Prediction and control method of VOC release amount of low temperature degumming process of composite materials for vehicles

Abstract

The present application relates to a method for predicting and controlling VOC release amount of low-temperature degumming process of vehicle composite material. The core is to realize high space-time resolution collection of interface stress fluctuation and temperature field in process environment by embedding micro piezoelectric film sensor array in the interface micro zone of hemp fiber bundle and gum, combined with multi-resolution singular spectrum analysis to extract key feature anchor points of degumming mechanism, further using principal component reverse projection, probability atlas construction and segmented quadratic response surface model to divide the process parameter space by physical meaning, realizing multi-mechanism segmented modeling of VOC release amount and online updating of adaptive model of new material system. The obtained effect is to significantly improve the adaptability of low-temperature degumming process to material difference, realize real-time prediction and dynamic precise control of VOC release in the degumming process of vehicle composite material, reduce environmental pollution and improve process efficiency and stability.

Description

Technical Field

[0001] This invention relates to the field of low-temperature degumming process of automotive composite materials and volatile organic compound (VOC) emission modeling technology, and in particular to a method for predicting and controlling VOC release from the low-temperature degumming process of automotive composite materials. Background Technology

[0002] With the development of new energy vehicles and the green transportation industry, the application of automotive composite materials is becoming increasingly widespread, especially environmentally friendly composite materials represented by hemp fiber reinforced materials. Low-temperature degumming is a crucial processing step in the preparation of automotive composite materials, and the emission of volatile organic compounds (VOCs) during this process has become a key indicator affecting the safety of the vehicle's interior environment and the compliance of the manufacturing process. Existing technologies mainly employ response surface methodology (RSM) based on process parameters—such as degumming temperature, time, and degumming agent concentration—to predict and control VOC release during the degumming process. Typical methods include traditional RSM and statistical regression modeling, which extract key influencing factors through experimental design and parameter scanning, and construct corresponding mathematical relationships.

[0003] In recent years, the industry has placed higher demands on the prediction accuracy, modeling efficiency, and generalization ability of VOC release. Existing technologies generally rely on combinations of static process parameters as modeling inputs, often employing manual or correlation-based feature selection methods. These methods fail to efficiently identify parameter combinations that significantly impact VOC release under the complex mechanical and thermal interaction mechanisms of the fiber-matrix interface. Due to a lack of in-depth understanding of the dynamic evolution of degumming mechanisms, the models have numerous and complex input variables, resulting in limited predictive capabilities and high consumption of experimental and computational resources. Furthermore, some studies have attempted to incorporate auxiliary techniques such as molecular simulation or image processing, but these all suffer from high technical barriers, cumbersome operating procedures, or massive data requirements, making large-scale industrial application difficult.

[0004] Therefore, there is an urgent need for innovative VOC emission prediction and control modeling methods that can overcome the limitations of static parameter-dependent modeling, achieve quantifiable and traceable feature anchoring of the dynamic evolution process of interface residual stress, improve the feature selection efficiency, prediction accuracy and interpretability of the model through mechanism-driven piecewise response surface modeling, significantly reduce experimental and computational resource consumption, and realize intelligent dynamic adaptation of materials and processes. Summary of the Invention

[0005] This application provides a method for predicting and controlling VOC release during the low-temperature degumming process of automotive composite materials, aiming to solve the problem mentioned in the background art of the inability to efficiently identify parameter combinations that significantly affect VOC release under complex interfacial mechanics and thermal interaction mechanisms.

[0006] The method for predicting and controlling VOC emissions from the low-temperature degumming process of automotive composite materials provided in this application specifically includes:

[0007] S1: In the experimental environment of low-temperature degumming process of automotive composite materials, the local stress fluctuation signal sequence and the corresponding temperature field time series data of the interface were obtained at the micro-region of the interface between the hemp fiber bundle and the adhesive.

[0008] S2: Process the local stress fluctuation signal sequence of the interface and the time series data of the temperature field in the corresponding region to obtain the set of stress evolution trajectory anchor points;

[0009] S3: Construct a high-dimensional trajectory feature vector based on the set of stress evolution trajectory anchor points, and process the high-dimensional trajectory feature vector to obtain the anchor point reachability probability map;

[0010] S4: Based on the probability map of the achievability of the anchor points, the full parameter space is divided into subdomains of stress relief dominance region, interface decoupling active region and colloidal volatilization window region, and a piecewise quadratic response surface model is independently constructed in each subdomain.

[0011] S5: After a new batch of materials enters the degumming process, the matching degree between the measured stress evolution trajectory anchor point and the historical anchor point attainability probability map is compared in real time, the deviation threshold between the measured trajectory and the standard trajectory is calculated, and it is determined whether the stress evolution law of the current material system has drifted.

[0012] S6: If the deviation threshold is determined to exceed the preset tolerance range, the coefficients of the piecewise quadratic response surface model are recalibrated.

[0013] S7: Based on the updated piecewise quadratic response surface model, it outputs the optimal degumming temperature, degumming time, and degumming agent concentration control commands, driving the low-temperature degumming equipment to perform dynamic process adjustment actions, and completing the accurate prediction and control of VOC release.

[0014] The method for predicting and controlling VOC emissions during the low-temperature degumming process of automotive composite materials provided in this application has the following beneficial effects:

[0015] This application enables real-time comparison of the measured anchor point trajectory with historical maps during the processing of new batches of materials. Once a deviation exceeding a set threshold is detected, it automatically triggers the re-division of subdomain boundaries and adaptive updates of local response surface model parameters, forming a closed-loop optimization mechanism of "perception-modeling-feedback-correction." This ensures that the model always conforms to the actual evolution path of the current material system. This mechanism avoids the cumbersome process of frequent recalibration or offline training required by traditional models, significantly enhancing the robustness and timeliness of the system, and shifting the control of the entire degumming process from passive response to active adaptation. In summary, this solution upgrades response surface analysis from a traditional empirical statistical tool to an intelligent decision engine with mechanism-guided capabilities by constructing an integrated framework of "observable trajectory—analyzable map—updateable model." This not only improves modeling efficiency and prediction accuracy but also provides a replicable and scalable technical paradigm for the processing of green, low-carbon, bio-based materials. Attached Figure Description

[0016] Figure 1 This is the main flowchart of the method for predicting and controlling VOC emissions in the low-temperature degumming process of automotive composite materials.

[0017] Figure 2 This is a sub-flowchart of a method for predicting and controlling VOC emissions from the low-temperature degumming process of automotive composite materials.

[0018] Figure 3 This is another sub-flowchart of the method for predicting and controlling VOC release during the low-temperature degumming process of automotive composite materials. Detailed Implementation

[0019] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0020] The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. Of course, these are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0021] like Figure 1 As shown, this application provides a method for predicting and controlling VOC emissions during the low-temperature degumming process of automotive composite materials, specifically including:

[0022] S1: In the experimental environment of low-temperature degumming process of automotive composite materials, the local stress fluctuation signal sequence and the corresponding temperature field time series data of the interface were obtained at the micro-region of the interface between the hemp fiber bundle and the adhesive.

[0023] S2: Process the local stress fluctuation signal sequence of the interface and the time series data of the temperature field in the corresponding region to obtain the set of stress evolution trajectory anchor points;

[0024] S3: Construct a high-dimensional trajectory feature vector based on the set of stress evolution trajectory anchor points, and process the high-dimensional trajectory feature vector to obtain the anchor point reachability probability map;

[0025] S4: Based on the probability map of the achievability of the anchor points, the full parameter space is divided into subdomains of stress relief dominance region, interface decoupling active region and colloidal volatilization window region, and a piecewise quadratic response surface model is independently constructed in each subdomain.

[0026] S5: After a new batch of materials enters the degumming process, the matching degree between the measured stress evolution trajectory anchor point and the historical anchor point attainability probability map is compared in real time, the deviation threshold between the measured trajectory and the standard trajectory is calculated, and it is determined whether the stress evolution law of the current material system has drifted.

[0027] S6: If the deviation threshold is determined to exceed the preset tolerance range, the coefficients of the piecewise quadratic response surface model are recalibrated.

[0028] S7: Based on the updated piecewise quadratic response surface model, it outputs the optimal degumming temperature, degumming time, and degumming agent concentration control commands, driving the low-temperature degumming equipment to perform dynamic process adjustment actions, and completing the accurate prediction and control of VOC release.

[0029] Step S1: In the experimental environment of the low-temperature degumming process for automotive composite materials, the local stress fluctuation signal sequence and the corresponding temperature field time series data of the micro-region at the interface between the hemp fiber bundle and the adhesive are acquired. Specifically, this includes:

[0030] S1.1: Based on the geometric constraints and thermodynamic boundary conditions of the experimental environment for low-temperature degumming process of automotive composite materials, the micro-region position of the hemp fiber bundle and the adhesive interface is calibrated in three dimensions, and a micro-region spatial topology mapping map containing parameters such as interface curvature radius, fiber orientation angle and adhesive layer thickness distribution is generated, which serves as the spatial reference for sensor array deployment.

[0031] The geometric dimensions and thermodynamic boundary conditions of the experimental chamber for low-temperature degumming of automotive composite materials were obtained as the basic constraint data for micro-area calibration.

[0032] Based on the microscopic morphological characteristics of the interface between hemp fiber bundles and gelatinous material, a three-dimensional optical profilometer is used to perform non-contact scanning of the interface region to generate high-density point cloud data.

[0033] The point cloud data is subjected to denoising and smoothing processes to remove outliers introduced by environmental vibrations and reconstruct a continuous three-dimensional surface model of the interface.

[0034] In the reconstructed three-dimensional surface model, the overlapping areas of the fiber axial and radial directions are identified, the normal vector distribution of each sampling point is calculated, and the extreme points of the interface curvature radius are determined.

[0035] The angle between the fiber principal axis direction vector and the reference coordinate system is extracted, and the fiber orientation angle parameter is calculated to characterize the spatial arrangement of the fiber in the matrix.

[0036] By combining microscopic section image analysis technology, the thickness distribution of the gelatinous layer at different locations is measured, and a gelatinous layer thickness gradient mapping matrix is ​​generated.

[0037] Spatial registration was performed on the interface curvature radius, fiber orientation angle, and gel layer thickness distribution parameters to construct a micro-region spatial topology mapping map under a unified coordinate system.

[0038] By using the micro-area spatial topology mapping map, the geometric and physical parameters from the previous step are transformed into a spatial reference for sensor deployment, enabling precise positioning of stress concentration areas.

[0039] S1.2: Using the micro-area spatial topology mapping, an embedded installation operation of a micro piezoelectric thin film sensor array is performed at the micro-area position of the interface between the hemp fiber bundle and the gel. The piezoelectric sensitive unit is attached to the interface stress concentration area by a micron-level positioning fixture, generating a sequence of original piezoelectric charge signals that can respond in real time to local stress fluctuations at the interface, which serves as the primary electrical signal output for sensing the mechanical state of the interface.

[0040] The interface curvature radius and fiber orientation angle data are read from the micro-area spatial topology mapping map to locate the coordinates of the stress concentration region at the interface between the hemp fiber bundle and the gelatinous layer. Based on the coordinates of the stress concentration region, motion control commands of a micron-level multi-axis positioning fixture are invoked to drive the piezoelectric sensing unit to approach the interface micro-area along the normal direction. At the instant the piezoelectric sensing unit contacts the interface, a preset micro-Newton level constant preload is applied to ensure a gapless physical bond between the sensor and the gelatinous layer surface. The contact impedance change during the bonding process is monitored. When the impedance value stabilizes within a set threshold range, the positioning fixture posture is locked, completing the embedded installation of the micro piezoelectric thin film sensor array. The data acquisition module is activated to capture the piezoelectric charge response excited by the microscopic deformation of the interface caused by the low-temperature gradient cooling at a high sampling rate. The captured raw charge signal is initially filtered to remove high-frequency electromagnetic interference noise and retain the low-frequency stress fluctuation component reflecting the change in the interface mechanical state. A raw piezoelectric charge signal sequence containing timestamp information is generated as a primary electrical signal output characterizing the dynamic evolution of local stress at the interface.

[0041] By using micron-level precise positioning and constant force bonding processing, the spatial topological reference of the previous step is transformed into a high-fidelity original piezoelectric charge signal sequence, achieving the expected technical effect of real-time and accurate perception of the interface mechanical state.

[0042] S1.3: Based on the impedance matching requirements of the original piezoelectric charge signal sequence, the charge amplification circuit connected to the output end of the piezoelectric thin film sensor array is subjected to gain calibration and frequency band shaping to generate a standardized interface local stress fluctuation voltage signal sequence with uniform dimensions and a signal-to-noise ratio that meets the requirements of dynamic monitoring, which serves as the standard mechanical input data for subsequent time series analysis.

[0043] The system receives the raw piezoelectric charge signal sequence generated by S1.2. This sequence has high impedance source characteristics and is susceptible to electromagnetic interference. A low-noise charge amplifier circuit is constructed to convert the high-impedance charge output of the piezoelectric thin-film sensor into a low-impedance voltage signal. A high-input-impedance operational amplifier is configured as a preamplifier stage, with its input impedance set to be greater than 10Ω. 12 Ohms, to minimize the signal source load effect.

[0044] The feedback capacitor of the charge amplifier circuit is precisely selected, and the gain factor is determined based on the sensitivity coefficient of the piezoelectric sensor to ensure that the weak stress fluctuation signal is linearly amplified to the optimal dynamic range of the data acquisition card. A parallel feedback resistor is introduced to provide a DC bias path to prevent amplifier saturation, and the time constant is set to be much larger than the maximum duration of the degumming process to ensure the complete preservation of the low-frequency stress drift signal.

[0045] Frequency band shaping was implemented, and a bandpass filter was designed to eliminate low-frequency environmental vibration noise and high-frequency radio frequency interference. The high-pass cutoff frequency was set to 0.1Hz to filter out baseline drift caused by slow temperature changes; the low-pass cutoff frequency was set to 10kHz to retain the high-frequency stress wave components at the moment of microcrack initiation. A Butterworth filter topology was employed to ensure the flatness of the amplitude-frequency response within the passband and to avoid signal phase distortion.

[0046] Perform a gain calibration operation by injecting a standard sine wave test signal and adjusting the amplifier gain coefficient to establish a linear mapping relationship between the output voltage and the known standard stress value.

[0047] The amplified voltage signal is subjected to anti-aliasing filtering before analog-to-digital conversion, and the sampling rate is set to more than 5 times the highest frequency of the signal to satisfy the Nyquist sampling theorem. A standardized interface local stress fluctuation voltage signal sequence with uniform dimension volts (V) and a signal-to-noise ratio higher than 60dB is generated.

[0048] By using charge amplification and frequency band shaping, the high-impedance original charge signal from the previous step is transformed into low-noise, wide-dynamic-range standardized voltage time-series data, enabling precise quantitative perception of the stress evolution in the micro-regions of the hemp fiber-gel interface, and providing a high-fidelity mechanical input benchmark for subsequent multi-resolution singular spectrum analysis.

[0049] S1.4: Simultaneously start the high-speed infrared thermal imaging acquisition system, and perform full-frame-rate thermal radiation capture operation on the same micro-region of the hemp fiber bundle and gel interface with the deployed sensor array. Generate the original grayscale image sequence of the temperature field of the corresponding region that is strictly timestamped with the standardized interface local stress fluctuation voltage signal sequence, as the primary optical data output for interface thermal state perception.

[0050] The hardware trigger interface of the high-speed infrared thermal imaging acquisition system is invoked. Based on the sampling clock reference of the standardized interface local stress fluctuation voltage signal sequence generated by S1.3, the exposure time and frame rate parameters of the image sensor are configured to ensure that the optical data acquisition and mechanical signal acquisition are strictly synchronized on the microsecond time scale.

[0051] The refrigeration unit is started to execute a gradient cooling program, while the optical shutter correction mechanism of the infrared thermal imaging lens is activated. Real-time full-frame-rate thermal radiation capture is performed on the micro-area location of the hemp fiber bundle and the gel interface to eliminate environmental background radiation interference and generate an original grayscale image data stream that reflects the micro-temperature distribution changes of the interface.

[0052] Hardware timestamp marking is performed on the acquired raw grayscale image data stream to accurately map the acquisition time of each frame image to the time axis coordinates of the standardized interface local stress fluctuation voltage signal sequence, and establish a spatiotemporal alignment index for the mechanical-thermal dual-mode data to ensure the accurate correspondence between stress abrupt change points and temperature field evolution points in subsequent analysis.

[0053] During the isothermal holding phase, a high frame rate acquisition mode is continuously maintained to record the transient thermal response characteristics of the gel layer during the residual stress release process. Buffer queue management technology is used to prevent data frame loss and generate a sequence of original grayscale images of the temperature field of the corresponding region that covers the entire degumming cycle and is strictly aligned with the timestamp of the mechanical signal.

[0054] By using the above-mentioned synchronous acquisition and spatiotemporal alignment processing method, the mechanical monitoring results of the previous step are transformed into thermal image data with precise time correlation, realizing the primary optical data output for interface thermal state perception, and providing a high-fidelity multiphysics coupling input benchmark for subsequent multi-resolution singular spectrum analysis.

[0055] S1.5: Based on the radiometric calibration parameters of the original grayscale image sequence of the temperature field in the corresponding region, perform non-uniformity correction and blackbody temperature inversion processing to convert grayscale values ​​into absolute temperature values ​​and reconstruct them into a two-dimensional matrix. Generate time series data of the temperature field in the corresponding region that is spatiotemporally and completely registered with the standardized interface local stress fluctuation voltage signal sequence, as the final fusion sensing dataset for multi-resolution singular spectrum analysis.

[0056] The system acquires the original grayscale image sequence and the factory-calibrated non-uniformity correction coefficient matrix output by the high-speed infrared thermal imaging acquisition system. It performs pixel-level offset compensation on each frame of original grayscale data to eliminate fixed pattern noise caused by the inconsistency of the response of each pixel in the focal plane array, and generates a corrected linear grayscale response matrix.

[0057] Based on Planck's blackbody radiation law, a nonlinear mapping relationship between grayscale values ​​and absolute temperature is established. By calling the preset ambient atmospheric transmittance parameters and emissivity correction factors, pixel-by-pixel inversion calculations are performed on the corrected linear grayscale response matrix to determine the absolute temperature distribution.

[0058] The absolute temperature values ​​obtained by inversion are reorganized into a two-dimensional temperature distribution matrix according to the spatial topology of the original image, ensuring that the matrix row and column indices correspond one-to-one with the physical coordinates in the micro-area spatial topology mapping map, thus forming a snapshot of the spatial distribution of the temperature field with clear geometric meaning.

[0059] The timestamp index of the standardized interface local stress fluctuation voltage signal sequence is read, and the two-dimensional temperature distribution matrix sequence is subjected to time axis alignment interpolation to fill the time gap caused by the difference in acquisition frequency, generating temperature field time series data that is strictly synchronized with the sampling time of the stress signal.

[0060] By using blackbody temperature inversion and non-uniformity correction, the original optical grayscale data from the previous step is transformed into a temperature field evolution matrix with physical dimensions and precise spatiotemporal registration. This enables the synchronous quantitative characterization of the interface thermal and mechanical states, providing high-fidelity multimodal fusion input data for subsequent multi-resolution singular spectrum analysis.

[0061] Step S2: Process the local stress fluctuation signal sequence at the interface and the corresponding regional temperature field time series data to obtain a set of stress evolution trajectory anchor points. Specifically, this includes:

[0062] S2.1: The interface local stress fluctuation signal sequence and the corresponding regional temperature field time series data are preprocessed. The continuously acquired raw data stream is divided into multiple overlapping time window segments using the sliding window truncation method to generate an initial signal segment set with time-frequency localization characteristics, providing standardized data input units for subsequent multi-scale decomposition.

[0063] S2.2: Perform multi-resolution singular spectral decomposition based on the initial signal segment set. By constructing a trajectory matrix and performing singular value decomposition, the mixed signal is separated into several principal component components with different oscillation frequencies and energy distributions, thereby generating a sequence of singular spectral components that characterize the contribution of different physical mechanisms, and achieving the initial separation of noise background and effective mechanism signal.

[0064] For the initial signal segment set with time-frequency localization characteristics generated by S2.1, perform multi-resolution singular spectrum decomposition to separate noise from effective mechanism signals.

[0065] For each time window segment, a standardized interface local stress fluctuation voltage signal sequence is constructed by defining the embedding dimension L and the window length W. This matrix preserves the temporal dependency structure of the signal by mapping the one-dimensional time series to a multi-dimensional phase space vector.

[0066] Singular value decomposition (SVD) is performed on the constructed trajectory matrix X to calculate the eigenvalues ​​and eigenvectors of its covariance matrix. The mixed signal is then projected onto an orthogonal basis through decomposition to extract principal component components representing different oscillation frequencies and energy distributions.

[0067] Based on the magnitude of the singular values, the components corresponding to the top r dominant singular values ​​are selected as the effective signal subspace, and the remaining components are assigned to the noise subspace. The decomposed matrix elements are then recombined into a one-dimensional time series using a diagonal averaging method, generating a sequence of singular spectral components representing the contributions of different physical mechanisms.

[0068] By using multi-resolution singular spectrum decomposition, the initial signal segment from the previous step is transformed into a singular spectrum component sequence with a significantly improved signal-to-noise ratio, thus achieving preliminary separation between the noise background and the effective mechanism signal reflecting the debonding mechanism transition.

[0069] S2.3: Apply the change point detection method to the singular spectral component sequence, calculate the second-order difference statistics of each component on the time axis to capture the energy change position, identify key turning points such as the initial stress relaxation start point, the critical point of microcrack initiation, the acceleration point of gel layer peeling and the precursor point of VOC volatilization peak, and generate a timestamp index list marked with degumming mechanism transition events.

[0070] The system receives the singular spectral component sequence after singular spectral decomposition and uses it as the input data source for the change point detection method.

[0071] For each Vickers heterospectral component sequence, a sliding time window is constructed, the local mean and variance of the data within the window are calculated, trend term interference is eliminated, and detrended stationary signal segments are generated.

[0072] A second-order difference operation is performed on the detrended signal segment. A dynamic threshold judgment criterion is set. When the absolute value of the second-order difference statistic exceeds three times the standard deviation of the preset noise baseline, the moment is marked as a potential energy mutation point.

[0073] By combining physical mechanism constraints, cluster analysis is performed on potential energy mutation points to remove isolated noise points and retain significant mutation clusters within continuous time periods.

[0074] Based on the temporal distribution characteristics of the mutation clusters, and matching the degumming process stages, the starting time of the first significant mutation cluster is marked as the initial stress relaxation starting point.

[0075] The moment when the stress amplitude first shows a step increase is marked as the critical point for microcrack initiation.

[0076] The abrupt change in temperature slope accompanied by high-frequency stress oscillations is marked as the acceleration point of gel layer peeling.

[0077] The last high-energy abrupt change before the stress signal energy rapidly decays is marked as the precursor point of the VOC volatile peak.

[0078] By integrating the key turning points identified above, a list of debonding mechanism transition events containing timestamp indexes and corresponding event types is generated.

[0079] By using change point detection and second-order difference statistics calculation, the singular spectral component sequence is transformed into a timestamp index list marked with debonding mechanism transition events, thus achieving accurate extraction from continuous signals to key points of discrete mechanisms.

[0080] S2.4: Extract the original signal feature values ​​at the corresponding time according to the timestamp index list, read the instantaneous peak value in the local stress fluctuation signal sequence of the interface as the stress amplitude, calculate the derivative of the temperature field time series data of the corresponding region at that time as the temperature slope, and count the time interval between two adjacent key turning points as the duration, thereby synthesizing a candidate anchor point dataset containing a triplet of stress amplitude, temperature slope and duration.

[0081] The system receives a time-stamped index list generated by S2.3, which includes the specific location markers on the time axis of the initial stress relaxation initiation point, the critical point of microcrack initiation, the acceleration point of gel layer peeling, and the precursor point of VOC volatilization peak.

[0082] Based on each key turning point in the timestamp index list, the corresponding data sampling point is located in the standardized interface local stress fluctuation voltage signal sequence, and the instantaneous voltage peak at that moment is extracted as a direct measure of the mechanical response intensity.

[0083] The extracted instantaneous voltage peak value is substituted into the piezoelectric sensor sensitivity calibration coefficient, and the true local stress amplitude of the interface at that moment is calculated through linear transformation, thus eliminating the dimensional deviation caused by the sensor gain difference.

[0084] The two-dimensional temperature distribution data at the same time stamp in the time series data of the temperature field in the corresponding area are read synchronously, and the average temperature value in the micro-area covered by the sensor is selected as the reference thermal state parameter.

[0085] The central difference method was used to calculate the rate of temperature change of adjacent time steps before and after the given moment, and the derivative function of temperature with respect to time was constructed to quantitatively characterize the thermal shock intensity and heat transfer rate during the degumming process.

[0086] The temperature slope is calculated using the following formula:

[0087]

[0088] in, For temperature slope, Absolute temperature For time, Index for the current time step, This represents the sampling time interval.

[0089] Traverse the timestamp index list, calculate the time difference between the current critical turning point and the previous critical turning point, and define the time difference as the duration of the current degumming mechanism stage, reflecting the speed of the physical process evolution.

[0090] The calculated stress amplitude, temperature slope, and duration are concatenated into vectors to form a triplet data structure describing a single degumming mechanism transition event.

[0091] Repeat the above extraction and calculation operations for all identified key turning points, assemble all triples in chronological order, and generate a candidate anchor point dataset containing complete dynamic features of the degumming cycle.

[0092] By using multi-source signal spatiotemporal alignment and differential feature extraction processing, the time index identified in the previous step is transformed into multi-dimensional physical feature data containing mechanical strength, thermal shock rate and evolution duration. This realizes the transformation from discrete time points to a description of continuous physical processes, providing standardized mechanism quantification input for the subsequent construction of high-dimensional trajectory feature vectors.

[0093] S2.5: Based on the candidate anchor point dataset, perform physical consistency verification and formatting encapsulation, remove abnormal data points that do not conform to thermodynamic constraints, and arrange the verified triplet data in a structured manner according to the degumming cycle order to finally generate a set of stress evolution trajectory anchor points used to characterize the dynamic evolution of the degumming process, thus completing the transformation from raw sensing data to mechanism feature anchor points.

[0094] like Figure 2 As shown, step S3 involves constructing a high-dimensional trajectory feature vector based on the set of stress evolution trajectory anchor points, and processing the high-dimensional trajectory feature vector to obtain an anchor point reachability probability map. Specifically, this includes:

[0095] S3.1: Based on the initial stress relaxation start point, microcrack initiation critical point, adhesive layer peeling acceleration point and VOC volatilization peak precursor point in the set of stress evolution trajectory anchor points, extract the appearance order code, relative time offset value and intensity attenuation ratio coefficient of each anchor point in the complete degumming cycle, and perform multi-dimensional feature splicing operation to construct a high-dimensional trajectory feature vector containing time series topology and mechanical response amplitude information.

[0096] Receive the set of stress evolution trajectory anchor points generated by step S2. This set includes the timestamp indexes of the initial stress relaxation start point, the critical point of microcrack initiation, the acceleration point of gel layer peeling, and the precursor point of VOC volatilization peak, as well as the corresponding mechanical and thermal triplet data.

[0097] Normalization is performed on the timestamp index in the anchor point set. Using the total duration of the degumming process as the base time unit, the relative temporal position of each key inflection point with respect to the starting time is calculated, and a numerical sequence of temporal offsets representing the timing of events is generated.

[0098] The stress amplitude change rate between two adjacent key inflection points is extracted. By calculating the ratio of the peak stress at the later anchor point to the peak stress at the previous anchor point, the energy attenuation characteristics during the interface damage accumulation process are quantified, and the strength attenuation ratio coefficient is generated.

[0099] The order in which each anchor point appears during the degumming cycle is encoded by integers. The initial stress relaxation start point is defined as the first sequence position, the microcrack initiation critical point is defined as the second sequence position, and so on, to construct a sequence encoding vector that reflects the evolution logic of the degumming mechanism.

[0100] The time series offset value, intensity attenuation ratio coefficient and sequential encoding vector are concatenated into multi-dimensional features. The Z-score normalization method is used to eliminate the numerical differences between different dimensions, and a high-dimensional trajectory feature vector containing information on time series topology and mechanical response amplitude is constructed.

[0101] Through the above feature extraction and splicing processing, discrete mechanism anchor points are transformed into structured high-dimensional trajectory feature vectors, realizing the digital representation of the dynamic evolution path of the degumming process and providing standardized input data for subsequent principal component inverse projection.

[0102] For example, in a low-temperature degumming experiment on a batch of hemp fibers, the total duration was set to 120 minutes. The initial stress relaxation initiation point was identified at minute 15, the microcrack initiation critical point at minute 45, the accelerated peeling point of the gelatinous layer at minute 80, and the precursor point of VOC volatilization at minute 105. The relative time offsets were calculated to be 0.125, 0.375, 0.667, and 0.875, respectively. The extracted stress amplitudes were 10 MPa, 25 MPa, 40 MPa, and 35 MPa, and the strength attenuation ratio coefficient was calculated. For example, the ratio from microcrack to peeling acceleration point was 40 / 25 = 1.6, and the ratio from peeling acceleration to VOC precursor was 35 / 40 = 0.875. The sequential encoding vector was [1, 2, 3, 4]. The temporal offset [0.125, 0.375, 0.667, 0.875], intensity attenuation ratio [1.6, 0.875], and sequential encoding [1, 2, 3, 4] are concatenated and normalized to generate a 10-dimensional high-dimensional trajectory feature vector. This vector accurately depicts the stress evolution trajectory under the current process conditions, significantly improving the completeness of the feature representation.

[0103] S3.2: Perform principal component analysis to reduce the dimensionality of the high-dimensional trajectory feature vector, calculate the covariance relationship between each principal component load matrix and the original process parameter space, and use the linear algebraic projection method to generate principal component projection components that characterize the sensitivity of key mechanical events to temperature, time and concentration, so as to eliminate feature redundancy and retain the core variation information that dominates the evolution of the degumming mechanism.

[0104] Receive a high-dimensional trajectory feature vector matrix constructed by step S3.1, which includes the occurrence order encoding, relative temporal offset, and intensity attenuation ratio coefficient. The number of rows in this matrix corresponds to the number of experimental samples, and the number of columns corresponds to the feature dimension.

[0105] Zero-mean processing is performed on the high-dimensional trajectory feature vector matrix. The arithmetic mean of each column of feature variables is calculated, and the original data is subtracted from the corresponding mean to eliminate the interference of dimensional differences on covariance calculation and generate a centered feature matrix.

[0106] The covariance matrix is ​​calculated based on the centered feature matrix. The linear correlation between the characteristics of different mechanical events is quantified by matrix transpose and self-multiplication, resulting in a symmetric positive definite covariance matrix, which reveals the coupling effect between temperature, time and concentration parameters.

[0107] The covariance matrix is ​​decomposed into eigenvalues, and the characteristic equation is solved to obtain the eigenvalues ​​and their corresponding eigenvectors. The magnitude of the eigenvalues ​​represents the amount of information carried by the principal components, and the eigenvectors define the direction of the new coordinate axes.

[0108] Based on the descending order of the eigenvalues, the top few eigenvectors with cumulative contribution rates exceeding a preset threshold are selected to form a projection matrix. Redundant noise components are eliminated, and the core variation information that dominates the evolution of the degumming mechanism is retained.

[0109] By using a projection matrix to map the centered feature matrix to a low-dimensional principal component space, principal component projection components that characterize the sensitivity of key mechanical events to process parameters are generated, thus achieving dimensionality reduction transformation from high-dimensional mechanism features to low-dimensional process space.

[0110] Principal component analysis (PCA) is used to reduce the dimensionality of the high-dimensional trajectory features from the previous step. This transforms them into low-dimensional principal component projections, eliminating feature redundancy while retaining core variation information. This provides simplified and high signal-to-noise ratio input data for subsequent inverse mapping to the original process parameter space.

[0111] For example, the high-dimensional trajectory feature vector matrix containing 12 features (4 anchor point triplets) is standardized with 50 samples. After calculating the covariance matrix, the cumulative contribution rate of the first three principal components obtained by eigenvalue decomposition is 92.5%. The first three eigenvectors are selected to construct a 3×12 projection matrix, mapping the original 12-dimensional data to 3-dimensional principal component scores. The first principal component mainly reflects the positive correlation between stress amplitude and degumming temperature; the second principal component reflects the negative correlation between duration and degumming agent concentration; and the third principal component characterizes the interaction between temperature slope and time. The dimensionality-reduced data retains most of the mechanical evolution information, significantly reducing the computational complexity of subsequent inverse mapping and improving the model's sensitivity to changes in process parameters.

[0112] S3.3: Construct an inverse mapping function based on the principal component projection components, and solve the dimensionality-reduced mechanical feature space coordinates in reverse to the three-dimensional original process parameter space composed of degumming temperature, degumming time, and degumming agent concentration. Calculate the likelihood estimate of the occurrence of the anchor point of the specific stress evolution trajectory corresponding to each process parameter combination point, and generate the original reachability probability distribution data characterizing the possibility of mechanical mechanism triggering under different process conditions.

[0113] It receives the coordinate data of the mechanical feature space after dimensionality reduction by principal component analysis and the covariance matrix of the original process parameter space as the input basis for the inverse mapping calculation.

[0114] A Bayesian inference-based inverse probability mapping model is constructed, defining degumming temperature, degumming time, and degumming agent concentration as three-dimensional random variable vectors, and treating the dimensionality-reduced principal component projection components as observational evidence.

[0115] The Gaussian kernel density estimation method is used to calculate the conditional probability density function of each process parameter combination point under the given principal component projection components, and to establish a nonlinear mapping relationship from the mechanical characteristic space to the process parameter space.

[0116] The discretized three-dimensional mesh space consisting of degumming temperature, degumming time and degumming agent concentration is traversed, and likelihood estimation is performed on each mesh node to generate the corresponding probability value.

[0117] The calculated probability values ​​are filled into a three-dimensional array to form the original accessibility probability distribution data that characterizes the possibility of mechanical mechanism triggering under different process conditions.

[0118] By using Bayesian inverse mapping and gridded likelihood estimation, the principal component projections from the previous step are transformed into probability distribution data in the three-dimensional process parameter space, achieving the expected technical effect of quantitative mapping from abstract mechanical features to specific process feasibility.

[0119] S3.4: Perform spatial interpolation smoothing and boundary constraint optimization on the original accessibility probability distribution data, eliminate the probability aberration regions caused by experimental noise, integrate the likelihood estimates of adjacent process parameter points to form a continuous probability density surface, and output an anchor point accessibility probability map with clear physical confidence boundaries, which serves as the quantitative basis for subsequent division of the stress relief-dominant region, the interface decoupling active region, and the colloid volatilization window region.

[0120] Receive the raw reachability probability distribution data output from step S3.3, which includes the likelihood estimate of the occurrence of stress evolution trajectory anchor points corresponding to discrete process parameter points.

[0121] The original discrete probability data is processed by Kriging interpolation to construct a spatial covariance function to quantify the correlation between adjacent process parameter points, and a preliminary model of a continuous probability density field covering the entire parameter space is generated.

[0122] For the local probability anomalous jump regions caused by experimental noise in the preliminary model, a Gaussian kernel smoothing filter is applied for convolution operation, and the smoothing bandwidth parameter is set to suppress high-frequency noise interference while preserving the global trend characteristics of the probability distribution.

[0123] By introducing physical confidence boundary constraints and based on the second law of thermodynamics and the activation energy threshold of hemp fiber degumming reaction, invalid data points whose probability density exceeds the boundary of the physical feasible region are eliminated, and the edge distortion of the probability surface is corrected.

[0124] By integrating the likelihood estimates of adjacent process parameter points after smoothing and boundary optimization, the probability density surface in the three-dimensional process parameter space is reconstructed to ensure the continuity and differentiability of the surface in the dimensions of degumming temperature, time, and concentration.

[0125] The output is an anchor point attainability probability map with clear physical confidence boundaries. This map intuitively represents the probability distribution of specific stress evolution mechanisms triggered under different process combinations, and serves as the quantitative basis for subdomain division in step S4.

[0126] By using spatial interpolation smoothing and physical boundary constraint optimization, the discrete likelihood estimate from the previous step is transformed into continuous probability density surface data that conforms to physical laws. This achieves the transformation from statistical probability to mechanism-driven process zoning benchmark, significantly improving the region division accuracy of subsequent response surface modeling.

[0127] Step S4: Based on the anchor point achievability probability map, the full parameter space is divided into subdomains: a stress relief-dominant region, an interface decoupling-active region, and a colloid volatile window region. Within each subdomain, a piecewise quadratic response surface model is independently constructed. Specifically, this includes:

[0128] S4.1: Based on the probability density surface distribution characteristics in the anchor point achievability probability map, contour tracing and gradient mutation detection processing are performed on the three-dimensional original process parameter space composed of degumming temperature, degumming time and degumming agent concentration to identify the boundary coordinate set where the probability density has a significant transition, and generate a subdomain partitioning threshold vector that defines the physical boundaries of the stress relief dominant area, the interface decoupling active area and the gum volatilization window area.

[0129] Receive the anchor point attainability probability map generated in step S3, which is a continuous probability density function surface defined in the three-dimensional space of degumming temperature, degumming time and degumming agent concentration.

[0130] Discretized mesh sampling is performed on the probability density surface to construct a three-dimensional matrix data containing the probability values ​​corresponding to the combination points of each process parameter, which serves as the basic data source for boundary detection.

[0131] The central difference method is used to calculate the partial derivatives of each sampling point in the three dimensions of temperature, time and concentration in the three-dimensional matrix, and to generate a gradient vector field representing the rate of change of probability density.

[0132] Based on the magnitude distribution of the gradient vector field, regions where the probability density changes drastically are identified, and a set of spatial coordinate points with high gradient magnitudes is selected by setting a gradient threshold.

[0133] Connectivity analysis is performed on the selected set of high-gradient coordinate points. The marching cubes method is used to trace isosurfaces and extract spatial curves connecting adjacent high-gradient points.

[0134] The extracted spatial curves are projected onto a two-dimensional process parameter plane to form closed or semi-closed geometric contour lines, thus initially defining the topological boundaries of regions dominated by different physical mechanisms.

[0135] The initially defined boundary contour lines are smoothed to eliminate jagged distortions caused by experimental noise, generating smooth and continuous subdomain boundaries.

[0136] Calculate the exact coordinates of the boundaries of each subdomain in the three-dimensional parameter space and encode them into structured data containing temperature thresholds, time thresholds, and concentration thresholds.

[0137] Through the above contour tracing and gradient mutation detection processing, the probability spectrum of the previous step is transformed into a subdomain partitioning threshold vector that defines the physical boundaries of the stress relief dominant region, the interface decoupling active region, and the colloidal volatilization window region, thereby realizing the quantitative mapping from the probability distribution to the physical subdomain boundary.

[0138] For example, the degumming temperature range was set to -40℃ to -10℃, the degumming time range to 10 min to 60 min, and the degumming agent concentration range to 1% to 5%. Mesh sampling was performed in this three-dimensional space with step sizes of 1℃, 1 min, and 0.5%, resulting in 31×51×9 data points. The gradient modulus of each point was calculated, and the gradient threshold was set to 0.8. Using the marching cubes method, two main dividing lines were identified: the first dividing line corresponds to a gradient abrupt change near a temperature of -25℃, a time of 30 min, and a concentration of 2.5%, defining the stress-relieving dominant region and the active interface decoupling region; the second dividing line corresponds to a gradient abrupt change near a temperature of -15℃, a time of 45 min, and a concentration of 4.0%, defining the active interface decoupling region and the volatile resin window region. The final output subdomain partitioning threshold vector is [(-25, 30, 2.5), (-15, 45, 4.0)], significantly improving the physical consistency of subdomain partitioning.

[0139] S4.2: Utilize the subdomain partitioning threshold vector to perform a logical mask segmentation operation on the full parameter space, mapping continuous combinations of process parameters to non-overlapping stress relief dominant regions, interface decoupling active regions, and colloid volatilization window regions, in order to generate a set of reconstructed subdomain spaces containing their own independent process parameter ranges and possessing clear physical mechanism attributes.

[0140] Receive the subdomain partitioning threshold vector generated by S4.1, which contains three-dimensional boundary coordinate parameters defining the stress relief dominant region, the interface decoupling active region, and the colloidal volatilization window region.

[0141] Construct a logical mask matrix with the same spatial resolution as the three-dimensional process parameters of degumming temperature, degumming time and degumming agent concentration, initialize the matrix element values ​​to zero, and establish a discretized mapping mesh in the full parameter space.

[0142] Traverse each grid node in the logic mask matrix and extract the combination of temperature, time, and concentration values ​​corresponding to the current node as the process state point to be determined.

[0143] The three-dimensional coordinates of the current process state point are compared item by item with the boundary conditions in the subdomain partitioning threshold vector, and multidimensional inequality logic operations are performed to determine the physical mechanism region to which the point belongs.

[0144] If the current point coordinates satisfy the boundary constraint conditions of the stress relief dominant region, then the element value at the corresponding position in the logic mask matrix will be marked as a first-class identifier, indicating that the region uses thermal stress relaxation as the main debonding mechanism.

[0145] If the current point coordinates fall within the threshold range of the interface decoupling active area, the corresponding element value will be updated to the second type identifier, representing the interface separation process dominated by microcrack initiation and propagation.

[0146] If the current point coordinates meet the determination rules for the volatile window region of colloids, the element value at the corresponding position will be set as a third-class identifier to represent the chemically dominant stage of colloid degradation and large-scale VOC release.

[0147] Perform connected component analysis on the labeled logical mask matrix to remove isolated noise points and smooth region boundaries, generating a set of reconstructed subdomain spaces with topological continuity.

[0148] By using a logical mask segmentation process, the threshold boundaries determined in the previous step are transformed into a set of reconstructed subdomain space data that are non-overlapping and have clear physical properties. This enables a precise mapping of the process parameter space to the mechanism-driven modeling region, providing a structured input basis for the independent construction of subsequent segmented response surface models.

[0149] For example, the degumming temperature range is set to -40℃ to -10℃, with a step size of 2℃; the degumming time is set to 30 min to 90 min, with a step size of 5 min; and the degumming agent concentration is set to 1% to 5%, with a step size of 0.5%. A logical mask matrix with a resolution of 16×13×9 is constructed. Based on the threshold vector output by S4.1, when the temperature is below -25℃ and the time is less than 45 min, it is determined to be a stress-relieving dominant region and marked as 1; when the temperature is between -25℃ and -15℃ and the concentration is greater than 3%, it is determined to be an active interface decoupling region and marked as 2; other high-temperature and high-concentration combinations are determined to be the gum volatilization window region and marked as 3. After connected component processing, a three-dimensional Boolean matrix containing three independent subdomains is generated, ensuring that each combination of process parameters uniquely belongs to a physical mechanism subdomain, eliminating boundary ambiguity, and significantly improving the convergence speed and prediction accuracy of subsequent local model fitting.

[0150] S4.3: For the historical experimental data within the reconstructed subdomain space set, extract the degumming efficiency index and volatile organic compound emission data under the corresponding subdomain, calculate the total release value of volatile organic compounds at each sample point and statistically analyze the distribution frequency of its chemical components, so as to generate a scalar value of total VOC and a scalar value of VOC component entropy that characterize the scale of pollution emissions and chemical complexity within a single subdomain.

[0151] The degumming process parameter combinations and synchronously acquired raw spectral data of volatile organic compound emissions were obtained for each historical experimental sample point within the reconstructed subdomain spatial set. Baseline correction and smoothing / denoising were performed on the raw spectral data to eliminate instrument noise interference, and the VOC component concentration values ​​corresponding to each characteristic peak were extracted. All detected VOC component concentration values ​​were summed to calculate the total VOC scalar value characterizing the total scale of pollution emissions.

[0152] The mole fraction distribution of each VOC component in the total emissions was statistically analyzed to construct a component probability distribution vector. Based on information theory principles, the Shannon entropy formula was used to calculate the VOC component spectral entropy value, quantifying the chemical complexity of the release process.

[0153] The calculated total VOC scalar value and the VOC component spectral entropy scalar value are correlated and stored to form a structured dataset containing dual response indicators. Through the above processing method, the subdomain partitioning results of the previous step are transformed into physically meaningful dual response value data, realizing a refined characterization of the pollution emission scale and chemical complexity of the degumming process, and providing standardized input for the subsequent construction of a high-precision piecewise response surface model.

[0154] S4.4: Based on the process parameter combination input variables in the reconstructed subdomain space set and the corresponding VOC total scalar value and VOC component spectrum entropy scalar value output response, perform least squares regression fitting operation independently in each subdomain to construct a quadratic polynomial mathematical expression containing linear terms, interaction terms and square terms, so as to generate local quadratic response surface model coefficient matrices that are respectively adapted to the stress relief dominant region, the interface decoupling active region and the colloid volatilization window region.

[0155] For the historical experimental data selected from the reconstructed subdomain space set, degumming temperature, degumming time, and degumming agent concentration were extracted as independent variable vectors, and the corresponding total VOC scalar value and VOC component spectral entropy scalar value were associated as dual response variables.

[0156] Construct a quadratic polynomial regression model structure that includes linear terms, interaction terms, and quadratic terms, and determine the dimension of the coefficient matrix to be solved, where there are 3 independent variables and a total of 10 undetermined coefficients, covering constant terms, linear terms, interaction terms, and quadratic terms.

[0157] The loss function is constructed using the least squares method, aiming to minimize the sum of squared residuals between the model's predicted values ​​and the actual measured values. The nonlinear optimization problem is transformed into a problem of solving a system of linear equations through differentiation.

[0158] The regression coefficient vector is calculated using the following formula:

[0159]

[0160] in, The design matrix consists of the process parameters for each experimental point and their combinations. The response vector contains the total VOC amount or spectral entropy value at the corresponding experimental point; Let be the coefficient matrix of the local quadratic response surface model to be determined. This indicates transpose.

[0161] The design matrix X is transposed and multiplied by itself to generate a symmetric positive definite matrix. Then, the inverse matrix of this matrix is ​​calculated to eliminate the influence of correlation between variables and ensure the uniqueness and stability of the solution.

[0162] By performing matrix multiplication on the inverse matrix, the transposed design matrix, and the response vector Y, the optimal regression coefficients in the current subdomain are directly calculated, thus completing the mapping from the original data to the mathematical model parameters.

[0163] The above fitting process was repeated for the stress relief-dominant region, the interface decoupling active region, and the colloidal volatilization window region, respectively, to generate three independent local quadratic response surface model coefficient matrices, thereby achieving accurate characterization under different physical mechanisms.

[0164] By using least squares regression fitting, the subdomain experimental data from the previous step is transformed into a coefficient matrix of a local quadratic response surface model that adapts to a specific physical mechanism, thereby achieving segmented and accurate modeling of VOC emissions and their chemical complexity.

[0165] S4.5: Integrate the coefficient matrix of the local quadratic response surface model with the corresponding subdomain boundary definition information, establish a switching logic rule that automatically routes to a specific subdomain model based on the current process parameter position, so as to generate a piecewise quadratic response surface model overall architecture that can dynamically call the corresponding prediction function according to real-time input conditions, and complete the multi-mechanism segmented accurate modeling of VOC release in the low-temperature degumming process of automotive composite materials.

[0166] Receive the coefficient matrices of the local quadratic response surface models generated in step S4.4, which are respectively adapted to the stress relief dominant region, the interface decoupling active region, and the colloidal volatilization window region, as well as the physical boundary definition information of each subdomain determined by steps S4.1 and S4.2.

[0167] A logical judgment routing table based on the spatial coordinates of process parameters is constructed, and a three-dimensional input vector consisting of degumming temperature, degumming time, and degumming agent concentration is compared and calculated with the subdomain boundary threshold in real time.

[0168] Define a piecewise function switching logic. When the input process parameter combination falls within the boundary range of the stress relief dominant zone, automatically load the VOC total prediction coefficient matrix and the component spectrum entropy prediction coefficient matrix corresponding to that region.

[0169] When the input process parameter combination falls within the boundary range of the interface decoupling active region, the system dynamically switches to the local response surface model parameter configuration of the interface decoupling active region to ensure the physical consistency between the mechanical mechanism and the mathematical model.

[0170] When the input process parameter combination falls within the boundary range of the colloid volatilization window, a dedicated prediction model for the high volatilization risk area is activated, focusing on enhancing the sensitivity to capture VOC release peaks.

[0171] A model output integration module is established to encapsulate the total VOC scalar value and the VOC component spectral entropy scalar value calculated by different subdomain models into a standard data format, forming a continuous and smooth global prediction surface.

[0172] Through the aforementioned segmented modeling and dynamic routing mechanism, the local model parameters of the previous step are transformed into a globally adaptable segmented quadratic response surface model architecture, enabling multi-mechanism segmented and accurate modeling of VOC release in the low-temperature degumming process of automotive composite materials.

[0173] Step S5: After a new batch of materials enters the degumming process, the matching degree between the measured stress evolution trajectory anchor point and the historical anchor point attainability probability map is compared in real time. The deviation threshold between the measured trajectory and the standard trajectory is calculated to determine whether the stress evolution law of the current material system has drifted. Specifically, this includes:

[0174] S5.1: Obtain the real-time interface local stress fluctuation signal sequence of the new batch of hemp fiber bundles during the low-temperature degumming gradient cooling stage by a micro piezoelectric thin film sensor array, and perform multi-resolution singular spectrum analysis on the real-time interface local stress fluctuation signal sequence to extract the set of measured stress evolution trajectory anchor points characterizing the current degumming mechanism transition state.

[0175] The real-time interface local stress fluctuation signal sequence of the new batch of hemp fiber bundles during the low-temperature degumming gradient cooling stage was acquired by a micro piezoelectric thin film sensor array and used as the raw time-domain data stream to be processed.

[0176] The real-time interface local stress fluctuation signal sequence is subjected to detrending processing, and the baseline drift component in the signal is removed by a polynomial fitting method to generate zero-mean stress fluctuation time series data that eliminates low-frequency thermal drift interference.

[0177] The zero-mean stress fluctuation time series data is segmented based on the sliding window mechanism. The window length is set to N and the step size is set to M. A set of local signal segments containing overlapping regions is constructed to preserve the local non-stationary characteristics of the signal.

[0178] A trajectory matrix is ​​constructed for each local signal segment, mapping the one-dimensional time series to a multi-dimensional phase space vector, forming a Hankel matrix structure that can reflect the dynamic evolution of the system.

[0179] Perform singular value decomposition on the Hankel matrix to calculate the principal component components after decomposition.

[0180] Based on the distribution characteristics of singular values, the components corresponding to the first few largest singular values ​​are selected as the effective signal subspace, and the noise subspace corresponding to the smaller singular values ​​is eliminated to achieve signal-noise separation.

[0181] An inverse mapping operation is performed on the reconstructed effective signal subspace to restore the multidimensional phase space vector to a one-dimensional time series, generating a pure stress evolution curve after noise reduction by multi-resolution singular spectrum analysis.

[0182] The change point detection method is used to calculate the second-order difference statistics of the pure stress evolution curve, identify the moment when the slope of the curve changes abruptly, and mark it as key event nodes such as the initial stress relaxation start point and the critical point of microcrack initiation.

[0183] Extract the instantaneous stress amplitude, duration, and time interval between adjacent nodes corresponding to each key event node, and combine them to form a set of measured stress evolution trajectory anchor points that characterize the current degumming mechanism transition state.

[0184] By combining multi-resolution singular spectrum analysis and change point detection, the original electrical signal from the previous step is transformed into a set of mechanism feature anchor points with clear physical meaning, thereby realizing feature extraction and quantitative characterization from microscopic mechanical response to macroscopic process state.

[0185] S5.2: Based on the occurrence order, relative temporal offset and intensity attenuation ratio of each anchor point in the measured stress evolution trajectory anchor point set, construct the current high-dimensional trajectory feature vector, and call the anchor point attainability probability map generated in the previous step as the standard reference benchmark to map the current high-dimensional trajectory feature vector into the domain space of the anchor point attainability probability map.

[0186] Extract the set of measured stress evolution trajectory anchor points obtained in step S5.1. This set includes four key physical event nodes: the initial stress relaxation initiation point, the critical point of microcrack initiation, the acceleration point of gel layer peeling, and the precursor point of VOC volatilization peak.

[0187] For each event node in the anchor point set, perform temporal feature encoding, read the absolute time value of each node on the degumming full-cycle time axis, calculate the time interval between adjacent nodes as the relative temporal offset, and construct a time topology sequence that reflects the speed of the degumming dynamic process.

[0188] Read the peak value of the standardized interface local stress fluctuation voltage signal corresponding to each anchor point, calculate the ratio of the current anchor point stress amplitude to the previous stage reference stress amplitude, and generate the strength attenuation ratio coefficient characterizing the interface damage accumulation rate.

[0189] The relative temporal offset values ​​and intensity attenuation ratio coefficients in the temporal topology sequence are concatenated into vectors and arranged according to the physical order of events to construct a high-dimensional trajectory feature vector containing multi-dimensional mechanical response information.

[0190] The anchor point attainability probability map generated in step S4 is invoked. This map defines the mapping relationship between the three-dimensional process parameter space consisting of degumming temperature, time, and concentration and the characteristics of each stress evolution trajectory.

[0191] Using the principal component inverse projection method, the constructed high-dimensional trajectory feature vector is input into the feature domain of the probability map. By solving the inverse operation of the linear transformation matrix, the projection coordinates of the measured trajectory in the original process parameter space are calculated.

[0192] Calculate the mapping confidence of the high-dimensional trajectory feature vector to the process parameter space. Based on the calculated mapping confidence, determine the specific landing position of the measured trajectory in the anchor point reachability probability map, and complete the semantic alignment from the mechanical feature space to the process parameter space.

[0193] By using principal component inverse projection, the high-dimensional trajectory features from the previous step are transformed into specific coordinate data within the domain of the anchor point attainability probability map, thus achieving spatial matching between the measured degumming state and the historical process benchmark, and providing a unified measurement basis for subsequent deviation calculation.

[0194] S5.3: Calculate the spatial difference between the mapped current high-dimensional trajectory feature vector and the corresponding standard trajectory feature vector in the anchor point reachability probability map using the Euclidean distance metric, and normalize the spatial difference to generate the deviation threshold between the measured trajectory and the standard trajectory.

[0195] Receive the current high-dimensional trajectory feature vector mapped to the domain space of the anchor point reachability probability map, and extract the measured feature components containing the dimensions of stress amplitude, temperature slope, and duration.

[0196] The standard trajectory feature vector of the corresponding process subdomain in the historical anchor point achievability probability map is used as the benchmark reference data to ensure that the measured data and the standard data are strictly aligned in the feature dimension.

[0197] The geometric distance between the measured feature vector and the standard feature vector in the multidimensional feature space is calculated using the Euclidean distance metric, which quantifies the degree of spatial difference between the two.

[0198] Obtain the maximum permissible characteristic fluctuation range of historical experimental data within the process subdomain, and determine the denominator coefficient used for normalization processing to eliminate the influence of different dimensions on deviation assessment.

[0199] Divide the calculated spatial difference value by the maximum allowable characteristic fluctuation range, perform linear normalization, and generate a dimensionless deviation threshold with a value range between 0 and 1.

[0200] By calculating and normalizing the Euclidean distance, the feature mapping results from the previous step are transformed into a standardized deviation threshold between the measured trajectory and the standard trajectory, thereby enabling a quantitative assessment of the drift degree of the stress evolution law of the new batch of materials.

[0201] For example, for the low-temperature degumming process of hemp fibers, a three-dimensional feature vector is constructed, including stress amplitude, temperature slope, and duration. The measured vector is [12.5, -0.8, 45], and the standard vector is [12.0, -0.75, 42]. The sum of squares of the differences in each dimension is calculated to be 0.25 + 0.0025 + 9 = 9.2525. The square root yields a spatial difference value of 3.04. The maximum allowable fluctuation range is set to 5.0. The normalized calculation of 3.04 / 5.0 yields a deviation threshold of 0.608. This threshold objectively reflects the degree of deviation of the current batch of material interface stress evolution from the standard model, providing a precise basis for subsequent judgment on whether to trigger a model update.

[0202] S5.4: Perform a logical comparison operation between the deviation threshold between the measured trajectory and the standard trajectory and the preset tolerance range. If the deviation threshold between the measured trajectory and the standard trajectory exceeds the preset tolerance range, a drift judgment signal for the stress evolution law of the material system is generated; otherwise, a process stability confirmation signal is generated.

[0203] The normalized deviation threshold between the measured trajectory and the standard trajectory output from step S5.3 is received and used as the input variable for logical judgment.

[0204] The process stability tolerance range parameter, which is pre-stored in the control system memory, is called. This parameter is determined based on the statistical distribution of stress evolution fluctuations in historical batches of materials during normal degumming, and includes two boundary values: an upper threshold and a lower threshold.

[0205] Perform a numerical comparison operation to compare the normalized deviation threshold with the upper and lower thresholds respectively, and determine whether the current deviation falls within the preset stable range.

[0206] If the normalization deviation threshold is greater than the upper threshold or less than the lower threshold, it is determined that the interfacial thermodynamic properties of the current new batch of hemp fiber material are significantly different from the historical benchmark model, and a high-level state material system stress evolution law drift determination signal is generated.

[0207] If the normalized deviation threshold is between the upper and lower thresholds, it is determined that the stress release mechanism of the current material system conforms to the expected physical laws, and a low-level process stability confirmation signal is generated.

[0208] Through the above logical comparison operation, the continuous deviation value is transformed into a discrete system state control signal, realizing the automatic identification of material batch consistency and providing a clear decision basis for subsequent model adaptive update or maintenance.

[0209] For example, the upper threshold of the process stability tolerance range is set to 0.15, and the lower threshold is set to -0.05. Low-temperature degumming monitoring is performed on a new batch of hemp fiber. The normalized deviation threshold calculated by S5.3 is 0.22. Comparing 0.22 with 0.15, since 0.22 is greater than 0.15, exceeding the upper limit of the stability range, the system immediately generates a drift judgment signal for the stress evolution law of the material system. This signal triggers the subdomain re-division mechanism in step S6, initiating the online model calibration process. The normalized deviation threshold of another batch of fiber is 0.08, falling between -0.05 and 0.15. The system generates a process stability confirmation signal, directly using the piecewise quadratic response surface model coefficients constructed in S4, without refitting, significantly shortening the process preparation time and ensuring the continuity of VOC prediction.

[0210] S5.5: Based on the drift judgment signal of the stress evolution law of the material system, trigger the subdomain re-division mechanism, or transmit the process stability confirmation signal to the dynamic process adjustment module to maintain the coefficient configuration of the current piecewise quadratic response surface model, and complete the final judgment of the adaptability of the current material system.

[0211] Receive the drift judgment signal of the stress evolution law of the material system or the process stability confirmation signal generated by S5.4 as the logical input condition for the execution of the sub-step.

[0212] If a process stability confirmation signal is received, the system reads the coefficient matrix and subdomain boundary definition data of the current piecewise quadratic response surface model, marks it as valid, and locks the parameter configuration.

[0213] The locked model parameters are transferred to the dynamic process adjustment module to maintain the operation of the low-temperature degumming equipment under the current optimal combination of process parameters, ensuring the continuity of VOC emission prediction.

[0214] If a drift determination signal for the stress evolution law of a material system is received, the system analyzes the deviation threshold value contained in the signal and the corresponding measured stress evolution trajectory anchor point feature vector.

[0215] Based on the judgment result that the deviation threshold exceeds the preset tolerance range, an enable instruction for the subdomain repartitioning mechanism is generated, activating the execution permission of the parameter space reconstruction module.

[0216] The enable command is packaged with the set of measured stress evolution trajectory anchor points and sent to the adaptive update interface described in step S6 to trigger the boundary dynamic offset correction process of the full parameter space.

[0217] Through the above logical branch processing, the judgment result of the previous step is transformed into control instructions for model preservation or model reconstruction, so as to realize the final judgment and closed-loop control of the adaptability of the new batch of material system.

[0218] For example, when a new batch of hemp fiber enters the degumming process, S5.3 calculates the Euclidean distance deviation between the measured trajectory and the standard trajectory to be 0.15, with a preset tolerance range of 0.05. Since 0.15 is greater than 0.05, S5.4 generates a drift judgment signal. The system then analyzes this signal and extracts a feature vector containing the timing offset of the microcrack initiation critical point +2.3s. The system generates a subdomain repartitioning enable command and sends the command and feature vector to module S6. Based on this, module S6 dynamically shifts the temperature boundary of the interface decoupling active region from -25℃ to -23.5℃ and refits the local response surface coefficients. Ultimately, the updated model reduces the error in predicting VOC release from the initial 12% to less than 3%, significantly improving the accuracy of process control.

[0219] Step S6: If the deviation threshold is determined to exceed a preset tolerance range, the coefficients of the piecewise quadratic response surface model are recalibrated. Specifically, this includes:

[0220] S6.1: Based on the deviation threshold calculation results of the measured stress evolution trajectory anchor point and the historical anchor point attainability probability map, the original subdomain boundary coordinates in the full parameter space are dynamically offset and corrected to generate an updated subdomain boundary set that reflects the mechanical response characteristics of the current material system.

[0221] The drift judgment signal of the stress evolution law of the material system and the corresponding deviation threshold data output from step S5 are received as the triggering condition and quantification basis for dynamic correction of the subdomain boundary.

[0222] Extract the set of original subdomain boundary coordinates that define the stress-relieving dominant region, the interface decoupling active region, and the colloid volatilization window region from the historical anchor point achievability probability map. This set contains isosurface parameters in three-dimensional space for degumming temperature, degumming time, and degumming agent concentration.

[0223] Based on the deviation threshold, the spatiotemporal position offset vector of each key stress evolution trajectory anchor point in the measured state and the standard state is calculated. The offset vector represents the mechanical response lag or lead caused by the component difference of the current batch of hemp fiber.

[0224] A boundary correction function is constructed using the weighted least squares method, which maps the spatiotemporal position offset vector to the original process parameter space, and calculates the dynamic compensation amount of each subdomain boundary in the directions of temperature axis, time axis and concentration axis.

[0225] The calculated dynamic compensation is superimposed onto the original subdomain boundary coordinates, and coordinate translation and rotation transformations are performed to eliminate parameter space distortion caused by material batch differences.

[0226] The transformed boundary coordinates are smoothed and filtered to remove local high-frequency jitter caused by sensor noise, generating an updated set of subdomain boundaries that is geometrically continuous and has clear physical meaning.

[0227] Through the above-mentioned boundary dynamic offset correction process, the deviation threshold of the previous step is transformed into an updated set of subdomain boundaries that reflects the mechanical response characteristics of the current material system, thereby realizing the rapid adaptation and accurate coverage of the piecewise response surface model to the new material system.

[0228] S6.2: Utilize the updated subdomain boundary set to perform a logical repartitioning operation on the original process parameter space, and remap the continuously changing combination of degumming temperature, degumming time, and degumming agent concentration to the stress-relieving dominant region, the interface decoupling active region, and the adhesive volatilization window region, so as to generate a reconstructed subdomain space with clear physical meaning that is suitable for the current batch.

[0229] Receive the updated subdomain boundary set generated by step S6.1, which contains the three-dimensional spatial coordinate thresholds of the stress relief dominant region, the interface decoupling active region, and the viscous volatile window region, reflecting the mechanical response characteristics of the current batch of hemp fiber materials.

[0230] A three-dimensional spatial grid of original process parameters based on degumming temperature, degumming time and degumming agent concentration is constructed. The continuously changing combination of process parameters is discretized into a finite number of sampling points to form a parameter matrix to be mapped.

[0231] Traverse each process parameter combination point in the parameter matrix and extract its corresponding temperature, time, and concentration values ​​as the basis for judgment.

[0232] The extracted process parameter values ​​are logically compared with the boundary coordinates in the updated subdomain boundary set to determine the relative position of the parameter point in three-dimensional space.

[0233] If the parameter point is located within the boundary of the stress relaxation-dominant region, then the point is assigned a first-class physical mechanism identifier and marked as a stress relaxation control mode.

[0234] If the parameter point is located within the boundary of the interface decoupling active region, then the point is assigned a second type of physical mechanism identifier and marked as a microcrack propagation control mode.

[0235] If the parameter point is located within the boundary of the colloidal volatilization window, then the point is assigned a third type of physical mechanism identifier and marked as VOC release control mode.

[0236] By using a logical mask segmentation method, all marked process parameter points are re-aggregated to generate three non-overlapping subdomains that cover the entire parameter space.

[0237] The boundary smoothness of the reconstructed subdomain space is verified to eliminate discrete noise caused by dynamic boundary offset, ensuring the continuity and physical meaning of the subdomain boundary.

[0238] Through the above logical repartitioning operation, the dynamic boundary obtained in the previous step is transformed into a reconstructed subdomain space with clear physical meaning that is suitable for the current batch, thereby realizing the accurate reconstruction of the input domain of the segmented response surface model and providing a structured data foundation for the adaptive update of the local model parameters.

[0239] For example, the degumming temperature range is set to -40℃ to -10℃, the degumming time range is 30 min to 90 min, and the degumming agent concentration range is 1% to 5%. The updated subdomain boundary set defines the stress-relieving dominant region as the area with a temperature below -30℃ and a time of less than 50 min; the interface decoupling active region as the area with a temperature between -30℃ and -20℃ or a time between 50 min and 70 min; and the gum evaporation window region as the area with a temperature above -20℃ or a concentration above 3%. The system traverses the parameter space consisting of 10×10×10 grid points, compares the coordinate point (-35℃, 40 min, 2%), determines that it belongs to the stress-relieving dominant region and assigns it label 1; compares the coordinate point (-25℃, 60 min, 2.5%), determines that it belongs to the interface decoupling active region and assigns it label 2; and compares the coordinate point (-15℃, 80 min, 4%), determines that it belongs to the gum evaporation window region and assigns it label 3. Ultimately, three independent reconstructed subdomains are generated, and the process parameters in each subdomain have unified physical mechanism attributes, which significantly improves the targeting and accuracy of subsequent VOC prediction models.

[0240] S6.3: For the newly acquired interface local stress fluctuation signal sequence and corresponding regional temperature field time series data in the reconstructed subdomain space, extract triplet feature variables containing stress amplitude, temperature slope and duration to construct a real-time experimental dataset for local model calibration.

[0241] The newly acquired interface local stress fluctuation voltage signal sequence and corresponding regional temperature field time series data within the reconstructed subdomain space are used as the original input source for feature extraction.

[0242] A peak detection method is applied to the local stress fluctuation voltage signal sequence at the interface. All sampling points on the time axis are traversed to identify local maxima and their corresponding time indices. The instantaneous voltage amplitude at that moment is extracted and converted into physical stress units to generate a stress amplitude scalar sequence.

[0243] Based on the time series data of the temperature field in the corresponding region, for the identified stress extreme moment index, the average temperature value in the micro-region at that moment is extracted, and the ratio of the temperature difference between adjacent time steps to the time interval is calculated to generate a temperature slope scalar sequence characterizing the thermal shock intensity.

[0244] The time interval between the timestamps of two adjacent key debonding mechanism transition events is statistically analyzed, and the cumulative duration from the initial stress relaxation start point to the current anchor point time is calculated to generate a time-dimensional scalar sequence characterizing the stress accumulation effect.

[0245] The three scalars extracted above—stress amplitude, temperature slope, and duration—are concatenated according to a preset feature vector structure to construct a triplet feature variable that includes mechanical strength, thermal change rate, and time cumulative effect.

[0246] The triplet feature variables are standardized to eliminate the influence of different units on model training, forming a real-time experimental dataset sample for local model calibration.

[0247] By using multimodal signal synchronous analysis and feature fusion processing, the raw sensing data from the previous step is transformed into structured triplet feature variables with clear physical meaning, realizing the quantitative characterization of the micromechanical-thermal coupling state of the degumming process, and providing high-fidelity data support for the adaptive update of the subsequent response surface model.

[0248] S6.4: Based on the real-time experimental dataset, the least squares method is used to perform iterative optimization calculations on the quadratic polynomial regression coefficients in each reconstructed subdomain to minimize the prediction residuals between the total VOC amount and the VOC component spectrum entropy value, thereby generating the updated piecewise quadratic response surface model coefficient matrix.

[0249] Receive the real-time experimental dataset within the reconstructed subdomain space. This dataset contains degumming temperature, degumming time, and degumming agent concentration as independent variables, and the corresponding total VOC and VOC component spectral entropy values ​​as dual response dependent variables.

[0250] For each independent reconstructed subdomain, a quadratic polynomial regression model structure is constructed, the model coefficient vector is set as the parameter to be optimized, and the coefficient matrix is ​​initialized as a zero vector or an estimated value based on the historical model.

[0251] Define the objective function of the least squares method, which aims to minimize the sum of squared residuals between the model predictions and the actual observations, and calculate the objective function value.

[0252] Take the partial derivatives of the objective function with respect to each regression coefficient, set the partial derivatives to zero to establish a normal system of equations, and use matrix operations to solve the linear system of equations to obtain the optimal regression coefficient estimates that minimize the sum of squared residuals.

[0253] A regularization term is introduced to constrain the coefficient matrix to prevent overfitting. The optimal regularization parameter is selected through cross-validation to ensure the model's generalization ability on unseen data.

[0254] The above iterative optimization process was performed on the total VOC response surface and the VOC component spectrum entropy response surface, respectively, to generate two independent sets of local quadratic response surface model coefficient matrices.

[0255] The optimized coefficient matrix is ​​associated with and stored with the subdomain boundary information to form a piecewise model parameter set that adapts to the stress evolution trajectory of the current material system.

[0256] By using the least squares iterative optimization calculation, real-time experimental data is transformed into an updated piecewise quadratic response surface model coefficient matrix, thereby achieving accurate adaptation of the model to the VOC release characteristics of the current batch of materials and improving prediction accuracy.

[0257] S6.5: Write the updated piecewise quadratic response surface model coefficient matrix into the model storage unit of the control system to replace the original old model parameters, so as to complete the online adaptive update of the piecewise quadratic response surface model and enable its output commands to accurately drive the low-temperature degumming equipment to perform dynamic process adjustment actions.

[0258] like Figure 3 As shown, step S7 involves outputting optimal degumming temperature, degumming time, and degumming agent concentration control commands based on the updated piecewise quadratic response surface model. This drives the low-temperature degumming equipment to perform dynamic process adjustment actions, achieving accurate prediction and control of VOC release. Specifically, this includes:

[0259] S7.1: Based on the updated piecewise quadratic response surface model outputting the optimal degumming temperature setpoint, optimal degumming time setpoint, and optimal degumming agent concentration setpoint, a multi-dimensional process control instruction set including the target temperature curve, target time threshold, and target concentration gradient is constructed to generate a standardized digital control signal sequence for driving the underlying actuators.

[0260] The updated setpoints for optimal degumming temperature, optimal degumming time, and optimal degumming agent concentration are received from the piecewise quadratic response surface model and used as initial control parameters. Based on the optimal degumming temperature setpoint, and considering the thermodynamic hysteresis characteristics of the low-temperature degumming process for hemp fibers, a smooth target temperature curve is constructed using cubic spline interpolation to eliminate the impact of abrupt changes on the refrigeration compressor unit. According to the time axis distribution of the target temperature curve, the time points corresponding to key temperature nodes are extracted to generate time-series data including heating rate, isothermal holding time, and cooling slope. Based on the optimal degumming time setpoint, an upper limit threshold for the total process time is set, and the time-series data is mapped to this threshold range to form a standardized time progress control vector. For the optimal degumming agent concentration setpoint, the volume ratio coefficient between the degumming agent and the cleaning medium is calculated, and the pulse frequency reference value of the metering pump is determined using a fluid dynamics model. The volume ratio coefficient is converted into a concentration gradient command that varies with time, ensuring high-concentration penetration in the active decoupling zone and low-concentration cleaning in the stress-relieving zone. The target temperature curve, time-based control vector, and concentration gradient command are spatiotemporally aligned to establish a synchronization correlation matrix among multidimensional process parameters. This synchronization correlation matrix is ​​then digitally encoded, discretizing continuous analog quantities into digital word-length sequences conforming to industrial communication protocol standards. These digital word sequences are encapsulated through standardized interfaces to generate standardized digital control signal sequences containing address codes, function codes, and data fields. This process transforms the model optimization results from the previous step into standard digital commands recognizable by the underlying actuators, achieving a precise mapping from theoretically optimal parameters to controllable engineering signals and providing a unified data benchmark for subsequent physical layer driving.

[0261] S7.2: The standardized digital control signal sequence is encapsulated and converted from digital to analog. The digital quantity is converted into an analog voltage signal or a pulse width modulation signal using an industrial fieldbus communication module to generate a physical layer control level that is compatible with the heating unit, timing control unit and fluid metering pump drive interface of the low-temperature degumming equipment.

[0262] S7.3: Based on the physical layer control level, the refrigeration compressor unit, electric heating compensation circuit and degumming agent injection valve of the low-temperature degumming equipment are driven to perform coordinated adjustment actions. The actual chamber temperature, cumulative processing time and solution mixing ratio are corrected in real time through the PID closed-loop feedback mechanism to form a dynamic degumming physical environment that meets the optimal combination of process parameters.

[0263] Receives physical layer control level signals generated by digital-to-analog conversion for the adapted refrigeration compressor unit, electric heating compensation circuit, and degumming agent injection valve.

[0264] The temperature setpoint component in the physical layer control level is analyzed and input into the low-temperature cavity PID temperature control method module to calculate the deviation between the current measured temperature and the target temperature.

[0265] Based on the aforementioned deviation, an incremental PID method is used to output the frequency conversion drive command for the refrigeration compressor and the power adjustment command for the electric heating compensation, thereby suppressing temperature overshoot and oscillation during the gradient cooling process.

[0266] The system monitors the cumulative processing time counter. When the optimal degumming time threshold output by the model is reached, a process stage switching signal is generated to terminate the current dynamic adjustment cycle.

[0267] Through the above-mentioned multi-actuator coordinated adjustment and PID closed-loop feedback mechanism, the control command of the previous step is transformed into a stable low-temperature degumming physical environment, thereby achieving precise control of VOC release from automotive composite materials.

[0268] S7.4: During the operation of the dynamic degumming physical environment, real-time temperature fluctuation data, actual time progress data and instantaneous flow concentration data are collected simultaneously. The deviations are compared and calculated with the target values ​​in the multi-dimensional process control instruction set to generate a real-time status monitoring feedback vector that characterizes the current process execution accuracy.

[0269] Receive real-time sensor data streams from the operation of the refrigeration compressor unit, electric heating compensation circuit, and fluid metering pump of the low-temperature degumming equipment, including the cavity temperature time sequence, cumulative processing timestamp, and instantaneous flow pulse signal of the degumming agent.

[0270] The cavity temperature time series is subjected to moving average filtering to eliminate high-frequency thermal noise interference, and a smooth temperature curve representing the actual thermal environment is extracted. Then, point-by-point difference operation is performed between the smooth temperature curve and the target temperature curve in the multi-dimensional process control instruction set to generate a temperature deviation component.

[0271] Based on the time difference between the cumulative processing timestamp and the preset degumming cycle threshold, the actual progress lag or advance of the current process stage is determined and normalized into a time schedule deviation scalar to quantify the execution accuracy in the time dimension.

[0272] The instantaneous flow pulse frequency collected by the turbine flow meter is converted into real-time concentration flow rate by combining the degumming agent density parameter. The total amount of degumming agent injected is calculated by integration and accumulation, and compared with the theoretical injection amount expected by the target concentration gradient model to generate a concentration flow rate deviation vector.

[0273] A three-dimensional status monitoring feedback vector space is constructed, and the temperature deviation component, time progress deviation scalar, and concentration flow rate deviation vector are linearly combined according to weighting coefficients to form a comprehensive error index characterizing the current process execution accuracy.

[0274] The modulus of the comprehensive error index is calculated using the Euclidean norm, and it is determined whether it is within the preset stability tolerance range. If it exceeds the tolerance, it is marked as a high deviation state, triggering the termination or correction logic in the subsequent S7.5 step.

[0275] By fusing multi-source sensor data and comparing deviations, the execution result of the physical layer control level in the previous step is transformed into a real-time status monitoring feedback vector that characterizes the current process execution accuracy, thus realizing closed-loop precise monitoring of key parameters such as temperature, time, and concentration during the low-temperature degumming process.

[0276] S7.5: Based on the real-time status monitoring feedback vector, determine whether the degumming process of the automotive composite material has reached the preset VOC release control endpoint. If the termination condition is met, automatically cut off the heat source and start the exhaust purification program to complete the fully automated low-temperature degumming process of a single low-VOC release automotive composite material.

[0277] Receive the real-time status monitoring feedback vector generated by S7.4. This vector includes the actual cavity temperature deviation at the current moment, the cumulative processing time progress, and the instantaneous concentration error of the degumming agent.

[0278] VOC release-related feature components are extracted from the feedback vector, including the attenuation rate of interface stress fluctuation amplitude and the change rate of temperature field gradient, as key indicators for judging the stage of degumming process.

[0279] The extracted feature components are input into the updated piecewise quadratic response surface model to calculate the predicted total VOC release rate and VOC component entropy under the current process conditions.

[0280] A dynamic threshold comparison method is used to compare the predicted VOC release rate with the preset low VOC emission control endpoint threshold, while verifying whether the VOC component spectrum entropy value is lower than the chemical complexity safety limit.

[0281] Construct a termination judgment logic function. When the predicted VOC release rate is below the threshold for multiple consecutive sampling periods and the component entropy value is stable, generate a degumming completion flag. The formula is as follows:

[0282]

[0283] in, To comprehensively determine the termination criterion index, for Predicted VOC release rate at any given time. for VOC component spectral entropy value at time t. and These are the weighting coefficients.

[0284] If the comprehensive termination criterion index is less than the preset threshold, it is determined that the VOC release control endpoint has been reached, triggering a shutdown command sequence.

[0285] A cut-off signal is sent to the heating control unit of the low-temperature degumming equipment to immediately stop the operation of the electric heating compensation circuit and block the heat source supply.

[0286] The control valve that simultaneously activates the exhaust purification process turns on the high negative pressure exhaust fan, drawing the residual volatile organic compounds in the cavity to the activated carbon adsorption filter.

[0287] After monitoring the pressure and gas concentration sensor data inside the chamber and confirming that the VOC concentration has dropped below the environmental safety standard, the exhaust system is shut off and the hatch is unlocked.

[0288] Through the above chain-like derivation and execution actions, real-time process monitoring data is transformed into clear shutdown and purification control commands, realizing automated closed-loop control of the entire process of low-temperature degumming of automotive composite materials, and ensuring that the VOC emission of the final product meets national standards.

[0289] For those skilled in the art, various other corresponding changes and modifications can be made based on the technical solutions and concepts described above, and all such changes and modifications should fall within the protection scope of the claims of this invention.

[0290] Unless otherwise defined, the technical or scientific terms used herein shall have the ordinary meaning as understood by one of ordinary skill in the art to which this application pertains. The terms “first,” “second,” “third,” and similar terms used in this patent application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms “an” or “a” and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms “comprising” or “including” and similar terms mean that the elements or objects preceding “comprising” or “including” encompass the elements or objects listed following “comprising” or “including” and their equivalents, and do not exclude other elements or objects. The “multiple” mentioned in the embodiments of this application refers to two or more. A and / or B indicate three possibilities: A; B; and A and B.

[0291] The above description is merely an exemplary embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and such modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for predicting and controlling VOC emissions during low-temperature degumming processes of automotive composite materials, characterized in that, Specifically, it includes: S1: In the experimental environment of low-temperature degumming process of automotive composite materials, the local stress fluctuation signal sequence and the corresponding temperature field time series data of the interface were obtained at the micro-region of the interface between the hemp fiber bundle and the adhesive. S2: Process the local stress fluctuation signal sequence of the interface and the time series data of the temperature field in the corresponding region to obtain the set of stress evolution trajectory anchor points; S3: Construct a high-dimensional trajectory feature vector based on the set of stress evolution trajectory anchor points, and process the high-dimensional trajectory feature vector to obtain the anchor point reachability probability map; S4: Based on the probability map of the achievability of the anchor points, the full parameter space is divided into subdomains of stress relief dominance region, interface decoupling active region and colloidal volatilization window region, and a piecewise quadratic response surface model is independently constructed in each subdomain. S5: After a new batch of materials enters the degumming process, the matching degree between the measured stress evolution trajectory anchor point and the historical anchor point attainability probability map is compared in real time, the deviation threshold between the measured trajectory and the standard trajectory is calculated, and it is determined whether the stress evolution law of the current material system has drifted. S6: If the deviation threshold is determined to exceed the preset tolerance range, the coefficients of the piecewise quadratic response surface model are recalibrated. S7: Based on the updated piecewise quadratic response surface model, it outputs the optimal degumming temperature, degumming time, and degumming agent concentration control commands, driving the low-temperature degumming equipment to perform dynamic process adjustment actions, and completing the accurate prediction and control of VOC release.

2. The method for predicting and controlling VOC release during the low-temperature degumming process of automotive composite materials according to claim 1, characterized in that, The acquisition of the local stress fluctuation signal sequence and the corresponding temperature field time series data at the micro-region of the interface between the hemp fiber bundle and the gel is achieved by simultaneously deploying a micro piezoelectric thin film sensor array and an infrared thermal imaging high-speed acquisition system at the micro-region of the interface between the hemp fiber bundle and the gel.

3. The method for predicting and controlling VOC release during the low-temperature degumming process of automotive composite materials according to claim 1, characterized in that, The process of processing the interface local stress fluctuation signal sequence and the corresponding regional temperature field time series data yields a set of stress evolution trajectory anchor points, including: Multi-resolution singular spectrum analysis is performed on the local stress fluctuation signal sequence of the interface and the time series data of the temperature field in the corresponding region to identify the initial stress relaxation initiation point, the critical point of microcrack initiation, the acceleration point of gel layer peeling and the precursor point of VOC volatilization peak, and to generate a set of stress evolution trajectory anchor points.

4. The method for predicting and controlling VOC release during the low-temperature degumming process of automotive composite materials according to claim 3, characterized in that, The set of anchor points for the stress evolution trajectory includes a triplet of stress amplitude, temperature slope, and duration.

5. The method for predicting and controlling VOC release during the low-temperature degumming process of automotive composite materials according to claim 1, characterized in that, The process of processing the high-dimensional trajectory feature vectors to obtain the anchor point reachability probability map is performed using the principal component inverse projection method.

6. The method for predicting and controlling VOC release during the low-temperature degumming process of automotive composite materials according to claim 1, characterized in that, The segmented quadratic response surface model uses both the total VOC amount and the VOC component spectrum entropy as dual response values.

7. The method for predicting and controlling VOC release during the low-temperature degumming process of automotive composite materials according to claim 6, characterized in that, S4 specifically includes: Based on the probability density surface distribution characteristics in the anchor point achievability probability map, the three-dimensional original process parameter space composed of degumming temperature, degumming time and degumming agent concentration is processed to identify the boundary coordinate set where the probability density has a significant transition, and to generate a subdomain partitioning threshold vector that defines the physical boundaries of the stress relief dominant area, the interface decoupling active area and the gum volatilization window area. The subdomain partitioning threshold vector is used to perform a logical mask segmentation operation on the full parameter space, mapping continuous process parameter combinations to non-overlapping stress relief dominant regions, interface decoupling active regions, and colloid volatilization window regions to generate a set of reconstructed subdomain spaces. For the historical experimental data within the reconstructed subdomain space set, the degumming efficiency index and volatile organic compound emission data under the corresponding subdomain are extracted and related calculations are performed to obtain the total VOC scalar value and the VOC component spectrum entropy scalar value. Based on the process parameter combination input variables in the reconstructed subdomain space set and the corresponding VOC total scalar value and VOC component spectrum entropy scalar value output response, the operation is performed independently in each subdomain to construct a quadratic polynomial mathematical expression containing linear terms, interaction terms and square terms, so as to generate local quadratic response surface model coefficient matrices that are adapted to the stress relief dominant region, the interface decoupling active region and the colloid volatilization window region respectively. By integrating the coefficient matrix of the local quadratic response surface model with the corresponding subdomain boundary definition information, a switching logic rule is established to automatically route to a specific subdomain model based on the current process parameter position, so as to generate a piecewise quadratic response surface model.

8. The method for predicting and controlling VOC release during the low-temperature degumming process of automotive composite materials according to claim 7, characterized in that, The process of processing the three-dimensional original process parameter space consisting of degumming temperature, degumming time, and degumming agent concentration specifically involves performing contour tracing and gradient mutation detection.

9. The method for predicting and controlling VOC release during the low-temperature degumming process of automotive composite materials according to claim 7, characterized in that, The process of extracting degumming efficiency indices and volatile organic compound emission data under corresponding subdomains and performing related calculations to obtain scalar values ​​for total VOCs and VOC component entropy values ​​includes: The degumming efficiency index and volatile organic compound emission data under the corresponding subdomains were extracted respectively. The total release of volatile organic compounds at each sample point was calculated and the distribution frequency of its chemical components was statistically analyzed to obtain the total VOC scalar value and the scalar value of the VOC component spectrum entropy value.

10. The method for predicting and controlling VOC release during the low-temperature degumming process of automotive composite materials according to claim 7, characterized in that, The independent computation within each subdomain is performed using least squares regression fitting.

Images