Foam monitoring and foam elimination and inhibition system in fermentation product inactivation process based on machine vision

Abstract

The invention discloses a fermentation product inactivation process foam monitoring and foam elimination and inhibition system based on machine vision, and relates to the technical field of industrial process control. The invention discloses a low-salt fermentation product inactivation system, and aims to solve the problems of pot overflow risk and low production efficiency caused by out-of-control foam growth in the inactivation process of a low-salt fermentation product, and poor adaptability and intervention lag of a traditional passive defoaming mode, and the low-salt fermentation product inactivation system comprises a multi-mode sensing module for obtaining state information of materials in an inactivation tank in real time, comprising a machine vision unit arranged at a window of the inactivation tank and used for collecting an image sequence of a foam area in the tank and extracting foam visual features; the digital twinning and predictive control module is used for receiving the data of the multi-mode sensing module; according to the method, a digital twinning advanced simulation mechanism is constructed, and a strategy of combining dynamic foam stability critical relation real-time tracking and model prediction control is adopted, so that active intervention and preventive inhibition on a foam generation source in the low-salt thick broad-bean sauce inactivation process are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial process control, and more particularly to a foam monitoring and suppression system based on machine vision for fermentation product inactivation process. BACKGROUND

[0002] In the industrial production of fermented foods such as soybean paste and douchi, inactivation (such as pasteurization or high-temperature sterilization) is a key process to ensure product safety and terminate fermentation. However, this process, especially for low-salt healthy products that have emerged in recent years, is always accompanied by a thorny problem - the uncontrollable generation and overflow of foam.

[0003] Low-salt formulations weaken the osmotic pressure and ionic strength of the system, causing protein, polysaccharide and other surface-active substances in the fermentation product to form stable and dense foam more easily when heated. Traditional coping strategies mainly rely on the experience of operators to observe, set fixed heating programs, or trigger the addition of defoaming agents or mechanical defoaming based on simple threshold values (such as a single liquid level switch).

[0004] These methods have significant limitations: first, manual observation is lagging and inaccurate, and cannot quantify the foam state in real time; second, fixed programs lack adaptability and cannot respond to changes in material rheology and foamability caused by differences in raw material batches and fermentation levels, and to ensure safety, they are often overly conservative, sacrificing heating efficiency and energy consumption; third, simple feedback control only responds passively after foam has already formed, which is a "band-aid" rather than a "root cause" solution, and excessive use of defoaming agents can affect product flavor and clean label requirements. Therefore, the existing technology lacks an intelligent control method that can accurately perceive the foam state, predict its evolution trend, and actively intervene in its source of generation, thereby maximizing inactivation efficiency while ensuring absolute safety. How to achieve "perceptible, predictable, and preventable" closed-loop precise control of foam in the inactivation process of low-salt fermented products has become a key technical bottleneck for improving the automation level, product quality, and economic efficiency of such food production. In view of this, we propose a foam monitoring and suppression system based on machine vision for fermentation product inactivation process. SUMMARY

[0005] The purpose of the present application is to provide a foam monitoring and suppression system based on machine vision for fermentation product inactivation process to solve the problems of overflow risk, low production efficiency, and poor adaptability and lagging intervention of traditional passive defoaming methods caused by uncontrollable growth of foam in the inactivation process of low-salt fermented products.

[0006] To solve the above technical problems, the present application provides the following technical scheme: a foam monitoring and suppression system based on machine vision for fermentation product inactivation process, comprising: a multi-modal perception module for acquiring material state information in the inactivation tank in real time, including a machine vision unit deployed at a viewing window of the inactivation tank, configured to collect image sequences of a foam region in the tank and extract foam visual features; a digital twin and predictive control module receiving data from the multi-modal perception module and performing the following operations: based on the received foam visual features, triggering in-depth monitoring of the material physicochemical state to acquire real-time multi-physical field data including temperature, pH value, material viscoelasticity, and micro-bubble nucleus distribution; based on the real-time multi-physical field data, driving a pre-constructed low-salt bean inactivation process digital twin model for advanced simulation, the digital twin model coupling heat transfer, fluid dynamics, bubble population balance, and surfactant transport mechanisms; through the advanced simulation, dynamically generating a foam stability critical relationship adapted to the current material state, the critical relationship relating at least temperature, heating rate, and pH value; based on the dynamically generated foam stability critical relationship, using a model predictive control algorithm to rollingly optimize a process parameter control sequence in a future time domain; a cooperative execution module receiving the process parameter control sequence issued by the digital twin and predictive control module and converting it into cooperative control instructions for the heating unit, stirring unit, and at least one defoaming execution mechanism to adjust the inactivation process; wherein, in the rapid heating phase, the system is configured to: according to the dynamically generated foam stability critical relationship, calculate and track the maximum safe heating rate under the current material state in real time, and through controlling the heating unit and the stirring unit, make the actual heating rate dynamically approach but not exceed the maximum safe heating rate.

[0007] The present application realizes active intervention and preventive inhibition of the foam generation source in the low-salt soybean paste inactivation process by constructing a digital twin advanced simulation mechanism based on machine vision triggering and fusion of multi-physical field data, and adopting a strategy combining real-time tracking and model predictive control of dynamic foam stability critical relationship. Specifically, the system uses the foam features captured by machine vision in real time as the initial signal, deeply fuses real-time process data such as temperature, pH, viscoelasticity, etc., drives a high-fidelity digital twin model to perform millisecond-level advanced simulation. The core output of this simulation is a dynamically generated "foam stability critical relationship", which is essentially a "safe heating rate boundary line" that changes in real time with the material state. In the key rapid heating stage, instead of executing a fixed or empirical heating curve, the system calculates and strictly tracks the "maximum safe heating rate" under the current state in real time according to this dynamic boundary, dynamically adjusts the heating and stirring power, so that the process path always operates close to the safety limit. This fundamental change upgrades the control logic from the traditional "foam appears-posterior elimination" to "state prediction-source prevention", thereby fundamentally eliminating the risk of uncontrolled growth and overflow of foam, while ensuring optimal heating efficiency within the safety boundary, solving the core problems of poor adaptability, delayed response and low efficiency in the background technology.

[0008] Preferably, the multi-modal perception module further comprises: a temperature sensing unit for monitoring the real-time temperature and heating rate of the material; a pH sensing unit for monitoring the real-time pH of the material; an ultrasonic sensing unit including an ultrasonic probe array arranged on the tank wall for emitting ultrasonic waves and receiving echo signals to invert the microbubble nucleus density and particle size distribution inside the material; an online rheological sensing unit for monitoring the complex viscosity of the material to characterize its viscoelastic state.

[0009] Preferably, the construction of the low-salt soybean paste inactivation process digital twin model includes a coupled three-dimensional heat transfer model, a three-dimensional computational fluid dynamics model, a population balance model, and an interface chemical model. wherein the population balance model is described by the following equation: ; wherein, represents the bubble number density function; is the characteristic size of the bubble; is the time variable; is the divergence operator; represents the flow field velocity vector; , , , are the generation, disappearance, breakage and coalescence rate terms of the bubbles, respectively.

[0010] The multi-physical field coupling simulation is realized in an ANSYS Fluent platform, a population balance model (PBM) is bidirectionally coupled with a flow and heat transfer module through a user-defined function (UDF), and an interface adsorption transport model is added to a component transport equation as a source term; the bubble coalescence efficiency coefficient and the initial value of the breakage frequency coefficient are obtained through high-speed photography and image analysis on the inactivation process of a typical batch of low-salt soyabean paste, and are obtained by fitting bubble size distribution evolution data. The interface adsorption isotherm parameters of the surface active substances (proteins and polysaccharides) are determined by combining literature values with material composition tests; Preferably, the digital twin and predictive control module is configured to utilize the real-time multi-physical field data to perform online correction on internal parameters of the digital twin model by solving the following parameter optimization problem: ; wherein, is a vector of internal parameters of the digital twin model to be optimized, including the bubble coalescence efficiency coefficient and the breakage frequency coefficient ; represents the current time; is a sliding time window length for parameter correction; is a state vector predicted by the digital twin model at the time under the parameter , including the predicted total volume of the foam and the average size of the bubbles; is a state vector actually measured at the time by the multi-modal perception module; is a weight matrix.

[0011] Preferably, the operation of dynamically generating the foam stability critical relationship is specifically: in the digital twin model, with the current real-time multi-physical field data as the initial condition, a plurality of different virtual heating rate trial values are set; for each , a lead simulation is run to predict the foam height evolution trajectory in a future set time period ; according to all the predicted trajectories, a critical heating rate that causes the foam height to first exceed a safety threshold is determined, which is expressed as: The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem: The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem:

[0012] The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem: The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem: The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem: The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem: The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem: The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem: The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem: The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem: The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem: The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem: The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem: The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem: The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem: The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem: The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem: The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem: The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem: The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem: The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem: The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem: The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem: The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem: The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem: The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem: The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem: The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem: The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem: The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem: The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem: The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem: The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem: The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem: The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem: The system further comprises a model predictive control algorithm configured to solve the following constrained optimization problem:

[0013] Preferably, the system is configured to control the entire inactivation process in stages, with each stage including at least a rapid heating stage and a heat preservation stage. During the heat preservation stage, the digital twin and predictive control module is configured to: based on a constant target temperature, find the pH value optimization range that makes the existing foam disintegration rate the fastest through the advanced simulation, and adjust the pH value of the material to the optimization range by controlling the pH adjuster addition device; The pH optimization range is achieved by minimizing the foam half-life. To determine: ; The half-life is determined by solving the foam decay kinetics model. Obtain; where, The optimal pH value or pH range to be determined; It is the rate constant; For model index; This is a constant representing the foam bursting rate related to pH.

[0014] Preferably, it also includes a fault diagnosis and fault tolerance control module, which is configured as follows: The key state variables predicted by the digital twin model are continuously compared with the actual measured values ​​of the multimodal sensing module; When the deviation continues to exceed the preset tolerance, an anomaly is determined and a fault-tolerant control strategy is triggered. The strategy includes switching to a robust PID control mode with the foam height extracted by the machine vision unit as the main feedback variable.

[0015] A foam control method for the inactivation process of low-salt fermented soybean paste includes the following steps: S1: Real-time acquisition of foam images inside the inactivation tank via machine vision unit, extraction of foam visual features; S2: In response to the aforementioned foam visual characteristics, simultaneously acquire data on the material's temperature, pH value, viscoelasticity, and microbubble core distribution; S3: Input the aforementioned foam visual features and multiphysics data into a pre-constructed digital twin model to perform advanced simulation and dynamically generate the critical relationship of foam stability under the current state; S4: Based on the aforementioned critical relationship of foam stability, the process parameter control sequence in the future time domain is continuously optimized using a model predictive control algorithm; S5: Execute the process parameter control sequence to coordinate the control of the heating unit, stirring unit and defoaming actuator; During the rapid heating phase, the maximum safe heating rate is calculated in real time based on the critical relationship of foam stability, and the actual heating rate is controlled to dynamically track the maximum safe heating rate.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention constructs a digital twin advanced simulation mechanism based on machine vision triggering and integrating multi-physics data. It employs a strategy combining real-time tracking of dynamic foam stability critical relationships with model predictive control to proactively intervene in and preventively suppress foam generation at the source during the inactivation process of low-salt fermented soybean paste. Specifically, the system uses real-time foam characteristics captured by machine vision as the initial signal, deeply integrating real-time process data such as temperature, pH, and viscoelasticity to drive a high-fidelity digital twin model for millisecond-level advanced simulation. The core output of this simulation is a dynamically generated "foam stability critical relationship," which is essentially a "safe heating rate boundary line" that changes in real-time with the material state. During the critical rapid heating phase, the system does not execute a fixed or empirical heating curve, but rather, based on this dynamic boundary, calculates and strictly tracks the "maximum safe heating rate" under the current state through a model predictive control algorithm, dynamically adjusting the heating and stirring power to ensure the process path always operates close to the safety limits. This fundamental change elevates the control logic from the traditional "bubble emergence - post-hoc elimination" to "state prediction - source prevention," thereby fundamentally eliminating the risk of uncontrolled bubble growth and overflow, while ensuring optimal heating efficiency within the safety boundary, and solving the core problems of poor adaptability, delayed response, and low efficiency proposed in the background technology.

[0017] 2. This invention also significantly improves the system's adaptability to complex and variable operating conditions and the reliability of state prediction by introducing a multimodal sensing and online correction mechanism for the digital twin model, laying a precise decision-making foundation for the aforementioned preventive control. The system-integrated multimodal sensors (vision, temperature, pH, ultrasound, rheology) provide synchronous, multidimensional sensing of the macroscopic morphology of foam and the microscopic physicochemical properties of materials. More importantly, using these real-time measurement data, the system continuously corrects the key internal parameters of the digital twin model (such as bubble coalescence and breakage efficiency coefficient) through online optimization algorithms, enabling the virtual model to adaptively "fit" the material characteristics and equipment status of the current batch. This closed loop of "sensing-correction-simulation" ensures the predictive accuracy of the digital twin advanced simulation. Compared with traditional methods that rely on static models or single sensor data, this solution effectively overcomes the uncertainties caused by raw material differences, sensor drift, or process disturbances, making the dynamically generated critical safety boundaries and the future state predictions required for model predictive control more accurate and reliable, thereby ensuring the effective execution and robustness of the preventive control strategy from the source.

[0018] 3. This invention also achieves refined and intelligent management of foam throughout the inactivation process and high system reliability by designing a phased adaptive defoaming collaborative strategy and a fault-tolerant mechanism. The system deconstructs the inactivation process into key stages such as rapid heating, heat preservation, and cooling, and implements differentiated optimal control based on the different foam-dominant mechanisms at each stage: the heating stage focuses on boundary tracking control to "inhibit formation"; the heat preservation stage uses simulation to find the optimal pH window to "promote disintegration" and utilizes the material's own physicochemical properties to accelerate foam dissipation; the cooling stage prevents secondary foaming caused by gas supersaturation. Simultaneously, the collaborative execution module intelligently selects and combines mechanical, acoustic, and chemical defoaming methods based on real-time process instructions and material rheological data. Furthermore, the built-in fault diagnosis and fault-tolerant module continuously compares digital twin predictions with actual measurements to promptly detect sensor or actuator anomalies and seamlessly switch to robust backup control mode. This end-to-end, multimodal, and adaptive collaborative control and safety design not only further consolidates and optimizes the foam suppression effect and reduces the reliance on single defoaming methods (especially chemical defoamers), aligning with the concept of high-quality food production, but also ensures the long-term stable and reliable operation of the entire intelligent system in complex industrial environments. Attached Figure Description

[0019] Figure 1 This is a diagram of the overall system architecture of the present invention; Figure 2 This is a schematic diagram of the digital twin and predictive control module of the present invention; Figure 3 This is a schematic diagram of the collaborative execution module of the present invention; Figure 4 This is a schematic diagram of the phased control strategy of the present invention; Figure 5 This is a flowchart illustrating the overall system workflow of the present invention. Figure 6 This is a flowchart illustrating the digital twin model construction and online calibration process of the present invention. Figure 7 This is a flowchart of the Model Predictive Control (MPC) optimization process of the present invention; Figure 8 This is a flowchart illustrating the phased control process of the present invention. Figure 9 This is a flowchart of the fault diagnosis and fault-tolerant control of the present invention. Detailed Implementation

[0020] Example 1: As Figures 1 to 4 As shown, this invention relates to a machine vision-based foam monitoring and defoaming system for the inactivation process of fermentation products, used in the inactivation process of low-salt fermented soybean paste, comprising: The multimodal perception module is used to acquire material status information inside the inactivation tank in real time. It includes a machine vision unit deployed at the window of the inactivation tank, which is used to collect image sequences of the foam area inside the tank and extract foam visual features. In an embodiment of the present invention, the multimodal sensing module further includes: Temperature sensing unit is used to monitor the real-time temperature and heating rate of the material; The pH sensing unit is used to monitor the real-time acidity and alkalinity of materials. An ultrasonic sensing unit includes an array of ultrasonic probes arranged on the tank wall for emitting ultrasonic waves and receiving echo signals to invert the nucleus density and particle size distribution of microbubbles inside the material. An online rheological sensing unit is used to monitor the complex viscosity of materials to characterize their viscoelastic state.

[0021] In an embodiment of the present invention, the machine vision unit includes an industrial camera and an image processor communicatively connected thereto, the image processor being configured to perform: The acquired image sequence is segmented to identify and quantify the height, coverage area, and texture features of the foam region; The instantaneous rate of change of foam height is calculated based on a continuous sequence of images; The height, coverage area, texture features, and rate of change of the foam region are used as one or more combinations of the foam visual features output.

[0022] The digital twin and predictive control module receives data from the multimodal sensing module and performs the following operations: Based on the received foam visual features, in-depth monitoring of the material's physicochemical state is triggered to acquire real-time multiphysics data including temperature, pH value, material viscoelasticity, and microbubble nucleus distribution. Based on the real-time multiphysics data, a pre-constructed digital twin model of the low-salt soybean inactivation process is driven to perform millisecond-level advanced simulation. The digital twin model couples heat transfer, fluid dynamics, bubble community equilibrium and surface active substance transport mechanism. In an embodiment of the present invention, the construction of the digital twin model of the low-salt soybean inactivation process includes: A three-dimensional heat transfer model describing the temperature field distribution and changes inside the inactivation tank was established; A three-dimensional computational fluid dynamics model describing the fluid flow state inside the tank under stirring is established; Establish a population equilibrium model to describe the processes of bubble nucleus generation, growth, aggregation, and breakup; Establish an interfacial chemical model to describe the adsorption and transport of protein and polysaccharide surfactants at the gas-liquid interface; The heat transfer model, computational fluid dynamics model, population equilibrium model, and interface chemistry model are coupled to form a multiphase flow multiphysics coupled simulation model.

[0023] The population equilibrium model describing the bubble evolution process is described by the following equation: ; In the formula: The bubble number density function represents the time... At that time, the bubble size was Bubble number concentration distribution; The characteristic size of the bubble (e.g., diameter); It is a time variable; is the divergence operator, representing the diffusion or transport effect in space; Represents the velocity vector of the flow field, describing the velocity distribution of the fluid inside the inactivation tank; This represents the rate of bubble formation, describing the rate at which new bubbles are formed due to gas supersaturation or the precipitation of dissolved gases. This represents the rate of bubble disappearance, describing the rate at which bubbles vanish as they dissolve back into the liquid from the gas. This represents the bubble breakup rate, which describes the rate at which a large bubble breaks into smaller bubbles due to fluid shear forces. This represents the bubble coalescence rate, which describes the rate at which bubbles collide and merge with each other, resulting in a decrease in the number of bubbles. Performance Analysis: This population equilibrium equation, as the core mechanistic model of the digital twin, quantifies the evolutionary dynamics of bubble swarms under complex flow and heat transfer environments at the microscopic scale. It mathematically describes the four key physical processes of bubble generation, dissipation, breakup, and coalescence, enabling the system to accurately predict the dynamic changes in bubble size distribution and quantity under different process conditions. By coupling this mechanistic model with macroscopic heat transfer and flow models, the digital twin possesses the ability to simulate the entire process of bubble generation, stabilization, and dissipation from first-principles calculations. This lays a solid physical foundation for the subsequent accurate generation of dynamic critical relationships and the implementation of advanced control strategies, enabling control decisions to transcend traditional experience-based threshold judgments and achieve true "model-driven" precision prevention.

[0024] In an embodiment of the present invention, the digital twin and predictive control module is configured to use the real-time multiphysics data to perform online correction of the internal parameters of the digital twin model through a data assimilation method. The internal parameters include at least one of the following: bubble coalescence efficiency coefficient, bubble breakage frequency coefficient, or surface tension temperature coefficient.

[0025] The online calibration is achieved by solving a parameter optimization problem, the objective function of which is to minimize the error between the model's predicted state and the sensor's measured state. ; In the formula: The vector of internal parameters of the digital twin model to be optimized typically contains parameters that are difficult to measure directly but are crucial to the model's accuracy, including the bubble coalescence efficiency coefficient. With the crushing frequency coefficient ; Indicates the current time; The length of the sliding time window used for parameter correction, i.e., considering the past... Data at each moment; For digital twin models in parameters Below the time The predicted state vector, including the predicted total foam volume and average bubble size, is a dimensionless normalized state vector. For at any time The state vector actually measured by the multimodal sensing module, and Corresponding to the dimension, it is a dimensionless normalized state vector; The weight matrix is ​​positive definite and is used to adjust the relative importance of errors of different state variables in the objective function. Its function is to balance the differences in magnitude of the state variables and ensure the objective function. It is a pure scalar; The signal normalization processing involved in this invention adopts the following method: for any physical quantity Its normalized value .in, and This represents the preset maximum and minimum values ​​of the physical quantity within the safe process range (e.g., temperature range of 60-100℃, foam height range of 0-tank safety height). Weight matrix. , , The initial values ​​are set empirically based on the importance of each normalized variable, and can be fine-tuned after the system is put into operation; Performance Analysis: This optimization formula forms the core of the online self-calibration function of the digital twin model. It dynamically optimizes and updates the model's internal mechanistic parameters by minimizing the weighted error between model predictions and multi-sensor measurements within a sliding time window. This process enables the general digital twin model, originally built based on typical operating conditions, to adaptively "fit" the unique properties of the specific batch of material being processed (such as rheological differences caused by different fermentation stages) and the actual operating state of the equipment. This continuous online calibration mechanism greatly enhances the system's robustness against raw material fluctuations, sensor drift, and process disturbances, ensuring the real-time reliability and accuracy of the digital twin prediction results, thus providing a reliable decision-making basis for feedforward and optimal control.

[0026] Through the aforementioned advanced simulation, a critical relationship for foam stability that is adapted to the current material state is dynamically generated. This critical relationship is at least related to temperature, heating rate, and pH value. In an embodiment of the present invention, the operation of dynamically generating the critical relationship for foam stability specifically includes: In the digital twin model, using the current real-time multiphysics data as the initial condition, multiple different virtual heating rate trial values ​​are set; For each of the proposed virtual heating rate values, the advanced simulation is run to predict the evolution trajectory of the foam state within a set future time period; Based on all predicted bubble state evolution trajectories, determine the critical heating rate that causes the bubble state to exceed the preset risk threshold; The critical heating rate is correlated with the current temperature and pH value to form the critical relationship for foam stability at the current moment.

[0027] Wherein, the critical heating rate The determination is based on the following criteria: during the simulation duration Inside, the total height of the foam First time exceeding the safety threshold The rate of temperature increase corresponding to this point is the critical value. This relationship can be expressed as: ; In the formula: This represents the critical heating rate calculated under the current material conditions, i.e., the maximum safe heating rate.

[0028] This represents the tentative value of the virtual heating rate set in the digital twin advanced simulation.

[0029] Indicates the current time Starting from the virtual heating rate Simulation was conducted, after experiencing The predicted foam height is obtained after the simulation duration.

[0030] This indicates a preset safety threshold for bubble height; exceeding this threshold is considered a risk of bubble runaway.

[0031] This refers to the current moment in the simulation calculation.

[0032] This indicates the time frame of the advanced simulation, i.e., how far into the future the bubble is predicted to be.

[0033] Performance Description: This formula is a concrete implementation of the dynamic foam critical line generation algorithm. It defines how the system actively and quantitatively explores and determines the safe operating boundary. By performing rapid parallel simulations on a series of virtual heating rates and evaluating their corresponding future foam risks, the system can accurately locate the boundary point between the "safe" and "risk" regions—the critical heating rate. This quantitative criterion transforms traditional qualitative experience (such as "slow heating") into precise numerical constraints based on real-time state data. This critical rate, as the core constraint of model predictive control, ensures that the optimization of the heating process is no longer a blind speed race, but rather an efficiency optimization within a dynamic safety corridor. This fundamentally achieves absolute safety in the production process while approaching the process limits.

[0034] In an embodiment of the present invention, the critical relationship of foam stability is expressed as a dynamically updated three-dimensional relationship surface or relationship mapping table, the three dimensions of which are temperature, heating rate and pH value, respectively. Any point on the surface defines the maximum safe heating rate allowed under the combination of temperature and pH value.

[0035] Based on the dynamically generated critical relationship of foam stability, a model predictive control algorithm is used to continuously optimize the process parameter control sequence in the future time domain, with the foam visual features as one of the key feedback variables in the process parameter control sequence. In embodiments of the present invention, the model predictive control algorithm is configured to perform the following optimization problem: The optimization objective is to minimize the sum of heating time and energy consumption over several future control cycles. Using the dynamically generated critical relationship of foam stability as the core constraint, it is ensured that the predicted foam state corresponding to the optimized process parameter control sequence is always lower than the preset risk level. The physical operating limits of the actuator are used as boundary constraints.

[0036] The mathematical description of the optimization problem to be solved in each control cycle is as follows: ; Constraints: System dynamic model (obtained by linearization or parameterization of the digital twin model); , ; , ; , ; In the formula: The control input sequence to be optimized, i.e., heating power, is typically determined by future... Heating power per control cycle composition, ; The objective function for Model Predictive Control (MPC) is defined as follows: For the prediction time domain, it represents the number of future prediction steps considered in the optimization problem; For control in the time domain, it represents the number of future control action steps to be optimized. ); For at any time For future moments Temperature forecast; The target temperature set for the process; The weighting coefficients (matrix) for the temperature tracking error term are used to adjust the degree of emphasis on temperature control accuracy. For the future The heating power value; This refers to the control increment of heating power, i.e., the power change between adjacent control cycles; The weighting coefficients (matrix) of the incremental penalty term are used to suppress drastic fluctuations in control actions and improve system stability. For at any time For future moments Predicted heating rate; The dynamic critical heating rate function, as determined above, is the current predicted temperature. and pH value The function, This indicates the value of the function at the predicted temperature and pH. , These are the lower and upper physical limits of the heating power actuator, respectively; , These are the lower and upper physical limits of the heating power change rate, respectively, used to ensure smooth operation of the actuator; objective function First item: The dimension is [temperature]. 2 Through weights (Dimension: [temperature]) -2 Make it dimensionless, objective function Second item: The dimension is [power]. 2 Weighting is required. (Dimension: [Power]) -2 This makes it dimensionless. The state and control variables in the objective function are normalized, and the weight coefficients are used to adjust the priority of each term after normalization. Therefore, the entire optimization problem is dimensionless. The simplified system dynamic model for MPC is obtained through the following steps: within the typical operating range set by the digital twin model, a pseudo-random sequence excitation signal is applied, and system input (heating power, stirring speed) and output (temperature, foam height) data are collected; a low-order state-space model or ARX model is obtained using a system identification algorithm (such as subspace identification method) as the prediction model for MPC. Performance Analysis: This optimization problem fully defines the decision-making logic of the model predictive controller. Its objective is to solve a series of optimal future control actions while satisfying multiple constraints. The objective function, while pursuing rapid and accurate achievement of the target temperature (first term), penalizes excessive control fluctuations to protect equipment and smooth the process (second term). Crucially, it incorporates dynamically generated bubble critical relationships as hard constraints into the optimization, ensuring that any process path guided by the adopted control sequence is absolutely safe in the prediction time domain. This architecture, which uses the "safety boundary" as the core constraint of the optimization problem, enables the system to automatically and in real-time calculate the optimal heating strategy that balances "efficiency," "safety," and "stability," achieving a leap from "feasible control" under safety conditions to "intelligent control" that pursues optimal comprehensive performance indicators.

[0037] In embodiments of the present invention, the process parameter control sequence includes: a timing instruction for the target heating power, target stirrer speed and steering mode, and target pH adjuster addition amount for multiple future time steps.

[0038] The collaborative execution module receives the process parameter control sequence issued by the digital twin and predictive control module and converts it into collaborative control instructions for the heating unit, stirring unit and at least one defoaming actuator to regulate the inactivation process. In an embodiment of the present invention, the cooperative execution module includes: A heating control unit is used to precisely adjust the power output of the heating unit according to the coordinated control command; A stirring control unit is used to adjust the rotation speed and direction of the stirring unit according to the cooperative control command, wherein the steering mode includes periodic forward and reverse rotation to enhance shearing; A defoaming actuator control unit is used to control at least one defoaming actuator, which includes one or more of an ultrasonic defoaming device, a defoaming agent metering and adding device, or a mechanical defoaming paddle.

[0039] In an embodiment of the present invention, the cooperative execution module is further configured as follows: Based on the process parameter control sequence and the material viscoelasticity data fed back by the online rheological sensing unit, different combinations of defoaming actuators are dynamically selected and activated. When the viscoelasticity of the material exceeds the first threshold, the mechanical defoaming paddle is activated first, supplemented by the ultrasonic defoaming device; When the foam visual characteristics indicate that the foam coverage exceeds the second threshold but the viscoelasticity is lower than the first threshold, the defoamer metering and adding device is activated preferentially.

[0040] During the rapid heating phase, the system is configured to: calculate and track the maximum safe heating rate under the current material state in real time based on the dynamically generated critical relationship of foam stability, and control the heating unit and the stirring unit to make the actual heating rate dynamically approach but not exceed the maximum safe heating rate.

[0041] Technical Solution Overview: This invention provides a machine vision-based foam monitoring and defoaming system for the inactivation process of fermentation products, particularly suitable for the inactivation process of low-salt fermented soybean paste. The system uses real-time machine vision monitoring of foam as the trigger and feedback core, integrating multi-physical field sensor data such as temperature, pH, and viscoelasticity to drive a digital twin model coupled with mechanisms of heat transfer, flow, and bubble evolution for millisecond-level advanced simulation. This simulation can dynamically generate a critical relationship for foam stability (such as a dynamic foam critical line) that precisely matches the current material state, and based on this, continuously optimize the process parameter control sequence through a model predictive control algorithm. During the critical rapid heating phase, the core function of the system is to calculate and track the maximum safe heating rate based on the current state in real time, and through coordinated control of the heating and stirring units, intelligently approach this safe limit to the actual heating rate, thereby fundamentally suppressing the uncontrolled generation of foam while ensuring inactivation efficiency. This system achieves closed-loop intelligent control from "sensing foam" to "predicting foam" and then to "preventing foam."

[0042] In an embodiment of the present invention, the system further includes a fault diagnosis and fault tolerance control module, which is configured to: The key state variables predicted by the digital twin model are continuously compared with the actual measured values ​​of the multimodal sensing module; When the deviation of the comparison continues to exceed the preset tolerance, the system is determined to be abnormal. The abnormality includes sensor data drift, actuator failure, or sudden change in material batch characteristics. Upon detecting an anomaly, a fault-tolerant control strategy is triggered, which includes switching to a robust PID control mode based on historical safety process data and issuing an alarm.

[0043] In an embodiment of the present invention, the robust PID control mode uses the foam height extracted by the machine vision unit as the main feedback variable, and the rotation speed of the stirring unit and / or the flow rate of the defoamer metering and adding device as control variables. Its control parameters are tuned based on a large amount of safety historical data.

[0044] In an embodiment of the present invention, the system is configured to control the entire inactivation process in stages, the stages including at least a rapid heating stage and a heat preservation stage; During the rapid heating phase, calculating and tracking the maximum safe heating rate is the core control task. During the heat preservation stage, the digital twin and predictive control module is configured to: based on a constant target temperature, find the pH value optimization range that maximizes the existing foam disintegration rate through advanced simulation, and adjust the pH value of the material to the optimization range by controlling the pH adjuster addition device.

[0045] In an embodiment of the present invention, the method for determining the pH value optimization range is as follows: In the digital twin model, the target insulation temperature is fixed, and the dynamic process of foam height decaying over time under different pH conditions is simulated. Calculate the half-life of foam height under various pH conditions; The pH range that minimizes the half-life is selected as the optimization range.

[0046] Wherein, the half-life Defined as the bubble height from the current value The time required for the decay to reach half its original value. The optimization objective is to find the pH value that minimizes the half-life, which can be mathematically expressed as: ; In the formula: The optimal pH value or pH range to be determined; Represents mathematical operators, indicating the search for functions that make a function When the minimum value is obtained, the corresponding value; The foam half-life is defined as the time required for the foam height to decrease from its initial value to half under specific constant pH conditions. It is a function of pH and is obtained by solving a foam decay kinetic model. , For time; The foam bursting rate constant is a pH-dependent constant, reflecting the intrinsic influence of pH on foam liquid film strength and bursting kinetics. Let be the rate constant, with dimensions of [time]. -2 ; The model index is used to describe whether the foam decay process conforms to first-order kinetics or a more complex mode. The two are obtained by fitting the simulation results of the digital twin model. Among them, the and The parameters are determined as follows: In the digital twin model, the target insulation temperature is fixed, and a series of different pH values ​​(e.g., 4.0, 4.5, ..., 6.0) are set to simulate and obtain the foam height. Over time The decay curves were obtained; for each decay curve, the kinetic model was fitted using the nonlinear least squares method to obtain the decay curves at that pH. Value; ultimately obtained Functional relationship with pH And take the curves obtained by fitting. The average value is used as the model index. ; The height of the foam; Effects Description: This set of formulas defines the optimization objective for active foam disintegration during the insulation stage. Through digital twin model simulation or historical data analysis, the system first establishes a quantitative relationship model between foam decay rate and pH value. Optimization Objective This clearly points to finding the pH environment that allows foam to dissipate naturally and most quickly. This signifies a shift in control strategy from "inhibiting formation" during the heating phase to "promoting disintegration" during the heat preservation phase. This is achieved by adjusting process parameters (…). By actively and precisely adjusting to this "optimal defoaming window," the system can maximize the use of the material's own physicochemical properties for defoaming, thereby significantly reducing reliance on external defoamers or strong mechanical intervention. This not only improves defoaming efficiency but also aligns with the production philosophy of high-quality, clean-label foods, demonstrating the profound value of this invention in process optimization and quality assurance.

[0047] In an embodiment of the invention, the system is further configured to control the cooling phase after inactivation is completed; During the cooling phase, the digital twin and predictive control module is configured to generate an optimized cooling rate curve based on an established secondary foaming prediction model for the cooling process, wherein the secondary foaming prediction model is associated with the cooling rate, the current pH value, and the supersaturation of dissolved gas. The collaborative execution module controls the flow rate of the cooling medium based on the optimized cooling rate curve.

[0048] In embodiments of the present invention, communication between the multimodal sensing module, the digital twin and predictive control module, and the cooperative execution module is achieved through industrial Ethernet or time-sensitive networks to ensure the real-time nature of control commands and status feedback.

[0049] Example 2: Figures 5 to 9 As shown, a foam control method for the inactivation process of low-salt fermented soybean paste is applied to a machine vision-based foam monitoring and defoaming system for the inactivation process of fermentation products, comprising the following steps: S1: Real-time acquisition of foam images inside the inactivation tank via machine vision unit, extraction of foam visual features; S2: In response to the aforementioned foam visual characteristics, simultaneously acquire data on the material's temperature, pH value, viscoelasticity, and microbubble core distribution; S3: Input the aforementioned foam visual features and multiphysics data into a pre-constructed digital twin model to perform advanced simulation and dynamically generate the critical relationship of foam stability under the current state; In another embodiment of the present invention, the step of dynamically generating the critical relationship for foam stability specifically includes: In a digital twin simulation environment, starting from the current state, multiple virtual heating rates are set for exploratory simulation. Assess the future bubble risk corresponding to each virtual warming rate; Determine the critical rate of temperature rise that will trigger an unacceptable risk of foaming, and correlate it with the current temperature and pH value to form an ever-updating safe operating boundary.

[0050] S4: Based on the aforementioned critical relationship of foam stability, the process parameter control sequence in the future time domain is continuously optimized using a model predictive control algorithm; S5: Execute the process parameter control sequence to coordinate the control of the heating unit, stirring unit and defoaming actuator; During the rapid heating phase, the maximum safe heating rate is calculated in real time based on the critical relationship of foam stability, and the actual heating rate is controlled to dynamically track the maximum safe heating rate.

[0051] As another embodiment of the present invention, a fault-tolerant step is also included: Monitor the deviation between digital twin predictions and actual sensor measurements; When the deviation exceeds the limit, switch to a robust backup control strategy based on historical data, and use the foam height monitored by machine vision as the core feedback variable for adjustment.

[0052] Example 3: Comparative Experiment and Effect Verification To verify the effectiveness of the system of the present invention, a comparative test was conducted for one month in the inactivation section of a low-salt broad bean paste production line with an annual output of 100,000 tons. The test subject was low-salt broad bean paste from the same fermentation batch and after the same pretreatment. The target inactivation temperature was 95℃, and the process required rapid heating from 60℃ to 95℃ and holding at that temperature for 20 minutes.

[0053] 1. Test System Configuration Inactivation vessel: 5m³ volume, jacketed heating, equipped with a variable frequency stirrer (speed range 0-100rpm).

[0054] This invention system (intelligent group): Sensing layer: A 2-megapixel industrial camera (sampling rate 10Hz) is installed at the top viewing window of the tank; a Pt100 temperature sensor, an online pH meter, an ultrasonic probe array (1MHz), and an online rotational rheometer probe are installed on the tank wall.

[0055] Digital twin: Built on ANSYS Fluent and a custom population equilibrium model (PBM), running on an edge computing server. Simulation step size is 100ms, and the lead simulation duration is ( Set to 120 seconds.

[0056] Controller: Employs a Model Predictive Control (MPC) algorithm generated using MATLAB / Simulink to predict the time domain. =30, control time domain =10, control cycle is 2 seconds.

[0057] Execution layer: variable frequency heater, variable frequency stirrer, ultrasonic defoaming device (28kHz), defoamer (food grade polydimethylsiloxane) metering pump, mechanical defoaming paddle.

[0058] Compared to System A (Fixed Program PID Group): Using the original factory equipment, a fixed temperature rise curve is set (60℃→95℃, taking 40 minutes). The heating power is controlled by PID based on a single temperature sensor. The defoamer pump is started after the foam height exceeds a fixed threshold (80% of the tank height).

[0059] Comparison System B (Experienced Human Group): Experienced operators manually adjust the heating power and defoamer addition based on observation through the viewing window.

[0060] 2. Experimental Design and Data Recording Ten batches of inactivation tests were conducted for each of the three control methods. The following key indicators were recorded and analyzed: Safety: Rate of overflow.

[0061] Efficiency: The actual average time to heat from 60°C to 95°C.

[0062] Economic and quality impact: Average amount of defoamer used per batch.

[0063] System performance: Mean absolute error (MAE) of bubble height prediction for the intelligent group, and frequency of online model correction.

[0064] 3. Experimental Results and Analysis The key experimental data are summarized in the table below: Table 1: Comparison of key indicators in the inactivation process of different control methods

[0065] Safety (overflow control): This invention's intelligent system achieves zero overflow. This is because the system dynamically generates a "safe heating rate boundary" (dynamic critical line) through digital twin advanced simulation and strictly tracks this boundary under MPC control. Figure 5 As shown in the simulation diagram, the traditional fixed program (dashed line) encroaches on the "foam risk zone" when encountering high-viscosity batches, while the intelligent system (solid line) can adjust its dynamic path in real time to avoid the risk zone. Both the PID group and the manual group experienced overflow because they could not predict and adapt to changes in materials.

[0066] Efficiency (heating time): Under the premise of absolute safety, the system of this invention reduces the average heating time to 28.5 minutes, which is about 28.8% more efficient than a fixed 40-minute program. This is thanks to the "boundary tracking" control strategy, which allows the heating process to proceed at the highest safe rate allowed by the current material state, rather than a conservative fixed value.

[0067] Economy and Quality (Defoamer Dosage): The defoamer dosage of this invention is only 15.3 mL / batch, which is reduced by approximately 82.4% and 70.8% compared to the PID group and the manual group, respectively. This is mainly attributed to two-stage optimized control: in the heating stage, foam generation is significantly reduced through source prevention; in the heat preservation stage, by optimizing and adjusting to the "pH window where foam is most easily destroyed" (approximately pH 5.2-5.4 in this experiment), the natural dissipation of foam is promoted, reducing the dependence on chemical defoamers and better meeting the requirements of the clean label.

[0068] System performance: The digital twin model's mean absolute error (MAE) for predicting bubble height is less than 3%, demonstrating the effectiveness of the multiphysics model and online correction mechanism. Model parameters are automatically corrected every 2-3 batches based on real-time data, ensuring long-term adaptability and prediction accuracy.

[0069] The specific experimental data provided in this embodiment demonstrate that the foam monitoring and defoaming system based on machine vision, digital twin, and model predictive control described in this invention can effectively solve the foam control problem in the deactivation process of low-salt fermented soybean paste. The system achieves a fundamental shift from passive response to proactive prevention, significantly improving heating efficiency and drastically reducing defoamer consumption while ensuring 100% production safety (zero overflow), thus verifying the advanced nature, effectiveness, and practical value of its technology.

[0070] The embodiments disclosed in this invention are preferred embodiments, but are not limited thereto. Those skilled in the art can easily understand the spirit of this invention based on the above embodiments and make different extensions and variations, but as long as they do not depart from the spirit of this invention, they are all within the protection scope of this invention.

Claims

1. A machine vision based foam monitoring and suppression system for a fermentation product inactivation process, characterized in that, Comprise: A multi-modal perception module for acquiring material state information in the inactivation tank in real time, including a machine vision unit deployed at the viewing window of the inactivation tank for collecting image sequences of the foam region in the tank and extracting foam visual features; A digital twin and predictive control module receiving data from the multi-modal perception module and performing the following operations: Based on the received foam visual features, triggering deep monitoring of the material physicochemical state, acquiring real-time multi-physical field data including temperature, pH value, material viscoelasticity and micro-bubble nucleus distribution; According to the real-time multi-physical field data, driving a pre-constructed low-salt bean inactivation process digital twin model to perform advanced simulation, the digital twin model coupling heat transfer, fluid dynamics, bubble population balance and surfactant transport mechanism; Through the advanced simulation, dynamically generating a foam stability critical relationship adapted to the current material state, the critical relationship at least relating temperature, heating rate and pH value; Based on the dynamically generated foam stability critical relationship, using a model predictive control algorithm to rollingly optimize the process parameter control sequence in the future time domain; A cooperative execution module receiving the process parameter control sequence issued by the digital twin and predictive control module and converting it into cooperative control instructions for the heating unit, stirring unit and at least one defoaming execution mechanism to adjust the inactivation process; Wherein, in the rapid heating stage, the system is configured to: according to the dynamically generated foam stability critical relationship, real-time calculation and tracking the maximum safe heating rate under the current material state, and through controlling the heating unit and the stirring unit, the actual heating rate is dynamically approximated but not exceeded.

2. A machine vision based fermentation product inactivation process foam monitoring and suppression system according to claim 1, characterized in that, The multi-modal perception module further comprises: A temperature sensing unit for monitoring the real-time temperature and heating rate of the material; A pH sensing unit for monitoring the real-time pH of the material; An ultrasonic sensing unit including an ultrasonic probe array arranged on the tank wall for emitting ultrasonic waves and receiving echo signals to invert the micro-bubble nucleus density and particle size distribution inside the material; An online rheological sensing unit for monitoring the complex viscosity of the material to characterize its viscoelastic state.

3. The foam monitoring and suppression system for a machine vision based inactivation process of a fermentation product according to claim 1, characterized in that The construction of the low-salt bean inactivation process digital twin model includes coupled three-dimensional heat transfer model, three-dimensional computational fluid dynamics model, population balance model and interface chemical model; Wherein, the population balance model is described by the following equation: ； wherein represents the bubble number density function; is the characteristic size of the bubble; is the time variable; is the divergence operator; represents the flow field velocity vector; , , , are the generation, disappearance, break-up and coalescence rate terms of the bubbles, respectively.

4. A machine vision based fermentation product inactivation process foam monitoring and suppression system according to claim 3, wherein, The digital twin and predictive control module is configured to use the real-time multi-physical field data to online correct the internal parameters of the digital twin model by solving the following parameter optimization problem: ； In the formula, The internal parameter vector of the digital twin model to be optimized includes the bubble coalescence efficiency coefficient. With the crushing frequency coefficient ; Indicates the current time; The length of the sliding time window used for parameter correction; For digital twin models in parameters Below the time The predicted state vector includes the predicted total foam volume and average bubble size; For at any time The state vector is obtained through actual measurement by the multimodal sensing module; This is the weight matrix.

5. The machine vision based fermentation product inactivation process foam monitoring and suppression system of claim 1, wherein, The operation of dynamically generating the foam stability critical relationship is specifically: In the digital twin model, a plurality of different virtual temperature rise rate trial values are set with current real-time multi-physical field data as initial conditions ; for each , run a forward simulation to predict the foam height evolution trajectory over a future set time period ;​ determining, from all predicted trajectories, a critical temperature rate of rise that causes the foam height to first exceed the safety threshold of the critical temperature rate of rise which is expressed by the relation: ； The In connection with the current temperature, pH value, the critical relationship of the foam stability at the current moment is formed, in which, is the current moment for simulation calculation.

6. A machine vision based fermentation product inactivation process foam monitoring and suppression system according to claim 5, wherein, The foam stability critical relationship is expressed as a dynamically updated three-dimensional relationship surface or mapping table, with temperature, heating rate and pH value as the three dimensions; the model predictive control algorithm is configured to solve the following constrained optimization problem: ； Constraint condition: system dynamic model (linearized or parameterized from digital twin model); ， ； ， ； ， ； wherein, is the control input sequence to be optimized; is the optimization objective function of model predictive control; is the prediction horizon; is the control horizon; is the temperature prediction value at time for future time ; is the target temperature set by the process; is the weight coefficient of the temperature tracking error term; is the heating power value at time for future time ; is the control increment of heating power; is the weight coefficient of the control increment penalty term; is the temperature ramp-up rate prediction value at time for future time ; are the physical lower and upper limits of the heating power actuator, respectively; , are the physical lower and upper limits of the heating power rate of change, respectively.

7. A machine vision based fermentation product inactivation process foam monitoring and suppression system according to claim 2, wherein, The cooperative execution module includes a heating control unit, a stirring control unit and a defoaming execution mechanism control unit; The defoaming execution mechanism control unit is configured to control at least one of an ultrasonic defoaming device, a defoaming agent metering and adding device, and a mechanical defoaming paddle. The cooperative execution module is further configured to dynamically select and activate different defoaming execution mechanism combinations according to the process parameter control sequence and the material viscoelasticity data fed back by the online rheological sensing unit.

8. The foam monitoring and suppression system for a machine vision based inactivation process of a fermentation product according to claim 1, characterized in that, The system is configured to control the whole inactivation process in stages, and the stages at least include a rapid heating stage and a holding stage. In the holding stage, the digital twin and predictive control module is configured to find the pH value optimization interval that makes the existing foam collapse rate fastest through the lead simulation based on a constant target temperature, and adjust the material pH value to the optimization interval through the control of the pH regulator adding device. wherein the pH-optimization interval is determined by minimizing the foam half-life : ； The half-life is obtained by solving a foam decay kinetic model obtained; wherein, is the optimal pH value or pH interval to be solved for; is the rate constant; is the model exponent; is the pH dependent foam collapse rate constant; is the foam height.

9. The machine vision based fermentation product inactivation process foam monitoring and suppression system of claim 1, wherein, Further comprising a fault diagnosis and fault-tolerant control module configured to: continuously compare the key state variables predicted by the digital twin model with the actual measured values of the multi-modal sensing module; when the deviation continuously exceeds the preset tolerance, it is determined that an abnormality occurs and a fault-tolerant control strategy is triggered, and the strategy includes switching to a robust PID control mode with the foam height extracted by the machine vision unit as the main feedback variable.

10. A method for foam control in a low-salt soy paste inactivation process, applied to a machine vision-based fermented product inactivation process foam monitoring and suppression system according to any one of claims 1 to 9, characterized in that, The method comprises the following steps: S1: acquiring foam images in the inactivation tank in real time through a machine vision unit, and extracting foam visual features; S2: in response to the foam visual features, synchronously acquiring temperature, pH value, viscoelasticity, and micro-bubble nucleus distribution data of the material; S3: inputting the foam visual features and multi-physical field data into a pre-constructed digital twin model, performing lead simulation, and dynamically generating a foam stability critical relationship under the current state; S4: based on the foam stability critical relationship, using a model predictive control algorithm to rollingly optimize the process parameter control sequence in the future time domain; S5: executing the process parameter control sequence to cooperatively control the heating unit, the stirring unit, and the defoaming execution mechanism; wherein in the rapid heating stage, the maximum safe heating rate is calculated in real time according to the foam stability critical relationship, and the actual heating rate is controlled to dynamically track the maximum safe heating rate.

Images