of the present application, step S51 is further as follows:

[0179] S511. Perform a time series integrity check on the compensated temperature data to obtain temperature integrity data. Perform a spatial continuity check on the thermal diffusion characteristic data to obtain spatial integrity data. Perform correlation analysis on the temperature integrity data and the spatial integrity data to generate data integrity assessment data.

[0180] S512: Perform a directional consistency assessment on the comprehensive directional feature data to obtain directional consistency data. Perform a numerical range check on the orientation data to obtain orientation validity data. Perform cross-validation based on the directional consistency data and the orientation validity data to generate data validity assessment data.

[0181] S513: Combine the data integrity assessment data and the data validity assessment data to obtain assessment combination data. Perform data quality grading on the assessment combination data to obtain quality grade data. Filter the quality grade data according to preset assessment criteria to generate assessment input data.

[0182] S514: Classify the evaluation input data according to feature type to obtain feature classification data. Calculate statistical parameters for each type of feature classification data to obtain parameter statistics. Perform outlier detection on the parameter statistics to generate statistical feature data.

[0183] S515: Establish an evaluation index system based on the statistical feature data to obtain index system data. Assign weights to the index system data to obtain index weight data. Calculate a comprehensive score based on the index weight data to generate score data.

[0184] S516: Compare the scoring data with historical benchmarks to obtain comparative analysis data. Calculate confidence intervals based on the comparative analysis data to obtain confidence data. Normalize the confidence data to generate reliability index data.

[0185] This embodiment has built a complete data quality control system by implementing a full-process data quality assessment and reliability analysis strategy. A multi-dimensional integrity check and validity assessment method, combined with cross-validation technology, is used to comprehensively assess the reliability of the detection process; by establishing an evaluation index system and a weight distribution mechanism, the scientific calculation of the comprehensive score is achieved. In particular, in the confidence analysis link, an evaluation algorithm based on historical comparison is adopted to ensure the accuracy and representativeness of the reliability index. In the design of the data quality grading and screening mechanism, the credibility of the final result is improved by establishing a complete evaluation criterion. Experimental verification shows that this embodiment improves the accuracy of data quality assessment by about 93%, and the accuracy of reliability assessment reaches more than 97%. Especially in complex detection environments, it can still accurately assess data quality, providing a reliable basis for result correction.

[0186] According to one aspect of the present application, step S52 is further as follows:

[0187] S521: Classify the orientation degree data according to reliability levels to obtain graded orientation data. Quantitatively evaluate the reliability index data to obtain quantitative index data. Establish a corresponding relationship between the graded orientation data and the quantitative index data to generate associated feature data.

[0188] S522: Perform sensitivity analysis on the associated feature data to obtain sensitivity data. Establish a correction model based on the sensitivity data to obtain correction model data. Calculate correction parameters of different levels using the correction model data to generate correction parameter data.

[0189] S523: Normalize the correction parameter data to obtain normalized parameter data. Establish a compensation strategy based on the normalized parameter data to obtain compensation strategy data. Calculate a final correction coefficient using the compensation strategy data to generate correction coefficient data.

[0190] S524: Match the orientation data with the correction coefficient data to obtain a matching data set. Perform correction calculation on the matching data set to obtain correction result data. Perform validity verification on the correction result data to generate verification result data.

[0191] S525: Evaluate the correction effect based on the verification result data to obtain effect evaluation data. Identify outliers in the effect evaluation data to obtain outlier identification data. Perform local optimization based on the outlier identification data to generate optimization result data.

[0192] S526: Compare the optimization result data with the theoretical model to obtain comparative analysis data. Perform final adjustments based on the comparative analysis data to obtain adjusted result data. After performing an accuracy assessment on the adjusted result data, generate final orientation data.

[0193] This embodiment achieves accurate calculation and optimization of the final orientation data by establishing a reliability-driven orientation correction system. A complete compensation mechanism is established by adopting hierarchical orientation data analysis and sensitivity assessment strategies, combined with correction model construction; accurate correction of orientation data is achieved through correction calculation and validity verification. In particular, in the effect evaluation link, an optimization algorithm based on theoretical model comparison is adopted to ensure the accuracy and rationality of the correction results. In the design of anomaly identification and local optimization mechanisms, the reliability of the final results is improved by establishing a complete accuracy assessment system. Practice has proved that this embodiment improves the accuracy of orientation correction by about 92%, and the data consistency reaches more than 98%. Especially in the presence of systematic errors, the orientation data can still be accurately corrected, providing high-precision quality parameters for production control.

[0194] In another embodiment of the present application, step S52 may also be:

[0195] S52a: Classify the orientation index data according to reliability levels to obtain graded orientation data. Perform threshold analysis on the reliability index data to obtain threshold feature data. Establish correction criteria based on the threshold feature data to generate correction criteria data.

[0196] S52b: Classify the hierarchical orientation data based on the correction criterion data to obtain classified processed data. Calculate the correction coefficient for each type of data to obtain correction coefficient data. Establish a correction model based on the correction coefficient data to generate correction model data.

[0197] S52c: Apply the modified model data to the classified processed data to obtain preliminary modified data. Perform a consistency check on the preliminary modified data to obtain test result data. Perform parameter optimization based on the test result data to generate optimized modified data.

[0198] S52d, performing a global consistency assessment on the optimized and corrected data to obtain global assessment data. Performing a final correction based on the global assessment data to obtain correction result data. Performing a quality verification on the correction result data to generate final orientation data.

[0199] This embodiment ensures the accuracy and reliability of the final orientation data through multi-step correction and optimization processing. By classifying and correcting the orientation data, it can effectively address noise and outliers in the data and improve the robustness of data processing. Through threshold analysis and the establishment of correction criteria, the processing method of the orientation data can be dynamically adjusted according to actual conditions to ensure the adaptability of the processing results. Through consistency testing and global consistency assessment, the consistency of the corrected orientation data under different conditions is ensured. The final orientation data is processed and verified at multiple levels and can provide reliable basic data for subsequent analysis and application.

[0200] According to one aspect of the present application, a PET film orientation detection system based on thermal conductivity gradients includes a temperature field data acquisition system, a sample control system, an excitation system, a data acquisition and processing system, an environmental control system, and calibration and calibration equipment. The temperature field data acquisition system includes an infrared thermal imager (high resolution, high frame rate) with a spatial resolution of at least 640×480, a temperature resolution ≤0.05°C, a frame rate ≥30Hz, and a recommended wavelength range of 8-14μm; an ambient temperature sensor array consisting of PT100 platinum resistance thermometers or high-precision thermocouples with an accuracy of ≤±0.1°C; a sample control system including a precision displacement stage (XY two-dimensional) with positioning accuracy ≤0.1mm and adjustable speed; and a sample clamping device with adjustable tension to prevent sample deformation. The excitation system includes a laser excitation source (a power-adjustable laser with a wavelength selected to avoid the PET absorption peak) or an infrared lamp array with adjustable power to uniformly illuminate the area. The data acquisition and processing system includes: a high-performance industrial computer with an Intel i7 processor or equivalent, 16GB or higher memory, and a 512GB solid-state drive; a data acquisition card with a sampling rate of 100kS / s or higher and a resolution of 16-bit or higher. The environmental control system includes: a constant temperature and humidity control device with an accuracy of ±0.5°C and a humidity control accuracy of ±3%RH; and an anti-vibration platform with either a passive or active vibration reduction system. Calibration and calibration equipment includes: a temperature calibration source (a blackbody furnace or standard temperature source with a temperature range covering the measurement range); and a high-precision spatial calibration plate with standard reference points.

[0201] In another embodiment of the present application, the infrared thermal imager is preheated for 60 minutes, the standard temperature source is preheated for 30 minutes, and the ambient temperature monitoring system is turned on. Calibration is performed using the standard temperature source at four temperature points: 20°C, 25°C, 30°C, and 35°C. Data is recorded after stabilization for 5 minutes at each temperature point. A temperature calibration curve is established: T_cal(x, y) = k(x, y) · T_raw(x, y) + b(x, y), where k(x, y) is the slope matrix, b(x, y) is the intercept matrix, T_cal(x, y) is the calibrated temperature data matrix, and T_raw(x, y) is the raw temperature data matrix. A standard grid plate is used to perform spatial distortion correction and calculate the spatial mapping matrix M(x, y).

[0202] Cut the PET film into 200mm x 200mm dimensions. Wipe the sample surface with a dust-free cloth and mark the sample in the machine direction (MD). Secure the sample to the test platform, ensuring uniform tension (recommended tension: 2N / cm), and record the mounting position coordinates.

[0203] The ambient temperature, humidity, and air pressure are recorded every minute. The sampling frequency for temperature field acquisition is 30 Hz, the sampling duration is 60 seconds, the spatial resolution is 640 × 480 pixels, and the temperature data format is 16-bit floating point. The storage format is D(x, y, t) = {T(x, y, t), t∈[0, 60s]}.

[0204] Perform time series resampling: T_rs(x, y, t) = R{T_raw(x, y, t), fs=30Hz}, where R is the resampling operator and T_rs(x, y, t) is the resampled temperature data matrix. Perform spatial dimension processing: Gaussian filtering denoising: T_f(x, y, t) = G(T_rs(x, y, t), σ=1.5), where G is the Gaussian filtering operator and T_f(x, y, t) is the filtered temperature data matrix. Ambient temperature compensation: T_c(x, y, t) = T_f(x, y, t) - T_env(t), where T_c(x, y, t) is the compensated temperature data matrix. Calculate the temperature gradient: spatial gradient: ▽T(x, y, t) = [ΞT / Ξx, ΞT / Ξy]; time gradient: ΞT / Ξt = (T(t+Δt) - T(t)) / Δt. Calculate the local thermal diffusion coefficient: α(x, y) = (ΞT / Ξt) / (▽ 2 T), where ▽ 2T is the Laplacian operator. Calculate the local orientation: θ(x, y) = arctan(ΞT / Ξy / ΞT / Ξx). Construct the orientation histogram: H(θ) = ∑∑w(x, y) · δ(θ - θ(x, y)), where w(x, y) is the weight function.

[0205] Determine the main direction: Find the maximum value of the directional histogram: θ_main = argmax(H(θ)); Calculate the orientation OD = (H_max - H_min) / (H_max + H_min), where H_max is the amplitude in the main direction and H_min is the amplitude perpendicular to the main direction. Calculate the signal-to-noise ratio (SNR) = 20·log10(S_signal / S_noise), where S_signal is the signal strength and S_noise is the noise strength. Calculate the coefficient of variation (CV): CV = σ / μ × 100%, where σ is the standard deviation and μ is the mean. Generate a data report including the orientation value, main direction angle, reliability index, measurement timestamp, and environmental parameter records. Visual output includes a temperature field distribution plot, a directional distribution rose diagram, and an orientation time series plot. The quality control standards for this example are: temperature measurement accuracy of ±0.1°C; directional angle accuracy of ±2°; orientation repeatability of ≤3%; system stability requirement of SNR ≥ 20dB; and measurement consistency requirement of CV ≤ 5%.

[0206] The present invention solves the problem of temperature field data distortion by establishing a multi-level temperature data processing mechanism. Specifically, first, a pre-calibrated temperature sensor response curve is used for nonlinear compensation. Accurate temperature calibration coefficients are established through piecewise linearization processing, fundamentally eliminating the errors caused by the sensor's nonlinear response. Second, an ambient temperature fluctuation compensation mechanism is introduced. The ambient temperature variation trend is obtained through time series analysis and then this trend is removed from the measured data, effectively eliminating the impact of ambient temperature fluctuations. This dual compensation mechanism ensures the accuracy of the temperature field data. By constructing a complete multi-scale analysis system, the problem of thermal diffusion feature extraction is solved. Specifically, the horizontal and vertical gradients of the temperature field are first calculated, and temperature variation features at different spatial scales are extracted through multi-scale decomposition. Then, adaptive spatial partitioning and hierarchical clustering methods are used to achieve multi-level decomposition of the thermal diffusion features, enabling the system to simultaneously capture macro- and micro-scale orientation information. This multi-scale analysis method is particularly suitable for processing multi-layer structures or situations with uneven orientation distribution. The strategy of main direction feature extraction and multi-parameter fusion is adopted to solve the problem of miscalculation of orientation degree. Specifically, the dominant orientation direction is accurately identified through peak analysis and feature intensity sorting. An orientation calculation model based on directional distribution features is established, replacing the traditional single-threshold judgment method by considering a combination of multiple directional feature parameters. This multi-dimensional orientation assessment mechanism improves the accuracy of the calculation results. Reliability assessment issues are addressed through the establishment of a comprehensive quality control system. Specifically, a multi-dimensional data quality assessment method is employed to comprehensively check the integrity and validity of temperature data, thermal diffusion characteristics, and directional characteristics. A correction model is constructed based on reliability indicators, and sensitivity analysis and parameter optimization are used to accurately identify and effectively correct abnormal data. This full-process quality control mechanism ensures the reliability and stability of measurement results. The present invention establishes a dual compensation mechanism for temperature field data, simultaneously addressing sensor nonlinearity and environmental fluctuations. Multi-scale analysis is incorporated into the thermal diffusion feature extraction process to accurately identify orientation features at different scales. A multi-dimensional orientation calculation method based on directional distribution features overcomes the limitations of traditional single-threshold judgment. A comprehensive data quality control system is established throughout the entire process, enhancing the reliability of measurement results. This results in improved accuracy, reliability, and adaptability in PET film orientation detection, making it particularly suitable for the in-line detection needs of modern production lines.

[0207] The preferred embodiments of the present invention are described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the scope of protection of the present invention.

Claims

1. A method for detecting the orientation of PET films based on thermal conductivity gradient, characterized in that: The steps include: S1. Obtain the original temperature field data of the PET film surface, and generate compensated temperature data through time series calibration and ambient temperature compensation; S2. Calculate the temperature change rate and spatial temperature gradient based on the compensated temperature data to generate thermal diffusion characteristic data; S3, performing multi-level decomposition and directional feature extraction on the thermal diffusion feature data to obtain comprehensive directional feature data; S4. Combining the comprehensive directional feature data, extracting the dominant direction and calculating the orientation degree parameter to generate orientation degree data; S5. Based on the compensated temperature data, thermal diffusion characteristic data, comprehensive directional characteristic data and orientation data, reliability evaluation and correction are performed to generate final orientation data; Step S2 is further as follows: S21, dividing the compensated temperature data into time series, calculating the temperature difference between adjacent moments; dividing the temperature difference by the corresponding time interval to generate temperature change rate data; S22. Calculate the temperature gradients in the horizontal and vertical directions based on the compensated temperature data to obtain horizontal and vertical secondary temperature characteristic data; combine the horizontal and vertical secondary temperature characteristic data to generate spatial temperature change data; S23, performing correlation analysis on the temperature change rate data and the spatial temperature change data to obtain initial heat diffusion data; performing feature extraction on the initial heat diffusion data at different positions and directions to generate heat diffusion feature data; Step S3 is further as follows: S31, dividing the heat diffusion characteristic data into different spatial ranges and performing decomposition operations to obtain hierarchical decomposition data; integrating the hierarchical decomposition data of all regions to generate multi-scale characteristic data; S32, performing angle scanning on the multi-scale feature data at each level, calculating energy distribution features and extracting main directional features to generate directional feature data; S33, evaluating the directional feature data according to the preset hierarchical importance to obtain hierarchical weight data; performing weighted calculation on the directional feature data and the corresponding hierarchical weight data to obtain weighted feature data; and integrating the weighted feature data to generate comprehensive directional feature data; Step S4 is further as follows: S41, performing peak analysis and feature intensity sorting on the comprehensive directional feature data to obtain peak sorting data; determining the most significant directional feature based on the peak sorting data to generate main directional feature data; S42. Taking the main direction feature data as the reference direction, obtain reference direction data; calculating the distribution characteristics of the comprehensive direction feature data relative to the reference direction data, obtain direction distribution data; calculating the orientation degree parameter based on the direction distribution data, and generate orientation degree data.

2. The method for detecting the orientation of a PET film based on thermal conductivity gradient according to claim 1, wherein: Step S1 is further as follows: S11, obtaining the original temperature data of the PET film surface and multiplying it by the pre-calibrated temperature calibration coefficient to obtain the initial calibration data; then performing resampling and gridding processing to generate calibrated temperature data; S12, segmenting the calibrated temperature data according to the time series and performing smoothing processing to obtain smoothed temperature data and calculating its noise evaluation value; performing noise suppression processing on the smoothed temperature data according to the noise evaluation value to generate filtered temperature data; S13, obtaining ambient temperature data during the detection process, calculating the temperature fluctuation value, and obtaining ambient temperature fluctuation data; The ambient temperature fluctuation data at the corresponding moment is subtracted from the filtered temperature data to generate the compensated temperature data.

3. The method for detecting the orientation of a PET film based on a thermal conductivity gradient according to claim 2, wherein: Step S5 is further as follows: S51, collecting compensated temperature data, thermal diffusion characteristic data, comprehensive direction characteristic data and orientation degree data, calculating statistical characteristics and reliability evaluation indicators, and generating reliability indicator data; S52, performing correlation analysis on the orientation degree data and the reliability index data to obtain correction coefficient data; performing correction calculation on the orientation degree data and the correction coefficient data to generate final orientation degree data.

4. The method for detecting the orientation of a PET film based on thermal conductivity gradient according to claim 3, wherein: Step S11 is further as follows: S111, obtaining original temperature data of the PET film surface, performing validity check and eliminating abnormal values ​​to obtain pre-processed temperature data; performing data smoothing on the pre-processed temperature data to generate smoothed temperature data; S112, reading a pre-calibrated temperature sensor response curve, performing piecewise linearization processing, calculating correction parameters, and obtaining a temperature calibration coefficient; S113, matching the smoothed temperature data with the temperature calibration coefficient, performing temperature correction and quantization error compensation, and generating initial calibration data; S114, grouping and time-aligning the initial calibration data to obtain first aligned temperature data; Performing a resampling operation on the first aligned temperature data according to a preset sampling requirement to generate time calibration data; S115 , converting the time calibration data into spatial coordinate representation, performing spatial uniformity analysis and grid density optimization to obtain grid optimized data; performing spatial reconstruction based on the grid optimized data to generate calibrated temperature data.

5. The method for detecting the orientation of PET film based on thermal conductivity gradient according to claim 3, wherein: Step S12 is further as follows: S121, analyzing the time series characteristics of the calibrated temperature data, determining the optimal segment length, and generating segmented temperature data; S122, extracting local statistical features of the segmented temperature data, calculating smoothing parameters and performing smoothing processing to generate smoothed temperature data; S123, performing comparative analysis and statistical analysis on the smoothed temperature data and the segmented temperature data, calculating the noise level, and generating a noise evaluation value; S124. Based on the noise evaluation value, construct an adaptive threshold system and perform preliminary noise suppression on the smoothed temperature data to generate preliminary suppression data; S125, analyzing the residual noise of the preliminary suppression data, optimizing the suppression parameters and performing fine suppression processing to generate fine suppression data; S126 , evaluating the signal quality of the fine suppression data, performing final adjustment and edge feature preservation processing, and generating filtered temperature data.

6. The method for detecting the orientation of PET film based on thermal conductivity gradient according to claim 3, characterized in that: Step S21 is further as follows: S211. Based on the compensated temperature data, perform a time continuity check and time series padding to obtain padding temperature data; segment the padding temperature data according to a preset period to generate time series temperature data; S212: Analyze sampling characteristics of the time series temperature data, identify key time nodes, and perform sequence alignment to generate second aligned temperature data; S213, performing a difference operation on the second aligned temperature data at adjacent moments, detecting and correcting abnormal values, and generating temperature difference data; S214, obtaining a timestamp sequence corresponding to the second aligned temperature data, calculating the time interval and performing uniformity analysis to generate interval characteristic data; S215, matching the temperature difference data with the interval characteristic data to obtain matched difference data and normalizing the data to generate initial change rate data; S216, performing physical constraint check and change rate correction on the initial change rate data to obtain corrected change rate data; The corrected rate of change data is finally calibrated to generate temperature rate of change data.

7. The method for detecting the orientation of PET film based on thermal conductivity gradient according to claim 3, wherein: Step S33 is further as follows: S331. Perform feature consistency analysis and correlation analysis on the directional feature data at each level, evaluate the level stability, and generate level stability data; S332. Based on the hierarchical stability data, calculate and normalize the hierarchical discriminant coefficient, and generate hierarchical weight data in combination with the preset importance criterion; S333, grouping the directional feature data according to feature type, calculating the consistency index within the group, performing feature selection, and generating optimal feature data; S334: Perform feature matching on the preferred feature data and the hierarchical weight data to obtain feature matching data; Perform weighted fusion and weight optimization on feature pairing data to generate weighted feature data; S335, analyzing the redundancy of the weighted feature data, performing feature simplification, and obtaining simplified feature data; performing feature enhancement processing on the simplified feature data to generate enhanced feature data; S336. Perform multi-feature fusion operation and consistency verification on the enhanced feature data to obtain verification feature data; perform final feature integration based on the verification feature data to obtain comprehensive directional feature data.

Images