Protein beverage production process regulation and control method and system based on digital twinning

Abstract

The invention discloses a protein beverage production process regulation and control method and system based on digital twinning, belongs to the technical field of food industry intelligent manufacturing, and aims at solving the problems that an existing protein beverage production quality detection method has hysteresis quality, and the production cost is low due to the limitation of a production environment. And real-time and even advanced risk early warning and multi-parameter adjustment cannot be realized in the production process of each working procedure, so that the improvement of the production efficiency and quality of the protein beverage is hindered. The method comprises the following steps: collecting protein beverage production data through a multi-element online monitoring unit, and performing feature extraction to obtain protein beverage quality features; analyzing the quality characteristics of the protein beverage in real time through a deep learning model, and outputting multiple key quality index estimated values; performing risk simulation based on the key quality index estimated value, and determining a risk quality index; and based on the risk quality index, performing analog calculation in the digital twin model, and determining an optimal parameter combination capable of avoiding the risk.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing technology in the food industry, and in particular to a method and system for controlling the production process of protein beverages based on digital twins. Background Technology

[0002] Current protein beverage production (such as soy milk, almond milk, and protein nutrient solutions) typically employs continuous production lines using ultra-high temperature instantaneous sterilization (UHT). The quality of the final product highly depends on the precise control of a series of physicochemical parameters in several key processes (such as blending, homogenization, sterilization, and filling), including: pH value, conductivity (indirectly reflecting ion concentration), temperature, processing time, protein content, and fat particle size.

[0003] However, several technical shortcomings exist in the quality inspection process: First, there is a lag in detection: the detection of key quality indicators, such as the degree of protein denaturation, fat globule size distribution, and microbial indicators, largely relies on offline sampling and testing, resulting in a large number of defective products by the time the results are returned. Second, there are environmental challenges: the high temperature, high humidity, and high conductivity of the production line environment place extremely high demands on the corrosion resistance, stability, and anti-interference capabilities of sensors, making it difficult for traditional single sensors to operate reliably. Third, there is isolated control: while existing PLC / DCS systems can control single parameters, such as temperature, they lack a comprehensive understanding and synergistic optimization of the coupling effects of multiple parameters. For example, controlling only the sterilization temperature while ignoring the flow rate of the material in the pipeline (i.e., the heating time) and pH value may still lead to incomplete sterilization or excessive protein denaturation. Fourth, there is poor risk predictability: it is impossible to provide early warning and intervention for potential quality risks such as microbial contamination outbreaks, protein flocculation, and fat floating during the production process.

[0004] In summary, existing quality testing methods for protein beverage production are outdated and, due to limitations in the production environment, cannot provide real-time or even advance risk warnings and multi-parameter adjustments during each production process, which seriously hinders the improvement of efficiency and quality in protein beverage production. Summary of the Invention

[0005] This invention provides a method and system for controlling the production process of protein beverages based on digital twins, which addresses the following technical problems: existing methods for detecting the quality of protein beverage production are lagging behind, and due to limitations in the production environment, they cannot achieve real-time or even early risk warning and multi-parameter adjustment in each production process, thus hindering the improvement of efficiency and quality in protein beverage production.

[0006] The embodiments of the present invention adopt the following technical solutions: On one hand, embodiments of the present invention provide a method for controlling the production process of protein beverages based on digital twins. The method includes: constructing a multi-element online monitoring unit in the protein beverage production line, including a pH measurement module, a conductivity measurement module, a component and particle size measurement module, and an auxiliary sensor module. The protein beverage production data is collected by the multi-dimensional online monitoring unit, and feature extraction is performed to obtain the quality characteristics of the protein beverage. Build and train deep learning models, and build a digital twin model of a protein beverage production line; The deep learning model is used to analyze the quality characteristics of the protein beverage in real time and output estimated values ​​of multiple key quality indicators. Based on the estimated values ​​of the key quality indicators, the protein beverage production data is input into the digital twin model for risk simulation to determine the risk quality indicators. Based on the aforementioned risk quality indicators, simulation calculations are performed in the digital twin model to determine the optimal parameter combination that can mitigate risks. The optimal parameter combination is sent to the PLC controller to control the protein beverage production line to adjust the production parameters in real time.

[0007] In one feasible implementation, a multi-source online monitoring unit is constructed in the protein beverage production line, specifically including: In the protein colloidal solution preparation process, a pH test electrode and a micro air pump with automatic pressure cleaning and online calibration functions are set up to form the pH measurement module; the pH test electrode is an antimony electrode or a glass composite electrode, and the electrode sheath is designed to be quick-removable; the micro air pump stores sterile air or CIP cleaning solution and sprays it onto the pH test electrode periodically to prevent protein and fat from adhering and contaminating the electrode membrane. In production processes where conductivity control is required, a four-electrode conductivity sensor is installed to form the conductivity measurement module. An industrial-grade near-infrared spectral probe and a light flow cell are installed on the outside of the equipment in the protein beverage preparation process to form the component and particle size measurement module. The near-infrared spectral probe is used to obtain various component indicators in the semi-finished liquid. At the same time, by analyzing the spectral scattering information, the particle size information can be indirectly deduced. In the heating process, a high-precision temperature sensor and an electromagnetic flow meter are set up to form the auxiliary sensor module, which is used to monitor the instantaneous temperature and flow rate of the material and calculate the heating time of the material.

[0008] In one feasible implementation, the protein beverage production data is collected by the multi-source online monitoring unit, and feature extraction is performed to obtain the quality characteristics of the protein beverage, specifically including: In the protein beverage production process, the real-time pH value of the protein colloidal solution is collected by the pH measurement module, the real-time conductivity value of the corresponding step is obtained by the conductivity measurement module, the near-infrared spectral information of the semi-finished solution is obtained by the composition and particle size measurement module, and the instantaneous temperature and flow rate of the protein solution during the heating process are obtained by the auxiliary sensor module, thus obtaining multi-source sensor data. The multi-source sensor data is filtered and noise-reduced, and timestamps are aligned to obtain the protein beverage production data. The protein beverage production data sequence is extracted based on a preset time window, and the quality characteristics of the protein beverage in the protein beverage production data sequence are extracted; wherein, the quality characteristics of the protein beverage include at least: solution pH, solution ion concentration, content of key components in the solution, heating temperature and heating time.

[0009] In one feasible implementation, constructing and training a deep learning model specifically includes: The deep learning model was trained using quality test indicators obtained from offline testing of protein beverages as labels and quality features of protein beverages collected simultaneously as training features. The deep learning model is deployed on an edge computing gateway to perform real-time analysis of online protein beverage production data, outputting estimated values ​​of multiple key quality indicators per second.

[0010] In one feasible implementation, constructing a digital twin model of a protein beverage production line specifically includes: A mechanistic model of key processes in a protein beverage production line was established based on the principles of reaction kinetics. Based on historical production and quality inspection data, a deep learning LSTM network is trained to learn the nonlinear mapping relationship between complex multi-parameter inputs and the final output of the system. Based on the use of CAD drawings and 3D scanning technology, a three-dimensional visualization model of each production equipment in the key processes of the protein beverage production line was constructed. The mechanistic model, the three-dimensional visualization model, and the nonlinear mapping relationship are combined to form the digital twin model; the digital twin model maintains data synchronization with the protein beverage production line through the OPC UA protocol.

[0011] In one feasible implementation, the deep learning model is used to analyze the quality characteristics of the protein beverage in real time and output estimated values ​​of multiple key quality indicators, specifically including: During the operation of the protein beverage production line, the acquired quality characteristics of the protein beverage are input into the deep learning model for analysis, and output results are obtained. The output results include estimated values ​​of multiple key quality indicators. Among them, the estimated values ​​of key quality indicators include at least: the solution protein content, the content of other nutrients, particle size, and bacterial content in each key process.

[0012] In one feasible implementation, based on the estimated values ​​of the key quality indicators, the protein beverage production data is input into the digital twin model for risk simulation to determine the risk quality indicators, specifically including: The estimated value of each key quality indicator is compared with the corresponding standard threshold to screen out the key quality indicators that do not meet the standards, and the non-compliant production process corresponding to the non-compliant key quality indicators is determined. The digital twin model is controlled to continuously simulate operation using the current real-time operating data of the protein beverage production line as the initial conditions; The protein beverage production data corresponding to the substandard production process is input into the digital twin model to extrapolate the quality indicators of the solution in each substandard production process within a future preset time period; wherein, the quality indicators include any one or more of the following: predicted sterilization intensity, predicted particle size, and predicted protein content; If at least one of the quality indicators fails to meet the standard, the failed quality indicator shall be identified as a risk quality indicator.

[0013] In one feasible implementation, based on the risk quality index, simulation calculations are performed in the digital twin model to determine the optimal parameter combination that can mitigate the risk, specifically including: Identify the production processes to be optimized corresponding to the aforementioned risk quality indicators; An optimizer based on Bayesian optimization or genetic algorithm is launched, with the optimization of the risk quality index as the objective function and the adjustable parameters of the production process to be optimized as variables. Several rapid simulation calculations are performed in the digital twin model to screen out the optimal parameter combination that can avoid risks.

[0014] In one feasible implementation, the optimal parameter combination is sent to a PLC controller to control the protein beverage production line to adjust production parameters in real time, specifically including: The optimal parameter combination is sent as an instruction to the PLC controller for the protein beverage production process. The PLC controller drives the corresponding actuator to complete parameter adjustments before a risk occurs. After the parameters were adjusted, the protein beverage production data for a period of time was captured again by the multivariate online monitoring unit and fed back to the digital twin model to verify the control effect and prepare for the next decision, thus forming a closed loop.

[0015] On the other hand, embodiments of the present invention also provide a protein beverage production process control system based on digital twins, the system comprising: The data feature extraction module is used to construct a multi-dimensional online monitoring unit in the protein beverage production line, including a pH measurement module, a conductivity measurement module, a component and particle size measurement module, and an auxiliary sensor module; the protein beverage production data is collected through the multi-dimensional online monitoring unit, and feature extraction is performed to obtain the quality characteristics of the protein beverage; The model building module is used to build and train deep learning models, as well as to build digital twin models of protein beverage production lines. The quality indicator prediction module is used to analyze the quality characteristics of the protein beverage in real time through the deep learning model and output estimated values ​​of multiple key quality indicators; based on the estimated values ​​of key quality indicators, the production data of the protein beverage is input into the digital twin model for risk simulation to determine risk quality indicators. The production parameter adjustment module is used to perform simulation calculations in the digital twin model based on the risk quality indicators to determine the optimal parameter combination that can avoid risks; and send the optimal parameter combination to the PLC controller to control the protein beverage production line to adjust the production parameters in real time.

[0016] Compared with the prior art, the protein beverage production process control method and system based on digital twin provided by the embodiments of the present invention has the following beneficial effects: 1. Real-time and forward-looking: The quality inspection process of protein beverage production is transformed from offline and delayed detection to online and real-time detection, and the control of production parameters is transformed from post-event and passive to pre-event and proactive, which significantly reduces the number of defective products and improves the operating efficiency of the production line.

[0017] 2. Multi-indicator synergy: It overcomes the limitations of single-parameter control and achieves global optimal decision-making based on the fusion of multiple indicators.

[0018] 3. High reliability: The integrated sensor and automatic cleaning function, designed for harsh environments, ensure the stability and accuracy of long-term measurements.

[0019] 4. Transparency and Explainability: Digital twin technology provides unprecedented visualization and insight into the production process, facilitating fault diagnosis and process optimization.

[0020] 5. Adaptive and self-optimizing: The system can continuously learn new production data and optimize its internal prediction and decision-making models, enabling the production line to continuously evolve. It no longer needs to rely on manual quality inspection and sampling inspection, thus improving the accuracy and comprehensiveness of protein beverage quality testing. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings: Figure 1 A flowchart of a protein beverage production process control method based on digital twins provided for embodiments of the present invention; Figure 2 This is a schematic diagram of a protein beverage production process control system based on digital twins, provided as an embodiment of the present invention. Detailed Implementation

[0022] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this invention.

[0023] This invention provides a method for controlling the production process of protein beverages based on digital twins, such as... Figure 1 As shown, the method for controlling the production process of protein beverages based on digital twins specifically includes steps S101-S106: S101. Construct a multi-dimensional online monitoring unit in the protein beverage production line, including a pH measurement module, a conductivity measurement module, a component and particle size measurement module, and an auxiliary sensor module.

[0024] Specifically, in the protein colloidal solution preparation process, a pH testing electrode and a micro air pump with automatic pressure cleaning and online calibration functions are set up to form a pH measurement module; the pH testing electrode is an antimony electrode or a glass composite electrode, and the electrode sheath is designed to be quick-removable; the micro air pump stores sterile air or CIP cleaning solution and sprays it onto the pH testing electrode periodically to prevent protein and fat from adhering and contaminating the electrode membrane.

[0025] In production processes requiring conductivity control, a four-electrode conductivity sensor is installed to form a conductivity measurement module. In the protein beverage preparation process, an industrial-grade near-infrared spectral probe and a flow cell are installed externally on the equipment to form a component and particle size measurement module. The near-infrared spectral probe is used to acquire various component indicators in the semi-finished liquid, and by analyzing spectral scattering information, particle size information can be indirectly deduced.

[0026] In the heating process, a high-precision temperature sensor and an electromagnetic flow meter are set up to form an auxiliary sensor module, which is used to monitor the instantaneous temperature and flow rate of the material and calculate the heating time of the material.

[0027] In one embodiment, in a peanut protein beverage production plant, a pH measurement module is installed on the outlet pipe of a gelling tank (used to prepare protein colloidal solutions). The module uses corrosion-resistant antimony electrodes, and its sheath is connected to the pipe via quick-release clamps, allowing for rapid removal for deep cleaning and calibration after each shift. A miniature air pump integrated into the sensor unit automatically sprays high-pressure sterile air every 15 minutes, effectively preventing the accumulation of peanut oil and protein colloids on the electrode membrane, ensuring the long-term stability and accuracy of pH measurements throughout the production batch.

[0028] In another embodiment, an industrial-grade online near-infrared (NIR) spectral probe is installed on the premixing tank (blending stage). The probe is mounted via a flange on a flow cell with a sapphire window, through which the premixed solution continuously flows. The NIR spectrometer acquires spectral data every 2 seconds. Using a pre-established PLS-R model, the system analyzes in real time the current liquid's fat content as 3.5%, protein content as 3.2%, and lactose content as 4.7%. Simultaneously, by analyzing the scattering characteristics of the spectrum in the short-wave near-infrared region, the model indirectly calculates the average particle size of the fat globules to be 0.8 μm and predicts that the product has excellent stability and is not prone to stratification.

[0029] S102. Collect protein beverage production data through a multi-dimensional online monitoring unit and extract features to obtain the quality characteristics of the protein beverage.

[0030] Specifically, in the protein beverage production process, the real-time pH value of the protein colloidal solution is collected through the pH measurement module, the real-time conductivity value of the corresponding stage is obtained through the conductivity measurement module, the near-infrared spectral information of the semi-finished solution is obtained through the composition and particle size measurement module, and the instantaneous temperature and flow rate of the protein solution during the heating process are obtained through the auxiliary sensor module, thus obtaining multi-source sensor data.

[0031] Furthermore, the multi-source sensor data is filtered and noise-reduced, and timestamps are aligned to obtain protein beverage production data.

[0032] Furthermore, protein beverage production data sequences are extracted based on a preset time window, and protein beverage quality characteristics are extracted from the protein beverage production data sequences; among which, protein beverage quality characteristics include at least: solution pH, solution ion concentration, content of key components in the solution, heating temperature and heating time.

[0033] In one embodiment, in the UHT sterilization section of a soymilk production line, the system simultaneously collects pH values ​​(6.8) from a pH electrode, conductivity values ​​(4.2 mS / cm) from a four-electrode conductivity sensor, full-spectrum data (1100-2300 nm) from an NIR spectrometer, temperature (138.5°C) from a PT100 temperature sensor, and flow rate (5000 L / h) from an electromagnetic flowmeter. The edge computing gateway first performs a moving average filter on this raw data to eliminate random noise and then assigns a unified timestamp to all data. The system extracts the data sequence within a 30-second sliding time window and extracts the average pH, maximum conductivity, principal component score of the NIR spectrum, average temperature, and total heating time of the material in the insulated pipe (calculated using flow rate and pipe volume) for that time period. These extracted features collectively constitute the quality feature vector of the protein beverage at that moment, which is then fed into a downstream deep learning model for analysis.

[0034] S103, Build and train a deep learning model, and build a digital twin model of a protein beverage production line.

[0035] Specifically, a deep learning model is trained using quality testing indicators obtained from offline protein beverage testing as labels and synchronously collected protein beverage quality features as training features. This deep learning model is then deployed on an edge computing gateway to perform real-time analysis of online protein beverage production data, outputting estimated values ​​for multiple key quality indicators per second.

[0036] In one embodiment, during the trial phase of the production line, manual sampling is performed simultaneously with data acquisition from online sensors every 5 minutes. The laboratory uses the Kjeldahl method to measure the actual protein content and a laser particle size analyzer to measure the actual particle size, recording these values ​​as labels. After collecting over 2000 valid data pairs, a high-precision prediction model is trained using the multi-source data from the sensors (features X) and laboratory measurements (labels Y) as the training set, employing the Gradient Boosting Decision Tree (GBDT) algorithm. This model is deployed on the edge computing gateway of the production line, enabling real-time reading of sensor data during production and outputting estimated protein content and particle size values ​​every second, which highly match the offline testing results.

[0037] Furthermore, a mechanistic model of the key processes in the protein beverage production line is established based on reaction kinetics principles. A deep learning LSTM network is trained using historical production and quality inspection data to learn the nonlinear mapping relationship between complex multi-parameter inputs and the system's final output. Three-dimensional visualization models of each production device in the key processes of the protein beverage production line are constructed using CAD drawings and 3D scanning technology. The mechanistic model, the three-dimensional visualization model, and the nonlinear mapping relationship are combined to form a digital twin model; this digital twin model maintains data synchronization with the protein beverage production line via the OPC UA protocol.

[0038] In one embodiment, to construct a digital twin model of the homogenization process, a mechanistic model of the homogenizer is first established based on fluid mechanics and crushing dynamics. This model describes the mathematical relationship between homogenization pressure, temperature, and outlet fat globule size. Secondly, the operating parameters (pressure, temperature, flow rate) and corresponding offline particle size inspection reports of the homogenizer over the past year are collected to train an LSTM network to learn the nonlinear changes in homogenization effect under specific material characteristics. Finally, a visualized digital twin of the homogenizer is constructed in the Unity 3D engine using its CAD 3D model. These three parts are connected to the real homogenizer control system via the OPCUA protocol, achieving virtual-real synchronization. The operator can view the pressure cloud map and fluid motion animation inside the virtual homogenizer on a computer and see the predicted outlet particle size in real time.

[0039] S104. Through a deep learning model, the quality characteristics of protein beverages are analyzed in real time, and estimated values ​​of multiple key quality indicators are output. Based on the estimated values ​​of key quality indicators, the production data of protein beverages are input into a digital twin model for risk simulation to determine risk quality indicators.

[0040] Specifically, during the operation of the protein beverage production line, the acquired quality characteristics of the protein beverage are input into a deep learning model for analysis, and the output results are obtained. The output results include estimated values ​​of multiple key quality indicators. Among them, the estimated values ​​of key quality indicators include at least: the solution protein content, the content of other nutrients, particle size, and bacterial content in each key process.

[0041] Furthermore, the estimated value of each key quality indicator is compared with the corresponding standard threshold to screen out the key quality indicators that do not meet the standards, and to determine the non-compliant production processes corresponding to the non-compliant key quality indicators.

[0042] Furthermore, the digital twin model continuously simulates operation using the current real-time operating data of the protein beverage production line as initial conditions. The protein beverage production data corresponding to the substandard production processes are input into the digital twin model to extrapolate the quality indicators of the solutions in each substandard production process within a preset future time period. These quality indicators include one or more of the following: predicted sterilization intensity, predicted particle size, and predicted protein content.

[0043] Furthermore, if at least one of the quality indicators fails to meet the standard, the failed quality indicator shall be identified as a risk quality indicator.

[0044] In one embodiment, during the pasteurization process of walnut milk, a deep learning model receives real-time quality characteristics from the production line (pH=6.9, conductivity=3.8 mS / cm, NIR spectral characteristics, T=72°C). The model instantly completes calculations and outputs estimated values ​​for several key quality indicators: protein content 3.0%, fat content 4.5%, estimated coliform count <10 CFU / mL, and D90 particle size 18μm. These estimated values ​​are immediately displayed on the central control screen for operator monitoring and simultaneously sent to the digital twin system for risk assessment.

[0045] S105. Based on risk quality indicators, perform simulation calculations in a digital twin model to determine the optimal parameter combination that can mitigate risks.

[0046] Specifically, the production processes to be optimized corresponding to the risk quality indicators are identified. An optimizer based on Bayesian optimization or genetic algorithms is launched, with the optimization of the risk quality indicators as the objective function and the adjustable parameters of the production processes to be optimized as variables. Several rapid simulation calculations are performed in the digital twin model to screen out the optimal parameter combination that can avoid risks.

[0047] In one embodiment, the system detected an estimated bacterial count of 50 CFU / mL at the UHT process outlet, exceeding the company's internal control standard (<10 CFU / mL). The digital twin system immediately identified this sterilization process as substandard. Using all current production data (including a high initial colony count, current temperature of 139°C, and flow rate of 5100 L / h) as initial conditions, the system projected the production status for the next 3 minutes. Simulation calculations showed that, based on the current parameters, the final product's F-value (sterilization intensity value) would only be 11.5 (below the standard of 12.0), failing to meet the predicted sterilization intensity. Therefore, the system identified "sterilization intensity (F-value)" as the most pressing risk quality indicator to address.

[0048] In a further embodiment, continuing the previous example, the process to be optimized corresponding to the risk indicator "sterilization intensity" is the UHT sterilization process. The system activates its built-in Bayesian optimizer, using "maximizing the F-value" as the objective function, and the steam regulating valve opening (controlling temperature) and feed pump frequency (controlling flow rate) as optimization variables. The digital twin model performs thousands of rapid simulation calculations within seconds, evaluating the predicted F-values ​​under different parameter combinations. The optimal parameter combination finally selected is: increasing the temperature from 139°C to 140.2°C, while slightly reducing the flow rate from 5100 L / h to 5050 L / h. This combination achieves a predicted F-value of 12.3 while minimizing production efficiency loss, effectively avoiding the risk of insufficient sterilization.

[0049] S106. Send the optimal parameter combination to the PLC controller to control the protein beverage production line to adjust the production parameters in real time.

[0050] Specifically, the optimal parameter combination is sent as an instruction to the PLC controller used in protein beverage production. The PLC controller then drives the corresponding actuator to adjust the parameters before any risk occurs.

[0051] Furthermore, the protein beverage production data over a period of time after parameter adjustment is captured again by the multivariate online monitoring unit and fed back to the digital twin model to verify the control effect and prepare for the next decision, forming a closed loop.

[0052] In one embodiment, the digital twin system sends the optimal parameter combination (temperature setpoint: 140.2°C, flow rate setpoint: 5050 L / h) to the production line's PLC via the OPC UA protocol as control commands. Upon receiving the commands, the PLC immediately takes action: the analog output module adjusts the opening of the steam regulating valve, and the frequency converter controls the feed pump to reduce its speed. The actuator completes the parameter adjustment within 30 seconds. After adjustment, the online sensors monitor in real time that the temperature has stabilized at 140.2°C and the flow rate has stabilized at 5050 L / h. Two minutes later, the system re-analyzes the estimated bacterial content at the UHT outlet, which has decreased to <5 CFU / mL. This successful result is fed back to the digital twin system, verifying the correctness of the control decision and used to optimize the next decision model, forming a perfect adaptive closed-loop control.

[0053] It should be noted that the specific data mentioned in the above embodiments are merely illustrative examples and are not intended to limit the implementation parameters of the present invention. The specific implementation parameters need to be adjusted and set in actual use.

[0054] In addition, embodiments of the present invention also provide a protein beverage production process control system based on digital twins, such as... Figure 2As shown, the protein beverage production process control system 200 based on digital twins specifically includes: The data feature extraction module 210 is used to construct a multi-dimensional online monitoring unit in the protein beverage production line, including a pH measurement module, a conductivity measurement module, a component and particle size measurement module, and an auxiliary sensor module; the protein beverage production data is collected through the multi-dimensional online monitoring unit, and feature extraction is performed to obtain the quality characteristics of the protein beverage; Model building module 220 is used to build and train deep learning models, as well as to build a digital twin model of a protein beverage production line; The quality indicator prediction module 230 is used to analyze the quality characteristics of the protein beverage in real time through the deep learning model and output estimated values ​​of multiple key quality indicators; based on the estimated values ​​of key quality indicators, the production data of the protein beverage is input into the digital twin model for risk simulation to determine risk quality indicators. The production parameter adjustment module 240 is used to perform simulation calculations in the digital twin model based on the risk quality indicators to determine the optimal parameter combination that can avoid risks; and send the optimal parameter combination to the PLC controller to control the protein beverage production line to adjust the production parameters in real time.

[0055] The various embodiments in this invention are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments of apparatus, devices, and non-volatile computer storage media are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0056] The foregoing has described specific embodiments of the present invention. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0057] The above description is merely an embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, various modifications and variations can be made to the embodiments of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the embodiments of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for controlling the production process of protein beverages based on digital twins, characterized in that, The method includes: A multi-dimensional online monitoring unit is constructed in the protein beverage production line, including a pH measurement module, a conductivity measurement module, a component and particle size measurement module, and an auxiliary sensor module; The protein beverage production data is collected by the multi-dimensional online monitoring unit, and feature extraction is performed to obtain the quality characteristics of the protein beverage. Build and train deep learning models, and build a digital twin model of a protein beverage production line; The deep learning model is used to analyze the quality characteristics of the protein beverage in real time and output estimated values ​​of multiple key quality indicators. Based on the estimated values ​​of the key quality indicators, the protein beverage production data is input into the digital twin model for risk simulation to determine the risk quality indicators. Based on the aforementioned risk quality indicators, simulation calculations are performed in the digital twin model to determine the optimal parameter combination that can mitigate risks. The optimal parameter combination is sent to the PLC controller to control the protein beverage production line to adjust the production parameters in real time.

2. The method for controlling the production process of protein beverages based on digital twins according to claim 1, characterized in that, Constructing a multi-dimensional online monitoring unit in the protein beverage production line, specifically including: In the protein colloidal solution preparation process, a pH test electrode and a micro air pump with automatic pressure cleaning and online calibration functions are set up to form the pH measurement module; the pH test electrode is an antimony electrode or a glass composite electrode, and the electrode sheath is designed to be quick-removable; the micro air pump stores sterile air or CIP cleaning solution and sprays it onto the pH test electrode periodically to prevent protein and fat from adhering and contaminating the electrode membrane. In production processes where conductivity control is required, a four-electrode conductivity sensor is installed to form the conductivity measurement module. An industrial-grade near-infrared spectral probe and a light flow cell are installed on the outside of the equipment in the protein beverage preparation process to form the component and particle size measurement module. The near-infrared spectral probe is used to obtain various component indicators in the semi-finished liquid. At the same time, by analyzing the spectral scattering information, the particle size information can be indirectly deduced. In the heating process, a high-precision temperature sensor and an electromagnetic flow meter are set up to form the auxiliary sensor module, which is used to monitor the instantaneous temperature and flow rate of the material and calculate the heating time of the material.

3. The method for controlling the production process of protein beverages based on digital twins according to claim 1, characterized in that, The protein beverage production data is collected by the multi-dimensional online monitoring unit, and feature extraction is performed to obtain the quality characteristics of the protein beverage, specifically including: In the protein beverage production process, the real-time pH value of the protein colloidal solution is collected by the pH measurement module, the real-time conductivity value of the corresponding step is obtained by the conductivity measurement module, the near-infrared spectral information of the semi-finished solution is obtained by the composition and particle size measurement module, and the instantaneous temperature and flow rate of the protein solution during the heating process are obtained by the auxiliary sensor module, thus obtaining multi-source sensor data. The multi-source sensor data is filtered and noise-reduced, and timestamps are aligned to obtain the protein beverage production data. The protein beverage production data sequence is extracted based on a preset time window, and the quality characteristics of the protein beverage in the protein beverage production data sequence are extracted; wherein, the quality characteristics of the protein beverage include at least: solution pH, solution ion concentration, content of key components in the solution, heating temperature and heating time.

4. The method for controlling the production process of protein beverages based on digital twins according to claim 1, characterized in that, Building and training deep learning models specifically includes: The deep learning model was trained using quality test indicators obtained from offline testing of protein beverages as labels and quality features of protein beverages collected simultaneously as training features. The deep learning model is deployed on an edge computing gateway to perform real-time analysis of online protein beverage production data, outputting estimated values ​​of multiple key quality indicators per second.

5. The method for controlling the production process of protein beverages based on digital twins according to claim 1, characterized in that, Constructing a digital twin model of a protein beverage production line, specifically including: A mechanistic model of key processes in a protein beverage production line was established based on the principles of reaction kinetics. Based on historical production and quality inspection data, a deep learning LSTM network is trained to learn the nonlinear mapping relationship between complex multi-parameter inputs and the final output of the system. Based on the use of CAD drawings and 3D scanning technology, a three-dimensional visualization model of each production equipment in the key processes of the protein beverage production line was constructed. The mechanistic model, the three-dimensional visualization model, and the nonlinear mapping relationship are combined to form the digital twin model; the digital twin model maintains data synchronization with the protein beverage production line through the OPC UA protocol.

6. The method for controlling the production process of protein beverages based on digital twins according to claim 1, characterized in that, The deep learning model is used to analyze the quality characteristics of the protein beverage in real time, and outputs estimated values ​​of several key quality indicators, including: During the operation of the protein beverage production line, the acquired quality characteristics of the protein beverage are input into the deep learning model for analysis, and output results are obtained. The output results include estimated values ​​of multiple key quality indicators. Among them, the estimated values ​​of key quality indicators include at least: the solution protein content, the content of other nutrients, particle size, and bacterial content in each key process.

7. The method for controlling the production process of protein beverages based on digital twins according to claim 1, characterized in that, Based on the estimated values ​​of the key quality indicators, the protein beverage production data is input into the digital twin model for risk simulation to determine the risk quality indicators, specifically including: The estimated value of each key quality indicator is compared with the corresponding standard threshold to screen out the key quality indicators that do not meet the standards, and the non-compliant production process corresponding to the non-compliant key quality indicators is determined. The digital twin model is controlled to continuously simulate operation using the current real-time operating data of the protein beverage production line as the initial conditions; The protein beverage production data corresponding to the substandard production process is input into the digital twin model to extrapolate the quality indicators of the solution in each substandard production process within a future preset time period; wherein, the quality indicators include any one or more of the following: predicted sterilization intensity, predicted particle size, and predicted protein content; If at least one of the quality indicators fails to meet the standard, the failed quality indicator shall be identified as a risk quality indicator.

8. The method for controlling the production process of protein beverages based on digital twins according to claim 1, characterized in that, Based on the aforementioned risk quality indicators, simulation calculations are performed in the digital twin model to determine the optimal parameter combination that can mitigate risks, specifically including: Identify the production processes to be optimized corresponding to the aforementioned risk quality indicators; An optimizer based on Bayesian optimization or genetic algorithm is launched, with the optimization of the risk quality index as the objective function and the adjustable parameters of the production process to be optimized as variables. Several rapid simulation calculations are performed in the digital twin model to screen out the optimal parameter combination that can avoid risks.

9. The method for controlling the production process of protein beverages based on digital twins according to claim 1, characterized in that, The optimal parameter combination is sent to the PLC controller to control the protein beverage production line to adjust production parameters in real time, specifically including: The optimal parameter combination is sent as an instruction to the PLC controller for the protein beverage production process. The PLC controller drives the corresponding actuator to complete parameter adjustments before a risk occurs. After the parameters were adjusted, the protein beverage production data for a period of time was captured again by the multivariate online monitoring unit and fed back to the digital twin model to verify the control effect and prepare for the next decision, forming a closed loop.

10. A protein beverage production process control system based on digital twins, characterized in that, The system includes: The data feature extraction module is used to construct a multi-dimensional online monitoring unit in the protein beverage production line, including a pH measurement module, a conductivity measurement module, a component and particle size measurement module, and an auxiliary sensor module; the protein beverage production data is collected through the multi-dimensional online monitoring unit, and feature extraction is performed to obtain the quality characteristics of the protein beverage; The model building module is used to build and train deep learning models, as well as to build digital twin models of protein beverage production lines. The quality indicator prediction module is used to analyze the quality characteristics of the protein beverage in real time through the deep learning model and output estimated values ​​of multiple key quality indicators; based on the estimated values ​​of key quality indicators, the production data of the protein beverage is input into the digital twin model for risk simulation to determine risk quality indicators. The production parameter adjustment module is used to perform simulation calculations in the digital twin model based on the risk quality indicators to determine the optimal parameter combination that can avoid risks; and send the optimal parameter combination to the PLC controller to control the protein beverage production line to adjust the production parameters in real time.

Images