Essence distillation sampling system and method

Abstract

This application discloses a flavor distillation sampling system and method, relating to the field of flavor preparation technology. The system includes: a multi-stage CCP control module integrated into key processes of the flavor production process, including but not limited to raw material receiving and acceptance, weighing and preparation, filtration and slag removal, and heating / stirring control; an online purification unit connected in series between the flavor production and packaging stages, comprising: a multi-stage filtration component adapted to different filter mesh sizes according to different flavor types; a foreign matter detection component using metal detection or X-ray detection to detect the presence of foreign matter residue before packaging; and a contamination prevention control unit including an inert gas protection chamber suitable for solid or powdered flavors. This application has the effect of improving the product quality of flavor preparation processes.

Description

Technical Field

[0001] This application relates to the field of fragrance preparation technology, and in particular to a fragrance distillation sampling system and method. Background Technology

[0002] Currently, flavor distillation sampling is a core step in quality control during the production process, requiring precise extraction and detection of volatile components (such as terpenes and esters). In traditional processes, after flavor raw materials are heated and distilled, samples must be taken and analyzed to determine the purity of the components, aroma intensity, and foreign matter residues, in order to ensure that the product meets food / cosmetic safety standards.

[0003] Existing flavor distillation sampling devices have a single-stage filter (such as 100 mesh) at the end of the distillation pipeline, which has insufficient interception efficiency. Furthermore, manual operation can easily introduce environmental pollutants (such as glove fragments and dust), leading to an increased rate of microbial contamination. As a result, there are defects in the flavor preparation process, such as poor product quality, which urgently need to be improved. Summary of the Invention

[0004] In order to improve the product quality of fragrance preparation process, this application provides a fragrance distillation sampling system and method.

[0005] Firstly, the objective of this invention is achieved through the following technical solution:

[0006] The flavor distillation sampling system includes:

[0007] A multi-stage CCP control module is integrated into key processes in the flavor production process, including but not limited to raw material receiving and acceptance, weighing and preparation, filtration and slag removal, and heating / stirring control.

[0008] An online purification unit, connected between the fragrance production and packaging stages, includes:

[0009] Multi-stage filtration components are used to adapt the corresponding filter mesh size according to different fragrance types;

[0010] Foreign object detection components employ metal detectors or X-ray inspection to detect the presence of foreign object residue before packaging;

[0011] The anti-pollution control unit includes an inert gas protection chamber suitable for solid or powdered fragrances.

[0012] By adopting the above technical solutions, the multi-stage CCP control module meets the requirements of precise process control and standardized production in flavor distillation sampling. By integrating raw material receiving and acceptance (CCP1), weighing and preparation (CCP2), filtration and slag removal (CCP3), and heating / stirring control (CCP4), it achieves real-time monitoring and automatic adjustment of key parameters (temperature, time, additive ratio) throughout the flavor production process, ensuring process stability. Standardized production helps improve the retention rate of flavor components. The online purification unit can be matched with the filter mesh size according to the flavor type, such as (100 mesh filter for liquid flavors, 10-20 mesh sieve for powdered flavors). Combined with multi-stage filtration components (metal filter + ceramic membrane + activated carbon), it ensures purity. Metal detection or X-ray detection can intercept foreign objects such as metal and glass before packaging. The anti-contamination control unit can fill nitrogen for solid and powdered flavors to inhibit oxygen reaction and reduce biological contamination. The overall flavor production process of this application has a high level of quality control, which is conducive to improving the product quality of flavor preparation process.

[0013] Secondly, the objective of this invention is achieved through the following technical solution:

[0014] A control method for a flavor distillation sampling system, the method comprising:

[0015] Obtain the distillation parameters of the flavor raw materials, and generate a dynamic sampling strategy based on the distillation parameters in the distillation sampling system;

[0016] Using each sampling time point as a basic unit, the distillation temperature and pressure are adjusted according to a preset first control factor to obtain real-time monitoring data of the distillate, and the corresponding component concentration distribution data are calculated based on the real-time monitoring data.

[0017] Using each sampling time point as the basic unit, the distillate flow rate and temperature gradient are analyzed according to the preset second control factor to generate distillation process stability assessment data.

[0018] Based on the component concentration distribution data and distillation process stability assessment data corresponding to each sampling time point, dynamic sampling operation is performed and the optimal sampling result is output.

[0019] By adopting the above technical solution, a multi-stage distillation model is established based on the fusion of distillation parameter data, generating an initial sampling strategy that includes temperature gradient and pressure compensation; the distillation temperature and pressure are adjusted in real time based on the first control factor (such as density change, surface tension change rate, and vacuum degree), and distillate data are collected simultaneously; a fluid dynamics model is constructed based on the second control factor (flow rate and temperature gradient), and the system stability evaluation index is calculated; then, by integrating the component concentration distribution and stability evaluation data, an adaptive sampling operation is triggered and the optimal result is output. This application triggers an adaptive sampling operation and outputs the optimal result by integrating the component concentration distribution and stability evaluation data.

[0020] In a preferred embodiment of this application: the step of obtaining distillation parameters for flavoring raw materials and generating a dynamic sampling strategy in the distillation sampling system based on the distillation parameters includes:

[0021] The distillation parameters are subjected to data fusion processing to obtain preprocessed data including material viscosity, boiling point curve and thermal sensitivity index;

[0022] A multi-stage distillation model is established based on the preprocessed data, and an initial sampling strategy is dynamically generated, including temperature gradient curves, pressure compensation values, and theoretical distillation range.

[0023] A multi-parameter sensor group installed at the top of the distillation column is used to collect distillate flow rate, temperature field distribution and component spectrum data in real time, and a dynamic sampling decision tree is generated in combination with the initial sampling strategy.

[0024] The dynamic sampling decision tree is optimized and adjusted using a fuzzy PID control algorithm to form a dynamic sampling strategy that includes sampling frequency, sampling threshold, and abnormal termination conditions.

[0025] By adopting the above technical solution, physical parameters (material viscosity, boiling point curve, and thermodynamic index) of distillation parameters are fused. By establishing a multi-stage distillation model, an initial sampling strategy including temperature gradient curve, pressure compensation value, and theoretical distillation range is generated. This solves the problem of the one-sidedness of single parameter control. Through the synergistic optimization of dynamic sampling decision tree and fuzzy PID algorithm, the response speed of the sampling strategy can be improved.

[0026] In a preferred embodiment of this application: the first control factor includes the distillate density change value, the distillate surface tension change rate, and the distillation vessel vacuum degree; the step of adjusting the distillation temperature and pressure according to the preset first control factor includes:

[0027] Calculate the change in distillate density, and identify the pre-distillation, main distillation, and tail distillation stages based on the change in distillate density;

[0028] When the rate of change of surface tension of the distillate exceeds the preset rate of change threshold, a three-level temperature compensation mechanism is triggered.

[0029] The vacuum level of the distillation vessel is calculated in real time, and the vacuum level is adjusted in real time through a differential pressure sensor to maintain a constant boiling point environment.

[0030] By adopting the above technical solution, the adjustment stage of the first control factor includes three control mechanisms: stage identification, dynamic compensation, and vacuum maintenance. The three-level temperature compensation mechanism effectively suppresses the risk of thermal decomposition of heat-sensitive components, while vacuum adjustment can significantly improve the component extraction efficiency.

[0031] In a preferred embodiment of this application, the component concentration distribution data is obtained through the following method:

[0032] Online component analysis of the distillate was performed using a miniature Fourier transform infrared spectrometer;

[0033] The migration patterns of characteristic aroma components were verified using gas chromatography-mass spectrometry.

[0034] Establish a three-dimensional mapping model of distillation time, component concentration, and distillation range.

[0035] By adopting the above technical solutions, the combination of infrared spectroscopy and chromatography enables cross-validation of component data, improving accuracy; the three-dimensional mapping model breaks through the limitations of traditional univariate analysis and supports the visualization and tracing of component migration patterns.

[0036] In a preferred embodiment of this application, the step of analyzing the distillate flow rate and temperature gradient based on a preset second control factor to generate distillation process stability assessment data specifically includes:

[0037] Construct a fluid dynamics model to calculate the Reynolds number and Mach number of the distillate;

[0038] The axial temperature distribution gradient of the distillation column was obtained using an infrared thermal imager.

[0039] Based on the distillate Reynolds number, the Mach number, and the axial temperature distribution gradient of the distillation column, the system stability assessment data are calculated by combining the flow rate fluctuation coefficient and the standard deviation of the temperature gradient.

[0040] By adopting the above technical solution, a refined stability assessment is carried out based on multi-parameter coupling. The stability assessment mechanism can reflect the uniformity of fluid state and temperature distribution during the distillation process in real time and provide early warning of potential instability factors.

[0041] In a preferred embodiment, this application further includes:

[0042] Dynamic pressure fluctuation compensation value is used to correct sampling errors caused by changes in ambient temperature.

[0043] A fraction transition zone identification algorithm is used to identify abrupt changes in distillate composition.

[0044] The emergency braking threshold system includes pressure surge protection, temperature over-limit fuse, and flow abnormality cut-off mechanism.

[0045] By adopting the above technical solutions and introducing a dynamic pressure fluctuation compensation mechanism, the system's anti-interference capability in complex environments is effectively improved, ensuring the accuracy of sampling data. The fraction transition range identification algorithm can accurately capture the inflection points of aroma component changes, improving the representativeness of the samples.

[0046] In a preferred embodiment of this application, the step of performing dynamic sampling and outputting the optimal sampling result includes:

[0047] Based on the dynamic sampling decision tree, a sampling operation verification model is established in conjunction with the three-dimensional mapping model, and the confidence level of distillate component migration is calculated.

[0048] When the emergency braking threshold system is detected to be triggered, the three-level temperature compensation mechanism and the flow rate fluctuation coefficient are coordinated to generate a dynamic sampling fault-tolerant path.

[0049] By using reinforcement learning algorithms to extract features from historical sampling data, an adaptive sampling knowledge base is constructed, which includes the weight of the distillation transition range, the vacuum degree decay compensation value, and the thermal sensitivity index decay factor. The optimal sampling scheme with confidence score is then output.

[0050] By adopting the above technical solutions, the sampling process has been transformed from a passive response to an active optimization, significantly improving the scientific and intelligent level of sampling decisions. By constructing a verification model and fault-tolerant path, the system has stronger robustness and self-repair capabilities. The learning-driven knowledge base continuously accumulates experience, enabling the system to automatically adjust the sampling strategy according to the characteristics of different raw materials.

[0051] In a preferred embodiment, this application further includes:

[0052] During dynamic sampling operations, a federated learning framework is used to synchronously optimize the weights of the dynamic sampling decision tree nodes, wherein:

[0053] Input the pressure fluctuation data of the distillation vessel into the generative adversarial network to generate a virtual pressure disturbance scenario;

[0054] By using a differential privacy mechanism, the locally optimized sampling frequency parameters are uploaded to a cloud-based collaborative database, enabling cross-device parameter federation.

[0055] By adopting the above technical solutions, federated learning enables collaborative optimization of data among multiple devices, improving the adaptability and generalization ability of the system model; generative adversarial networks expand the diversity of training data and enhance the system's ability to cope with complex working conditions.

[0056] In summary, this application includes at least one of the following beneficial technical effects:

[0057] 1. Multi-stage quality control of the entire flavor production process has been achieved. In particular, the introduction of CCP control mechanism in key links such as raw material acceptance, weighing and preparation, filtration and slag removal and heating and stirring has significantly improved process controllability and product quality consistency.

[0058] 2. It achieves refined and intelligent management of the fragrance distillation process. Through the generation of dynamic sampling strategies, sampling becomes more targeted and representative, improving the accuracy of aroma component extraction. Combined with dual control factors (the first control factor is used to adjust process parameters, and the second control factor is used to evaluate process stability), high-precision separation and capture of complex volatile fragrance components can be achieved. Attached Figure Description

[0059] Figure 1 This is a framework diagram of a flavor distillation sampling system according to one embodiment of this application;

[0060] Figure 2 This is a flowchart of a control method applied to a flavor distillation sampling system according to an embodiment of this application. Detailed Implementation

[0061] The present application will be further described in detail below with reference to the accompanying drawings.

[0062] In one embodiment, such as Figure 1 As shown, this application discloses a flavor distillation sampling system, which includes a multi-stage CCP control module, an online purification unit, and a contamination prevention control unit. The multi-stage CCP control module is integrated into key processes in the flavor production process, including but not limited to raw material receiving and acceptance, weighing and preparation, filtration and slag removal, and heating / stirring control. Raw material receiving and acceptance involves setting up a QR code scanning terminal to automatically compare against a supplier qualification database and quickly screening the compliance of raw material components using a near-infrared spectrometer. Weighing and preparation uses a high-precision electronic scale, combined with an MES system to record the proportions of each batch of raw materials. Filtration and slag removal involves installing a pressure sensor to monitor the filter pressure difference (set threshold 0.3 MPa), and initiating a backwashing procedure when the pressure difference exceeds the limit. Heating control, during oil / water phase preparation, uses a PLC to achieve temperature gradient control, and stirring control controls the rotation speed of the stirring apparatus.

[0063] The multi-stage filtration components include edible flavorings made of 316L stainless steel with a three-stage filtration system (50μm stainless steel pre-filter, 10μm ceramic membrane filter, and 1μm PTFE membrane filter for fine filtration); synthetic flavorings use a high-temperature resistant PTFE filter (continuous operating temperature 260℃) to prevent degradation of heat-sensitive components, with a filtration accuracy of 0.1μm. Natural flavorings include a ceramic membrane filter (0.2μm pore size) to prevent plant fiber residue.

[0064] The pollution prevention and control unit includes an inert gas protection chamber. The gas configuration of the inert gas protection chamber is food-grade nitrogen (purity 99.999%), which enters the chamber after being stabilized by a pressure reducing valve (accuracy 0.01MPa). An online oxygen content monitoring probe (detection limit <1ppm) is set up. When the oxygen concentration >5ppm, an audible and visual alarm is triggered. The inert gas protection chamber uses an inflatable silicone sealing strip (compression permanent deformation <20%), which, together with a solenoid valve, enables the chamber to open and close quickly (<10 seconds).

[0065] In one embodiment, such as Figure 2 As shown, this application also discloses a control method for a flavor distillation sampling system. This control method for a flavor distillation sampling system is applied to the aforementioned flavor distillation sampling system, and specifically includes the following steps:

[0066] S1: Obtain the distillation parameters of the flavor raw materials, and generate a dynamic sampling strategy in the distillation sampling system based on the distillation parameters.

[0067] In this embodiment, distillation parameters include material viscosity (unit: mPa·s), boiling point curve (e.g., limonene boiling point range 176-178℃), and thermistivity index (HPI, characterizing the thermal stability of volatile components). The dynamic sampling strategy includes a temperature gradient curve (e.g., linear change from 200℃ at the top of the column to 180℃ at the bottom), pressure compensation value (adjustable within ±0.05MPa range), and theoretical distillation range (e.g., initial boiling point 155℃ / final boiling point 185℃).

[0068] Specifically, step S1 includes:

[0069] S11: Perform data fusion processing on the distillation parameters to obtain preprocessed data including material viscosity, boiling point curve and thermal sensitivity index.

[0070] In this embodiment, the viscosity of the material refers to the flow resistance characteristics of the flavor raw material (unit: mPad·s), such as the viscosity of terpineol at 25°C being 1.15 mPa·s; the boiling point curve refers to the boiling characteristic curve of the raw material components as a function of temperature; the thermodynamic index (HPI) is an indicator used to characterize the thermal stability of volatile components.

[0071] Specifically, a Kalman filter fusion algorithm is used to fuse data using year, boiling point, and thermosensitivity index as state vectors, in order to integrate multi-source heterogeneous data (such as raw material property data, real-time monitoring data, and historical experience data) into a unified dataset.

[0072] S12: Establish a multi-stage distillation model based on preprocessed data and dynamically generate an initial sampling strategy that includes temperature gradient curves, pressure compensation values, and theoretical distillation range.

[0073] In this embodiment, the multi-stage distillation model is a mathematical model that simulates the distillation process in stages; the temperature gradient curve refers to the temperature distribution at different heights within the distillation column; the pressure compensation value refers to the pressure correction amount set to offset environmental disturbances. The theoretical distillation range is the expected distillation interval calculated based on the feed boiling point curve (e.g., initial boiling point 155℃ / final boiling point 185℃).

[0074] Specifically, based on the preprocessed data in S11, Aspen Plus was used for modeling. The number of trays in the model (theoretical trays N=20), reflux ratio (R=1.5), feed position (mid-section feed), the slope of the simulated temperature gradient curve (0.5℃ / min), and the pressure compensation formula (ΔP=0.02×ΔT) were set. The temperature control logic in the initial sampling strategy is as follows: when the ambient temperature change is detected to be >2℃, pressure compensation is triggered (e.g., if the current temperature is 25℃ → the set temperature is 23℃, the compensation ΔP=0.04MPa).

[0075] S13: Real-time acquisition of distillate flow rate, temperature field distribution, and component spectrum data is achieved by a multi-parameter sensor group installed at the top of the distillation column, and dynamic sampling decision tree is generated by combining the initial sampling strategy.

[0076] In this embodiment, the multi-parameter sensor group is a cluster of sensors integrating flow rate, spectrum, and temperature detection functions. The temperature field distribution is the temperature gradient at different positions along the distillation column axis (z-axis). The component spectral data are molecular vibrational characteristic peaks (wavenumber range 4000-400 cm⁻¹) obtained through a miniature Fourier transform infrared spectrometer (FTIR). -1 ).

[0077] Specifically, a sensor array is deployed, including a mass flow meter, an FTIR spectrometer, and a thermocouple array sensor. The mass flow meter measures the distillate flow rate (accuracy ±0.1 mL / min, range 0-50 mL / min); the FTIR spectrometer has a scanning frequency of 10 Hz and detects characteristic peaks (such as limonene at 1380 cm⁻¹). -1The characteristic peak intensity is measured; the thermocouple array consists of K-type thermocouples (accuracy ±1℃) arranged every 20cm along the column. When FTIR detects a sudden increase of 10% in the characteristic peak intensity of limonene (e.g., from 500 to 550), the sampling frequency is increased to once per minute; when the thermocouple detects local overheating (e.g., column bottom temperature > 185℃), a termination sampling command is automatically generated.

[0078] S14: The fuzzy PID control algorithm is used to optimize and adjust the dynamic sampling decision tree to form a dynamic sampling strategy that includes sampling frequency, sampling threshold and abnormal termination conditions.

[0079] In this embodiment, the fuzzy PID control algorithm is a control method based on fuzzy logic to dynamically adjust PID parameters (Kp, Ki, Kd). In the fuzzy PID control algorithm, the error (e) is divided into fuzzy sets (e.g., NB = negative large, PS = positive small). The sampling frequency is the number of samples taken within a specified time interval.

[0080] Specifically, in the large fuzzy rule base of the fuzzy PID control algorithm, the error partitioning rule is as follows: the fuzzy set of error e is set to {NB, NS, ZE, PS, PB} (corresponding range: e < -2, -2 ≤ e < 0, 0 ≤ e < 2, 2 ≤ e < 4, e ≥ 4); the parameter adjustment rule is as follows: when e ∈ NB, Kp = 0.6, Ki = 0.1, Kd = 0.3; when e ∈ PB, Kp = 1.2, Ki = 0.5, Kd = 0.8.

[0081] Parameter optimization during the adjustment includes: when the distillate flow rate fluctuates by more than 15% per unit time, the Kd value is increased by 0.2 to suppress overshoot. The output constraints for the optimization adjustment are: limiting the sampling volume threshold (single sample ≤ 5 mL) and abnormal termination conditions (such as pausing sampling when the pressure rises by more than 0.1 MPa / s). Abnormal termination conditions can also be customized according to the actual flavor type and process flow.

[0082] S2: Using each sampling time point as the basic unit, adjust the distillation temperature and pressure according to the preset first control factor to obtain real-time monitoring data of the distillate, and calculate the corresponding component concentration distribution data based on the real-time monitoring data.

[0083] In this embodiment, the first control factor includes the distillate density change (Δρ), the surface tension change rate, and the vacuum degree of the distillation vessel (absolute pressure ≤ 10 kPa). The component concentration distribution data is quantified using a three-dimensional mapping model (X-axis: time, Y-axis: distillation range, Z-axis: concentration).

[0084] Specifically, step S2 includes:

[0085] S21: Calculate the change in distillate density and identify the pre-distillation, main distillation, and tail distillation stages based on the change in distillate density.

[0086] In this embodiment, the distillate density change value refers to the difference between the real-time density and the initial density of the distillate (unit: g / cm³), reflecting the change in the volatility of the components;

[0087] Specifically, an oscillating tube densitometer (accuracy ±0.001 g / cm³, measurement frequency 1 Hz) is used to obtain the distillate density value in real time. During the pre-distillation stage, Δρ < 0.02 g / cm³ indicates that light components have not been distilled off in large quantities. The main distillation pinch-off is defined as 0.02 g / cm³ < Δρ ≤ 0.08 g / cm³, indicating concentrated distillation of the target components. During the tail distillation stage, Δρ > 0.08 g / cm³ indicates the distillation of high-boiling impurities. In practical applications, the stage identification method is as follows: when Δρ exceeds 0.02 g / cm³ three times consecutively, it is determined that the main distillation stage has begun. During the main distillation stage, if the rate of increase of Δρ slows down (e.g., the rate of change of distillate density < 0.01 g / cm³·min), the distillate density value is further determined. -1 This triggers the main distillation extension logic.

[0088] S22: When the rate of change of surface tension of distillate exceeds the preset rate of change threshold, a three-level temperature compensation mechanism is triggered.

[0089] In this embodiment, the surface tension change rate (Δγ / γ0) is the relative change of the real-time surface tension with respect to the initial value (unit: %). Surface tension is measured using a platinum plate method surface tension meter, and noise is eliminated using Kalman filtering.

[0090] For example, when Δγ / γ0 > 15% (e.g., γ changes from 72 to 61 mN / m), level three compensation is triggered; the compensation is executed as follows:

[0091] Level 1 compensation: Increase PID output by ΔT = 2℃;

[0092] Secondary compensation: The frequency of the vacuum pump inverter is increased from 50Hz to 52.5Hz;

[0093] Level 3 compensation: Open the pressure relief valve (50% opening) to maintain the boiling point at 155℃.

[0094] S23: Calculates the vacuum level of the distillation vessel in real time and adjusts the vacuum level of the distillation vessel in real time through a differential pressure sensor to maintain a constant boiling point environment.

[0095] In this embodiment, the vacuum degree of the distillation vessel (P_abs) is the absolute pressure of the distillation vessel (unit: kPa), for example, a target value of 10 kPa (corresponding to the boiling point depression effect). The differential pressure sensor measures ΔP = P_set - P_abs and outputs a compensation signal, where P_set is the set target value that the flavor distillation sampling system wants to maintain.

[0096] Specifically, the vacuum degree adopts a fuzzy PID control algorithm (the corresponding fuzzy rule base is Kp=0.8, Ki=0.1, Kd=0.3 when e∈NB), and the dynamic calculation formula is: ΔP_comp=Kp×e+Ki×∫edt+Kd×ė, where ΔP_comp is the pressure compensation amount; e is the deviation between the set value and the actual value; and ė is the rate of change of error over time.

[0097] In this embodiment, the constant boiling point is maintained according to the Antoine equation (log 10 P = AB / (T+C) , where A reflects the vapor pressure characteristics of a substance within a specific temperature range; B is related to the latent heat of vaporization and molecular weight of the substance, determining the rate of vapor pressure change with temperature; C is the correction factor for the temperature range offset, ensuring the accuracy of the equation's fit within the specific temperature interval; A is dimensionless, and the units of B and C are °C. When P_abs = 10 kPa, T = 155 °C. If ΔP = ±0.5 kPa is detected, the vacuum pump power adjustment is triggered (e.g., if ΔP = +0.5 kPa, the frequency of the vacuum pump's inverter is reduced by 1%).

[0098] S3: Using each sampling time point as the basic unit, analyze the distillate flow rate and temperature gradient according to the preset second control factor to generate distillation process stability assessment data.

[0099] In this embodiment, the second control factor includes distillate flow rate (Reynolds number Re), temperature gradient (ΔT / Δz, where z is the axial coordinate), and flow rate fluctuation coefficient. The stability assessment data includes dynamic pressure compensation value (ΔP_comp), fraction transition range identification algorithm (such as entropy change detection), and emergency braking threshold (such as ΔP > 0.2MPa triggering circuit breaker).

[0100] Specifically, step S3 includes:

[0101] S31: Construct a fluid dynamics model to calculate the Reynolds number and Mach number of the distillate.

[0102] In this embodiment, the Reynolds number (Re) is a dimensionless number representing the fluid flow state, and the formula is Re=( ×u×D) / ,in Where is density (kg / m³), u is flow velocity (m / s), and D is pipe diameter (m). The dynamic viscosity (Pa·s) is defined as follows: Re < 2000 indicates laminar flow, Re > 4000 indicates turbulent flow, and flow in between is considered transitional. The Mach number (Ma) is the ratio of flow velocity to the local speed of sound, expressed as Ma = u / c, where c = γ × R × T is the speed of sound (m / s), γ is the specific heat ratio, R is the gas constant, and T is the temperature (K). A threshold of Ma < 0.3 is set for subsonic flow to avoid shock wave generation.

[0103] Specifically, the fluid dynamics model employs CFD simulation (ANSYS Fluent) with a structured mesh (mesh size Δx = 1 mm) to ensure boundary layer accuracy. The solver parameters are configured as follows: Realizable k-ε model for turbulence; time step Δt = 0.01 s; convergence criterion: residual < 1e-6. An alarm is triggered when Re < 1000 (laminar flow), indicating that insufficient flow velocity may lead to decreased component separation efficiency.

[0104] S32: Obtain the axial temperature distribution gradient of the distillation column using an infrared thermal imager.

[0105] In this embodiment, the temperature gradient (∇T) refers to the rate of temperature change per unit distance (°C / cm), and the formula is ∇T=ΔT / Δz, reflecting the uniformity of heat distribution within the distillation column. Axial temperature distribution refers to the temperature change curve along the height of the distillation column (z-axis).

[0106] Specifically, an infrared thermal imager, such as a FLIR T650sc (thermal sensitivity <20mK, spatial resolution 1.3mm), can be installed on the side wall of the distillation column (field of view covers the entire column). After noise is eliminated by Kalman filtering, the temperature data is used to calculate the gradient, that is, to calculate ∇T segment by segment along the column height (example: the gradient from the top of the column to the bottom of the column is 3℃ / cm for every 20cm).

[0107] Furthermore, if ∇T>5℃ / cm, it is determined to be local overheating (such as a sudden change in gradient near the column bottom due to coking), triggering the adjustment of vacuum pump power (such as reducing the pumping speed to slow down the evaporation rate).

[0108] S33: Based on the distillate Reynolds number, Mach number, and axial temperature distribution gradient of the distillation column, combined with the flow rate fluctuation coefficient and the standard deviation of the temperature gradient, the system stability assessment data is calculated.

[0109] In this embodiment, the velocity fluctuation coefficient is the ratio (%) of the standard deviation of the velocity to the average velocity, reflecting the stability of the velocity. Temperature gradient standard deviation ( ∇T) refers to the statistical dispersion (°C / cm) of the axial temperature gradient of the distillation column, which measures the temperature uniformity.

[0110] Specifically, the standard deviation of flow velocity Flow velocity fluctuation coefficient = ,in, The number of samples; This represents the average flow velocity. denoted as the actual flow rate of the distillate during the i-th sampling; i is the sampling identifier.

[0111] The temperature standard deviation is used to quantify the uniformity of the axial temperature distribution in a distillation column, and its mathematical expression is as follows:

[0112] ,in, For the first Local temperature gradient value at the measuring point; Let n be the average temperature gradient across all measuring points, and n1 be the total number of measuring points.

[0113] Furthermore, the stability score is comprehensively evaluated by integrating multiple key parameters (such as flow rate fluctuation coefficient, temperature gradient standard deviation, etc.), using the following weighted scoring model:

[0114] S= Where S is the final stability score, with a value range of 0 to 1. The closer to 1, the more stable the system. To allow for the standard deviation of the maximum flow velocity fluctuation; The maximum allowable temperature gradient standard deviation; This represents the pressure fluctuation value. The maximum allowable pressure fluctuation threshold; , and The weighting coefficients for each indicator satisfy... + + =1.

[0115] S4: Based on the component concentration distribution data and distillation process stability assessment data corresponding to each sampling time point, perform dynamic sampling operation and output the optimal sampling result.

[0116] In this embodiment, performing dynamic sampling and outputting the optimal sampling result includes:

[0117] S41: Based on the dynamic sampling decision tree, a sampling operation verification model is established by combining a three-dimensional mapping model, and the confidence level of distillate component migration is calculated.

[0118] In this embodiment, the dynamic sampling decision tree is a decision logic tree generated based on real-time sensor data and historical rules; the three-dimensional mapping model is a mathematical model that maps distillation time (t), distillation range (T*), and component concentration (C) to C=f(t, T*); the component migration confidence refers to the quantification of the migration reliability of the target component in the distillate through a deep neural network (DNN) (scoring range 0-1).

[0119] Specifically, during the construction of the 3D mapping model, the distillate flow rate, temperature field distribution (from an infrared thermal imager), and component spectral data (FTIR) are first input into the LSTM network to generate dynamic decision tree node weights for 3D mapping training. Historical data is used to train the DNN model (input: time, distillation range; output: concentration), with MSE (mean squared error) as the loss function. Confidence is calculated by outputting the transfer confidence score through a Softmax layer (example: when the limonene concentration prediction error is <5%, the confidence score is ≥0.9). The logical verification of the 3D mapping model is as follows: calculate the root mean square error (RMSE) between the predicted concentration and the GC-MS measured value; if RMSE ≤ 0.05, the verification is successful.

[0120] S42: When the emergency braking threshold system is detected to be triggered, the three-level temperature compensation mechanism and the flow rate fluctuation coefficient are coordinated to generate a dynamic sampling fault-tolerant path.

[0121] In this embodiment, the emergency braking threshold system includes sudden pressure rise (ΔP / Δt > 0.2 MPa / s), temperature exceeding limit (T > set value + 10℃), and abnormal flow rate (…). >20%) Level 3 trigger condition.

[0122] Specifically, the three-level temperature compensation mechanism includes:

[0123] Level 1 compensation: PID parameter fine-tuning (±2℃);

[0124] Secondary compensation: Vacuum pump power adjustment (±10%);

[0125] Level 3 compensation: Emergency pressure relief valve opens (ΔP=0.15MPa).

[0126] Threshold detection includes pressure detection and temperature detection. The emergency braking threshold system includes: When the pressure rises suddenly: when ΔP / Δt > 0.2MPa / s, the third-level compensation is triggered (pressure relief valve opening 50%), and the PID output increases by ΔT = 3℃; When the temperature exceeds the limit: if T > set value + 10℃, the second-level compensation is activated (the vacuum pump power of the distillation column is increased by 15%), and the heating power is reduced by 10%.

[0127] Collaborative correction of flow velocity fluctuation coefficient refers to dynamically adjusting the differential term coefficient (Kd) of the PID based on the flow velocity fluctuation coefficient.

[0128] S43: Feature extraction is performed on historical sampling data using reinforcement learning algorithms to construct an adaptive sampling knowledge base that includes the weight of the distillation transition range, the vacuum degree decay compensation value, and the thermal sensitivity index decay factor, and outputs the optimal sampling scheme with confidence scores.

[0129] In this embodiment, the reinforcement learning algorithm refers to the use of a deep Q-network (DQN) for policy optimization, and the reward function is designed as a weighted combination of component recovery rate and energy consumption; the weight of the distillation transition range is assigned to different distillation segments based on historical data (e.g., the weight of the initial boiling point to the main boiling point is 0.6, and the weight of the tail boiling point is 0.4); the vacuum degree decay compensation value refers to the dynamically adjusted compensation amount based on the vacuum pump aging curve (e.g., a decrease of 0.05 kPa every 100 hours).

[0130] Specifically, the reinforcement learning algorithm trains the DQN model based on manually labeled data with optimal sampling parameters (such as sampling time, frequency, and quantity). During DQN model training, the reward function is R = 0.7 × η_recovery rate + 0.3 × (1 − energy consumption / baseline energy consumption), and the learning rate is 0.001.

[0131] Specifically, dynamic weight adjustment refers to adjusting the fraction weights based on the real-time distillation stage (initial distillation / main distillation / tail distillation) (e.g., increasing the weight of the main distillation section to 0.8). The calculation formula for attenuation compensation is: vacuum compensation value = 0.01 × t (t is the running time in hours). After the DQN model is trained, the action with the highest Q value is selected, and a sampling instruction with confidence score is output (e.g., "Sample immediately, confidence score 0.85") to output the optimal sampling scheme.

[0132] Furthermore, a control method applied to a flavor distillation sampling system also includes:

[0133] Dynamic pressure fluctuation compensation value is used to correct sampling errors caused by changes in ambient temperature.

[0134] A fraction transition zone identification algorithm is used to identify abrupt changes in distillate composition.

[0135] The emergency braking threshold system includes pressure surge protection, temperature over-limit fuse, and flow abnormality cut-off mechanism.

[0136] In this embodiment, the dynamic pressure fluctuation compensation value refers to the value based on changes in ambient temperature ( The dynamically corrected pressure compensation amount ΔP_comp is calculated using the following formula:

[0137] ΔP_comp= ,in, The proportionality coefficient (unit: kPa / ℃) represents the pressure compensation directly generated when the ambient temperature changes by 1℃. The change in ambient temperature (unit: °C); The integral coefficient is given by kPa·s / ℃.

[0138] The fraction transition range identification algorithm refers to locating the abrupt change nodes in distillate composition by detecting abrupt changes in component concentration gradients (second derivative method). The formula is as follows:

[0139] Mutation point = ,in, The second derivative of component concentration measures the acceleration of the rate of concentration change and reflects the intensity of abrupt changes in component distillation. The mutation threshold is the critical value used to determine whether a component has undergone a significant mutation.

[0140] In this embodiment, the emergency braking threshold system further includes:

[0141] ① Pressure surge protection: When ΔP / Δt>0.2MPa / s, the pressure relief valve opening is triggered to 50%.

[0142] ② Temperature over-limit fuse: When T>set value+10℃, the heating power is turned off and nitrogen cooling is started.

[0143] ③ Abnormal flow rate cut-off: When the standard deviation of the flow rate is >20%, the feed is stopped and the pump speed is reset to a safe value.

[0144] In one embodiment, a control method applied to a flavor distillation sampling system further includes:

[0145] When performing dynamic sampling operations, a federated learning framework is used to synchronously optimize the weights of the dynamic sampling decision tree nodes. Specifically, the pressure fluctuation data of the distillation vessel is input into the Generative Adversarial Network to generate a virtual pressure disturbance scenario. The locally optimized sampling frequency parameters are uploaded to the cloud collaborative database through a differential privacy mechanism to achieve cross-device parameter federated aggregation.

[0146] Specifically, Federated Learning refers to a distributed machine learning framework that allows multiple devices to collaboratively train models, with data remaining local and only model parameters shared. Generative Adversarial Networks (GANs) are deep learning models composed of a generator and a discriminator, used to generate realistic virtual data; Differential Privacy protects data privacy by adding noise, ensuring that individual data cannot be reverse-engineered.

[0147] Specifically, historical pressure fluctuation data from the distillation vessel is collected and normalized. The generator of the Generative Adversarial Network (GAN) is input with a random noise vector (100 dimensions) and outputs a virtual pressure sequence (1000 units in length). The network structure consists of fully connected layers (512→256→128) + LeakyReLU activation function. The discriminator of the GAN is input with the real / virtual pressure sequences and outputs the probability of authenticity (values ​​ranging from 0 to 1). The corresponding network structure consists of an LSTM layer (128 units), a fully connected layer (64→1), and a Sigmoid activation function. The loss function uses Wasserstein GAN (WGAN) loss with a gradient penalty term. The virtual pressure sequence generated by the GAN (e.g., a sudden ΔP = ±1.2 kPa) is input into a decision tree model to simulate extreme conditions. This application uses reinforcement learning (DQN) to train the weights of the decision tree nodes, improving the sampling stability under abnormal pressure.

[0148] Specifically, when the differential privacy mechanism aggregates parameters in a federated manner, the optimization object of local parameter optimization is the sampling frequency parameter of each device (such as the initial value of 1 time / minute); the objective function is to minimize the sum of local sampling error (MSE) and privacy loss, where the privacy loss is associated with the privacy protection strength coefficient as a weighting coefficient, and the privacy protection strength coefficient is set to 0.5.

[0149] Specifically, the steps to achieve privacy include:

[0150] ① Noise addition: Add Gaussian noise to the locally optimized parameters (e.g., sampling frequency f = 1.2 times / minute): in, The sampling frequency after adding noise; =0.1 is the standard deviation of Gaussian noise (calculated based on a privacy budget ε=2.0); f is the original sampling frequency; N(0, The mean is 0 and the variance is . A Gaussian distribution is used to generate noise.

[0151] ② Privacy budget allocation: An adaptive σ adjustment strategy is adopted to ensure that the total ε ≤ 3.0.

[0152] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0153] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0154] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A control method for a flavor distillation sampling system, characterized in that, The flavor distillation sampling system includes: A multi-stage CCP control module is integrated into the key processes of the flavor production process, including raw material receiving and acceptance, weighing and preparation, filtration and slag removal, and heating or stirring control. An online purification unit, connected between the fragrance production and packaging stages, includes: Multi-stage filtration components are used to adapt the corresponding filter mesh size according to different fragrance types; Foreign object detection components employ metal detectors or X-ray inspection to detect the presence of foreign object residue before packaging; The pollution prevention control unit includes an inert gas protection chamber suitable for solid or powdered fragrances; The method includes: Obtain the distillation parameters of the flavor raw materials, and generate a dynamic sampling strategy based on the distillation parameters in the distillation sampling system; Using each sampling time point as a basic unit, the distillation temperature and pressure are adjusted according to a preset first control factor to obtain real-time monitoring data of the distillate, and the corresponding component concentration distribution data are calculated based on the real-time monitoring data. Using each sampling time point as the basic unit, the distillate flow rate and temperature gradient are analyzed according to the preset second control factor to generate distillation process stability assessment data. Based on the component concentration distribution data and distillation process stability assessment data corresponding to each sampling time point, dynamic sampling operation is performed and the optimal sampling result is output. The first control factor includes the distillate density change value, the distillate surface tension change rate, and the distillation vessel vacuum degree. Adjusting the distillation temperature and pressure according to the preset first control factor includes: Calculate the change in distillate density, and identify the pre-distillation, main distillation, and tail distillation stages based on the change in distillate density; When the rate of change of surface tension of the distillate exceeds the preset rate of change threshold, a three-level temperature compensation mechanism is triggered. The vacuum level of the distillation vessel is calculated in real time, and the vacuum level is adjusted in real time through a differential pressure sensor to maintain a constant boiling point environment. The second control factor includes distillate flow rate, temperature gradient, and flow rate fluctuation coefficient; stability assessment data includes dynamic pressure compensation value, fraction transition zone identification algorithm, and emergency braking threshold. The step of analyzing the distillate flow rate and temperature gradient based on a preset second control factor to generate distillation process stability assessment data includes: Construct a fluid dynamics model to calculate the Reynolds number and Mach number of the distillate; The axial temperature distribution gradient of the distillation column was obtained using an infrared thermal imager. Based on the distillate Reynolds number, Mach number, and axial temperature distribution gradient of the distillation column, the system stability assessment data are calculated by combining the flow rate fluctuation coefficient and the standard deviation of the temperature gradient.

2. The control method for a flavor distillation sampling system according to claim 1, characterized in that, The process of obtaining distillation parameters for flavoring raw materials and generating a dynamic sampling strategy in the distillation sampling system based on these parameters includes: The distillation parameters are subjected to data fusion processing to obtain preprocessed data including material viscosity, boiling point curve and thermal sensitivity index; A multi-stage distillation model is established based on the preprocessed data, and an initial sampling strategy is dynamically generated, including temperature gradient curves, pressure compensation values, and theoretical distillation range. A multi-parameter sensor group installed at the top of the distillation column is used to collect distillate flow rate, temperature field distribution and component spectrum data in real time, and a dynamic sampling decision tree is generated in combination with the initial sampling strategy. The dynamic sampling decision tree is optimized and adjusted using a fuzzy PID control algorithm to form a dynamic sampling strategy that includes sampling frequency, sampling threshold, and abnormal termination conditions.

3. The control method for a flavor distillation sampling system according to claim 1, characterized in that, The component concentration distribution data was obtained through the following methods: Online component analysis of the distillate was performed using a miniature Fourier transform infrared spectrometer; The migration patterns of characteristic aroma components were verified using gas chromatography-mass spectrometry. Establish a three-dimensional mapping model of distillation time, component concentration, and distillation range.

4. The control method for a flavor distillation sampling system according to claim 3, characterized in that, The method also includes: Dynamic pressure fluctuation compensation value is used to correct sampling errors caused by changes in ambient temperature. A fraction transition zone identification algorithm is used to identify abrupt changes in distillate composition. The emergency braking threshold system includes pressure surge protection, temperature over-limit fuse protection, and flow abnormality cut-off mechanism.

5. The control method for a flavor distillation sampling system according to claim 4, characterized in that, The process of performing dynamic sampling and outputting the optimal sampling result includes: Based on the dynamic sampling decision tree, a sampling operation verification model is established in conjunction with the three-dimensional mapping model, and the confidence level of distillate component migration is calculated. When the emergency braking threshold system is detected to be triggered, the three-level temperature compensation mechanism and the flow rate fluctuation coefficient are coordinated to generate a dynamic sampling fault-tolerant path. By using reinforcement learning algorithms to extract features from historical sampling data, an adaptive sampling knowledge base is constructed, which includes the weight of the distillation transition range, the vacuum degree decay compensation value, and the thermal sensitivity index decay factor. The optimal sampling scheme with confidence score is then output.

6. The control method for a flavor distillation sampling system according to claim 2, characterized in that, The method also includes: During dynamic sampling operations, a federated learning framework is used to synchronously optimize the weights of the dynamic sampling decision tree nodes, wherein: Input the pressure fluctuation data of the distillation vessel into the generative adversarial network to generate a virtual pressure disturbance scenario; By using a differential privacy mechanism, the locally optimized sampling frequency parameters are uploaded to a cloud-based collaborative database, enabling cross-device parameter federation.

Images