Preparation method of carbonized ginseng

Abstract

The application discloses a carbonized ginseng preparation method, and relates to the technical field of traditional Chinese medicine processing, wherein a closed type program-controlled carbonization reactor is used as a core equipment, and a three-stage gradient pyrolysis process is sequentially performed on ginseng decoction pieces under the protection of an inert atmosphere formed by one kind of gas selected from nitrogen and carbon dioxide. Through multi-sensor fusion and a partial least squares-back propagation neural network intelligent judgment algorithm, the carbonization end point of "carbon storage" is converted from subjective sensory judgment into objective quantitative digital automatic judgment. The batch-to-batch variation coefficient of the content of Rg3, the concentration of carbon points and the carbonization degree and other key quality attributes of the finished product can be controlled within 5%, and the quality of batches is highly consistent.

Description

Technical Field

[0001] This invention relates to the field of traditional Chinese medicine processing technology, specifically a method for preparing carbonized ginseng. Background Technology

[0002] Charred medicines are a special type of processed medicine commonly used in traditional Chinese medicine. After ginseng is charred, it is transformed from a tonic that greatly replenishes vital energy into a hemostatic medicine with the effects of replenishing qi and controlling blood. Clinically, it is used to treat critical conditions such as metrorrhagia, hematochezia, and massive bleeding and collapse caused by qi deficiency. The correlation between its clinical efficacy and the degree of charring has been confirmed by traditional Chinese medicine books and modern pharmaceutical research.

[0003] Currently, the closest preparation method to this invention in industrial production is the traditional iron pot open-flame charring method: dried ginseng slices are placed in an iron charring pot, and the pot is heated to 200°C to 300°C by an open flame or electric heating device. The operator continuously stirs the ginseng with a shovel and judges whether the processing endpoint has been reached by observing whether the surface color of the ginseng turns charred black, whether the color of the cut surface inside turns charred yellow, and whether a charred aroma is produced. The ginseng is then taken out and spread out to cool naturally. Some improved solutions use automated drum charring machines to replace manual iron pots, but the endpoint determination still depends on human experience, and the heating environment is still an open oxygen condition. The fundamental problem has not been solved.

[0004] The aforementioned traditional process suffers from the following four core technological defects: Firstly, the determination of the processing endpoint relies entirely on the senses of the operators. Due to factors such as individual differences, fatigue levels, and years of experience, the actual carbonization degree of each batch of ginseng fluctuates significantly. The batch-to-batch differences in the content of rare saponins and the amount of nano carbon dots generated in the finished product are significant, which cannot meet the requirements of modern pharmacopoeia for the uniformity and traceability of the quality of Chinese herbal medicine slices.

[0005] Secondly, traditional open-flame frying uses continuous heating in a single temperature zone without the ability to control the temperature in stages. The two competing reactions of ginsenoside deglycosylation (generating rare saponins Rg3 and Rh2) and further thermal decomposition and degradation of rare saponins are superimposed and cannot be decoupled. The net accumulation of rare saponins is far lower than the theoretical value, and a large number of precious active ingredients are deeply degraded.

[0006] Third, open-air roasting is carried out in an aerobic environment, where some materials undergo localized aerobic combustion. Carbon precursors (polysaccharides, proteins) are oxidized and consumed, resulting in reduced carbon point formation and uneven particle size distribution. At the same time, incomplete aerobic combustion produces carcinogenic byproducts such as benzo[a]pyrene, which seriously threaten product safety.

[0007] Fourth, open-flame roasting generates a large amount of fumes containing volatile organic compounds that are emitted into the production workshop without proper organization, endangering the occupational health of the workers and failing to meet current air pollutant emission standards and green manufacturing requirements for traditional Chinese medicine.

[0008] Modern pharmaceutical research has confirmed that under heating conditions of 100℃ to 250℃, the glycosyl groups (glucose, rhamnose, etc.) attached to the C-3, C-6, and C-20 positions of the saponin nucleus undergo sequential breakage due to the thermal energy exceeding the C-O glycosidic bond energy, gradually generating rare saponins Rg3 (losing one C-20 glucose group), Rh2 (losing two glycosyl groups), and finally the aglycone. Thermal processing parameters are key variables controlling the direction and efficiency of saponin conversion. Recent studies have shown that nano-carbon dots with particle sizes of 2nm to 10nm isolated and identified from traditional charred medicinal materials such as schizonepeta charcoal, typha pollen charcoal, and sanguisorba charcoal are crucial for promoting the efficacy of these materials. One of the key material bases for hemostatic activity is that its activity mechanism may be related to the activation of coagulation cascade reactions by oxygen-containing functional groups on the surface of carbon dots. Near-infrared spectroscopy (NIR) technology has been recommended by the US FDA, EU EMA and China NMPA for online real-time monitoring of pharmaceutical production processes. The quantitative analysis model established by combining partial least squares (PLS) with NIR spectroscopy has been widely commercialized in pharmaceutical content determination and other fields. However, the above technologies have not yet been introduced into the field of traditional Chinese medicine carbonization processing. The technical problems in the existing ginseng carbonization process, such as the digital determination of the processing endpoint, the directional conversion and regulation of effective components, and the controllable preparation of functional nano carbon dots, have not been fundamentally solved. Summary of the Invention

[0009] The technical problem to be solved by this invention is to overcome the technical defects of traditional ginseng charring process, such as reliance on subjective sensory judgment of the processing endpoint, inability to directionally and controllably prepare effective components (rare saponins and functional nano carbon dots), production of carcinogenic polycyclic aromatic hydrocarbon byproducts by aerobic combustion, and pollution of the production environment. The invention provides a carbonized ginseng preparation method that can achieve digital determination of the processing endpoint, simultaneous and directional enrichment of rare saponins and functional nano carbon dots, and a safe and environmentally friendly process.

[0010] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows: A method for preparing carbonized ginseng, using a closed-loop, programmable carbonization reactor as the core equipment, involves sequentially performing a three-stage gradient pyrolysis process on ginseng slices under an inert atmosphere selected from nitrogen and carbon dioxide, including: The first stage is vacuum-assisted low-temperature dehydration: the gas pressure in the sealed reactor is reduced to 20 kPa to 50 kPa, and the temperature of the chamber is increased to 100°C to 120°C at a heating rate of 3°C / min to 5°C / min and held for 30 to 60 minutes. The completion of this stage is determined by the thermogravimetric sensor showing a stable mass loss rate of 10% to 15% and a significant decrease and stabilization of the OH stretching vibration absorption peaks near 1450 nm and 1940 nm in the near-infrared spectrum. Once completed, the system automatically switches to inert atmosphere positive pressure protection and enters the second stage. The second stage is thermal deglycosylation: the temperature is increased to 150℃ to 180℃ at a heating rate of 2℃ / min to 5℃ / min, and kept at this temperature for 60 to 90 minutes under positive pressure protection in an inert atmosphere, which drives the sequential thermal breakage of CO glycosidic bonds in ginseng macromolecular saponins, causing the main saponins such as Rb1 and Re to be converted into rare saponins Rg3 and Rh2 in turn. The third stage is critical carbonization short-time thermal shock: the temperature is increased to 220℃ to 250℃ at a heating rate of 3℃ / min to 8℃ / min, and thermally shocked for 10 to 20 minutes under the protection of an inert atmosphere and positive pressure, so that ginseng polysaccharide and protein undergo Maillard reaction and carbonization condensation under anaerobic pyrolysis conditions, and functional nano carbon dots with a particle size of 2nm to 8nm are generated in situ. During the third stage, the processing status is monitored in real time through a multi-sensor fusion intelligent endpoint determination system. When the processing endpoint is determined to be reached, the liquid nitrogen quenching subsystem is automatically triggered to reduce the cavity temperature to below 50°C within 60 seconds, thus precisely terminating the pyrolysis reaction. The gauge pressure of the inert atmosphere positive pressure protection is 0.01 MPa to 0.05 MPa.

[0011] The beneficial effects of this invention are as follows: First, by using multi-sensor fusion and a partial least squares-backpropagation neural network intelligent judgment algorithm, the final determination of "carbon-based properties" is transformed from subjective sensory judgment to objective and quantitative digital automatic judgment. The batch-to-batch variation coefficient of key quality attributes such as Rg3 content, carbon point concentration, and degree of carbonization in the finished product can be controlled within 5%, achieving a high degree of consistency in quality between batches.

[0012] Secondly, through a three-stage gradient heating process and precise temperature-time control, the reaction efficiency of converting macromolecular saponins into rare saponins is significantly improved. The Rg3 content in the finished product is more than 3 times higher than that of the traditional charcoal roasting method, and the Rh2 content is more than 5 times higher. At the same time, the conversion state is precisely locked by liquid nitrogen quenching to prevent excessive degradation and fully tap the value of the active ingredients of precious medicinal materials.

[0013] Third, the inert atmosphere protection switches the processing mode from aerobic combustion to anaerobic pyrolysis, driving the in-situ controllable synthesis of carbon dots. The product contains functional nano carbon dots with uniform particle size (2nm to 8nm) and stable fluorescence properties, and the benzo[a]pyrene content is reduced to below 5μg / kg, greatly improving product safety.

[0014] Fourth, the fully enclosed reactor design, combined with a centralized exhaust gas treatment system using activated carbon adsorption and catalytic oxidation, eliminates fugitive VOC emissions, meets the green manufacturing standards for traditional Chinese medicine, and improves the occupational health environment for operators. Attached Figure Description

[0015] Figure 1 A schematic diagram of the overall structure of the carbonized ginseng preparation device system; Figure 2 Here is the overall flowchart of the partial least squares-backpropagation neural network fusion intelligent endpoint determination algorithm; Figure 3 This is a flowchart of the overall process for preparing carbonized ginseng. Detailed Implementation

[0016] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0017] I. Overall Structure of the Carbonized Ginseng Preparation Device System like Figure 1 As shown, the carbonized ginseng preparation device of the present invention consists of six functional subsystems, and the components of each subsystem and their mutual signal / material transmission relationships are as follows.

[0018] The inert atmosphere supply subsystem is responsible for maintaining an oxygen-free inert atmosphere in the sealed cavity throughout the entire processing. The gas storage tank uses a mass flow controller (MFC) to precisely control the inlet flow rate (typically 0.5L / min to 2L / min), and an oxygen content detector inside the chamber monitors the O2 concentration in real time. The heating program is only allowed to start after the O2 concentration is confirmed to be below 0.5%. When vacuum-assisted dehydration is required, the vacuum pump reduces the pressure inside the chamber to 20 kPa to 50 kPa.

[0019] The sealed carbonization reactor is the physical core of the device. It uses a 304 stainless steel sealed cavity (pressure bearing capacity 0.5MPa) and has a built-in zoned programmable electric heating component to achieve gradient temperature control with an accuracy of ±1℃. The inner wall of the cavity integrates a high-temperature resistant NIR fiber optic probe (sapphire window, operating temperature 350℃), a spring suspension wire micro thermogravimetric sensor (resolution 0.1mg), a tail gas sampling pipeline, and a top liquid nitrogen injection nozzle, realizing the integration of "processing-detection-rapid cooling".

[0020] The multi-source sensor detection subsystem converts the raw signals from various sensor probes into digital data streams, corresponding to a near-infrared spectrometer (acquiring the complete spectrum from 1000nm to 2500nm, acquiring data every 30 seconds, accumulating 32 frames per scan and averaging to suppress random noise, with a spectral resolution of 8cm). -1), thermogravimetric data acquisition module (sampling frequency 0.1Hz) and VOCs / CO complex gas detector (photoionization detector to measure total VOCs concentration, range 0ppm to 10000ppm); A non-dispersive infrared sensor is used to measure CO concentration, with a range of 0% to 100%. The sampling frequency is one data point every 30 seconds.

[0021] The intelligent judgment and control unit is the core of the system's calculation and decision-making. The industrial control computer runs data preprocessing software and a partial least squares-backpropagation neural network fusion algorithm to convert multi-sensor data into processing status scores in real time. It then sends precise control commands to the heating components and liquid nitrogen solenoid valves through a programmable logic controller (PLC).

[0022] After receiving the PLC trigger command, the liquid nitrogen quenching subsystem opens the solenoid valve within 0.5 seconds, and liquid nitrogen is evenly sprayed into the cavity through the top annular nozzle in an atomized manner, reducing the cavity temperature to below 50°C within 60 seconds.

[0023] The exhaust gas treatment subsystem sequentially purifies the exhaust gas from the processing process by activated carbon adsorption (removing high-boiling-point VOCs) and catalytic oxidation at 250℃ (completely oxidizing low-boiling-point VOCs) before discharging it from the emission outlet that meets the standards. The system also continuously verifies the emission compliance through an online VOCs concentration monitor.

[0024] II. Detailed Explanation of the Three-Stage Gradient Pyrolysis Process like Figure 3 As shown, the core innovation of this invention is the three-stage gradient pyrolysis process, in which the temperature-time windows of the three stages are precisely coupled with the corresponding key chemical reactions.

[0025] Raw material pretreatment and loading The ginseng raw materials are cleaned, uniformly sliced ​​(3mm to 5mm thick), and graded by particle size. The moisture content is controlled to 10% to 15% through low-temperature drying. The initial mass m0 (unit: g) is accurately weighed and recorded. The ginseng slices are loaded into a closed carbonization reactor, the reactor is closed, and the inert atmosphere supply subsystem is started. Through multiple cycles of "vacuuming-filling with inert gas", the oxygen content in the chamber is reduced to below 0.5%. The real-time reading of the oxygen content detector is used as a hard safety gate to enter the heating program.

[0026] Phase 1 - Vacuum-assisted low-temperature dehydration After the oxygen content in the cavity is qualified, the vacuum pump is started to maintain the gas pressure in the cavity at 20 kPa to 50 kPa (the boiling point of water is correspondingly reduced to about 60°C to 82°C). At the same time, the cavity temperature is raised to 100°C to 120°C at a rate of 3°C / min to 5°C / min and kept at this temperature. The purpose of this stage is to fully remove the residual free water (initial water content of about 10% to 15%) and bound water in the ginseng slices under conditions below the starting temperature of saponin thermal conversion, without damaging the saponin structure.

[0027] Thermogravimetric sensor outputs mass loss rate in real time The calculation formula is as follows: ; In the formula, for The mass loss rate at any given time, expressed as a percentage (%), is a quantitative indicator of the degree of macroscopic carbonization during the processing. When the processing procedure is started ( The initial mass of the ginseng sample, in grams (g); for The real-time mass of the ginseng sample in the reactor at any given time, expressed in grams (g). This is the time elapsed from the start of the preparation process to the current moment, in minutes (min).

[0028] mass loss rate (i.e.) Regarding time The first derivative of the rate is also an important parameter used to judge the activity level of the reaction at each stage: the rate peak indicates that the reaction is vigorous at the current stage, and the rate approaching zero indicates that the reaction is approaching completion.

[0029] When the thermogravimetric sensor displays When the OH stretching vibration absorption peak intensity near 1450 nm and 1940 nm in the near-infrared spectrum is significantly reduced and tends to stabilize, the system determines that dehydration is complete, automatically releases the vacuum, switches to inert atmosphere positive pressure protection (gauge pressure 0.01 MPa to 0.05 MPa), and enters the second stage.

[0030] Phase Two – Thermal Insulation and Deglycosylation The temperature was increased from the final temperature of the first stage to 150°C to 180°C at a heating rate of 2°C / min to 5°C / min, and held at that temperature for 60 to 90 minutes under positive pressure protection of an inert atmosphere.

[0031] This temperature range represents the kinetic optimum for the thermal breakage of CO glycosidic bonds in ginsenoside macromolecular saponins (Rb1, Re, etc.). Within the temperature range of 150℃ to 180℃, the CO glycosidic bonds connecting the sugar moiety and the saponin nucleus in ginsenoside macromolecular saponins (Rb1, Re, etc.) undergo thermal breakage, gradually releasing the sugar moiety and generating rare saponins Rg3 and Rh2. Therefore, this temperature range can preferentially drive the sequential shedding of sugar moiety. The specific transformation pathways are: Rb1 (tetrasaccharide) → Rd (trisaccharide) → Rg3 (disaccharide) → Rh2 (monosaccharide), and Re → Rg1 → Rg2 → Rh1.

[0032] In this stage, the changes in the intensity of the CO bond characteristic absorption peak from 1650 nm to 1800 nm in the near-infrared spectrum and the phased increase in CO release from the exhaust gas detector are used as key online monitoring indicators of the reaction progress. The partial least squares spectral regression model continues to run in this stage to predict the progress of saponin conversion in real time, providing data basis for entering the third stage.

[0033] The third stage – critical carbonization short-time thermal shock The temperature was increased from the final temperature of the second stage to the critical carbonization zone of 220°C to 250°C at a heating rate of 3°C / min to 8°C / min, while maintaining positive pressure protection under an inert atmosphere, and subjected to short-term thermal shock for 10 to 20 minutes.

[0034] At 220℃, the thermal decomposition initiation temperature of ginseng polysaccharide (with α-1,4-glucan as the main chain) is exceeded (approximately 200℃). The polysaccharide chain segments undergo dehydration and breakage, while Maillard reaction and further carbonization and condensation occur between the polysaccharide and ginseng protein, generating nano carbon dots in situ. Due to the inert atmosphere isolating oxygen, the carbonization path of the organic precursor is pure pyrolysis and condensation rather than aerobic combustion, which can directionally generate nano carbon dots with uniform particle size (2nm to 8nm), while completely avoiding the formation of polycyclic aromatic hydrocarbons such as benzo[a]pyrene.

[0035] The key control point at this stage is that the heat preservation time should not be too long: if rare saponins (accumulated in the second stage) continue to be exposed to 220℃ to 250℃ for a long time, their saponin aglycone structure will also undergo further cleavage and degradation. Therefore, this stage must be carried out within the time window when the carbon point generation reaches its peak and the rare saponins have not yet significantly degraded, and the liquid nitrogen quenching should be precisely triggered by the intelligent endpoint determination system.

[0036] III. Multi-sensor fusion intelligent endpoint determination system like Figure 2 As shown, the endpoint determination algorithm of this invention consists of a two-level calculation model: The first layer is a partial least squares spectral regression layer, which reduces and compresses high-dimensional NIR spectra and transforms them into quantitative predictions of key quality attributes. The second layer is a backpropagation neural network fusion layer, which fuses partial least squares predictions with thermogravimetric and gas sensor data to output a comprehensive processing status score and make an endpoint determination decision. The complete algorithm is divided into an "offline training stage" and an "online inference stage".

[0037] (a) Offline training phase Constructing a multigradient processing experiment calibration dataset Under controlled experimental conditions, ginseng slices of the same batch and quality (thickness 3mm to 5mm, moisture content 15%, batch feed amount 100g) were used as experimental subjects. Multiple sets of heating experiments were designed according to a three-stage gradient pyrolysis process. Samples were taken at multiple time points covering the entire process of the three stages (with adjacent sampling points spaced 5 to 10 minutes apart). The sample size at each time point was 3g to 5g. For each sampling point, the original NIR diffuse reflectance spectral vector (dimension p, p is approximately 1024 wavenumber points in this device) and the mass loss rate were recorded simultaneously. And exhaust gas detection data vector (including three indicators: CO concentration, total VOCs concentration, and VOC peak concentration).

[0038] Subsequently, offline analysis was performed on samples from each sampling point: the content of rare saponin Rg3 was accurately determined by high performance liquid chromatography (HPLC, Agilent ZORBAX SB-C18 column, acetonitrile-water gradient elution, 203 nm UV detection wavelength). (mg / g, dry basis) and Rh2 content (mg / g, dry basis); The carbon dot particle size distribution was observed by transmission electron microscopy (TEM), and the fluorescence intensity of the carbon dot aqueous extract was measured by fluorescence spectroscopy (excitation wavelength 360 nm) and converted into carbon dot concentration using a pre-established working curve. (mg / g, dry basis); The carbonizability index was calculated using a specific NIR absorbance ratio (absorbance ratio at 1020 nm / 2000 nm, pre-calibrated by scanning electron microscopy energy dispersive spectroscopy). (Dimensionless, value range 0 to 1), and finally construct a calibration dataset with no less than 80 sample points.

[0039] NIR spectral preprocessing Before inputting the raw NIR diffuse reflectance spectral data into the model, the following preprocessing operations must be performed sequentially: Baseline correction: Select 3 to 5 reference bands with no obvious characteristic absorption at each end of the spectrum (in this invention, 1000nm to 1020nm and 2480nm to 2500nm are selected as reference intervals), perform linear interpolation on the two reference points to form a baseline, and subtract the baseline value point by point from the entire spectrum to eliminate the additive system noise introduced by instrument drift and changes in ambient temperature.

[0040] Standard Normal Variable Transform (SNV): For each baseline-corrected spectral vector, the mean of the spectral vector is subtracted and then divided by the standard deviation of the spectral vector, so that the mean of each spectrum after transformation is 0 and the standard deviation is 1. This eliminates the multiplicative scattering effect caused by changes in the bulk density and surface state of ginseng slices under high temperature conditions, and makes the spectra collected at different time points comparable.

[0041] Savitzky-Golay first derivative smoothing: For the spectral vector processed by SNV, the first derivative spectrum is calculated using a Savitzky-Golay polynomial smoothing filter (window width of 13 sampling points, polynomial order of 2) to enhance the resolution of overlapping peaks, further eliminate residual baseline drift, and strengthen the spectral peak position drift signal caused by changes in saponin structure.

[0042] After the above three preprocessing steps, the N preprocessed spectra are combined into a preprocessed spectral matrix. (N rows × p columns) serves as the input for subsequent partial least squares modeling.

[0043] Establish a partial least squares spectral regression model Partial least squares model with preprocessed spectral matrix The predictor variable matrix consists of a response variable matrix composed of three key quality attributes. (Each column corresponds to) , , Using a dimension of N rows × 3 columns as the modeling target, establish a quantitative regression relationship from NIR spectra to saponin content and carbon point concentration: ; In the formula, The matrix represents the predicted values ​​of three key quality attributes for N sample points (dimension N rows × 3 columns), with each column corresponding to the predicted values ​​of Rg3 content, Rh2 content, and carbon point concentration, respectively. The preprocessed NIR spectral matrix (dimension N rows × p columns); The partial least squares regression coefficient matrix (p rows × 3 columns) is calculated from the training set data. Each column corresponds to a weight vector of the regression relationship between a quality attribute and each wavenumber point of the spectrum.

[0044] Leave-one-out cross-validation (LOOCV) is employed, with the criterion of minimizing the root mean square error (RMSECV) of predictions on the validation set. RMSECV is calculated for each latent variable from A=1 to A=30, and the value of A corresponding to the stationary RMSECV value is selected (in practice, A is usually taken as 6 to 12). The nonlinear iterative partial least squares (NIPALS) algorithm is then used to calculate... The root mean square error of prediction (RMSEP) and coefficient of determination are calculated using an independent validation set (20% to 25% of the samples randomly drawn from the total dataset, which are not used in the modeling process). ,Require A partial least squares model can be considered to meet the requirements for online application only if its accuracy is not lower than 0.92 and its relative prediction error (RPE) does not exceed 8%.

[0045] Construct and train a backpropagation neural network fusion model The task of the backpropagation neural network is to fuse the predicted mass attributes output by the partial least squares model with thermogravimetric and gas sensor data to output a comprehensive processing status score. (Values ​​range from 0 to 100 on a percentage scale, reflecting the degree of coupling between saponins and carbon dots) and predicted carbonization degree. (Value range: 0 to 1).

[0046] The network structure is a backpropagation neural network with a three-layer fully connected structure. The configuration of each layer is as follows: The input layer contains 7 neurons, corresponding to 3 partial least squares predictions (…). , , ), 1 thermogravimetric mass loss rate ( The input data is normalized to the [0,1] interval by Min-Max normalization and three exhaust gas detection indicators (CO concentration, total VOCs concentration, and VOC peak concentration). The first hidden layer contains 14 neurons, and the activation function is a rectified linear unit (ReLU). The second hidden layer contains 7 neurons, and the activation function is also the rectified linear unit (ReLU), which serves as a feature convergence layer from high dimension to low dimension; The output layer contains two neurons, which output... and The activation function uses the Sigmoid function to constrain the output range.

[0047] The true label of the comprehensive processing status score is comprehensively evaluated by experts in the field of traditional Chinese medicine processing based on the HPLC and TEM results of each sampling point (maximum score 1.0 point, taking into account the coupling status of rare saponin content and carbon point concentration). The carbonization degree of the true label is determined offline. value.

[0048] Adam optimization algorithm (momentum parameter) is used , The initial learning rate is set to Combined with a cosine annealing learning rate decay strategy; Set the batch size to 16; The loss function uses mean squared error; Set an early stopping strategy: terminate training when the validation set loss no longer decreases after 20 consecutive training epochs, and save the network weight parameters at this time. The model performance was evaluated using 5-fold cross-validation, and the coefficient of determination for each fold validation set was required. Not less than 0.90.

[0049] (ii) Online Inference Phase During the third stage of critical carbonization thermal shock in the production batch, the system performs the following operations in a 30-second cycle: acquiring the current moment from the NIR probe. The original diffuse reflectance spectral vector (dimension p); Read the current mass from the thermogravimetric data acquisition module. And calculate in real time according to the aforementioned formula. ; Read the current exhaust gas data vector (3D) from the gas detector.

[0050] The acquired raw NIR spectra were subjected to baseline correction, standard normal variable transformation, and Savitzky-Golay first derivative processing, which were performed sequentially with parameters identical to those used in the offline training phase, to obtain preprocessed spectral vectors. (1×p row vectors).

[0051] Will Compared with the saved partial least squares regression coefficient matrix Perform vector-matrix multiplication to obtain the online predicted values ​​of each key quality attribute at the current time: ; In the formula, for Online predicted value of rare saponin Rg3 content at any given time, in mg / g; for Online predicted values ​​of rare saponin Rh2 content at any given time, in mg / g; for Online predicted value of nano carbon dot concentration at any given time, in mg / g; for The preprocessed NIR spectrum row vector (dimension 1×p) at time step. The partial least squares regression coefficient matrix (dimension p×3) saved for the offline training phase.

[0052] Partial least squares prediction output value With quality loss rate The exhaust gas data vector is concatenated with the exhaust gas data vector in the same dimension to form a 7-dimensional fused input vector. After Min-Max normalization, it is input into a trained backpropagation neural network for forward propagation calculation, outputting the comprehensive processing status score at the current moment. (Values ​​range from 0 to 100) and predicted carbonization (Value range 0 to 1), the calculation time on the industrial control computer is no more than 50ms, which fully meets the real-time requirement of a 30-second acquisition cycle.

[0053] Endpoint determination and liquid nitrogen quenching triggering based on the three-condition coupling criterion The system determines the processing endpoint based on the criterion that the following three conditions must be met simultaneously. These three conditions cover three dimensions: "maximizing saponin activity," "maximizing carbon point formation," and "preventing overcharging risks." They complement each other and form multiple safeguards. Condition 1 (Comprehensive Score Peak Criterion): Comprehensive Processing Status Score The data showed a decrease over three consecutive monitoring cycles (i.e., 90 seconds) in the historical monitoring sequence, confirming the formation of a clear peak. And currently Not higher than 95%; Condition 2 (Overcarbonization Control Criteria): Current predicted carbonization value It does not exceed the preset upper limit threshold of carbonization of 0.65; Condition 3 (Quality Loss Rate Range Criterion): Current Quality Loss Rate It falls within the suitable carbonization quality loss range of 20% to 35%.

[0054] When the above three conditions are met simultaneously within the same detection cycle, the system sends a liquid nitrogen rapid cooling trigger electrical signal to the PLC controller. After receiving the trigger signal, the PLC opens the solenoid valve of the liquid nitrogen storage tank within 500ms. Liquid nitrogen is evenly sprayed into the sealed reactor cavity in an atomized manner through the top annular nozzle. The cavity temperature drops rapidly to below 50°C within 60 seconds, accurately terminating all pyrolysis chemical reactions and realizing the digital locking of the "carbonized and preserved" processing state.

[0055] System reset and full electronic batch record archiving After the quenching process is completed and the cavity temperature drops to a safe range, the system automatically records the complete process data curves for this batch (including NIR spectral time series database, thermogravimetric mass loss rate curve, exhaust gas detection curve, time series of predicted values ​​of each key quality attribute, endpoint trigger time, and three-condition satisfaction status), generating an encrypted electronic batch record that conforms to the 21 CFR Part 11 data integrity specification. This provides a complete data foundation for quality traceability and continuous model iteration and optimization. When non-conforming products appear in subsequent batches, the cause of the deviation can be traced from the electronic batch record, and the actual measured values ​​can be added to the training set as new samples, triggering continuous learning and updating of the model, forming a closed-loop self-optimization system. IV. Specific Implementation Examples Example 1 Take 300g of dried Changbai Mountain ginseng slices (slice thickness 3mm, moisture content 12%) and accurately weigh the initial mass. g, evenly fill the sealed carbonization reactor, and start. The inert atmosphere supply subsystem introduces an inert atmosphere at a rate of 1.5 L / min via a mass flow controller. Simultaneously perform vacuuming and filling. The cycle is repeated three times until the O2 concentration in the cavity drops to 0.4%. The system automatically confirms that the safety gate has passed and allows the heating program to start.

[0057] The first stage of vacuum-assisted cryogenic dehydration was performed: the vacuum pump was started to maintain the internal pressure at 30 kPa, and the chamber temperature was increased to 110°C at a rate of 4°C / min and held. After about 45 minutes, the thermogravimetric sensor showed that the mass loss rate stabilized at 12.8%, and the intensity of the OH absorption peaks near 1450 nm and 1940 nm in the NIR spectrum decreased to 35% of the reference value and tended to stabilize. The system determined that dehydration was complete, automatically released the vacuum, and was refilled. When the gauge pressure reaches 0.02 MPa, switch to the second stage.

[0058] The second stage of thermal deglycosylation was performed: the temperature was increased to 165°C at a rate of 3°C / min. After 75 minutes of heat preservation under positive pressure protection, the online prediction of the partial least squares model showed that the predicted value of Rg3 content reached 0.38 mg / g and the predicted value of Rh2 content reached 0.09 mg / g at about 60 minutes. The CO release in the exhaust gas showed a phased increasing trend, meeting the conditions for entering the third stage.

[0059] Perform the third stage of critical carbonization short-time thermal shock: heat to 235℃ at a rate of 5℃ / min, and... Under positive pressure protection, thermal shock is performed. Simultaneously, full-speed multi-sensor acquisition and partial least squares-backpropagation neural network fusion algorithm are started for real-time inference. The system monitors in cycles of 30 seconds. After about 12 minutes of thermal shock, the system detects that three conditions are met at the same time: the comprehensive processing status score has decreased for three consecutive cycles (90 seconds), and the current value is 91% of the peak value (below the 95% threshold). Predicted carbonization value =0.58 (below the threshold of 0.65); Quality loss rate =27.3% (in the range of 20% to 35%), the system sends a liquid nitrogen quenching trigger electrical signal to the PLC, and opens the liquid nitrogen solenoid valve within 0.4 seconds. The chamber temperature drops to 42°C within 55 seconds, and the pyrolysis reaction is precisely terminated.

[0060] After the product is removed from the oven, quality inspection is carried out: HPLC determination shows that the Rg3 content is 0.52 mg / g (dry basis) and the Rh2 content is 0.11 mg / g (dry basis). The concentration of carbon dots was determined to be 0.31 mg / g (dry basis) by fluorescence spectroscopy. TEM observations showed that the carbon dots had a particle size distribution between 3 nm and 7 nm. The benzo[a]pyrene content was determined to be 3.2 μg / kg by GC-MS. The quality loss rate was 27.3%. The product has a dark brown surface and a yellowish-brown interior, with a complete fiber structure. All of the above quality indicators meet the product quality standards of this invention, verifying the technical feasibility of the method of this invention.

[0061] Example 2 (Alternative to Microwave-Assisted Heating) The closed reactor's programmed electrothermal heating method is replaced with a closed microwave-assisted heating method. Utilizing the selective coupling heating characteristics of microwaves (frequency 2.45 GHz) on polar molecules, energy is directly applied to the polar groups inside the ginseng, significantly improving heating efficiency and volume uniformity. Microwave output power (unit: W) is used instead of temperature as the direct control parameter. Closed-loop feedback control of the cavity temperature is achieved through an embedded fiber optic thermometer (immune to microwave interference). The overall framework of inert atmosphere protection and the three-stage gradient heating program remains unchanged. The intelligent endpoint determination system is also exactly the same as in Example 1. The PLC control actuator is replaced by an electrothermal power controller with an output power regulator of a microwave solid-state power source, while the rest remains unchanged. This implementation method can shorten the time to reach the target temperature at each stage by 40% to 60%, and the total processing cycle is shortened by about 30% to 50% compared to Example 1. This is beneficial for completing the endpoint response within the precise time window of the rare saponin accumulation peak, further reducing batch-to-batch differences.

[0062] V. Characteristics and Quality Standards of the Carbonized Ginseng Product Prepared by This Invention The carbonized ginseng product prepared by this invention has the following characteristic quality indicators that distinguish it from traditional charred products (see table below): Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make possible changes and modifications without departing from the spirit and scope of the present invention. Therefore, any modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for preparing carbonized ginseng, characterized in that, Using a closed-loop, programmable carbonization reactor as the core equipment, and under the protection of an inert atmosphere formed by a gas selected from nitrogen and carbon dioxide, a three-stage gradient pyrolysis process is sequentially performed on ginseng slices: The first stage is vacuum-assisted low-temperature dehydration: the gas pressure in the sealed reactor is reduced to 20 kPa to 50 kPa, the temperature of the chamber is increased to 100°C to 120°C at a heating rate of 3°C / min to 5°C / min and held for 30 to 60 minutes. The completion of this stage is determined by the following criteria: the mass loss rate is stable at 10% to 15% as indicated by the thermogravimetric sensor, and the intensity of the OH stretching vibration absorption peaks near 1450 nm and 1940 nm in the near-infrared spectrum decreases to less than 40% of the initial acquisition intensity, and the peak intensity change does not exceed 5% within three consecutive acquisition cycles. After completion, the system automatically switches to inert atmosphere positive pressure protection and enters the second stage. The second stage is thermal deglycosylation: the temperature is increased to 150℃ to 180℃ at a heating rate of 2℃ / min to 5℃ / min, and kept at this temperature for 60 to 90 minutes under positive pressure protection in an inert atmosphere to drive the sequential thermal breakage of CO glycosidic bonds in ginsenoside macromolecular saponins. The third stage is critical carbonization short-time thermal shock: the temperature is increased to 220°C to 250°C at a heating rate of 3°C / min to 8°C / min, and thermally shocked for 10 to 20 minutes under positive pressure protection of an inert atmosphere, so that ginseng polysaccharides and proteins undergo Maillard reaction and carbonization condensation under anaerobic pyrolysis conditions. During the third stage, the processing status is monitored in real time through a multi-sensor fusion intelligent endpoint determination system. When the processing endpoint is determined to be reached, the liquid nitrogen quenching subsystem is automatically triggered to reduce the cavity temperature to below 50°C within 60 seconds, thereby terminating the pyrolysis reaction. The gauge pressure of the inert atmosphere positive pressure protection is 0.01 MPa to 0.05 MPa.

2. The method for preparing carbonized ginseng according to claim 1, characterized in that, The ginseng slices are ginseng slices with a thickness of 3mm to 5mm and a moisture content of 10% to 15%; Before the first stage begins, the oxygen content in the sealed reactor chamber is reduced to below 0.5% by evacuation and inert gas circulation replacement. The real-time reading of the oxygen content detector in the chamber is used as the safety gate for entering the heating process.

3. The method for preparing carbonized ginseng according to claim 1, characterized in that, In the second stage, the intensity changes of the CO bond characteristic absorption peak from 1650 nm to 1800 nm in the near-infrared spectrum and the phased increase of CO release in the exhaust gas detector are used as online monitoring indicators of the degree of saponin deglycosylation reaction. Simultaneously, the saponin conversion progress is predicted in real time by the partial least squares spectral regression model.

4. The method for preparing carbonized ginseng according to claim 1, characterized in that, The multi-sensor fusion intelligent endpoint determination system includes a high-temperature resistant near-infrared spectroscopy probe, a micro-thermogravimetric sensor, and a VOCs / CO composite gas detector integrated into the inner wall of the sealed reactor cavity, as well as an intelligent determination and control unit running a partial least squares-backpropagation neural network fusion algorithm. During the third stage of thermal shock, the endpoint determination system cyclically collects multi-sensor data in 30-second cycles, calculates the comprehensive processing status score and carbonization degree prediction value in real time, and triggers liquid nitrogen quenching when the following three conditions are simultaneously met within the same detection cycle: Condition 1: The comprehensive processing status score has reached its peak after three consecutive monitoring cycles in the historical monitoring sequence and then started to decline, and the current comprehensive processing status score is not higher than 95% of that peak value; Condition 2: The current predicted carbonization value does not exceed 0.65; Condition 3: The current quality loss rate is between 20% and 35%.

5. The method for preparing carbonized ginseng according to claim 4, characterized in that, The partial least squares-backpropagation neural network fusion algorithm is constructed through the following offline training method: a multi-gradient processing experiment is carried out under controlled conditions according to a three-stage gradient pyrolysis process. Samples are taken at multiple time points covering the entire process of the three stages. Near-infrared raw diffuse reflectance spectra, mass loss rate and exhaust gas detection data are recorded synchronously at each sampling point. The contents of rare saponins Rg3 and Rh2 are determined offline by high performance liquid chromatography, the concentration of nano carbon dots is determined by fluorescence spectroscopy, and the carbonization index is calculated by near-infrared absorbance ratio. A calibration dataset of no less than 80 sample points is constructed. The raw near-infrared spectra were preprocessed sequentially with baseline correction, standard normal variable transformation, and Savitzky-Golay first-order derivative smoothing. Using the preprocessed spectral matrix as the predictor variable and the contents of rare saponins Rg3, Rh2, and carbon point concentration as the response variables, the optimal number of latent variables was determined using leave-one-out cross-validation. The regression coefficient matrix was calculated using a nonlinear iterative partial least squares algorithm to establish a partial least squares spectral regression model. The coefficient of determination R on the independent validation set is required. 2 Not less than 0.92; A 7-dimensional input vector is constructed by concatenating the predicted mass attributes from the partial least squares model output, the thermogravimetric mass loss rate, and exhaust gas detection data. The processing status score, as determined by expert evaluation, is used as the label. A three-layer fully connected backpropagation neural network with 7 neurons in the input layer, 14 neurons in the first hidden layer, 7 neurons in the second hidden layer, and 2 neurons in the output layer is trained using the Adam optimization algorithm and an early-stop strategy. Five-fold cross-validation is required to validate the coefficient of determination R0 of each fold. 2 Not less than 0.

90.

6. A carbonized ginseng preparation apparatus, characterized in that, include: The main body of the sealed carbonization reactor adopts a pressure-bearing sealed cavity and has a built-in zoned programmable electric heating component to achieve gradient temperature control with an accuracy of ±1℃. The inert atmosphere supply subsystem includes an inert gas storage tank, a mass flow controller, and an in-cavity oxygen content detector; The multi-source sensor detection subsystem includes a high-temperature resistant near-infrared spectroscopy probe integrated into the inner wall of the cavity, a spring-loaded micro-thermogravimetric sensor, and a VOCs / CO complex gas detector. Intelligent decision and control unit, including industrial control computer and programmable logic controller running partial least squares-backpropagation neural network fusion algorithm; The liquid nitrogen quenching subsystem includes a liquid nitrogen storage tank, a solenoid valve, and a top annular nozzle, wherein the solenoid valve has a response time of no more than 0.5 seconds; The exhaust gas treatment subsystem includes an activated carbon adsorption device and a catalytic oxidation purification device connected in series. The above subsystems are integrated into one unit through signal lines and material pipelines.

7. The carbonized ginseng preparation apparatus according to claim 6, characterized in that, The high-temperature resistant near-infrared spectral probe adopts a sapphire optical window diffuse reflection fiber optic probe with a rated operating temperature of not less than 350℃ and a spectral acquisition range of 1000nm to 2500nm. A complete spectrum is acquired every 30 seconds and each scan accumulates 32 frames and takes the average value. The spring-loaded micro-thermogravimetric sensor has a resolution of 0.1 mg and a sampling frequency of 0.1 Hz. The operating temperature of the catalytic oxidation purification device is 250℃.

8. A carbonized ginseng, characterized in that, Prepared by the method according to any one of claims 1 to 5, the carbonized ginseng contains: a rare saponin Rg3 content of not less than 0.5 mg / g (dry basis), a rare saponin Rh2 content of not less than 0.1 mg / g (dry basis), and a nano carbon dot concentration of not less than 0.3 mg / g (dry basis), wherein the nano carbon dots have a particle size of 2 nm to 8 nm; The content of benzo[a]pyrene shall not exceed 5 μg / kg; The quality loss rate is 20% to 35%; The surface is dark brown, the interior is yellowish-brown, and the fibrous structure is intact.

9. The carbonized ginseng according to claim 8, characterized in that, The inter-batch coefficient of variation for the key quality attributes of rare saponin Rg3 content, nano carbon dot concentration, and degree of carbonization in the carbonized ginseng does not exceed 5%.

Images