A method for monitoring dynamic changes of monoterpenes in dry process of Amomum villosum based on near infrared spectroscopy

Abstract

The application provides a kind of grass fruit drying process monoterpenes component dynamic monitoring method based on near infrared spectroscopy, it is related to the technical field of agricultural products deep processing, this method includes: before drying, extract moisture content and particle size characteristics by initial spectrum, combined with clustering algorithm to automatically homogenize heterogeneous raw materials grading;In drying, through the pre-built chemometric school model, periodically analyze the spectrum to obtain the 1,8-eucalyptol fresh degree of quality characterization and material water activity of process characterization;Finally, execute dynamic control process, calculate aroma diffusion entropy to quantify quality risk, and calculate time axis drag amount to measure efficiency deviation, then vector synthesis protect incense and catch-up control signal, generate intelligent baking entropy reduction spoon to adjust operating parameters.This method makes dynamic trade-off between quality deterioration risk and drying efficiency lag, realizes the precise protection of heat-sensitive components and the coordinated optimization of production cycle.

Description

Technical Field

[0001] This invention relates to the field of agricultural product deep processing technology, specifically to a method for dynamic monitoring of monoterpenoid components in the drying process of cardamom based on near-infrared spectroscopy. Background Technology

[0002] In modern agriculture and the food industry, drying is a fundamental and crucial unit operation. Its fundamental purpose is to effectively inhibit microbial growth and adverse biochemical reactions by reducing the water activity of materials, thereby extending the shelf life of products and giving them a specific commercial form. For materials like cardamom, which are rich in volatile substances and medicinal components, the quality of the drying process not only affects their physical stability but also directly determines their core economic value, namely the final retention rate of monoterpenoid compounds, represented by the key medicinal component 1,8-cineole.

[0003] Under uniform heating conditions, small-sized cardamom pods with low moisture content prematurely complete the moisture evaporation and cooling stage, causing a rapid increase in their own temperature. This leads to the high-temperature decomposition or volatilization of monoterpenoid components, resulting in a "dry and charred" inferior product. Meanwhile, large-sized cardamom pods with high moisture content remain insufficiently dried, posing a risk of mold and spoilage. Current technology cannot detect and respond to this internal "out-of-sync" phenomenon within the material; its blind and extensive control mode directly leads to a severe decline in overall product quality and value loss. Furthermore, in existing technologies, when processing raw materials with uneven particle size and initial moisture content, sometimes the pursuit of production efficiency results in small-sized, low-moisture particles experiencing instantaneous moisture evaporation during continuous heating. This not only causes them to overheat and char rapidly but also leads to severe loss of monoterpenoid components due to intense water vapor evaporation. Meanwhile, large-sized particles remain insufficiently dried. Summary of the Invention

[0004] The purpose of this invention is to provide a method for dynamic monitoring of monoterpenoid components in the drying process of cardamom based on near-infrared spectroscopy, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A method for dynamic monitoring of monoterpenoid components in the drying process of cardamom based on near-infrared spectroscopy, the specific steps of which include: S1, in the initial stage of the drying process, performing an initial near-infrared spectral scan on all cardamom materials to be dried to obtain initial spectral data; S2. Analyze the initial spectral data and, based on the spectral characteristics related to the initial moisture content and particle size, identify and classify at least one batch of target cardamom with similar initial states. S3. During the subsequent drying process, near-infrared spectral sequence data of the target cardamom batches are periodically collected and analyzed to obtain in real time the freshness of 1,8-cineole, which characterizes the core value of monoterpenoid compounds, and the water activity of the material, which characterizes its dehydration potential. S4. Execute the dynamic control process, which is executed cyclically within each heating time window, including: S41. For the target batch of cardamom, calculate the quantitative indicators of its quality deterioration trend, namely the aroma dissipation entropy. S42. Based on the aroma dissipation entropy and combined with preset control rules, generate a smart baking entropy reduction spoon for the current heating window. S43. Apply the intelligent baking entropy reduction spoon to adjust the operating parameters of the drying equipment until the dynamic control process meets the preset termination conditions.

[0006] Furthermore, the method employs a pre-built chemometric correction model, which is used to perform the identification and segmentation in step S2 and the analysis in step S3. The chemometric correction model is established by collecting spectral data from standard samples with known initial moisture content, particle size, and monoterpene content.

[0007] Furthermore, the specific process of identifying and classifying the target cardamom batches in step S2 includes: extracting feature information related to water absorption and light scattering effects from the initial spectral data; processing the feature information using a clustering algorithm; and identifying one or more groups of cardamom materials with the minimum intra-class variance as the target cardamom batches.

[0008] Furthermore, in step S3, the freshness of 1,8-cineole was obtained by calculation based on the quantitative relationship between the spectrum and the CH bond vibration of monoterpenoid compounds; the water activity of the material was obtained by calculation based on the quantitative relationship between the spectrum and the OH bond vibration of water.

[0009] Furthermore, the aroma escaping entropy calculated in step S41 is calculated as follows: when the water activity of the target cardamom batch is lower than the preset critical threshold, the negative change rate of 1,8-cineole freshness within the current heating window is multiplied by a heat input factor characterizing the current heating intensity to obtain the final aroma escaping entropy.

[0010] Furthermore, step S41 also includes calculating the time axis drag amount, which characterizes the difference between the drying process and the preset target. The method for obtaining this is to compare the current material water activity of the target cardamom batch with the target water activity potential that should be achieved at the current moment according to the preset baseline drying curve, and the absolute value of the difference between the two is defined as the time axis drag amount.

[0011] Furthermore, the intelligent baking entropy reduction spoon generated in step S42 consists of an aroma-protecting entropy suppression factor and a beat-following pulse; by vector synthesis of the aroma-protecting entropy suppression factor and the beat-following pulse, the final intelligent baking entropy reduction spoon is formed.

[0012] Furthermore, the generation logic of the aroma entropy suppression factor is as follows: in response to the aroma dissipation entropy exceeding the preset quality warning line, a negative control signal is generated to reduce heat input, and the signal strength is positively correlated with the degree of aroma dissipation entropy exceeding the limit.

[0013] Furthermore, the generation logic of the beat catch-up pulse is as follows: in response to the time axis drag amount being greater than the preset efficiency tolerance range and the aroma dissipation entropy being lower than the quality warning line, a positive control signal aimed at increasing heat input is generated, and its signal strength is positively correlated with the degree of time axis drag amount.

[0014] Furthermore, the operating parameters adjusted in step S43 include heating power or hot air temperature; the termination condition of the dynamic control process is: the water activity of the target cardamom batch reaches the target endpoint, and its 1,8-cineole freshness enters a convergent and stable state within multiple consecutive heating time windows.

[0015] Compared with the prior art, the beneficial effects of the present invention are: This invention utilizes near-infrared spectroscopy and clustering algorithms before drying to rapidly and non-destructively classify cardamom raw materials with individual differences online, achieving the effect of dividing heterogeneous materials into multiple homogeneous batches with uniform internal states. This allows for the application of customized drying strategies to different batches, fundamentally avoiding the problems of "over-drying" and "under-drying" that exist in traditional mixed drying processes, laying a material foundation for improving the overall uniformity and pass rate of the finished product.

[0016] This invention constructs a high-precision chemometric correction model and applies it to real-time spectral analysis during the drying process to obtain key internal state indicators such as "1,8-cineole freshness," which characterizes the core value, and "material water activity," which characterizes the drying process. This method achieves closed-loop feedback control of the drying process, changing the traditional coarse-grained open-loop mode that relies solely on external temperature and time parameters. It directly links control decisions to the actual quality and dehydration state of the material, thus providing more precise protection for heat-sensitive active ingredients.

[0017] This invention assesses the risk of quality degradation by quantifying "aroma dissipation entropy" and measures drying efficiency by calculating "time-axis drag." It then vector-synthesizes a "fragrance-protecting entropy suppression factor" and a "cycle-following pulse" to form a unified control signal, achieving a dynamic and synergistic balance between the conflicting goals of ensuring finished product quality and pursuing production efficiency. This control strategy enables the system to make nuanced quantitative decisions within each control cycle, rationally following the production cycle while mitigating quality risks, thus optimizing overall output efficiency. Attached Figure Description

[0018] Figure 1 This is a schematic diagram illustrating the core logistics entities, scenarios, and technical routes of this invention; Figure 2 This is a schematic diagram illustrating the execution process of steps S1-S4 of the present invention. Detailed Implementation

[0019] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0020] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0021] Example 1: Please see Figure 1 This invention provides a technical solution: a method for dynamic monitoring of monoterpenoid components during the drying process of cardamom based on near-infrared spectroscopy, the specific steps of which include: S1. In the initial stage of the drying process, a preliminary near-infrared spectral scan is performed on all the cardamom materials to be dried to obtain initial spectral data. S2. Analyze the initial spectral data and, based on the spectral characteristics related to the initial moisture content and particle size, identify and classify at least one target cardamom batch with similar initial states. The specific process of identifying and classifying the target cardamom batch in step S2 includes: extracting feature information related to moisture absorption and light scattering effects from the initial spectral data; processing the feature information using a clustering algorithm, and determining one or more groups of cardamom materials with the minimum intra-class variance as the target cardamom batch.

[0022] In this embodiment, please refer to Figures 1 to 2 This invention provides a technical solution: a method for dynamic monitoring and regulation of monoterpenoid components in the drying process of cardamom based on near-infrared spectroscopy.

[0023] The practical application scenario of this embodiment is an automated grading and precision drying production line for cardamom in a large-scale spice processing center. A batch of newly harvested cardamom exhibits significant individual heterogeneity (different sizes and initial moisture contents) due to differences in its origin, harvesting time, and plant part. The goal is to rapidly and non-destructively classify this batch of mixed materials online before it enters the large-scale continuous drying equipment, dividing it into multiple "homogeneous batches" with highly consistent internal states. This allows for the application of optimal, customized drying strategies to different batches, avoiding the problems of "over-baking" or "under-drying" from the source.

[0024] Steps S1-S2: Automated grading of cardamom materials based on initial spectra; S1-S2 integrates the complete process from raw data acquisition and preprocessing to feature extraction, ultimately achieving batch classification of materials. Its internal implementation is as follows: S11. Production Line High-Throughput Spectral Data Acquisition: In this embodiment, cardamom material is laid out in a single layer on a conveyor belt moving at a uniform speed (preferably 0.5 m / s) and passes sequentially through a "spectral acquisition dark box" device. A Fourier transform near-infrared linear array spectrometer system is deployed inside the spectral acquisition dark box. Two rows of halogen lamps arranged symmetrically at a 45-degree angle provide uniform and stable broadband illumination to the central area of ​​the conveyor belt. The spectrometer's acquisition lens is perpendicular to the top of the conveyor belt, operating in online scanning mode, with a spectral acquisition range covering 900 nm to 2500 nm. As the cardamom passes below the lens, the spectrometer continuously acquires multiple spectral lines at high speed (preferably 500 lines / s), each line containing the spectral information of multiple independent cardamom individuals on the cross-section of the conveyor belt. A hyperspectral image is generated in real time; its spatial dimension corresponds to the width and length of the conveyor belt, while the spectral dimension records detailed absorbance data for each pixel (representing a very small cardamom surface area) in the 900-2500 nm range.

[0025] S12. Spectral Data Preprocessing and Correction: Hyperspectral images are susceptible to interference from various factors such as instrument noise, baseline drift, and particle scattering. To extract accurate chemical information, the following preprocessing is required: First, identify and remove "bad spots" caused by detector anomalies in the hyperspectral image. The Savitzky-Gorye smoothing algorithm is used to process each spectral curve. This algorithm effectively filters out high-frequency random noise by fitting data points within a local window using a polynomial, while preserving the true peak shape characteristics of the spectrum to the maximum extent. In this embodiment, a window width of 15 data points and a polynomial order of 2 are selected. Standard normal transformation is used to correct the hyperspectral image. By performing independent standardization processing on each spectral curve (subtracting its own mean and then dividing by its own standard deviation), the light scattering effect caused by differences in the size, shape, and surface roughness of individual cardamom fruits can be effectively eliminated, allowing the absorbance of the spectrum to more accurately reflect the content of its internal chemical components (water, organic matter).

[0026] S21. Two-dimensional core feature extraction: After preprocessing, two core feature dimensions are extracted from each spectral curve representing a single cardamom fruit or its local region to characterize its initial state: The area of ​​the water characteristic peak is denoted as A. water This feature is used to quantify the initial moisture content of cardamom. Moisture (H2O) exhibits a very strong OH-bonding absorption peak near 1940 nm in the near-infrared spectrum. This invention characterizes moisture content by calculating the integrated area under this absorption peak. Compared to using only peak height, the integrated area is less sensitive to peak shape changes and is more robust. The calculation formula is as follows: Where Abs(λ) is the absorbance at wavelength λ, and dλ is the differential symbol in calculus, representing an "infinitely small wavelength increment". Simply put, the interval [1900nm, 1980nm] is divided into countless extremely narrow "strips," each with a width of dλ and a height equal to the absorbance Abs at the current wavelength. The integration process involves summing the areas of all these infinitely many "strips" to obtain the total area. This indicates that the integration is performed on the wavelength. water The higher the value, the higher the initial moisture content of the cardamom.

[0027] Baseline drift index, denoted as I bs This feature indirectly reflects the particle size of cardamom. Although SNV correction has largely eliminated scattering, the residual overall baseline height is still related to particle size. Generally, larger cardamom particles cause more significant baseline drift. This invention selects a wavelength band with no obvious chemical absorption information (set to 1100nm-1200nm) and calculates its average absorbance as the baseline drift index.

[0028] Here, mean{…} represents the mathematical operation of calculating the average, which will calculate the arithmetic mean of all values ​​within the curly braces {}; find all absorbance values ​​Abs(λ) in the wavelength range of 1100nm to 1200nm, then calculate the average of these absorbance values, and assign this average to I. bs I bs A higher value generally indicates a larger cardamom particle size. Ultimately, each individual cardamom pod (or a representative region thereof) scanned on the production line is converted into a two-dimensional feature vector: V=[A water ,I bs ].

[0029] S22. Batch segmentation based on K-Means clustering: The K-means (K-Means) clustering algorithm is used to perform unsupervised classification on the two-dimensional feature vector V of all cardamom fruits, automatically dividing them into K "target cardamom fruit batches" with highly similar internal states. First, the value of K is set: K is the preset number of batches. In this embodiment, according to the production process requirements, K=3 is set, aiming to divide the cardamom fruits into three batches: "high moisture - large particle size", "medium state", and "low moisture - small particle size".

[0030] K-Means clustering process: S221. Randomly select 3 from all two-dimensional feature vectors as the initial cluster centers (centroids) C1, C2, and C3.

[0031] S222. For each two-dimensional feature vector V of a grass fruit, calculate its Euclidean distance to the three centroids C1, C2, C3, and assign it to the cluster represented by the nearest centroid.

[0032] S223. After all the two-dimensional eigenvectors have been assigned, recalculate the centroid of each cluster (i.e., the average value of all two-dimensional eigenvectors within the cluster) to obtain new C1', C2', C3'.

[0033] S224. Repeat steps S222 and S223 until the position of the centroid no longer changes after each iteration (or the change is less than a preset convergence threshold), at which point the clustering process ends.

[0034] S23. After clustering, all cardamom belonging to the same cluster are defined as the same "target cardamom batch". The final centroid coordinates of each cluster, i.e., [A... watercenter ,I bscenter This becomes the "digital ID card" defining the characteristics of this batch. Table 1 below shows the classification process for some samples.

[0035] Table 1: Example of batch partitioning of cardamom based on K-Means clustering (K=3) Finally, the batch assignment information for each cardamom pod is transmitted in real time to the pneumatic sorting unit at the end of the production line. As the cardamom pods fall from the conveyor belt, the sorting unit uses high-speed airflow to precisely blow them into the corresponding collection bins based on their batch labels (e.g., bin 1 corresponds to batch 1, bin 2 to batch 2, and bin 3 to batch 3). This completes the automated, non-destructive online grading of the mixed cardamom material, providing a homogeneous material basis for the precise and differentiated drying control in subsequent steps S3 and S4.

[0036] The technical principle of this embodiment is as follows: before drying, the initial spectrum of the cardamom material is obtained non-destructively using near-infrared spectroscopy, and the moisture content (moisture characteristic peak area A) is extracted. water ) and particle size (baseline drift index I) bs The invention identifies relevant characteristics and uses the K-Means clustering algorithm to achieve automated homogenization and grading of heterogeneous raw materials. Subsequently, during the drying process of specific batches, near-infrared spectroscopy is used for dynamic monitoring, and the "1,8-cineole freshness" (characterizing quality retention) and the "material water activity potential" (characterizing dryness degree) are analyzed in real time. This invention constructs a closed-loop dynamic control process, calculating the quality deterioration risk (aroma dissipation entropy S). aroma ) and drying efficiency deviation (time axis drag D) time This generates a comprehensive regulatory signal (intelligent baking entropy reduction spoon K). entropy This is used to adjust the operating parameters of the drying equipment in real time.

[0037] Compared to existing technologies, the advantages of this invention are reflected in the following aspects: Through pre-processed automated grading, customized drying strategies are matched to material batches in different initial states, fundamentally solving the problem of "over-drying" and "under-drying" coexisting in traditional mixed drying, thus improving the uniformity between finished product batches. Process monitoring and feedback control based on indicators directly related to economic value, such as "1,8-cineole freshness," changes the previous extensive mode that relied solely on temperature and time, making the protection of heat-sensitive active ingredients more precise. The dynamic control mechanism can moderately catch up with drying efficiency (compensating for time drag) while ensuring quality (suppressing aroma entropy), achieving a better balance between finished product quality and production cycle, improving the automation level of the drying process and the overall output value.

[0038] Please refer to Figure 1 , Figure 1The core physical entity / scenario of the invention: The physical carrier of the entire invention is an automated industrial drying production line. Its core scenario includes: a conveyor belt transporting cardamom raw materials of varying sizes and textures; a near-infrared (NIR) spectral scanner located above the conveyor belt, serving as the data acquisition source; a pneumatic sorting device following the scanner, which blows the cardamom into different collection bins based on the scan results, achieving physical grading; and a large industrial drying chamber where the graded cardamom undergoes precise closed-loop controlled drying. This drying chamber will be shown in cross-sectional / hidden-face views to reveal the internal drying process.

[0039] Figure 1 The isometric view at the top visually illustrates the hardware system upon which this method relies. The cardamom raw materials of varying sizes on the conveyor belt, the near-infrared spectroscopy scanner, and the subsequent pneumatic sorting device and collection bins collectively constitute the physical implementation of steps S1 (initial spectral scanning) and S2 (identifying and classifying target cardamom batches). This section clearly expresses the core idea of ​​this invention: automated homogenization and grading of heterogeneous raw materials using spectral technology. The cross-sectional view of the industrial drying chamber provides a scene for the subsequent drying process. In the diagram, BATCH1, BATCH2, and BATCH3 represent batch 1 (bin 1), batch 2 (bin 2), and batch 3 (bin 3), respectively, representing three batches: "high moisture - large particle size," "medium state," and "low moisture - small particle size." Figure 1 The lower part of the technical roadmap precisely reveals the core control logic of this invention. The first flowchart, "Initial Spectral Grading," summarizes the content of steps S1 and S2. The second flowchart, "Process Dynamic Monitoring," corresponds to step S3, where the system periodically collects spectral data in the drying chamber to obtain real-time information on eucalyptol freshness and material water activity. The third and fourth flowcharts—"Compound Index Calculation" and "Entropy Reduction Closed-Loop Control"—correspond in detail to the internal process of step S4: the system first executes step S41 to calculate compound indices such as aroma dissipation entropy, then executes steps S42 and S43 to generate an "intelligent baking entropy reduction spoon" and adjust drying parameters, forming a complete closed-loop control. The entire diagram tightly integrates physical execution with logical decision-making, comprehensively illustrating the entire process of this invention from raw material grading to precise dynamic control.

[0040] Example 2 Before performing online classification in step S2 and dynamic monitoring in step S3, a high-precision and robust chemometric correction model must be constructed in advance. This model serves as a mathematical bridge connecting "spectral data (X variable)" and "physicochemical properties of the material (Y variable)." Its construction process involves a rigorous offline experimental procedure, implemented as follows: S201. Construction of the Standard Sample Set and Precise Determination of Physicochemical Values: This step aims to obtain "standard answer" data for training the chemometrics calibration model. During the cardamom harvest season, over 500 cardamom samples were collected from plants of different production areas and maturity levels. During collection, it was deliberately ensured that the samples covered extreme sizes (from very small to very large fruits) and moisture contents (from freshly picked fruits to semi-dried fruits).

[0041] S202. Each sample is individually numbered and immediately divided into two parts. One part is used for spectral acquisition, and the other part is sent to the laboratory to determine its "ground true value" using national or industry standard methods: Initial moisture content (%), calculated by measuring the weight difference before and after drying using a 105℃ oven drying method. Particle size (mm): Using a digital vernier caliper with an accuracy of 0.01mm, the three orthogonal axes (length, width, and height) of each cardamom pod are measured, and the geometric mean is taken as the equivalent particle size. 1,8-Cineole content (mg / g): Gas chromatography-mass spectrometry (GC-MS) is used. After solvent extraction and purification of the sample, it is injected into the GC-MS system, and the peak area of ​​1,8-cineole at a specific retention time is used for precise quantification by referring to a standard curve. Water activity: Using a high-precision water activity meter (such as a capacitive or resistive sensor), the equilibrium relative humidity of the sample is measured under constant temperature conditions. This value is the material water activity (α). w ).

[0042] S203. The other half of the samples used for spectral acquisition in step 202 are scanned using a Fourier transform near-infrared spectrometer identical to that used in the online system (S11), acquiring spectral data in the 900-2500 nm range. Subsequently, the spectral data undergoes the same preprocessing procedure as in S12 (defect removal, SG smoothing, SNV correction). Finally, each processed spectral data is aligned with its corresponding four physicochemical values ​​(moisture content, particle size, 1,8-cineole content, and material water activity) to form a large dataset containing "spectral-physicochemical value" pairs.

[0043] S204. This invention uses Partial-Least-Squares-Regression (PLSR) as the core algorithm to construct the chemometric correction model. The PLSR algorithm excels at handling the multicollinearity problem commonly found in spectral data (i.e., data at different wavelengths are highly correlated) and can simultaneously model multiple Y variables (physicochemical values). Dataset partitioning: The "spectral-physicochemical value" paired dataset is randomly divided into training and validation sets at a 3:1 ratio using the Kennard-Stone algorithm. On the training set, the preprocessed spectral matrix is ​​used as input (X), and the four physicochemical value matrices are used as output (Y), to execute the PLSR algorithm. Through iteration, the algorithm finds a series of "latent variables" that can explain the variance of both the spectral matrix X and the physicochemical value matrix Y to the greatest extent possible, thereby establishing a robust regression relationship from X to Y.

[0044] S205, Performance Evaluation and Final Determination of the Chemometric Correction Model; Calculations are performed using the training set data. This measures how well the model fits the data used to train it. It reflects how well the chemometric correction model "learns." All samples from the training set are used, with n... c indivual.

[0045] For each sample i in the training set, obtain its true value y. i and model predictions .

[0046] Calculate the sum of squared residuals of the training set, denoted as SS. rec,c : ; Calculate the total sum of squares of the training set, denoted as SS. tot,c : ; The corrected coefficient of determination is denoted as . The specific formula is as follows: ; For each sample j in the validation set, the number of samples is set to n. p Obtain its true value y j and model predictions .

[0047] Calculate the sum of squared residuals on the validation set, denoted as SS. rec,p : ; Calculate the total sum of squares of the validation set, denoted as SS. tot,p : ; Verify the coefficient of determination, denoted as . The specific formula is as follows: ; The predictive performance of the chemometric corrected model was evaluated by adjusting and validating the coefficient of determination to ensure its generalization ability. The evaluation metrics are shown in Table 2 below.

[0048] Table 2: Examples of performance evaluation metrics for chemometrics-corrected positive model The chemometric correction model is finally determined and deployed to the online system for real-time analysis of S2 and S3 only when the validation determination coefficients of all predicted targets are higher than 0.90 and the RMSEP is within the acceptable process error range. The technical principle of this embodiment lies in constructing and validating a high-precision chemometric calibration model through a rigorous offline process. This process first constructs a standard sample set covering a wide range of particle sizes and moisture contents. Secondly, two parallel analyses are performed on each sample: one uses standard methods (such as GC-MS or oven drying) to accurately determine its 1,8-cineole content, water activity, and other true physicochemical values; the other acquires its near-infrared spectrum. Then, the paired "spectral-physicochemical value" datasets are used to establish a mathematical model using the partial least squares regression (PLSR) algorithm. Finally, the model's predictive accuracy and generalization ability are evaluated and confirmed through an independent validation set and rigorous performance metrics (such as a validation determination coefficient Rp² > 0.90).

[0049] Compared to existing methods with unclear model building processes or insufficient validation, the advantages of this invention are reflected in: ensuring the accuracy and reliability of the decision-making basis for subsequent online grading and dynamic control; providing high-quality training data for the model through systematic sample collection and high-precision physicochemical value calibration; effectively avoiding the "overfitting" problem of the model by using an independent validation set, ensuring that the model still has stable predictive performance when faced with new samples not seen on the production line; and successfully transforming and solidifying the analytical capabilities of time-consuming and destructive laboratory analysis methods into a near-infrared spectroscopy model that can be applied rapidly, non-destructively, and online, providing core technical support for the realization of the entire automated process.

[0050] Example 3 S3. During the subsequent drying process, near-infrared spectral sequence data of the target cardamom batches are periodically collected and analyzed to obtain in real time the freshness of 1,8-cineole, which characterizes the core value of monoterpenoid compounds, and the water activity of the material, which characterizes its dehydration potential. The method employs a pre-built chemometric correction model, which is used to perform the identification and segmentation in step S2 and the analysis in step S3. The chemometric correction model is established by collecting spectral data from standard samples with known initial moisture content, particle size, and monoterpene content.

[0051] In step S3, the freshness of 1,8-cineole was obtained by calculation based on the quantitative relationship between the spectrum and the CH bond vibration of monoterpenoid compounds; the water activity of the material was obtained by calculation based on the quantitative relationship between the spectrum and the OH bond vibration of water.

[0052] Step S3: Dynamic monitoring and data examples of key indicators in the drying process: After completing the homogenization grading of cardamom in S2, the cardamom from the same batch is sent to the drying equipment. During the drying process, step S3 is executed by using a near-infrared probe deployed inside the drying chamber, or by periodically taking out a small number of samples for rapid scanning. S31: Calculation of core indicators in dynamic monitoring; using the constructed chemometric correction model, the periodically collected spectra are analyzed in real time to calculate two core dynamic indicators: 1,8-cineole freshness (%): This indicator is not a direct predicted value, but a secondary calculation result based on the 1,8-cineole content, used to intuitively reflect the retention rate of the core value component.

[0053] The chemometric correction model predicts the real-time eucalyptol content (mg / g) based on the current spectrum. This is then compared to the initial eucalyptol content of the batch at the start of drying (0 hours). 1,8-Cyclocarpine freshness = (current 1,8-cineole content / initial 1,8-cineole content) × 100%; 1,8-cineole freshness starts at 100% and decreases over drying time, making it crucial for assessing whether the drying process "preserves quality." Material water activity (α...) w This index is directly predicted from the spectrum by the chemometric calibration model. It reflects the ability of "free water" to escape from the material and is a core physical quantity for judging the drying rate and endpoint. Its value range is from 0 to 1, and the closer it is to 0, the drier the material.

[0054] S32. Dynamic monitoring data representation, for example, Table 3 below shows a set of typical data recorded during dynamic monitoring of "Batch 3 (High Moisture - Large Particle Size)" cardamom in the drying equipment for the "Batch 3 (High Moisture - Large Particle Size)" segment in S2.

[0055] Table 3: Example of dynamic monitoring data during the drying process of cardamom in batch 3 The essence of the status interpretation and corresponding control commands is an expert decision-making system based on preset rules, rather than an artificial intelligence capable of free creation. Its core task is to transform complex real-time monitoring data into executable decisions that meet process requirements. The underlying principle is to systematically engineer the wisdom of human experts and a large amount of experimental data into a rigorous, predictable, automated "brain center." The construction first clarifies its closed-loop working logic: receiving input, processing it, and generating output. The input is real-time quantitative data from the front-end analysis model, mainly including eucalyptol freshness (%) and material water activity (α). w The three core indicators are: input data, drying time (in hours), and drying time. The core of the processing is a "rule engine" that internally stores a "rule base," continuously matching the input data with the rules in the library at high speed. The output consists of two parts: first, a "status interpretation" text describing the current material status for operators' reference; and second, explicit "corresponding control instructions" that can directly drive the equipment.

[0056] The actual operation of the expert decision-making system relies on solidifying the above knowledge into specific corresponding control command logic rules, and achieving fully automatic closed-loop control through integration with hardware.

[0057] The implementation structure of the rules is clear and straightforward, triggering corresponding control commands by setting conditions. The following is an example of a constructed control rule demonstrating how decision-making is performed: Rule 1: Initial stage of rapid dehydration; when (drying time < 1 hour) and (material water activity > 0.95), the status interpretation is generated as: "Initial status: material is full and moisture is sufficient...", and the control instruction is generated as: "First instruction: start high temperature rapid dehydration mode (set temperature 100℃, high wind speed)".

[0058] Rule 2: Mid-stage of quality balance; when (material water activity < 0.7) and (material water activity > 0.4); the generated state interpretation is: "Mid-stage dehydration period: drying rate slows down..." and the control instruction is generated: "Second instruction: appropriately reduce the drying temperature (set temperature 85℃) to slow down the loss of volatile oils. Rule 3: Stop the drying endpoint stage; when (material water activity ≤ 0.20); generate the status interpretation: "Endpoint judgment: water activity has reached the target threshold..." and generate the control instruction: "Third instruction: Stop drying and enter the cooling stage". Rule 4: Quality Protection Abnormal Alarm Stage; When (1,8-cineole freshness < 90%) and (material water activity > 0.30); the status interpretation is: "Warning: Volatile oil loss too fast!", and the control instruction is generated: "Fourth instruction: Immediately execute the protection procedure (significantly reduce the temperature to 70℃)..." The technical principle of this embodiment lies in the periodic acquisition of near-infrared spectra during the drying process of homogenized cardamom batches, and the real-time analysis using a pre-established chemometric correction model. This process calculates two core dynamic indicators: one is the "material water activity (α)" which directly reflects the dehydration process. w The first method involves comparing the real-time 1,8-cineole content with the initial value to obtain the "1,8-cineole freshness (%)", which provides a direct assessment of quality retention. These real-time quantitative indicators are input into an expert decision-making system based on preset rules. This system automatically matches and generates clear and executable equipment control instructions according to the different ranges in which the indicators are located.

[0059] Compared to existing technologies, the beneficial effects of this invention are reflected in the following aspects: It achieves real-time "visualization" of the internal state of the drying process, transforming the traditional time-temperature-dependent "open-loop" control into a "closed-loop" control based on feedback from the actual physicochemical state of the material. It introduces the relative indicator of "1,8-cineole freshness," providing a unified and intuitive benchmark for evaluating and protecting core value components, making quality control objectives clearer. Furthermore, through a rule-based expert system, human experience and process knowledge are solidified into automated decision-making logic, reducing reliance on operator experience and ensuring the consistency and repeatability of drying strategies across different batches, thus contributing to the stabilization and improvement of final product quality.

[0060] Example 4 S4. Execute the dynamic control process, which is executed cyclically within each heating time window, including: S41. For the target batch of cardamom, calculate the quantitative indicators of its quality deterioration trend, namely the aroma dissipation entropy. The aroma escaping entropy calculated in step S41 is calculated as follows: when the water activity of the target cardamom batch is lower than the preset critical threshold, the negative change rate of 1,8-cineole freshness within the current heating window is multiplied by a heat input factor characterizing the current heating intensity to obtain the final aroma escaping entropy. Step S41 also includes calculating the time axis drag amount, which characterizes the difference between the drying process and the preset target. The method for obtaining this amount is to compare the current material water activity of the target cardamom batch with the target water activity potential that should be achieved at the current moment according to the preset baseline drying curve. The absolute value of the difference between the two is defined as the time axis drag amount.

[0061] S42. Based on the aroma dissipation entropy and combined with preset control rules, generate a smart baking entropy reduction spoon for the current heating window. The intelligent baking entropy reduction spoon generated in step S42 consists of an aroma protection entropy suppression factor and a beat-following pulse; the final intelligent baking entropy reduction spoon is formed by vector synthesis of the aroma protection entropy suppression factor and the beat-following pulse.

[0062] The generation logic of the beat catch-up pulse is as follows: in response to the time axis drag amount being greater than the preset efficiency tolerance range and the aroma dissipation entropy being lower than the quality warning line, a positive control signal aimed at increasing heat input is generated, and its signal strength is positively correlated with the degree of time axis drag amount.

[0063] The generation logic of the aroma entropy suppression factor is as follows: in response to the aroma emission entropy exceeding the preset quality warning line, a negative control signal is generated to reduce heat input, and the signal strength is positively correlated with the degree of aroma emission entropy exceeding the limit.

[0064] S43. Apply the intelligent baking entropy reduction spoon to adjust the operating parameters of the drying equipment until the dynamic control process meets the preset termination conditions.

[0065] The operating parameters adjusted in step S43 include heating power or hot air temperature; the termination condition of the dynamic control process is: the water activity of the target cardamom batch reaches the target endpoint, and its 1,8-cineole freshness enters a convergent and stable state within multiple consecutive heating time windows.

[0066] Specific data example: The present invention enters the final dynamic control process (S4). This process aims to achieve the optimal balance between drying efficiency and finished product quality, and its core is an intelligent decision-making and control closed loop that is executed cyclically within each "time heating window" (e.g., set to 10 minutes).

[0067] S41. Real-time quantitative calculation of composite state indicators: At the beginning of each heating time window, the method first calculates two key composite state indicators based on the real-time data obtained in S3, which are used to guide control decisions, including: S411. Calculate the aroma dissipation entropy, denoted as: S aroma This indicator aims to quantify the risk of degradation of the core aroma quality of cardamom under the current heating intensity. It is not a thermodynamic entropy, but rather borrows the concept to characterize the rate of quality "disordering" or "dissipation." Aroma dissipation entropy only occurs in the middle and late stages of material dehydration, i.e., when the material's water activity (a) reaches a certain level. w If the value is below a preset critical threshold, it is denoted as a. wcrit (In this embodiment, the value is set to 0.70) to initiate the calculation. This is because before this point, water evaporation is the primary process, and high temperature has relatively little impact on the volatile oils. The aroma dissipation entropy S at the current time t is... aroma The formula for calculating (t) is: in, V represents the change in 1,8-cineole freshness within the current heating window. euca (t) V euca (t Δt); Since freshness decreases, this value is negative, so taking its negative value yields a positive rate of change Δt, which represents the length of the heating window (e.g., 10 minutes or 600 seconds). The heat input factor is a dimensionless, normalized value characterizing the current heating intensity, calculated as follows: Among them, P current (t) is the current heating power of the equipment, P max This is the maximum heating power of the equipment; Example: The freshness of the previous window was set to 94.1%. The freshness of the current window is 93.5%. Therefore, the change in eucalyptol freshness is ΔV. euca =93.5% 94.1% = 0.6%. The length of each time window is set at the beginning of the example, set to 10 minutes. Thermal input factor H factor The current heating power is 80% of the maximum power. Therefore, this dimensionless factor H... factor =0.8.

[0068] Substitute into the calculation: ; S412. Calculate the time axis drag, denoted as D. time This indicator is used to quantify the degree of deviation of the current drying process from an idealized "baseline drying curve" and is the core benchmark for measuring drying efficiency.

[0069] The baseline drying curve is denoted as a. w,ref (t): This is a pre-defined curve representing the optimal drying path. Based on the initial state of the batch of cardamom, it defines the ideal water potential target to be achieved at each time t. This curve can be obtained by fitting experimental data, specifically through exponential decay curve fitting. Among them, a w,start and a w,end denoted as the initial and target water activity potentials, respectively, and k is the drying rate constant; It should be noted that the calibration method for the drying rate constant k includes: during the system development phase, technicians will conduct a large number of drying experiments on all typical batches that may be divided in S2 (such as "high moisture - large particle size", "low moisture - small particle size", etc.). For each typical batch, the process verified as optimal will be used for drying, and its water activity potential a will be recorded throughout the process. wThe data changing with time t will be collected (t, a w Data points were used to perform nonlinear fitting on the following exponential decay model using mathematical software (such as MATLAB's CurveFittingToolbox or Python's SciPy library): The software automatically calculates the k-value that best fits the experimental data. The characteristics of different batches, their optimal k-values, and corresponding initial / termination water activity potentials are stored in a parameter lookup table, as shown in Table 4. In actual production, the system's workflow is as follows: A new batch of cardamom enters the drying equipment. The system executes step S2, identifying it as, for example, a "high moisture - large particle size" batch. The control system immediately looks up the corresponding k value for this batch from the parameter lookup table, obtaining k=0.35. Using this k value, along with the batch's actual initial water activity potential (e.g., 0.98) and target water activity potential (0.20), the system instantly generates a "baseline drying curve" for this drying task. ; In the subsequent S4 dynamic control process, the system will continuously compare the real-time measured aw(t) with this baseline curve aw,ref(t) to calculate the "time axis drag".

[0070] The time axis drag amount is denoted as D. time Calculate the time axis drag D at the current time t. time Calculated using the following formula: in, This represents the actual measured water activity of the material at the current time t. This is the ideal water potential target that should be achieved at the current time t, based on the baseline curve.

[0071] Example: At the 4th hour of drying, according to the baseline curve, the material water activity should be 0.55. However, the actual measured value is 0.60. Therefore, the time axis drag is D. time =|0.60 0.55 | = 0.05. The larger this value is, the more the drying progress lags behind the preset cycle time.

[0072] S42, Intelligent Baking Entropy Reduction Spoon, denoted as K entropy K entropy It is a comprehensive regulatory signal that vector-synthesizes two regulatory factors with opposing objectives, aiming to dynamically balance the conflicting goals of "catching up with efficiency" and "protecting quality." (Fragrance entropy suppression factor F)protect The generation logic is as follows: when the risk of quality degradation exceeds the safe range, a negative regulatory signal is generated to "apply the brakes." The trigger condition is: the aroma dissipation entropy S. aroma Greater than the preset quality warning line S alert (For example, set to 0.04% / min).

[0073] in, This is the first proportional gain constant, and the magnitude of the proportional gain constant is proportional to the degree of risk exceeding the limit.

[0074] The beat catches up with the pulse, denoted as P. catchup The generation logic is as follows: when drying efficiency lags behind and quality is relatively safe, a positive control signal is generated to "accelerate" the process. The trigger condition is the time axis drag amount D. time Greater than the efficiency tolerance interval D tol (For example, set to 0.02), and the aroma dissipation entropy S aroma Below the quality warning line S alert .

[0075] Calculation formula: in, This is the second proportional gain constant; The final synthesis of the intelligent baking entropy reduction spoon K entropy It is the algebraic sum of the two factors mentioned above, and it represents the net adjustment to the current heating power.

[0076] This value can be positive, negative, or zero, providing precise guidance for the direction and intensity of the next round of regulation.

[0077] At the end of each heating window, the system applies the calculated entropy reduction key to update the device operating parameters P for the next window. next . Among them, P next This is the target heating power for the next window. The system will also ensure P... next Always within the safe operating range of the equipment; Assuming the total power of the drying equipment is 50kW, we map its power control range (0-100%) to the output range of the control signal. Using the above tuning method, for "Batch 3 (High Moisture - Large Particle Size)" cardamom, we obtain the following set of specific gain constants with good performance: First proportional gain constant (K... p =800; Physical meaning: when the aroma dissipates, the entropy S aroma Quality warning line Salert When the value exceeds 0.01% / min, the fragrance entropy inhibition factor F protect This will generate a control signal of -800 × 0.01 = -8. This is equivalent to reducing the heating power by 8%. This level of response allows for timely intervention without causing system instability.

[0078] Second proportional gain constant (K) d =300; Physical meaning: When the time axis drag is D time When the deviation of water activity reaches 0.05 (as in the case of the 4th hour in the example), as long as the quality is safe (S aroma ≤S aler ), beat chasing pulse P catchup This will generate a control signal of +300 × 0.05 = +15. This is equivalent to increasing the heating power by 15%, using a firm but not overly aggressive approach to catch up with the preset drying cycle. This set of values ​​(K... p =800,K d =300) reflects a control strategy of "quality first, efficiency second": the system's response to quality risks (coefficient 800) is more sensitive and stronger than its response to efficiency lags (coefficient 300). This is entirely in line with the core requirements of high-quality spice drying.

[0079] The S4 loop control process does not run indefinitely; it will automatically terminate when the following two preset termination conditions are met: Drying endpoint reached: material water activity a w (t) reaches or falls below the final target value a w,end (Set to 0.20).

[0080] Quality status convergence: 1,8-Cineole freshness V euca It enters a steady state within multiple consecutive heating time windows (set to 3), meaning the absolute value of its rate of change is consistently less than a very small convergence threshold. stable (Set to 0.01% / min). This indicates that at the current low temperature, the quality no longer deteriorates significantly.

[0081] The technical principle of this embodiment lies in constructing a dynamic control process that is executed cyclically within each time window. This process first calculates two core composite indicators: one is the "aroma dissipation entropy (S)," which quantifies the risk of quality degradation. aroma The first is the "time-axis drag" which correlates the rate of change in 1,8-cineole freshness with heat input intensity; the second is the "time-axis drag" that quantifies the deviation in drying efficiency. timeThe system compares the real-time water activity with a preset baseline curve. Then, based on these two indicators, it generates two opposing control factors: a negative "aroma protection entropy suppression factor" is generated to reduce heat input when aroma evaporation entropy exceeds the limit; a positive "beat-catching pulse" is generated to increase heat input when drying progress lags behind but quality is safe. Finally, these two factors are algebraically summed to synthesize a net control signal, "Intelligent Baking Entropy Reduction Spoon (K)". entropy This is used to precisely adjust the heating power for the next period until the preset termination conditions are met.

[0082] Compared to existing technologies, the advantages of this invention are reflected in the following: by using two original quantitative indicators, "aroma dissipation entropy" and "time axis drag," the mutually restrictive goals of "quality" and "efficiency" are transformed into calculable and comparable engineering parameters. The proposed "intelligent baking entropy reduction spoon" control logic abandons the either-or switching mode in traditional control, and achieves a dynamic and synergistic balance between the two goals of "quality preservation" and "efficiency improvement" through vector synthesis. The control system is no longer a simple tracking or protection system, but can make delicate and quantitative decisions that best suit the current interests based on real-time risk and efficiency assessments in each control cycle, thereby seeking reasonable production efficiency while ensuring the quality of the finished product.

[0083] It should be noted that all calculation formulas in this application employ regression analysis, including but not limited to machine learning algorithms, to deeply analyze the collected parameters and identify their natural trends and interrelationships. Specialized software, such as Python's Scikit-learn library or the R language, is used to automatically generate mathematical models that match the data. Then, cross-validation and other methods are used to objectively evaluate the model performance, and continuous feedback and optimization are combined to ensure that the created formulas truly reflect the inherent laws of the data, thereby guaranteeing their effectiveness and accuracy. In all calculation formulas in this application, the parameters in each formula undergo dimensionless processing within a consistent range to ensure that different physical quantities are compared on the same scale; dimensionless processing techniques include, but are not limited to, min-max-normalization and Z-score standardization. The algorithm of this invention is implemented as a Python script. Before executing the core logic, the program first executes a data loading module (e.g., using the widely used p and as libraries in Python) configured to read the aforementioned spreadsheet file and load its contents into the program's working memory (e.g., a DataFrame data structure). Subsequent algorithm steps will directly query and retrieve the required configuration parameters from this memory data structure.

[0084] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for dynamic monitoring of monoterpenoid components during the drying process of cardamom based on near-infrared spectroscopy, characterized in that, The specific steps include: S1. In the initial stage of the drying process, a preliminary near-infrared spectral scan is performed on all the cardamom materials to be dried to obtain initial spectral data. S2. Analyze the initial near-infrared spectral data and, based on the spectral characteristics related to the initial moisture content and particle size, identify and classify at least one batch of target cardamom with the same initial state. S3. During the subsequent drying process, near-infrared spectral sequence data of the target cardamom batches are periodically collected and analyzed to obtain in real time the freshness of 1,8-cineole, which characterizes the core value of monoterpenoid compounds, and the water activity of the material, which characterizes its dehydration potential. S4. Execute the dynamic control process, which is executed cyclically within each time heating window, including: S41. For the target batch of cardamom, calculate the quantitative index of its quality deterioration trend, specifically the aroma dissipation entropy. S42. Based on the aroma dissipation entropy and combined with preset control rules, generate a smart baking entropy reduction spoon for the current heating window. S43. Apply the intelligent baking entropy reduction spoon to adjust the operating parameters of the drying equipment until the dynamic control process meets the preset termination conditions.

2. The method for dynamic monitoring of monoterpenoid components in the drying process of cardamom based on near-infrared spectroscopy according to claim 1, characterized in that: The method employs a pre-built chemometric correction model, which is used to perform the identification and segmentation in step S2 and the analysis in step S3. The chemometric correction model is established by collecting spectral data from standard samples with known initial moisture content, particle size, and monoterpene content.

3. The method for dynamic monitoring of monoterpenoid components in the drying process of cardamom based on near-infrared spectroscopy according to claim 1, characterized in that: The specific process of identifying and classifying the target cardamom batches in step S2 includes: extracting feature information related to water absorption and light scattering effects from the initial spectral data; processing the feature information using a clustering algorithm; and identifying one or more groups of cardamom materials with the minimum intra-class variance as the target cardamom batches.

4. The method for dynamic monitoring of monoterpenoid components in the drying process of cardamom based on near-infrared spectroscopy according to claim 1, characterized in that: In step S3, the freshness of 1,8-cineole was obtained by calculation based on the quantitative relationship between the spectrum and the CH bond vibration of monoterpenoid compounds; the water activity of the material was obtained by calculation based on the quantitative relationship between the spectrum and the OH bond vibration of water.

5. The method for dynamic monitoring of monoterpenoid components in the drying process of cardamom based on near-infrared spectroscopy according to claim 1, characterized in that: The aroma escaping entropy calculated in step S41 is calculated as follows: when the water activity of the target cardamom batch is lower than the preset critical threshold, the negative change rate of 1,8-cineole freshness within the current heating window is multiplied by a heat input factor characterizing the current heating intensity to obtain the final aroma escaping entropy.

6. The method for dynamic monitoring of monoterpenoid components in the drying process of cardamom based on near-infrared spectroscopy according to claim 1, characterized in that: Step S41 also includes calculating the time axis drag amount, which characterizes the difference between the drying process and the preset target. The method for obtaining this amount is to compare the current material water activity of the target cardamom batch with the target water activity potential that should be achieved at the current moment according to the preset baseline drying curve. The absolute value of the difference between the two is defined as the time axis drag amount.

7. The method for dynamic monitoring of monoterpenoid components in the drying process of cardamom based on near-infrared spectroscopy according to claim 1, characterized in that: The intelligent baking entropy reduction spoon generated in step S42 consists of an aroma protection entropy suppression factor and a beat-following pulse; the final intelligent baking entropy reduction spoon is formed by vector synthesis of the aroma protection entropy suppression factor and the beat-following pulse.

8. The method for dynamic monitoring of monoterpenoid components in the drying process of cardamom based on near-infrared spectroscopy according to claim 1, characterized in that: The generation logic of the aroma protection entropy suppression factor is as follows: in response to the aroma emission entropy exceeding the preset quality warning line, a negative control signal is generated to reduce heat input, and the signal strength is positively correlated with the degree of aroma emission entropy exceeding the limit.

9. The method for dynamic monitoring of monoterpenoid components in the drying process of cardamom based on near-infrared spectroscopy according to claim 1, characterized in that: The generation logic of the beat catch-up pulse is as follows: in response to the time axis drag amount being greater than the preset efficiency tolerance range and the aroma dissipation entropy being lower than the quality warning line, a positive control signal aimed at increasing heat input is generated, and its signal strength is positively correlated with the degree of time axis drag amount.

10. The method for dynamic monitoring of monoterpenoid components in the drying process of cardamom based on near-infrared spectroscopy according to claim 1, characterized in that: The operating parameters adjusted in step S43 include heating power or hot air temperature; the termination condition of the dynamic control process is: the water activity of the target cardamom batch reaches the target endpoint, and its 1,8-cineole freshness enters a convergent and stable state within multiple consecutive heating time windows.

Images