Multistage membrane integrated intelligent milk concentration and storage device and quality control system thereof

Abstract

The invention relates to the technical field of dairy product processing, in particular to a multi-stage membrane integrated intelligent milk concentration and storage device and a quality control system thereof, and particularly relates to an efficient intelligent milk concentration and storage solution combining a multi-stage membrane separation technology, multispectral analysis, multi-sensor fusion, deep reinforcement learning and a digital twinning technology. According to the multi-stage membrane integrated intelligent milk concentration and storage device, three-stage gradient membrane separation and double closed-loop circulation design are fused, efficient and fine separation of milk is achieved, and the concentration efficiency and the quality consistency are improved. The multi-spectral analysis and multi-sensor fusion technology is integrated, the viscosity, the pH value, the water activity and the conductivity of the concentrated solution are monitored in real time, and the quality is guaranteed in an omnibearing and multi-dimensional mode. The system intelligently adjusts the parameters of the filtering membrane through multispectral signals and sensor data, and ensures the stable quality of the concentrated solution. According to the device, the intelligent level of milk concentration storage is effectively improved, an innovative solution is provided for the food processing industry, quality control is enhanced, and the product competitiveness is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of dairy processing, and particularly to a multi-stage membrane integrated intelligent milk concentration and storage device and its quality control system, and more particularly to an efficient intelligent milk concentration and storage solution combining multi-stage membrane separation technology, multi-spectral analysis, multi-sensor fusion, deep reinforcement learning, and digital twin technology. Background Art

[0002] Milk concentration technology has important application value in the field of dairy product processing. Traditional milk concentration methods mainly include thermal evaporation concentration and freeze concentration. However, thermal evaporation concentration has high energy consumption and is prone to quality problems such as denaturation of milk proteins and caramelization of sugars; freeze concentration has the disadvantages of complex process and high cost. In recent years, membrane separation technology has been widely used in the field of milk concentration due to its low energy consumption, high separation efficiency, and ability to retain milk nutrients.

[0003] However, there are still many technical bottlenecks in the existing membrane separation milk concentration technology: First, a single-stage membrane separation system is difficult to achieve fine separation of milk components, affecting the concentration efficiency and quality; second, membrane fouling is the main problem in the membrane separation process, resulting in a decrease in membrane flux and shortening of membrane life; third, there is a lack of real-time and comprehensive quality monitoring means, unable to ensure the stability of the quality of concentrated milk; fourth, the regulation of membrane separation parameters mainly relies on experience, making it difficult to achieve dual optimization of energy consumption and product quality; fifth, traditional storage methods cannot manage concentrated milk dynamically according to quality, easily leading to quality fluctuations and resource waste.

[0004] Although there have been some technical solutions in the industry to attempt to solve the above problems, such as using a series of multi-membrane systems to improve separation efficiency, using an automatic cleaning system to extend membrane life, and applying various sensors to monitor product quality. However, there is currently no system that can organically integrate multi-stage membrane technology, intelligent monitoring, automatic cleaning, intelligent optimization control, and intelligent storage management to achieve the intelligentization and high efficiency of the entire process of milk concentration and storage. Summary of the Invention

[0005] The object of the present invention is to provide a multi-stage membrane integrated intelligent milk concentration and storage device and its quality control system to solve the problems of low membrane separation efficiency, serious membrane fouling, incomplete quality monitoring, difficult parameter optimization, and non-intelligent storage management existing in the prior art.

[0006] The present invention proposes a multi-stage membrane integrated intelligent milk concentration and storage device, including: It includes: a milk storage tank; a primary filtration membrane module, whose outlet is connected to the milk storage tank to form a primary concentration circulation path; a secondary filtration membrane module, which is connected to the primary filtration membrane module to form a secondary concentration circulation path; a tertiary filtration membrane module, which is connected to the secondary filtration membrane module; a multispectral analyzer, which acquires the multispectral signals of the concentrated liquids at all levels; multiple sensors, which acquire the viscosity, pH value, water activity and conductivity of the concentrated liquids at all levels; a storage unit, which acquires the quality of the concentrated liquid based on the multispectral signals and the sensor signals, and adjusts the parameters of the filtration membrane module.

[0007] Preferably, the primary filtration membrane module is a microporous filter with a filtration accuracy of 0.2 μm.

[0008] Preferably, the secondary filtration membrane module is a membrane stack with a pore size of 100 nm.

[0009] Preferably, the tertiary filtration membrane module is a membrane stack with a pore size of 50 nm.

[0010] Preferably, it further includes: a terminal storage tank for storing the tertiary concentrated liquid output by the tertiary filtration membrane module, wherein the mass of the concentrated liquid with a quality not exceeding the prefabricated quality parameter stored in the terminal storage tank is not greater than the difference between the set maximum liquid storage volume and the mass of the concentrated liquid with a quality exceeding the prefabricated quality parameter.

[0011] Preferably, it further includes: an on-line cleaning unit, which determines the position of membrane fouling based on the multispectral signal of the terminal storage tank, and generates a control signal to close the membrane module at the previous level before the position of membrane fouling and open the membrane module at the next level after the position of membrane fouling, so that the milk is concentrated in the membrane module at the previous level before the position of membrane fouling.

[0012] Preferably, it further includes: a deep reinforcement learning unit, which acquires the optimal membrane flux and optimal membrane operating pressure at each stage based on the membrane flux, membrane operating pressure of each filtration membrane module and the parameters of the concentrated liquids at all levels, wherein the deep reinforcement learning unit acquires the comprehensive energy consumption index and milk quality at each stage based on the membrane flux and the membrane operating pressure.

[0013] Preferably, it further includes: a virtual simulation unit, which constructs a digital twin system of the primary concentration circulation path and the secondary concentration circulation path, and acquires the optimal parameters of at least one filtration membrane module based on the optimized parameters of the digital twin system.

[0014] Preferably, it further includes: an edge computing unit, which constructs a digital twin system of the primary concentration circulation path and the secondary concentration circulation path, and acquires the optimal process parameters based on the optimized parameters of the digital twin system and the parameters of the concentrated liquids at all levels, wherein the process parameters include the positions and detection periods of the sensors.

[0015] Quality control system for a multi-stage membrane integrated intelligent milk concentration and storage device, comprising: A data acquisition module for acquiring multi-spectral signals, viscosity, pH value, water activity and conductivity data of concentrated liquids at all levels of the multi-stage membrane concentration device; A data processing module, connected to the data acquisition module, for performing feature extraction and anomaly detection based on the multi-spectral signals and the parameter data; A deep learning module, connected to the data processing module, for establishing a milk quality prediction model and evaluating the quality of the concentrated liquid based on the prediction model; A parameter optimization module, connected to the deep learning module, for adjusting the operating parameters of the multi-stage membrane concentration device based on the quality evaluation results; An on-line cleaning module, connected to the parameter optimization module, for automatically adjusting the working state of the membrane module and executing a cleaning program when membrane fouling is detected.

[0016] The multi-stage membrane integrated intelligent milk concentration and storage device and its quality control system of the present invention have the following beneficial effects: 1. Through the design of a three-stage gradient membrane separation system and a double-closed-loop circulation path, efficient and fine separation of milk components is achieved, improving the concentration efficiency and quality consistency; 2. Integrating multi-spectral analysis and multi-sensor fusion technology, real-time monitoring of the quality of concentrated milk in all directions and multi-dimensions is realized, providing comprehensive data support for quality control; 3. Adopting a deep reinforcement learning algorithm, adaptive optimization of parameters such as membrane flux and operating pressure is achieved, balancing energy consumption and product quality; 4. The on-line cleaning unit based on multi-spectral signal analysis can accurately identify the location and type of membrane fouling, realize intelligent cleaning, and extend the membrane life; 5. By integrating virtual simulation and edge computing technologies, a digital twin system is constructed to optimize process parameters and improve the overall efficiency of the system; 6. An intelligent storage management mechanism based on quality is established to realize quality grading storage and predictive management of concentrated milk. Description of the Drawings

[0017] Figure 1 It is a schematic diagram of the system structure of the multi-stage membrane integrated intelligent milk concentration and storage device of the present invention; Figure 2 It is a schematic diagram of the structure of the first-stage concentration circulation path of the present invention; Figure 3 It is a schematic diagram of the structure of the second-stage concentration circulation path of the present invention; Figure 4 It is a schematic diagram of the structure of the deep reinforcement learning unit of the present invention; Figure 5 Schematic diagram of the working process of the online cleaning unit of the present invention; Figure 6 Schematic diagram of the relationship between the virtual simulation unit and the edge computing unit of the present invention; Figure 7 Schematic diagram of the module composition of the quality control system of the present invention; Figure 8 Schematic diagram of the working principle of multi-spectral analysis and multi-sensor fusion of the present invention; Figure 9 Schematic diagram of the membrane fouling detection and cleaning strategy selection process of the present invention; Figure 10 Schematic diagram of the working process of the deep learning model for predicting the quality of milk of the present invention. Detailed implementation manners

[0018] Please refer to the appended Figures 1-10 drawings, and the present invention will be further described in detail below in conjunction with the drawings and embodiments.

[0019] As Figure 1 shown, the multi-stage membrane integrated intelligent milk concentration and storage device of the present invention mainly includes a milk storage tank 1, a primary filtration membrane module 5, a secondary filtration membrane module 7, a tertiary filtration membrane module 11, a multi-spectral analyzer 17, a plurality of sensors, and a storage unit 20.

[0020] The milk storage tank 1 is used to store the milk to be processed, and the milk is transported to the primary filtration membrane module 5 through a pipeline 3. The primary filtration membrane module 5 has an outlet, and the primary dialysis fluid flowing out through this outlet is communicated with the milk storage tank 1 to form a primary concentration circulation path. Preferably, the primary concentration circulation path includes the primary filtration membrane module 5, the first outlet of the primary concentrated liquid passing through the filtration membrane module 5, the milk storage tank 1, the second outlet of the primary filtration membrane module 5, the first outlet of the secondary filtration membrane module 7, and the secondary concentrated liquid, which are sequentially communicated and form a closed loop.

[0021] The secondary filtration membrane module 7 is communicated with the primary filtration membrane module 5, its inlet is fluidly communicated with the first outlet of the primary filtration membrane module 5, and the first outlet and the second outlet are respectively fluidly communicated with the second outlet of the primary filtration membrane module 5 through pipelines 9 and 10 to form a secondary concentration circulation path. Preferably, the secondary concentration circulation path includes the secondary filtration membrane module 7, the second outlet of the secondary concentrated liquid filtration membrane module 7, the second outlet of the secondary filtration membrane module 7, and the first outlet of the tertiary filtration membrane module 11, which are sequentially communicated and form a closed loop.

[0022] The tertiary filtration membrane module 11 is in communication with the secondary filtration membrane module 7, and its inlet is in fluid communication with the second outlet of the secondary filtration membrane module 7. The first outlet and the second outlet are respectively in fluid communication with the first outlet of the secondary filtration membrane module 7 via pipelines 13 and 14. The tertiary concentrate is in fluid communication with the terminal storage tank 19 through the second outlet of the tertiary filtration membrane module 11.

[0023] A multispectral analyzer 17 is provided in the system for acquiring multispectral signals of the primary concentrate, the secondary concentrate, and the tertiary concentrate. In an embodiment of the present invention, the multispectral analyzer 17 can adopt a near-infrared spectrometer with a wavelength range of 400 - 2500 nm, covering the characteristic absorption peaks of the main components in milk such as protein, fat, and lactose.

[0024] Multiple sensors are distributed at key positions in the system for acquiring the viscosity, pH value, water activity, and conductivity of the primary concentrate, the secondary concentrate, and the tertiary concentrate. Preferably, these sensors include a micro rotary viscometer, a pH electrode sensor, a water activity sensor based on the capacitance principle, and a four-electrode conductivity sensor.

[0025] The storage unit 20 is electrically connected to the multispectral analyzer 17 and the multiple sensors, receives the multispectral signals and the multiple sensor signals, and obtains the quality of each concentrate based on these signals. When a concentrate with a quality exceeding the pre-set quality parameter is detected, the storage unit 20 stores the concentrate and the corresponding multispectral signal. In particular, during the process of the storage unit 20 acquiring a concentrate with a quality exceeding the pre-set quality parameter, if the amount of this concentrate is not higher than the set liquid storage amount threshold, the storage unit 20 adjusts the parameters of at least one of the primary filtration membrane module 5, the secondary filtration membrane module 7, and the tertiary filtration membrane module 11 based on the multispectral signal of this concentrate until the amount of the concentrate with a quality exceeding the pre-set quality parameter reaches the set maximum liquid storage amount.

[0026] In a preferred embodiment of the present invention, the primary filtration membrane module 5 is a microporous filter with a filtration accuracy of 0.2 μm. This accuracy selection is based on the size distribution of large particulate matter and microorganisms in milk, which can effectively remove bacteria and impurities while retaining valuable components in milk. The microporous filtration membrane is usually made of polyvinylidene fluoride (PVDF) or polysulfone (PS) materials, having good mechanical strength and chemical stability.

[0027] The secondary filtration membrane module 7 is a membrane stack with a pore size of 100 nm. This pore size can effectively retain macromolecular proteins in milk while allowing small molecules to pass through. In an embodiment of the present invention, the membrane stack adopts a multi-layer flat ceramic membrane design, with a flow channel interval provided between each layer to increase the effective filtration area and improve the filtration efficiency.

[0028] The tertiary filtration membrane module 11 is a membrane stack with a pore size of 50 nm. This pore size design can further finely separate small molecule proteins and lactose in milk, achieving a highly concentrated effect. Preferably, the membrane stack is made of polyethersulfone (PES) or modified polyacrylonitrile (PAN) materials, having excellent pressure resistance and selective permeation performance.

[0029] The terminal storage tank 19 is used to store the tertiary concentrate output by the tertiary filtration membrane module 11. In an embodiment of the present invention, the terminal storage tank 19 adopts a double-layer stainless steel structure, with the inner layer being food-grade 316L stainless steel, the outer layer being provided with a thermal insulation layer, and equipped with a temperature control system to maintain the storage environment within the optimal temperature range of 4 ± 0.5 °C. In particular, the mass of the concentrate stored in the terminal storage tank 19 that does not exceed the prefabricated quality parameters is not greater than the difference between the set maximum storage volume and the mass of the concentrate that exceeds the prefabricated quality parameters. This design ensures the efficient use of the storage space while maintaining the stability of the product quality.

[0030] In an embodiment of the present invention, as described in claim 6, the device further includes an on-line cleaning unit 27. The on-line cleaning unit 27 determines the location of membrane fouling at the terminal storage tank 19 based on the multi-spectral signals of the terminal storage tank 19, and generates a control signal to close the membrane module at the previous stage before the location of membrane fouling and open the membrane module at the next stage after the location of membrane fouling, so that the milk is concentrated in the membrane module at the previous stage before the location of membrane fouling.

[0031] The on-line cleaning unit 27 uses an innovative spectral analysis algorithm to detect membrane fouling. This unit identifies different types of membrane fouling by analyzing the multi-spectral signals of the concentrate in the terminal storage tank 19, especially the characteristic changes in the 1100 - 1800 nm wavelength band. For example, protein deposition is manifested as enhanced characteristic absorption at 1450 nm and 1650 nm, while mineral scaling is manifested as enhanced characteristic absorption at 1200 nm. When membrane fouling at a specific location is detected, the on-line cleaning unit 27 triggers an intelligent cleaning strategy to achieve targeted cleaning of a specific membrane module through pipeline switching, while ensuring that the rest of the system continues to operate, greatly improving the overall efficiency of the system.

[0032] The device of the present invention further includes a deep reinforcement learning unit. This unit obtains the optimal membrane flux and the optimal membrane operating pressure at each stage based on parameters such as the membrane flux at the first outlet of the primary filtration membrane module 5, the membrane flux at the first outlet of the secondary filtration membrane module 7, the membrane flux at the first outlet of the tertiary filtration membrane module 11, the membrane operating pressure, the viscosity, pH value, water activity, and conductivity of the milk at each stage outlet.

[0033] The deep reinforcement learning unit adopts an advanced Q-learning algorithm to establish a state-action-reward mapping relationship: , wherein: represents the system state vector, which is composed of parameters such as feed flow rate, solute concentration of each stage of concentrate, temperature, etc.; represents the action, that is, the adjustment value of the membrane operating pressure; represents the immediate reward, based on the comprehensive evaluation of energy consumption and product quality; represents the discount factor, with a value range of 0.85 - 0.95, to balance the current reward and long-term benefits; represents the maximum Q value that can be obtained in the next state; represents the next state; represents the optimal action in the next state; represents the exploration term, with an initial value set to 0.3 and gradually decreasing to 0.05 as the learning process progresses.

[0034] Through continuous learning and optimization, this algorithm enables the system to minimize energy consumption while ensuring product quality. In particular, the deep reinforcement learning unit obtains the comprehensive energy consumption index and milk quality at each stage based on the membrane flux and membrane operating pressure of the primary filtration membrane module 5, secondary filtration membrane module 7, and tertiary filtration membrane module 11. The calculation formula for the comprehensive energy consumption index (CEI) is: , wherein: represents the weight coefficient of the i-th stage membrane module, usually set to [0.4, 0.35, 0.25]; represents the operating pressure (MPa) of the i-th stage membrane module; represents the flow rate (L / h) of the i-th stage membrane module; represents the reference flow rate, usually set to 100 L / h.

[0035] The milk quality score (MQS) comprehensively considers protein content, fat content, pH value stability, and microbial safety, and the calculation formula is: , wherein: and respectively represent the actual and target protein contents (%); and respectively represent the actual and target fat contents (%); and respectively represent the actual and target pH values; and respectively represent the actual and reference microbial colony counts (CFU / mL); is the weight coefficient, usually set to [0.35, 0.25, 0.2, 0.2].

[0036] By comprehensively optimizing CEI and MQS, the system achieves the best balance between energy consumption and quality. Experimental data show that compared with traditional control methods, this system can reduce energy consumption by 25 - 35% while improving product quality consistency by 40 - 50%.

[0037] The device of the present invention further includes a virtual simulation unit 34. The virtual simulation unit 34 constructs digital twin systems for the primary concentration circulation path and the secondary concentration circulation path, and obtains the optimal parameters of at least one of the primary filtration membrane module 5, the secondary filtration membrane module 7, and the tertiary filtration membrane module 11 based on the optimized parameters of the digital twin system.

[0038] The virtual simulation unit 34 uses a high-precision computational fluid dynamics (CFD) model and a mass transfer model to construct a digital twin environment highly corresponding to the physical system. This environment can simulate fluid behavior, membrane fouling dynamics, and mass transfer efficiency under different operating conditions, providing a virtual test platform for parameter optimization. In the preferred embodiment of the present invention, the core equations of the virtual simulation model include: 1. Hydrodynamics equation: , 2. Mass transfer equation: , 3. Membrane flux equation: , Where: represents fluid density (kg / m3); represents fluid velocity vector (m / s); represents pressure (Pa); represents dynamic viscosity (Pa•s); represents body force (N / m3); represents solute concentration (mol / m³); represents diffusion coefficient (m2 / s); represents reaction term; represents membrane flux (L / m2 h); represents membrane permeability coefficient; represents transmembrane pressure difference (MPa); represents reflection coefficient; represents osmotic pressure difference (MPa); represents the resistance caused by membrane fouling.

[0039] Based on these models, the virtual simulation unit 34 optimizes the operating parameters of the membrane module through Monte Carlo simulation and sensitivity analysis, including flow rate, pressure, temperature, cleaning cycle, etc. The optimized parameters are transmitted to the physical system to achieve closed-loop control of the combination of virtual and real.

[0040] The device of the present invention further includes an edge computing unit 35. The edge computing unit 35 also constructs digital twin systems for the primary concentration circulation path and the secondary concentration circulation path, and obtains the optimal process parameters based on the optimized parameters of the digital twin system and the parameters of each stage of concentrated liquid, including the position and detection cycle of each sensor.

[0041] The edge computing unit 35 adopts a lightweight neural network and a fuzzy logic controller to achieve real-time processing and decision-making close to the data source. This unit is divided into three-layer structure: data layer, analysis layer and control layer. The data layer is responsible for real-time data acquisition and preprocessing; the analysis layer performs feature extraction, anomaly detection and trend prediction; the control layer generates control instructions according to the analysis results to adjust the system parameters.

[0042] In particular, the edge computing unit 35 obtains the optimal process parameters based on the multi-spectral signals of the concentrated liquid with quality exceeding the pre-set quality parameters, the physical and chemical parameters of the concentrated liquid, the physical and chemical parameters of each stage of concentrated liquid and milk, and the optimized parameters of the digital twin system. The process parameter optimization adopts a multi-objective genetic algorithm, and the objective function is set as: , where: represents the energy consumption index; represents the quality index; represents the time index; is the weight coefficient, usually set as [0.4, 0.4, 0.2].

[0043] The optimization of the sensor position is based on the principle of maximizing information entropy to ensure obtaining the maximum system information with the fewest sensors. The optimization of the detection cycle comprehensively considers data timeliness and computing resource consumption, and is generally set within the range of 10 - 30 seconds per time for key parameters and 60 - 120 seconds per time for non-key parameters.

[0044] In an embodiment of the present invention, examples of the optimal positions and detection cycles of each sensor are as follows: Viscosity sensor: Located at the outlet of each stage of membrane module, detection cycle 15 seconds; pH value sensor: Located at the outlet of each stage of membrane module and the end storage tank, detection cycle 30 seconds; Water activity sensor: Located at the end storage tank, detection cycle 60 seconds; Conductivity sensor: Located at the outlet of each stage of membrane module, detection cycle 20 seconds.

[0045] The present invention also provides a quality control system for a multi-stage membrane integrated intelligent milk concentration and storage device, including a data acquisition module, a data processing module, a deep learning module, a parameter optimization module, and an on-line cleaning module.

[0046] The data acquisition module is used to obtain multi-spectral signals, viscosity, pH value, water activity, and conductivity data of the concentrated solutions at all levels of the multi-stage membrane concentration device. In this embodiment, the data acquisition module adopts a distributed sensing network architecture, including a front-end sensor, a signal conditioning circuit, and a data acquisition unit. The multi-spectral signal acquisition uses a near-infrared spectrometer in the range of 400 - 2500 nm, with a sampling interval of 2 nm and a sampling speed of 30 spectra per minute. The acquisition of physical and chemical parameters uses corresponding professional sensors to form a unified data stream.

[0047] The data processing module is connected to the data acquisition module and is used to perform feature extraction and anomaly detection based on the multi-spectral signals and parameter data. The feature extraction adopts a method combining principal component analysis (PCA) and wavelet transform to extract key features from the original data. The anomaly detection is based on an improved local outlier factor (LOF) algorithm, and the calculation formula is: , Where: represents the set of k nearest neighbors of point p; represents the reachable distance from point p to point o represents the local reachable density of point o; represents the number of k nearest neighbors.

[0048] When the LOF value exceeds the threshold of 1.5, the system determines it as an anomaly and triggers an early warning mechanism.

[0049] The deep learning module is connected to the data processing module and is used to establish a milk quality prediction model and evaluate the quality of the concentrated solution based on the prediction model. This module uses a long short-term memory network (LSTM) to construct a time series prediction model. The network structure includes an input layer, 2 LSTM hidden layers (each with 64 neurons), 1 fully connected layer, and an output layer. The model training uses the mean square error (MSE) as the loss function, and the Adam optimizer is used for weight update. The learning rate is set to 0.001, and the batch size is 32.

[0050] The quality evaluation system comprehensively considers four dimensions: nutritional components, microbial safety, sensory characteristics, and storage stability, and generates a comprehensive quality score (QCS): , Where: represents the nutritional score (0 - 100); represents the microbial safety score (0 - 100); represents the sensory score (0 - 100); Indicates the stability score (0 - 100); is the weight coefficient, usually set to [0.3, 0.3, 0.2, 0.2].

[0051] When the QCS exceeds 85 points, the system determines that the quality of the concentrate is excellent; 75 - 85 points is determined as qualified; below 75 points is determined as unqualified and needs to be processed or used at a lower level.

[0052] The parameter optimization module is connected to the deep learning module and is used to adjust the operating parameters of the multi - stage membrane concentration device based on the quality assessment results. This module uses a method combining model predictive control (MPC) and adaptive neuro - fuzzy inference system (ANFIS) to achieve dynamic optimization of parameters. The core of MPC is to solve the following optimization problem: , , where: represents the control input; represents the system output; represents the reference trajectory; represents the prediction horizon length; represents the control horizon length; and respectively represent the weight matrices of the output error and the control input; and respectively represent the state - transition function and the output function; represents the constraints on the control input and its rate of change.

[0053] The online cleaning module is connected to the parameter optimization module and is used to automatically adjust the working state of the membrane module and execute the cleaning program when membrane fouling is detected. This module identifies different types of membrane fouling based on the membrane flux decay rate and the change in the characteristics of the permeating substances, and selects the corresponding cleaning strategy.

[0054] Membrane fouling detection is based on the following criteria: When the membrane flux drops by more than 15% and the transmembrane pressure difference increases by more than 10%, it is determined as membrane fouling; When the absorption at a specific wavelength (such as 1450 nm) in the multi - spectral signal increases by more than 20%, it is determined as a specific type of fouling (such as protein fouling).

[0055] The cleaning strategy is automatically selected according to the fouling type: For protein fouling: Use an alkaline cleaning solution with pH 10.5, temperature 45°C, and circulate for 15 minutes; For mineral scaling: Use an acidic cleaning solution with pH 2.5, temperature 40°C, and circulate for 20 minutes; For the biofilm: Use a disinfectant solution containing 100 ppm sodium hypochlorite, at a temperature of 25 °C, and circulate and clean for 10 minutes.

[0056] During the cleaning process, the system automatically adjusts the working state of the membrane modules, closes the feed valve of the contaminated membrane module, turns on the cleaning circulation pump, and at the same time keeps other membrane modules running normally, minimizing the impact of cleaning on production.

[0057] In practical applications, the multi-stage membrane integrated intelligent milk concentration and storage device and its quality control system of the present invention can achieve efficient concentration and high-quality storage of milk. Taking the treatment of 100 liters of raw milk as an example, the system can complete the concentration process within 3 hours, produce 30 liters of high-quality concentrated milk, with a protein retention rate of over 95% and a fat retention rate of over 98%. The energy consumption is reduced by over 60% compared to the traditional thermal evaporation method. The intelligent cleaning and parameter optimization functions of the system extend the service life of the membrane modules from the traditional 3 - 6 months to 12 - 18 months, significantly reducing the maintenance cost and downtime.

[0058] Through the quality monitoring system integrating multi-spectral analysis and multi-sensor fusion, the quality consistency of the concentrated milk is significantly improved, and the coefficient of variation between batches is reduced from 8 - 15% in the traditional process to 2 - 5%. The intelligent storage management mechanism based on quality can predict the shelf life of the concentrated milk according to its actual quality characteristics and conduct hierarchical storage, increasing the resource utilization efficiency by over 30%.

[0059] In summary, the multi-stage membrane integrated intelligent milk concentration and storage device and its quality control system of the present invention, through the organic combination of multi-stage membrane technology, multi-spectral analysis, multi-sensor fusion, deep reinforcement learning, and digital twin technology, realize the intelligentization and high efficiency of the whole process of milk concentration and storage, solve the problems of low efficiency, serious membrane fouling, incomplete quality monitoring, difficult parameter optimization, and non-intelligent storage management existing in the prior art, and provide an innovative technical solution for the dairy processing field.

[0060] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. Multi-stage membrane integrated intelligent milk concentration and storage device, characterized in that, Comprising: A milk storage tank; A primary filtration membrane module, whose outlet is connected to the milk storage tank to form a primary concentration circulation path; A secondary filtration membrane module, which is connected to the primary filtration membrane module to form a secondary concentration circulation path; A tertiary filtration membrane module, which is connected to the secondary filtration membrane module; A multispectral analyzer for acquiring multispectral signals of concentrated solutions at all levels; Multiple sensors for acquiring the viscosity, pH value, water activity and conductivity of concentrated solutions at all levels; A storage unit for obtaining the quality of the concentrated solution based on the multispectral signal and the sensor signal, and adjusting the parameters of the filtration membrane module.

2. The multi-stage membrane integrated intelligent milk concentration and storage device according to claim 1, wherein The primary filtration membrane module is a microporous filter with a filtration accuracy of 0.2 μm.

3. The multi-stage membrane integrated intelligent milk concentration and storage device according to claim 1, characterized in that The secondary filtration membrane module is a membrane stack with a pore size of 100 nm.

4. The multi-stage membrane integrated intelligent milk concentration and storage device according to claim 1, wherein The tertiary filtration membrane module is a membrane stack with a pore size of 50 nm.

5. The multi-stage membrane integrated intelligent milk concentration and storage device according to claim 1, characterized in that, Further comprising: A terminal storage tank for storing the tertiary concentrated solution output by the tertiary filtration membrane module, wherein the mass of the concentrated solution stored in the terminal storage tank with a quality not exceeding the prefabricated quality parameter is not greater than the difference between the set maximum storage volume and the mass of the concentrated solution with a quality exceeding the prefabricated quality parameter.

6. The multi-stage membrane integrated intelligent milk concentration and storage device according to claim 1, characterized in that, Further comprising: An on-line cleaning unit for determining the position of membrane fouling based on the multispectral signal of the terminal storage tank, and generating a control signal to close the membrane module at the previous level before the position of membrane fouling and open the membrane module at the next level after the position of membrane fouling, so that the milk is concentrated in the membrane module at the previous level before the position of membrane fouling.

7. The multi-stage membrane integrated intelligent milk concentration and storage device according to claim 1, characterized in that Further comprising: A deep reinforcement learning unit for obtaining the optimal membrane flux and optimal membrane operating pressure at each stage based on the membrane flux, membrane operating pressure of each stage of the filtration membrane module and the parameters of the concentrated solutions at all levels, wherein the deep reinforcement learning unit obtains the comprehensive energy consumption index and milk quality at each stage based on the membrane flux and the membrane operating pressure.

8. The multi-stage membrane integrated intelligent milk concentration and storage device according to claim 1, characterized in that, Further comprising: A virtual simulation unit for constructing digital twin systems of the primary concentration circulation path and the secondary concentration circulation path, and obtaining the optimal parameters of at least one filtration membrane module based on the optimized parameters of the digital twin systems.

9. The multi-stage membrane integrated intelligent milk concentration and storage device according to claim 1, characterized in that Further comprising: An edge computing unit for constructing digital twin systems of the primary concentration circulation path and the secondary concentration circulation path, and obtaining the optimal process parameters based on the optimized parameters of the digital twin systems and the parameters of the concentrated solutions at all levels, wherein the process parameters include the positions and detection periods of each sensor.

10. A quality control system for a multi-stage membrane integrated intelligent milk concentration and storage device according to any one of claims 1-9, comprising: A data acquisition module for acquiring multispectral signals, viscosity, pH value, water activity and conductivity data of concentrated solutions at all levels of a multi-stage membrane concentration device; A data processing module connected to the data acquisition module for performing feature extraction and anomaly detection based on the multispectral signal and the parameter data; A deep learning module connected to the data processing module for establishing a milk quality prediction model and evaluating the quality of the concentrated solution based on the prediction model; A parameter optimization module connected to the deep learning module for adjusting the operating parameters of the multi-stage membrane concentration device based on the quality evaluation result; An online cleaning module, connected to the parameter optimization module, is used to automatically adjust the working state of the membrane module and execute a cleaning program when membrane fouling is detected.

Images