Intelligent optimization and control system for biomimetic multi-stage extraction process of soy isoflavones

Abstract

The application discloses an intelligent optimization and control system for a soybean isoflavone biomimetic multistage extraction process, and relates to the technical field of food processing and automatic control.The system comprises a raw material physical property online sensing unit, a biomimetic extraction environment construction unit, a multistage self-adaptive extraction execution unit, a dynamic interface mass transfer monitoring unit, a biomimetic kinetics intelligent optimization unit and a solvent closed-loop circulation and resource recycling unit.The application realizes real-time sensing and self-adaptive adjustment of interface tension, absorbance and mass transfer rate in the extraction process through integration of near-infrared spectrum scanning, biomimetic transport model prediction and cascade countercurrent extraction control, solves the problem of unstable yield caused by fluctuation of raw material components, and significantly improves isoflavone extraction purity through digital twin evolution and dynamic correction of solvent polarity, reduces solvent and energy consumption, and ensures green, efficient, safe and reliable production process.

Description

Technical Field

[0001] This invention relates to the field of food processing and automation control technology, and in particular to an intelligent optimization and control system for a biomimetic multi-stage extraction process of soybean isoflavones. Background Technology

[0002] In the field of the health industry and the development of bioactive substances, soy isoflavones, as a key component with significant physiological functions such as preventing cardiovascular diseases, anti-oxidation, and improving osteoporosis, have always been a research focus in the efficient extraction and purification of natural products. With the increasing market demand for high-purity bioactive extracts, traditional extraction processes are evolving towards greener, more intelligent, and biomimetic approaches. This process involves complex solvent selection, mass transfer equilibrium, and the dynamic migration of target molecules in multi-level environments. Its core lies in constructing a complex kinetic system that can simulate biological physiological environments and achieve precise component separation, thereby maximizing the preservation of the target product's bioactivity and improving resource conversion efficiency.

[0003] Among them, the intelligent optimization and control system for the biomimetic multi-stage extraction process of soybean isoflavones serves as a key means to improve extraction accuracy and process stability. It aims to achieve real-time monitoring and closed-loop regulation of the entire extraction process through the deep integration of biomimetic strategies and intelligent algorithms. This system constructs a biomimetic extraction logic capable of adapting to complex and variable environments by collaboratively optimizing solvent interfacial tension, material ratio, flow field distribution, and kinetic parameters during the multi-stage extraction process. This not only requires the system to be able to sense fluctuations in upstream raw materials but also to achieve dynamic balance control of the multi-stage extraction gradient and mass transfer efficiency to ensure consistent final product quality throughout the entire production cycle.

[0004] Current technologies for soybean isoflavone extraction still face multiple challenges in terms of intelligence and biomimetic environmental adaptability. While some control schemes can achieve coordinated parameter adjustment by collecting physical field data, they are mostly designed for specific supercritical conditions and do not adequately consider the biomimetic kinetics of multi-stage liquid-phase extraction, limiting their applicability in complex processes simulating biological environments. Furthermore, existing control logic largely relies on linear feedback of physical parameters, lacking in-depth analysis of the dynamic evolution of the solvent interface and multi-stage coupling effects during extraction. This results in unsatisfactory response flexibility and energy utilization when handling large-scale, multi-stage tasks. In addition, traditional mechanical quantitative devices can only achieve static control of upstream materials, lacking real-time monitoring and intelligent feedback capabilities throughout the extraction process. They cannot effectively cope with interference from fluctuations in raw material components, and the lack of automatic optimization mechanisms for extraction gradients and dynamic equilibrium directly leads to limited mass transfer efficiency and excessive solvent consumption. These shortcomings collectively restrict the intensive and intelligent development of the isoflavone extraction industry. Therefore, an optimized intelligent optimization and control system for the biomimetic multi-stage extraction process of soybean isoflavones is desired. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent optimization and control system for the biomimetic multi-stage extraction process of soybean isoflavones, in order to solve the problems in the prior art such as insufficient consideration of biomimetic kinetic characteristics, lack of dynamic evolution analysis of liquid phase multi-stage extraction, weak system sensitivity to raw material fluctuations, limited mass transfer efficiency, and large solvent consumption.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: A smart optimization and control system for a biomimetic multi-stage extraction process of soybean isoflavones includes: The raw material property online sensing unit performs continuous non-destructive testing on incoming soybean raw materials, and acquires a set of digital features of raw materials including initial isoflavone concentration, cell wall integrity, particle size distribution and surface water activity. The biomimetic extraction environment construction unit adjusts the hydrogen ion concentration index, ionic strength, and surfactant addition amount of the extraction solvent system in real time based on the digital feature set of raw materials output by the online raw material property sensing unit, simulating the molecular transport environment in a living organism. The multi-stage adaptive extraction execution unit constructs a physical topology consisting of multiple extraction vessels connected in series, and dynamically adjusts the feed-liquid ratio, stirring shear intensity, and extraction residence time of each stage according to the real-time concentration gradient in each extraction vessel. The dynamic interface mass transfer monitoring unit detects the refractive index, absorbance, conductivity and solvent interfacial tension of the extract online. By analyzing the diffusion rate of substances at the solvent interface, it determines the dynamic equilibrium state of mass transfer in each extraction process in real time. The biomimetic dynamics intelligent optimization unit receives production parameters from each unit, constructs a digital twin system based on a biomimetic transport model, and generates control setpoint correction instructions for each front-end execution unit based on the deviation between the purity of the output and the energy consumption index. The solvent closed-loop recycling and resource recovery unit performs multi-effect evaporation and membrane module separation on the waste mother liquor after extraction to recover high-purity ethanol solvent and monitor the polarity index of the recovered solvent to ensure the consistency of solvent quality during the recycling process.

[0007] In one embodiment of the present invention, the online raw material property sensing unit integrates a near-infrared spectrometer and an online laser particle size analyzer. The near-infrared spectrometer performs spectral scanning at a transparent window in the material conveying pipeline, with a scanning frequency set to once per minute, acquiring spectral data in the wavelength range of 1000 nm to 2500 nm. The system converts the spectral characteristics into a mass percentage value of soybean isoflavones using a preset partial least squares model. The online laser particle size analyzer utilizes the principle of laser diffraction to measure the particle size distribution of the pulverized material in real time, generating a particle size feature vector with the median diameter as the core. The raw material digital feature set also includes moisture content data measured by a capacitive moisture meter.

[0008] The biomimetic extraction environment construction unit includes a high-precision reagent metering pump set and an online pH transmitter. Based on the assessment results of the raw material cell wall integrity, the system instructs the pH transmitter to monitor the hydrogen ion concentration index of the extraction solvent in real time. When the raw material cell wall density exceeds a preset threshold of 20%, the control unit adds a specific concentration of a cellulase and pectinase mixture to the solvent via the high-precision reagent metering pump to achieve pre-softening of the biomass structure. Simultaneously, the system adjusts the mixing ratio of ethanol and deionized water to control the solvent polarity index within a preset range of 3.5 to 5.2 to match the solubility characteristics of different isoflavone isomers.

[0009] The multi-stage adaptive extraction unit employs a cascaded countercurrent extraction mode. Each extraction vessel is equipped with an independent variable frequency stirring motor and a heat transfer oil jacket temperature control system. The system's control logic is as follows: the first-stage extraction vessel receives the raw material with the highest concentration and uses the lower-concentration extract from the second-stage extraction vessel as a solvent; this continues until the final extraction vessel uses pure solvent. Ultrasonic flow meters are deployed inside each extraction vessel to monitor the feed-to-liquid ratio in real time. When the total flavonoid concentration in the first-stage extract reaches 85% of its saturation solubility, the system instructs the variable frequency stirring motor to increase its speed to 120-180 revolutions per minute, thereby enhancing shear force to disrupt the boundary layer and improve the instantaneous mass transfer coefficient.

[0010] The dynamic interface mass transfer monitoring unit deploys an online fiber optic spectral probe at the overflow port of each extraction vessel. This probe monitors the absorbance changes of the extract at characteristic wavelengths of 260 nm and 310 nm in real time. The system calculates the rate of change of absorbance over time; when the rate of change is below a preset threshold of 0.01 for three consecutive minutes, the extraction stage is considered to have reached quasi-equilibrium. Furthermore, the dynamic interface mass transfer monitoring unit utilizes an online interfacial tensiometer based on the pendant drop principle to monitor the tension evolution at the solvent interface. When a sudden change in interfacial tension exceeds 10%, the system automatically identifies it as severe phase emulsification or impurity dissolution and immediately triggers a slag discharge frequency adjustment command.

[0011] The biomimetic dynamics intelligent optimization unit operates within a high-performance computing server, and its core control logic is based on a nonlinear Bayesian inference algorithm. This unit acquires real-time data streams of raw material properties, temperature, pressure, flow rate, and interfacial tension, and combines this with the historical best production batch's trajectory to construct a high-dimensional state-space control matrix. The system calculates the Euclidean distance between the current state point and the optimal trajectory to predict the product concentration fluctuation trend over the next 30 minutes. When the predicted value deviates from the set target by more than 5%, the optimization unit uses a heuristic search algorithm to generate corrected control setpoints, including increasing the temperature of the third-stage extraction vessel by 2 degrees Celsius and increasing the circulation reflux ratio by 10%. The correction instructions are issued to the PLC execution unit via the fieldbus protocol. The solvent closed-loop circulation and resource recovery unit includes a triple-effect falling film evaporator and a ceramic ultrafiltration membrane module. The extracted mother liquor first passes through the ceramic ultrafiltration membrane to remove large molecular impurities such as proteins and polysaccharides, and then enters the triple-effect falling film evaporator. The system monitors the vacuum level in the evaporation chamber using a pressure transmitter and adjusts the steam flow rate to maintain a constant evaporation rate. The recovered ethanol vapor is condensed by a condenser and enters a solvent storage tank. The storage tank is equipped with a conductivity sensor and a refractometer to determine the ethanol purity of the recovered solvent online. When the purity is below 95%, the system automatically activates the reflux branch of the distillation column for secondary purification.

[0012] Furthermore, the system also includes an intelligent flow field adjustment module, which has adjustable-angle baffles installed on the inner walls of each extraction vessel and driven by a stepper motor. When the viscosity of the extraction system increases by 15% due to the increase in material concentration, the system automatically adjusts the baffle angle to change the turbulence distribution inside the vessel, eliminate low-speed dead zones, and ensure sufficient contact between the solid and liquid phases.

[0013] The temperature control of the multi-stage adaptive extraction unit adopts an inter-stage gradient setting scheme. The first-stage extraction temperature is set at 45 degrees Celsius, and each subsequent stage increases by 5 degrees Celsius until it does not exceed 65 degrees Celsius. This gradient design aims to use temperature differences to drive molecular diffusion while avoiding the thermosensitive degradation of soy isoflavones caused by high temperatures.

[0014] The dynamic interface mass transfer monitoring unit also includes a solvent polarity dynamic correction algorithm. This algorithm calculates the required solvent parameters in real time based on the online detection of the proportion of isoflavone monomer types. If the proportion of daidzein increases, the system will automatically increase the percentage of ethanol in the solvent, with the increase controlled between 2% and 5% each time, until the absorbance recovers.

[0015] The system integrates a production safety early warning subsystem. This subsystem monitors the concentration of ethanol vapor in the production environment and the internal pressure of each extraction vessel in real time. When the ambient ethanol concentration exceeds 20% of the lower explosive limit or the pressure inside the vessel exceeds 0.2 MPa, the system immediately initiates an emergency venting and depressurization procedure, cuts off all electric heating devices, and simultaneously sends a fault code to the higher-level management platform via industrial Ethernet.

[0016] In one embodiment of the present invention, a material buffer chamber is provided between the online raw material property sensing unit and the multi-stage adaptive extraction execution unit. The buffer chamber is equipped with a load sensor to mitigate flow fluctuations caused by uneven yields in the upstream crushing process. The system predicts the sustainable operating time of the downstream extraction unit by reading the rate of change of the load sensor, and adjusts the frequency converter of the screw feeder accordingly to achieve constant control of the feed rate.

[0017] The system's control interface uses a digital dashboard display. The dashboard presents real-time extraction efficiency curves at each stage, solvent recovery rate, energy consumption level, and real-time fluctuations of key control parameters. Historical data is stored locally in 1-second intervals, and the system supports retrospective analysis of production process trajectories for any time period within the past 365 days.

[0018] Furthermore, the biomimetic dynamics intelligent optimization unit introduces a self-learning mechanism during algorithm iteration. The system uses the output feedback after each execution of a correction instruction as a positive or negative incentive to automatically adjust the neuron connection weights of the internal control model. After evolution through 100 to 200 production batches, the system can establish an optimal extraction strategy library for soybean raw materials from different origins.

[0019] The solvent closed-loop circulation and resource recovery unit is equipped with an online chemical oxygen demand (COD) analyzer at the discharge outlet. This analyzer collects a water sample every 30 minutes to analyze the organic matter content of the discharged wastewater. If the value exceeds 100 mg / L, the system automatically closes the discharge valve and diverts the wastewater to an anaerobic digester for harmless treatment, ensuring that the entire production process meets green and environmentally friendly standards.

[0020] The actuator of the intelligent flow field adjustment module uses an explosion-proof stepper motor with a motor housing protection rating of no less than IP65. The motor and main controller are connected via shielded twisted-pair cable and equipped with a surge protector to suppress interference from the complex industrial electromagnetic environment on signal transmission, ensuring control accuracy within 0.5 degrees.

[0021] The multi-stage adaptive extraction unit employs a combination design of a three-bladed propeller and a six-bladed open turbine. The bottom blades are responsible for agitating the settled material, while the upper blades generate axial circulation. The system calculates the fluid viscosity based on real-time collected torque data, and automatically increases the output torque via a frequency converter when the torque increases by 10%.

[0022] The probe surface of the dynamic interface mass transfer monitoring unit is treated with a nano-anti-fouling coating. This coating has extremely high oleophobic and hydrophobic properties, preventing soybean protein or lipids from adhering to and forming scale on the fiber optic probe surface. The system performs an automatic air-flushing cleaning every 4 hours, using clean compressed air at 0.2 MPa to blow away any potential deposits on the probe surface, ensuring the long-term accuracy of optical detection.

[0023] In one embodiment of the present invention, the distributed control architecture of the system consists of one central industrial control computer, four field control cabinets, and several intelligent sensor terminals. The central industrial control computer is responsible for running high-level optimization algorithms and persistently storing data; the field control cabinets perform millisecond-level process logic control through PLCs; and the intelligent sensor terminals realize local data acquisition and signal conditioning. Efficient and transparent data transmission is achieved between each level through an industrial bus network that meets real-time requirements.

[0024] Compared with the prior art, the beneficial technical effects of the present invention are as follows: This invention overcomes the limitations of traditional extraction processes that rely solely on adjusting a single physical parameter by constructing a biomimetic extraction environment building unit. The system can simulate the complex molecular transport mechanisms in living organisms, creating an excellent physicochemical environment for the dissolution of soy isoflavone molecules by dynamically adjusting the hydrogen ion concentration index, ionic strength, and surfactants. This biomimetic logic not only improves the dissolution rate of the target product but also maximizes its biological activity. Experimental data show that compared with traditional control methods, this system increases the yield of soy isoflavones by more than 22% within the same time period.

[0025] This invention introduces a dynamic interface mass transfer monitoring unit, which enables real-time and accurate perception of the micro-dynamic evolution of the extraction process. By monitoring the solvent interfacial tension and absorbance change rate online, the system can accurately determine whether each extraction stage has reached thermodynamic equilibrium, thereby avoiding energy waste and impurity dissolution caused by ineffective stirring and over-extraction. The introduction of this unit reduces the system's error in determining the extraction endpoint to within 3 minutes, significantly improving the compactness of the production cycle.

[0026] This invention utilizes a biomimetic dynamic intelligent optimization unit to achieve closed-loop automatic optimization of the entire process; the control model based on nonlinear Bayesian inference can autonomously extract control laws from massive production data, effectively coping with external interference caused by fluctuations in raw material composition; this adaptive optimization mechanism enables the system to ensure that the purity deviation of the final product is controlled within 0.5% even when the soybean isoflavone content fluctuates by 15%, greatly improving the stability and consistency of product quality.

[0027] This invention achieves extremely high resource conversion efficiency and environmental friendliness through the coordinated operation of a multi-stage adaptive extraction execution unit and a solvent closed-loop circulation and resource recovery unit. The cascade countercurrent extraction mode, combined with precise dynamic adjustment of the feed-liquid ratio, reduces solvent consumption per unit product by more than 30%. At the same time, the integrated high-efficiency solvent recovery system ensures that the recycling rate of ethanol solvent reaches more than 98%, significantly reducing production costs and emissions of volatile organic compounds.

[0028] This invention features a high degree of system integration, with comprehensive safety warning and intelligent flow field adjustment functions. Through a digital twin system and enhanced flow field control, it effectively solves the dead zone phenomenon and mass transfer bottleneck in high-concentration extraction systems. The system's self-learning capability enables the production process to be continuously optimized with the increase of batches, realizing the transformation from an experience-driven to a knowledge-driven production mode. The entire system has a rigorous structure and clear logic, providing a mature, reliable, and highly innovative solution for the industrial-scale precise extraction of soybean isoflavones. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the overall technical architecture of the intelligent optimization and control system for the biomimetic multi-stage extraction process of soybean isoflavones proposed in this invention. Figure 2 This is a schematic diagram of the core principle framework of intelligent optimization and closed-loop control based on a biomimetic dynamics model proposed in this invention. Detailed Implementation

[0030] The features and exemplary embodiments of various aspects of the present invention will now be described in detail. To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely intended to explain the present invention and not to limit the present invention. For those skilled in the art, the present invention can be practiced without some of these specific details. The following description of the embodiments is merely to provide a better understanding of the present invention by illustrating examples of the invention.

[0031] Example 1 Please refer to the attached document. Figure 1 This embodiment discloses an intelligent optimization and control system for a biomimetic multi-stage extraction process of soybean isoflavones. Through multi-level hardware integration and software algorithm coupling, it achieves fully digital control of the soybean isoflavone extraction process. The system comprises an online raw material property sensing unit, a biomimetic extraction environment construction unit, a multi-stage adaptive extraction execution unit, a dynamic interface mass transfer monitoring unit, a biomimetic kinetics intelligent optimization unit, and a solvent closed-loop circulation and resource recovery unit. Millisecond-level data interaction between the units is achieved via industrial Ethernet and fieldbus technology, ensuring the coordinated operation and dynamic optimization of the entire extraction process chain.

[0032] The raw material property online sensing unit is deployed at the feed front of the production line, and its main task is to continuously scan the quality of incoming soybean raw materials. This unit integrates a near-infrared spectrometer, an online laser particle size analyzer, and a capacitive moisture meter. The near-infrared spectrometer is installed at the quartz glass transparent window of the material conveying pipeline. Its light source emits a continuous spectrum with wavelengths ranging from 1000 nm to 2500 nm, penetrating a uniformly thick material layer. The spectral acquisition frequency is set to once per minute, with each acquisition including multiple scans and averaging to eliminate non-uniformity caused by material flow. The partial least squares model running within the system performs second-derivative processing and multivariate scattering correction on the original spectrum, thereby accurately retrieving the initial mass percentage value of soybean isoflavones. Simultaneously, the online laser particle size analyzer uses the principle of laser diffraction to capture the scattered light signal of particles during material descent, calculates the particle size distribution of the material using Mie scattering theory, and extracts the median diameter as the core feature. The capacitive moisture meter acquires surface water activity data in real time by measuring the change in dielectric constant as the material flows through the electrode plates. The data output from the aforementioned sensors are aggregated into a digital feature set of the raw materials, including the initial concentration of isoflavones, cell wall integrity assessment values, particle size distribution feature vectors, and surface water activity. This feature set is transmitted in real time to subsequent units as the original basis for setting process parameters.

[0033] A material buffer silo is installed between the online raw material property sensing unit and the multi-stage adaptive extraction execution unit. A high-precision load cell is installed at the bottom of this silo to monitor the material level in real time. By reading the rate of change of the load cell, the system calculates the difference between the current front-end yield and the back-end extraction consumption rate, thereby predicting the sustainable operating time of the extraction unit. When the material level falls below the preset 30% safety level, the system automatically adjusts the frequency converter of the screw feeder to smooth flow fluctuations and achieve constant feed rate control.

[0034] The biomimetic extraction environment construction unit receives a digital feature set of raw materials from the front end, with its core objective being to create the optimal physicochemical environment for the release of soybean isoflavones. This unit includes a high-precision reagent metering pump set, an online pH transmitter, an ion strength regulating valve, and a surfactant injection assembly. The system determines the degree of material damage based on the assessment results of cell wall integrity. When the cell wall density of the raw material exceeds a preset threshold of 20%, the control unit instructs the high-precision reagent metering pump to inject a composite enzyme solution of cellulase and pectinase mixed in a 1:1 ratio into the extraction solvent. The online pH transmitter monitors the hydrogen ion concentration index of the solvent in real time. By adding trace amounts of citric acid or sodium hydroxide solution to the solvent circulation loop, the pH is maintained in the weakly acidic range of 5.0 to 6.5 to simulate the acid-base balance environment in a living organism. Furthermore, the system controls the mixing ratio of ethanol and deionized water by adjusting the frequency converter pump ratio to ensure that the solvent polarity index remains stable between 3.5 and 5.2. This multi-parameter coupling regulation mechanism aims to minimize the energy barrier for isoflavone molecules to escape from the biomass matrix, thereby increasing their dissolution rate.

[0035] The multi-stage adaptive extraction unit employs a cascaded countercurrent extraction mode based on physical topology. This unit consists of a first-stage extraction vessel, a second-stage extraction vessel, and a third-stage extraction vessel connected in series. Each extraction vessel is equipped with an independent variable-frequency stirring motor, a heat transfer oil jacket temperature control system, an ultrasonic flow meter, and an intelligent flow field adjustment module. The intelligent flow field adjustment module has four adjustable-angle baffles on the inner wall of the vessel, driven by an explosion-proof stepper motor with a protection rating of 65. The system calculates the viscosity of the current extraction system based on real-time collected torque data. When the viscosity increases by 15% due to increased material concentration, the stepper motor automatically adjusts the baffle angle to change the turbulence distribution within the vessel and eliminate dead zones. The stirring blades adopt a combination design of a three-blade propeller and a six-blade open turbine. The bottom blades are responsible for lifting the precipitated material particles upwards, while the upper blades generate strong axial circulation. The system's inter-stage temperature control logic uses a gradient setting scheme: the first-stage extraction temperature is set at 45 degrees Celsius, increasing by 5 degrees Celsius for each subsequent stage. This gradient design utilizes the thermal diffusion force generated by the temperature difference, which, together with the concentration gradient, drives the migration of isoflavone molecules. The first-stage extraction vessel receives the raw material and uses the extract from the second-stage extraction vessel as a solvent, while the final extraction vessel is circulated with high-purity recycled ethanol solvent. This configuration greatly improves solvent utilization.

[0036] The dynamic interface mass transfer monitoring unit is crucial for achieving precise process control. Each extraction vessel's overflow pipe is equipped with an online fiber optic spectral probe, an online conductivity meter, and an online interfacial tensiometer based on the pendant drop principle. The online fiber optic spectral probe scans the absorbance of the extract at characteristic wavelengths of 260 nm and 310 nm in real time. The system's internal logic calculates the rate of change of absorbance over time in real time. When the rate of change is below a preset threshold of 0.01 for three consecutive minutes, the system determines that the extraction process at that stage has reached thermodynamic quasi-equilibrium and immediately issues a discharge command. The online interfacial tensiometer captures the morphological changes of solvent droplets and calculates the tension evolution at the solvent interface in real time. When the abrupt change in interfacial tension exceeds 10%, the system identifies excessive dissolution of impurities or phase emulsification and immediately issues a slag discharge frequency adjustment command to the execution layer. To ensure long-term accuracy of optical detection, the probe surface is covered with a nano-anti-fouling coating with oleophobic and hydrophobic properties. Every 4 hours, the system automatically cleans the probe surface with 0.2 MPa clean compressed air to remove potential protein deposits.

[0037] Please refer to the attached document. Figure 1 and appendix Figure 2The biomimetic dynamics intelligent optimization unit, running within a high-performance computing server, serves as the central decision-making brain of the entire system. Based on a nonlinear Bayesian inference algorithm, this unit constructs a multidimensional digital twin system encompassing raw material properties, temperature, pressure, flow rate, and interfacial tension. The core of this digital twin system lies in modeling the state evolution of the extraction process as a Bayesian nonlinear state-space model, where the process state equation and observation equation describe the extraction kinetics evolution and sensor measurement relationships, respectively. The Bayesian nonlinear state-space model treats system parameters as random variables and assigns them prior distributions, rather than fixed unknown constants as in traditional control methods. Iterative updates to the posterior distribution of parameters are performed using real-time measurement data acquired online, enabling the digital twin model to adaptively track the dynamic changes in the extraction process. In terms of state-space modeling, the system uses the raw material property feature set as the initial state vector, and the temperature, pressure, feed-to-liquid ratio, stirring speed, and solvent interfacial tension within each extraction vessel constitute the process state vector, constructing a high-dimensional nonlinear control matrix. The raw material property feature set includes the initial isoflavone concentration, cell wall integrity, particle size distribution, and water activity. The system employs a nonlinear Bayesian inference algorithm—specifically, the sequential Monte Carlo method—to perform online state inference and parameter identification on the state-space model. Within each control cycle, the posterior distribution of the system state is sampled and estimated, thereby analyzing in real-time the coupling relationships between state variables and their weighted impact on mass transfer efficiency. Based on the real-time updated state estimates, the system constructs a control matrix in the state space and calculates the weighted Euclidean distance between the current operating state and the historical optimal production trajectory. To quantify the kinetic characteristics of the extraction process, the system applies the following biomimetic interface mass transfer rate equation: In the above formula, This represents the mass transfer flux of isoflavone molecules at the solvent interface. Indicates the overall mass transfer coefficient. This represents the total interfacial area of ​​the solid-liquid contact. This represents the saturated solubility of soy isoflavones in the current solvent system. This represents the real-time concentration of the extract. As the apparent activation energy, is the gas constant, and This represents the absolute temperature of each extraction vessel. Using this formula, the system can analyze the decay of the mass transfer driving force in real time. When it is predicted that the product concentration will deviate from the target by more than 5% within the next 30 minutes, the biomimetic dynamics intelligent optimization unit uses a heuristic search algorithm to generate correction instructions, such as instructing the variable frequency stirring motor of the third-stage extraction vessel to increase its speed to 150 revolutions per minute, or increasing the setpoint of the heat transfer oil jacket temperature control system by 2 degrees Celsius. The system has a self-learning mechanism, which can automatically update the connection weights of the internal control model based on the output feedback after each adjustment.

[0038] The solvent closed-loop circulation and resource recovery unit achieves a green closed loop in the production process. This unit includes a ceramic ultrafiltration membrane module, a triple-effect falling film evaporator, a distillation column, and a solvent storage tank. The waste mother liquor after extraction first enters the ceramic ultrafiltration membrane module, utilizing its extremely high chemical stability and mechanical strength to retain proteins, polysaccharides, and macromolecular impurities under a pressure of 0.3 MPa. The clarified liquid then enters the triple-effect falling film evaporator for solvent recovery. The system monitors the vacuum level in each evaporation chamber via a pressure transmitter and maintains a constant evaporation rate by adjusting the steam valve opening. The recovered ethanol vapor is condensed by a condenser and enters the solvent storage tank. The tank is equipped with a conductivity sensor and a refractometer. When the ethanol purity of the recovered solvent is detected to be below 95%, the system automatically switches the solenoid valve to divert the solvent to the reflux branch of the distillation column for secondary purification. Furthermore, a solvent polarity dynamic correction algorithm adjusts the blending ratio after ethanol recovery in real time based on the online detected isoflavone monomer ratio. If the proportion of daidzein in the raw materials increases, the system will instruct the proportion of ethanol in the temporary storage tank to be increased, with the increase controlled at about 3% each time, until the absorbance reading of the downstream monitoring unit returns to the normal range.

[0039] The system integrates a production safety early warning subsystem to ensure operational safety in complex industrial environments. This subsystem deploys ethanol vapor concentration sensors at key locations in the workshop and retrieves real-time data from the internal pressure transmitters of each extraction vessel. When the ethanol concentration in the environment reaches 20% of the lower explosive limit or the pressure inside the vessel exceeds 0.2 MPa, the production safety early warning subsystem immediately triggers emergency shutdown logic. This logic includes: cutting off the power to all electric heating devices, opening the emergency venting and pressure relief solenoid valves, activating the forced ventilation system throughout the workshop, and simultaneously sending a fault code to the central industrial control computer. All safety events and control commands are recorded in a distributed database at 1-second intervals, supporting digital traceability of the process trajectory at any production node over the past 365 days.

[0040] The system's distributed control architecture ensures high reliability of management and control. The central industrial control computer is responsible for running high-level optimization algorithms and storing big data; four field control cabinets achieve millisecond-level process control via programmable logic controllers (PLCs); and intelligent sensor terminals distributed throughout the pipelines enable local signal conditioning of the data. Transparent transmission between levels is achieved through an industrial bus protocol that meets real-time requirements. At the discharge outlet, the system is also equipped with an online chemical oxygen demand (COD) analyzer. This analyzer collects water samples every 30 minutes, analyzing the organic load of the discharged wastewater using the potassium dichromate method. If the value exceeds 100 mg / L, the system automatically closes the main discharge valve and instructs a pneumatic pump to switch the wastewater to an anaerobic digester for biodegradation treatment, ensuring that the entire soybean isoflavone extraction process meets environmental standards.

[0041] Regarding system stability, the motors of the multi-stage adaptive extraction actuators are connected to the main controller using shielded twisted-pair cables, along with high-performance surge protectors, effectively suppressing electromagnetic interference caused by the high-frequency switching of the inverter. The stepper motor housings of all actuators undergo anodizing treatment, achieving industrial-grade corrosion resistance. Through this deep integration of hardware and software, the system achieves intelligent control throughout the entire process from raw material input to product output, significantly reducing solvent loss and improving the extraction purity and production efficiency of isoflavones.

[0042] Example 2 This embodiment, based on the extraction of soybean isoflavones, has undergone process adaptation and system parameter fine-tuning for black soybean raw materials from a specific origin. Black soybean raw materials, due to the high concentration of anthocyanins and melanin in their seed coats, are more sensitive to the extraction environment. Please refer to the appendix. Figure 1 In this embodiment, the online raw material property sensing unit has been enhanced with a visible light absorption spectroscopy module to assess the characteristic peak intensity of seed coat pigments while detecting isoflavone content. The scanning frequency of the near-infrared spectroscopy analyzer has been increased to once every 30 seconds to capture fluctuations in raw material components more precisely.

[0043] The biomimetic extraction environment construction unit, tailored to the specific characteristics of black soybean material, presets the solvent polarity index to a range of 3.8 to 5.5. This is because the proportion of isoflavone aglycones in black soybeans differs significantly from that in ordinary yellow soybeans. The system introduces trace amounts of nonionic surfactants via a reagent metering pump assembly to reduce the contact angle between the solvent and the material surface. The online pH transmitter's setpoint is locked at 5.2 to prevent excessive acidity from causing structural degradation of anthocyanins.

[0044] In this embodiment, the multi-stage adaptive extraction unit increases the base speed of the variable frequency stirring motor. Due to the higher viscosity of black soybeans after grinding, the stirring torque is increased by 12% compared to yellow soybeans. The intelligent flow field adjustment module increases the shear force field intensity inside the vessel by adjusting the baffle angle. The inter-stage gradient setting scheme is modified as follows: the extraction temperature for the first stage is 48 degrees Celsius, increasing by 3 degrees Celsius for each subsequent stage, with the maximum temperature limited to below 60 degrees Celsius, in order to protect the heat-sensitive active ingredients using mild thermodynamic conditions.

[0045] Please refer to the attached document. Figure 1 and appendix Figure 2 In this embodiment, the biomimetic dynamics intelligent optimization unit employs a source-specific strategy library trained through reinforcement learning. For efficiency monitoring during solvent recovery, the system applies the following solvent recovery performance evaluation index formula: In the above formula, This represents the mass balance efficiency of solvent recovery. and These represent the volume of recovered liquid produced by the condenser and the volume of the original liquid entering the evaporator, respectively. and For the corresponding real-time density value, and This is the ethanol mass fraction measured by a refractometer. The system dynamically adjusts the steam supply pressure of the triple-effect falling film evaporator by calculating this value in real time. When the value is below 98%, the system automatically starts the heat leakage diagnosis subroutine to determine whether there is scaling on the condenser tube by analyzing the data fluctuations of the pressure transmitter.

[0046] In this embodiment, the solvent closed-loop circulation and resource recovery unit enhances the consistent control of the polarity of the recovered solvent. Because a significant amount of fine colloidal substances are generated during black soybean extraction, the cleaning frequency of the ceramic ultrafiltration membrane module is increased. The system monitors the pressure difference across the membrane module; when the pressure difference exceeds 0.15 MPa, it automatically switches to the standby membrane module and performs in-situ chemical cleaning on the fouled membrane module. This configuration ensures the continuity of the black soybean extraction process, resulting in a final soy isoflavone product with a purity exceeding 92% and a 40% reduction in pigment impurities.

[0047] Example 3 This embodiment discloses an intelligent optimization and control system for a biomimetic multi-stage extraction process of soybean isoflavones. Based on a high degree of automation, it focuses on enhancing the precise identification and adaptive control of the proportions of different isoflavone monomers. Please refer to the appendix. Figure 1 The fiber optic online spectral probe of the dynamic interface mass transfer monitoring unit not only monitors absorbance at a fixed wavelength but also introduces a full-spectrum scanning mode. This mode can identify the concentration distribution of daidzein, genistein, and their corresponding aglycones in solution.

[0048] The biomimetic extraction environment construction unit dynamically adjusts the hydrogen ion concentration index of the solvent based on real-time feedback of the monomer ratio. Research has found that when the proportion of genistein increases, appropriately increasing the ionic strength of the solvent helps to enhance its diffusion rate. The system replenishes the circulating solvent with a certain proportion of saline solution via an ionic strength regulating valve, with the adjustment range between 0.05 mol / L and 0.2 mol / L. Simultaneously, the temperature control scheme of the multi-stage adaptive extraction execution unit is also linked, with the temperature of the third-stage extraction vessel dynamically fine-tuned according to the genistein dissolution curve, achieving an adjustment accuracy of 0.1 degrees Celsius.

[0049] In this embodiment, the biomimetic dynamics intelligent optimization unit activates a deep neural network module. This module structures data from the past 1000 production batches, establishing a nonlinear mapping between isomer ratios and mass transfer efficiency. The system predicts component shift trends for the next hour. If the prediction indicates a decrease in the proportion of daidzein, the optimization unit automatically instructs the first-stage extraction vessel to increase the residence time by 5% and fine-tunes the pump flow rate via a frequency converter to offset the yield decrease caused by component changes.

[0050] The solvent closed-loop circulation and resource recovery unit incorporates an electrodialysis desalination component to handle high-ionic-strength solvents. Before the mother liquor enters the triple-effect falling film evaporator, electrolyte ions are removed from the solvent via electrodialysis to prevent scaling on the evaporator tube walls, which would affect heat transfer efficiency. The top temperature of the distillation column is controlled at 78.2 degrees Celsius, with a temperature control error of no more than 0.05 degrees Celsius, ensuring that the purity of the produced ethanol remains above 98%. The solvent polarity dynamic correction algorithm in the temporary storage tank is even more refined, dynamically allocating weights based on monomer polarity differences to ensure that each batch of solvent used is precisely matched to the chemical characteristics of the current material.

[0051] In this embodiment, the production safety early warning subsystem incorporates anomaly diagnosis based on machine hearing. Through acoustic sensors installed on the top of each extraction vessel, the system can capture the operating sound patterns of the variable frequency stirring motor under different loads. When an abnormal shift in the sound pattern frequency is detected, or periodic mechanical impact sounds are detected, the system determines that there may be a mechanical hazard in the baffle angle adjustment mechanism or the stirring shaft, and issues a secondary maintenance warning in advance, preventing unplanned downtime due to mechanical failure. The digital dashboard includes real-time pie charts of component distribution and monomer yield trend prediction curves, allowing operators to intuitively grasp the microscopic dynamics of the extraction process and achieving deeper-level industrial production decision support.

[0052] Example 4 This embodiment describes the integrated application of the system in a large-scale industrial park, which achieves collaborative optimization of multiple extraction production lines through central scheduling logic. Please refer to the appendix. Figure 1 In this system, the online raw material property sensing units operate in parallel at the feed inlets of each production line. All digital feature sets of the raw materials are aggregated into a global database center. The central industrial control computer automatically allocates different batches of raw materials to specific extraction units based on the load of each production line.

[0053] In this scenario, the multi-stage adaptive extraction execution unit demonstrates exceptional flexibility. If a second-stage extraction vessel on a production line malfunctions, the system automatically adjusts the process topology, temporarily switching from the original cascade countercurrent extraction mode to an enhanced two-stage high-throughput extraction mode, and recalculates the control setpoints through a biomimetic dynamics intelligent optimization unit. At this time, the system instructs the first-stage extraction vessel to increase the stirring shear intensity to 180 revolutions per minute and adjust the solvent ratio from 1:10 to 1:15 to compensate for the mass transfer loss caused by the reduced number of stages. This self-healing control logic significantly enhances the plant's resilience.

[0054] The solvent closed-loop circulation and resource recovery unit achieves balanced solvent scheduling across the entire plant. The recovered high-purity ethanol not only supplies this production line but is also allocated across lines based on their real-time polarity requirements. Secondary steam generated by the triple-effect falling film evaporator is recovered and reused via heat pipe technology to provide heat energy for the solvent preheating process in the biomimetic extraction environment construction unit. This comprehensive energy utilization reduces the steam consumption per ton of product across the entire park by 25%.

[0055] The data from the dynamic interface mass transfer monitoring unit is used not only for process determination on this line but also as an evolutionary training set for the plant-wide digital twin system. By comparing the dynamic equilibrium state of mass transfer for the same raw materials on different production lines, the system identifies performance deviations in each piece of equipment caused by long-term operation. If the mass transfer rate constant of extraction vessel No. 3 is detected to be consistently below 5% of the average level, the system automatically lists it as a priority maintenance target. The production safety early warning subsystem is linked to the park's fire control center. Once the ambient ethanol concentration triggers a Level 1 alarm, the park-wide venting and depressurization network will coordinate its actions to ensure the overall safety of the storage tank area and the production area.

[0056] Example 5 This embodiment discloses an intelligent optimization and control system for a biomimetic multi-stage extraction process of soybean isoflavones, which demonstrates excellent performance in extracting high-purity daidzein. Due to the extremely low solubility of daidzein in water, the system incorporates a supercritical pre-regulation loop within its biomimetic extraction environment construction unit. When the raw material's digital feature set indicates that the daidzein content exceeds 35% of the total isoflavones, the system instructs the solvent system to enter a semi-critical high-pressure state. By increasing the pressure to 5 MPa and injecting a specific co-solvent, the migration ability of daidzein molecules is enhanced.

[0057] The first-stage extraction vessel of the multi-stage adaptive extraction unit is designed for high pressure and corrosion resistance. The variable frequency stirring motor uses magnetic coupling drive, completely solving the dynamic sealing problem under high pressure. The system calculates the overall mass transfer coefficient in real time based on the interface mass transfer rate equation. Due to the increase in pressure, the activation energy... As the sensitivity to temperature changes, the biomimetic dynamic intelligent optimization unit automatically adjusts its internal regression coefficients. Experiments show that under high-pressure assisted conditions, the extraction time of daidzein is shortened from 4 hours to 1.5 hours, without a significant increase in solvent usage.

[0058] The dynamic interface mass transfer monitoring unit utilizes a sapphire-encapsulated fiber optic online spectrometer for high-pressure environments. The probe's optical path is automatically compensated to ensure the accuracy of absorbance data under pressure fluctuations. The solvent closed-loop circulation and resource recovery unit incorporates a dedicated pressure energy recovery turbine device. When the supercritical extractant undergoes a pressure relief operation before entering the evaporator, the system converts pressure energy into electrical energy via the turbine, feeding it back to the plant's power grid. The condensate heat energy from the triple-effect falling film evaporator is also recovered a second time via a plate heat exchanger to maintain a constant return water temperature in the extraction vessel's thermal oil jacket temperature control system.

[0059] In this specific embodiment, the solvent polarity dynamic correction algorithm employs a more advanced partial least squares discriminant analysis model. This model can accurately identify trace amounts of impurity monomers based on the fingerprint spectrum of the recovered solvent and remove them through the side-stream sampling function of the distillation column. The ethanol purity in the temporary storage tank is increased to 99% analytical grade to meet the stringent solvent quality requirements of high-purity daidzein. A digital dashboard simultaneously presents a three-dimensional cloud map of monomer purity distribution, demonstrating the real-time concentration gradient evolution process from the solid phase to the liquid phase.

[0060] Example 6 This embodiment focuses on enhancing the system's anti-interference capabilities and self-healing function during the extraction of soybean isoflavones, particularly in situations involving power system fluctuations or extremely uneven physical properties of raw materials. Please refer to the appendix. Figure 1 The raw material property online sensing unit integrates a 3D machine vision component. This component scans the material flow pattern at the screw feeder outlet in real time. If large agglomerates are detected, the system automatically instructs the online pulverizer upstream of the execution unit to increase its speed. The feature vector measured by the online laser particle size analyzer is mapped in real time to the dynamic torque compensation value of the stirring motor inside the extraction vessel.

[0061] In this embodiment, the multi-stage adaptive extraction execution unit incorporates pulsed stirring logic. The system does not operate at a constant speed continuously, but rather based on the mass transfer rate fed back by the dynamic interface mass transfer monitoring unit. When the absorbance increases slowly, the variable frequency stirring motor generates a 2 Hz speed pulse for a short period, with a maximum speed of 220 revolutions per minute. This transient high shear force effectively breaks the already stable liquid film boundary layer, reactivating the mass transfer process. The execution of this logic is automatically triggered by the biomimetic dynamic intelligent optimization unit based on Bayesian inference results.

[0062] The solvent closed-loop circulation and resource recovery unit has added a solvent quality lifecycle assessment module. This module predicts solvent degradation trends by recording the number of cycles, polarity shift history, and impurity accumulation curves for each batch of solvent. When the assessment results show that the solvent quality has declined to a critical point, the system will automatically schedule deep distillation purification of the entire solvent volume in the next maintenance cycle. Simultaneously, the conductivity sensor in the solvent storage tank can identify trace amounts of organic acids introduced during the extraction process and trigger the ion exchange resin tower to perform online acid removal.

[0063] In terms of safety and compliance, the online chemical oxygen demand (COD) analyzer integrates automatic sampling and dilution functions to meet the instantaneous detection needs of high-concentration organic wastewater. The system features a dedicated environmental compliance page on a digital dashboard, recording and publicly displaying wastewater treatment parameters and the percentage of wastewater discharged in compliance with standards in real time. The neuron weight adjustment mechanism within the biomimetic dynamic intelligent optimization unit adds an "energy consumption sensitivity" factor. When calculating correction instructions, the algorithm automatically avoids the operating conditions with the lowest energy efficiency ratio, ensuring the system always operates at the Pareto optimal frontier. This deep combination of engineering optimization and biological logic gives this system a high level of technological leadership in the field of intelligent extraction of soybean isoflavones.

[0064] In summary, this embodiment constructs a highly closed-loop, adaptive, and environmentally friendly intelligent control system through the extreme refinement and dynamic coupling of each unit. The system not only achieves efficient multi-level extraction at the physical level but also realizes microscopic simulation and macroscopic control of molecular migration processes at the information level, perfectly solving the core pain points of limited mass transfer and fluctuation perception in the background technology.

[0065] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape, and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. An intelligent optimization and control system for a biomimetic multi-stage extraction process of soybean isoflavones, comprising: The raw material physical property online sensing unit performs continuous non-destructive testing on incoming soybean raw materials. It integrates a near-infrared spectrometer, an online laser particle size analyzer, and a capacitive moisture meter, and outputs a digital feature set of raw materials including initial isoflavone concentration, cell wall integrity, and material particle size distribution. The biomimetic extraction environment construction unit adjusts the hydrogen ion concentration index, ionic strength, and surfactant addition amount of the extraction solvent system according to the digital feature set of the raw materials. The biomimetic extraction environment construction unit includes a high-precision reagent metering pump group and an online pH transmitter, which controls the solvent polarity index within the range of 3.5 to 5.2 by adjusting the mixing ratio of ethanol and deionized water. The multi-stage adaptive extraction execution unit adopts a cascaded countercurrent extraction mode, consisting of a physical topology composed of multiple extraction vessels connected in series. Each extraction vessel is equipped with an independent variable frequency stirring motor, a heat transfer oil jacket temperature control system, and an intelligent flow field adjustment module including adjustable angle baffles. The multi-stage adaptive extraction execution unit adjusts the feed-liquid ratio and stirring shear intensity according to the real-time concentration gradient in each extraction vessel. The dynamic interface mass transfer monitoring unit detects the refractive index, absorbance and solvent interfacial tension of the extract online. An online fiber optic spectral probe and an online interfacial tension meter are deployed at the overflow port of each extraction vessel. The dynamic equilibrium state of mass transfer is determined by analyzing the diffusion rate of substances at the solvent interface. The biomimetic dynamics intelligent optimization unit operates based on a nonlinear Bayesian inference algorithm, constructs a digital twin system based on a biomimetic transport model, and generates control setpoint correction instructions in reverse based on the deviation of output indicators. The solvent closed-loop recycling and resource recovery unit includes a triple-effect falling film evaporator and a ceramic ultrafiltration membrane module to recover high-purity ethanol solvent and monitor the polarity index of the recovered solvent.

2. The intelligent optimization and control system for the biomimetic multi-stage extraction process of soybean isoflavones according to claim 1, characterized in that, A material buffer chamber is provided between the online raw material property sensing unit and the multi-level adaptive extraction execution unit; the material buffer chamber is equipped with a load sensor to read the load change rate and adjust the frequency of the screw feeder according to the load change rate.

3. The intelligent optimization and control system for the biomimetic multi-stage extraction process of soybean isoflavones according to claim 1, characterized in that, The temperature control of the multi-stage adaptive extraction execution unit adopts an inter-stage gradient setting scheme; the extraction temperature of the first stage is set to 45 degrees Celsius, and each subsequent stage increases by 5 degrees Celsius, with a maximum of 65 degrees Celsius.

4. The intelligent optimization and control system for the biomimetic multi-stage extraction process of soybean isoflavones according to claim 1, characterized in that, The probe surface of the dynamic interface mass transfer monitoring unit is treated with a nano anti-pollution coating and is automatically air-cleaned every 4 hours.

5. The intelligent optimization and control system for the biomimetic multi-stage extraction process of soybean isoflavones according to claim 1, characterized in that, The near-infrared spectrometer is used to acquire spectral data from 1000 nm to 2500 nm and convert the spectral data into a mass percentage value of soybean isoflavones; the line laser particle size analyzer uses the principle of laser diffraction to determine the particle size distribution of the pulverized material and generate a particle size feature vector with the median diameter as the core; the capacitive moisture meter acquires surface water activity data by measuring the change in dielectric constant when the material flows through the electrode plate.

6. The intelligent optimization and control system for the biomimetic multi-stage extraction process of soybean isoflavones according to claim 1, characterized in that, It also includes an intelligent flow field adjustment module, which drives an explosion-proof stepper motor to automatically adjust the angle of the baffle inside the vessel when the viscosity of the extraction system increases by 15%.

7. The intelligent optimization and control system for the biomimetic multi-stage extraction process of soybean isoflavones according to claim 1, characterized in that, It also includes a production safety early warning subsystem, which activates an emergency venting and depressurization procedure when the ambient ethanol concentration exceeds 20% of the lower explosive limit or the pressure inside the vessel exceeds 0.2 MPa.

8. The intelligent optimization and control system for the biomimetic multi-stage extraction process of soybean isoflavones according to claim 1, characterized in that, The solvent closed-loop circulation and resource recovery unit is equipped with an online chemical oxygen demand analyzer at the discharge outlet; when the analysis result exceeds 100 mg / L, the system automatically diverts the wastewater to the anaerobic digester.

9. The intelligent optimization and control system for the biomimetic multi-stage extraction process of soybean isoflavones according to claim 1, characterized in that, The stirring blades in the multi-stage adaptive extraction execution unit adopt a combination design of three-blade propulsion and six-blade open turbine, which automatically increases the output torque based on real-time torque data.

Images