Grass-fruit adaptive drying system based on digital twinning and model predictive control

Abstract

The application discloses a grassfruit self-adaptive drying system based on digital twinning and model predictive control, and relates to the technical field of grassfruit drying.The application comprises: a multi-modal perception network arranged at key nodes of a grassfruit drying production line, which is used for collecting raw material initial quality data, drying process physical quantity data and actuator state data in real time; a grassfruit drying process digital twinning body running on a control platform, which is a high-fidelity virtual mapping of a physical drying system, and has a drying dynamics reduced-order model with multi-physical field coupling built-in; and a model predictive controller in communication connection with the multi-modal perception network and the digital twinning body, which is used for performing feedforward initialization and online rolling optimization; so as to solve the problems that an existing drying system cannot perceive and respond to initial differences of materials, process control is lagged and isolated, and quality, energy consumption and efficiency are difficult to be cooperatively optimized.

Description

Technical Field

[0001] This invention relates to the field of cardamom drying technology, specifically, to an adaptive cardamom drying system based on digital twin and model predictive control. Background Technology

[0002] In the large-scale processing industry of cardamom, drying is the most crucial yet vulnerable link that determines its final quality, safety, and commercial value. As an important medicinal and edible raw material, the drying process of cardamom not only needs to effectively reduce moisture for long-term storage, but also needs to preserve its unique aroma components, color, appearance, and internal active substances to the greatest extent possible. However, the current mainstream processing mode, especially the one widely used by farmers and small and medium-sized processing plants in the vast producing areas, is still the traditional, earthen drying method represented by earthen stoves and coal-fired drying rooms. This process relies heavily on the personal experience of the operators, judging the degree of dryness by visual observation and touch, and using direct or indirect heating with open flames. This method has a series of fundamental defects that are difficult to overcome: First, the fumes produced by combustion come into direct contact with the material, which easily leads to the adsorption of strong carcinogens such as benzo(a)pyrene on the surface of the cardamom, and brings a distinct smoky taste, which seriously violates food safety and medicinal material standards; Second, due to the lack of precise control over temperature, humidity, wind speed and time, the drying process fluctuates drastically, often resulting in uneven drying of the material, local over-drying or wet core, and causing severe enzymatic and non-enzymatic browning, resulting in a dark color and poor appearance of the finished product; Third, traditional processes generally lack standardized pre-treatment procedures, such as efficient cleaning, color protection and uniform spreading, resulting in product residues, excessive microorganisms, high loss rate, and extremely poor quality stability between batches.

[0003] Although some mechanized drying equipment using electric or air-source heat pumps has emerged in recent years, mitigating pollution problems to some extent and achieving basic temperature and humidity settings, its technological logic has not undergone a fundamental leap. These systems mostly remain at the level of single-point or simple multi-point feedback control, operating based on fixed and rigid "recipes," unable to perceive and respond to the inherent differences in the initial moisture content, maturity, and composition of each batch of raw materials. Their control behavior is "remedial," meaning adjustments are only made after sensors detect deviations from set values, exhibiting significant lag. More importantly, the entire drying process remains a "black box," lacking in-depth understanding and modeling of the internal moisture migration, quality transformation kinetics, and heat and mass transfer fields within the equipment. Therefore, existing technological solutions cannot achieve adaptive optimization drying for the characteristics of different batches of raw materials, making it difficult to simultaneously pursue optimal drying efficiency and energy consumption while ensuring optimal quality (color and aroma retention), and even more difficult to achieve end-to-end digital quality traceability and continuous process optimization from raw materials to finished products. Therefore, developing an intelligent drying system that can deeply integrate multi-dimensional perception and has process prediction and adaptive optimization capabilities has become an urgent technological need to break through the current bottlenecks in the processing industry of cardamom and similar agricultural products. Summary of the Invention

[0004] The purpose of this invention is to provide an adaptive drying system for cardamom based on digital twins and model predictive control, so as to solve the problems of existing drying systems being unable to sense and respond to initial material differences, lagging and isolated process control, and difficulty in coordinating and optimizing quality, energy consumption and efficiency.

[0005] To solve the above problems, the present invention employs the following technical means: An adaptive drying system for cardamom based on digital twin and model predictive control includes: A multimodal sensing network is deployed at key nodes in the cardamom drying production line to collect raw material initial quality data, drying process physical quantity data, and actuator status data in real time. The digital twin of the cardamom drying process runs on the control platform and is a high-fidelity virtual mapping of the physical drying system. It has a built-in reduced-order model of drying dynamics with multi-physics field coupling. The model prediction controller, which is communicatively connected to the multimodal sensing network and the digital twin, is used to perform feedforward initialization and online rolling optimization; The system is configured to perform the following steps: Step S1: Based on the initial quality data collected by the multimodal sensing network, simulation and optimization are performed in the digital twin to generate the optimal process parameter setting trajectory for the current batch of cardamom. Step S2: During the drying process, the model predictive controller executes in a fixed control cycle: acquiring the current system state, using the corrected model in the digital twin to predict the future time domain system behavior, and solving the constrained optimization problem to obtain the optimal control sequence in the future control time domain; Step S3: Implement the first control quantity in the optimal control sequence, and based on actuator feedback and system output feedback, correct the model parameters of the digital twin through the state estimator to achieve closed-loop adaptive control.

[0006] Preferably, the multimodal sensing network includes: The hyperspectral imaging unit is installed in the raw material pretreatment section to acquire hyperspectral image data of the cardamom surface. I h ( λ, x, y ); An online near-infrared spectroscopy unit, installed in the raw material pretreatment section, is used to sample and obtain the near-infrared absorption spectrum of cardamom. A nir ( λ ); An array-type microwave moisture sensor, arranged laterally above the conveyor belt in at least one critical temperature zone inside the dryer, is used for online measurement of the dielectric constant distribution of the material layer. ε n ( t And invert the moisture distribution along the width direction. M distribution ( n , t ); An electronic nose sensor array, arranged inside the dryer's exhaust duct, is used to monitor the response vector of volatile organic compounds in the exhaust gas. R gas ( t ).

[0007] Furthermore, the digital twin of the cardamom drying process includes: Three-dimensional geometric and mesh model based on physical dryer structure; The multiphysics coupled computation kernel is used to solve the CFD control equations describing the hot air flow, heat and mass transfer in the dryer, the DEM equations describing the movement and accumulation of cardamom particles, and the thin-layer drying kinetic equations describing the diffusion of moisture inside the cardamom. The thin-layer drying kinetic equation includes a moisture diffusion equation: ; in D eff The effective moisture diffusion coefficient; and the drying rate equation: .

[0008] Furthermore, in step S1, converting the initial quality data into digital twin model parameters includes: Surface moisture distribution retrieved from hyperspectral data W s0 ( x , y (and near-infrared prediction of internal moisture) W i0 Calculate the average initial moisture content of the material: ; Where α is a preset weighting coefficient; Based on the sugar (S0) and starch (C0) content predicted by near-infrared spectroscopy, the effective moisture diffusion coefficient of the model is corrected: ; in f 1 This is a preset association function.

[0009] Furthermore, in step S1, generating the optimal setpoint trajectory in the digital twin specifically involves: constructing and solving for the future control input sequence. U For a finite-time optimization problem with decision variables, the objective function J is: ; in, W ( t f To predict the final moisture content, W target For the target moisture content, P total For total power consumption, t f Drying time, σ W ( k () represents the standard deviation of the predicted moisture content distribution. w 1 to w 4 These are the weighting coefficients. Q This is the weight matrix.

[0010] Furthermore, the online rolling optimization in step S2 is performed in each control cycle. k Execution, including: State acquisition: Obtain the current measurement output vector from the multimodal sensing network. y m ( k ); State and parameter estimation: Using an extended Kalman filter, state estimation is performed based on the previous time step. Control input u ( k-1 ) and current measurements y m ( k Update the joint estimate of the current state. The augmented state vector x a Includes system status x and model parameters θ ; Optimization solution: Estimate based on the current state As initial conditions, based on the corrected model Solve the following optimization problem to obtain the optimal control sequence. U * ( k ): ; in, For the model's predicted output, r ( k + i The setpoint trajectory is from step S1. N p To predict the time domain, N c To control the time domain.

[0011] Furthermore, step S3 includes: Control command output: The optimal control sequence U * ( k The first control variable in ) U * ( k | k (This is) sent to the physical actuator; Feedback correction: in the next cycle k +1, using the new measurement value y m ( k +1), the model parameters are corrected through the update step of the extended Kalman filter: ; in, For prior state prediction, K ( k +1) is the Kalman gain. h ( ) represents the observation model; Model update: The estimated new parameters Update the prediction model of the digital twin.

[0012] Furthermore, the system also includes a global collaborative scheduling module, which is configured to dynamically adjust the production cycle of the pretreatment unit based on the digital twin's prediction of the drying completion time of the current batch and the real-time status of the downstream cooling unit and temporary storage silo, so as to achieve optimized matching of overall production capacity and energy consumption.

[0013] Furthermore, the optimization problem solved in the online rolling optimization must satisfy the following constraints: System dynamic constraints: ; Input constraints: ; Input rate of change constraint: ; Output soft constraints: .

[0014] Furthermore, through the aforementioned adaptive drying control method, the fluctuation range of the moisture content of the cardamom at the drying endpoint is controlled within a certain range. Within, and the drying uniformity index σ W Compared to fixed parameter control, it reduces costs by more than 30%.

[0015] The present invention has the following beneficial effects during use: By sensing the initial components of raw materials using hyperspectral and near-infrared technologies, and combining steam color protection with personalized drying path planning based on digital twins, color and aroma are locked in from both the source and the process. In particular, the volatile organic compound spectrum monitored by the electronic nose is used as control feedback, with aroma retention as one of the optimization goals, ensuring that the finished product has a bright red and uniform color and a rich and pure unique flavor, significantly improving the product grade.

[0016] The entire process involves closed-loop hot air circulation drying without open flames, eliminating benzo(a)pyrene contamination and the source of smoky odor. Simultaneously, the model predictive control algorithm precisely controls the moisture content fluctuation range at the drying endpoint to an extremely small extent, and significantly improves drying uniformity through proactive optimization of moisture distribution.

[0017] This approach breaks away from the traditional model that relies on fixed formulas and post-event feedback. It utilizes initial quality data for feedforward optimization, customizing drying strategies for each batch of materials. Through online rolling optimization and extended Kalman filtering to correct model parameters in real time, the control system can proactively adjust the process, actively adapting to changes in material properties and external disturbances, achieving adaptive and robust control.

[0018] The model predictive controller treats the dryer as a coupled multi-input multi-output system, comprehensively optimizing multiple variables such as temperature, humidity, wind speed, and conveyor belt speed. While satisfying complex constraints, it dynamically seeks the balance point of quality, energy consumption, and time, achieving global optimal performance that cannot be achieved by a single PID control loop. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0020] Therefore, the following detailed description of embodiments of the present invention is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0021] It should be noted that, unless otherwise specified, the embodiments and features described in this invention can be combined with each other.

[0022] This embodiment presents an adaptive drying system for cardamom based on digital twins and model predictive control. A closed-loop control architecture integrating physical sensing, virtual simulation, intelligent decision-making, and precise execution is constructed. This enables the collaborative operation of a multimodal sensing network, a digital twin of the cardamom drying process, and a model predictive controller. The multimodal sensing network is deployed at key nodes of the production line, integrating advanced sensors such as hyperspectral imaging, online near-infrared spectroscopy, arrayed microwave moisture sensing, and electronic noses to collect comprehensive data from the initial quality of the raw materials (surface / internal composition) to process physical quantities (temperature, humidity, moisture distribution, volatile gases). The digital twin is a high-fidelity virtual mapping of the physical dryer, its core integrating computational fluid dynamics, discrete element method, and thin-layer drying dynamics models, capable of simulating the complex coupled processes of hot air flow, particle movement, and internal moisture diffusion. The model predictive controller acts as the brain, utilizing the sensing data and the digital twin model to perform feedforward initialization and online rolling optimization, and driving the actuators (heaters, fans, conveyor belts, etc.).

[0023] In this embodiment, the specific configuration of the multimodal sensing network is further defined. A hyperspectral imaging unit is installed at the raw material inlet to acquire surface spectral images. I h ( λ , x , y ), and online near-infrared spectroscopy unit sampling to obtain internal absorption spectra. Anir ( λ Inside the dryer, in the critical temperature zone, an array of microwave moisture sensors is arranged laterally to measure the moisture distribution along the width of the material layer in real time. M distribution ( n , t An electronic nose sensor array is installed at the exhaust vent to monitor the volatile organic compound fingerprint of the exhaust gas. R gas ( t These sensors together constitute the real-time sensing capability of the internal / external / gas three-dimensional state of materials.

[0024] In this embodiment, the construction of a digital twin of the cardamom drying process is its key technological foundation. It establishes a three-dimensional geometric model based on the physical equipment and simulates the real process by solving a set of coupled governing equations. Its core drying kinetics are described by the moisture diffusion equation within the material: ; in, D eff The effective moisture diffusion coefficient is related to both moisture content and temperature; meanwhile, surface evaporation is described by the convective mass transfer equation: ; To meet the requirements of real-time control, the high-fidelity model needs to be reduced in order to obtain a state-space model for model predictive control.

[0025] The control method in this embodiment begins with feedforward initialization and setpoint optimization. The system utilizes initial sensing data (surface moisture W) s0 Internal moisture W i0 Sugars (S0), starch = Customize the parameters of the digital twin model. For example, calculate the average initial moisture content: ; And correct the diffusion coefficient: ; Subsequently, in the digital twin, with future control sequences U Let be the variable, and solve a finite-time optimization problem with the objective function . J A comprehensive consideration of quality, energy consumption, time, and uniformity: ; The optimal process parameter trajectory obtained from the solution will serve as a reference target for subsequent online control.

[0026] During the drying process, the system enters an online rolling optimization phase. In each control cycle... kThe model prediction controller performs the following steps: First, obtain the current measurement value. y m ( k Secondly, the extended Kalman filter is invoked, based on the state estimate from the previous time step. Control input u ( k - 1 ) and current measurements y m ( k ), jointly estimate the current system state and model parameters Next, using the corrected model... Starting from the current state, solve a constrained rolling time-domain optimization problem: ; To track the setpoint trajectory generated by feedforward optimization r ( k + i This process penalizes drastic changes in the control variable, thereby obtaining the optimal control sequence in the future control time domain. U * ( k ).

[0027] After optimization calculations are completed, the system executes control command output and feedback correction. Only the first control variable in the optimal sequence is processed. U * ( k | k The data is sent to the physical actuator. In the next cycle, the new sensor measurements are used. y m ( k The update steps of the extended Kalman filter are as follows: ; Calculate the updated model parameters This information is then fed back to the digital twin model, enabling online adaptive correction of the model. This closed loop ensures that the control system can continuously compensate for model errors and external disturbances.

[0028] In this embodiment, the overall system also includes a global collaborative scheduling function. Based on the digital twin's prediction of the drying completion time of the current batch, and combined with the real-time status of downstream cooling, temporary storage, color sorting and other units, the central control platform dynamically adjusts the production cycle of the pretreatment (such as cleaning and color protection) units, thereby achieving the global optimal matching of the entire production line's capacity and energy consumption, and avoiding waiting or congestion between processes.

[0029] In the aforementioned online rolling optimization process, the optimization problem to be solved must satisfy a series of physical and technological constraints. These constraints include: determined by the corrected model. The system dynamic equations described are: ; Hard input constraints of the actuator: ; and its rate of change constraint: ; And soft output constraints on key process variables (such as temperature and humidity): ; To enhance the robustness of control.

[0030] The technical solution in this embodiment can ultimately control the moisture content fluctuation range of the dried cardamom within the target value. W target Within ±0.5%, and through the analysis of moisture distribution. σ W The proactive optimization significantly improves drying uniformity compared to traditional fixed parameter control, thereby achieving high-quality, high-efficiency, and low-energy-consumption intelligent production while ensuring optimal color and aroma retention.

[0031] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An adaptive drying system for cardamom based on digital twin and model predictive control, characterized in that, include: A multimodal sensing network is deployed at key nodes in the cardamom drying production line to collect raw material initial quality data, drying process physical quantity data, and actuator status data in real time. The digital twin of the cardamom drying process runs on the control platform and is a high-fidelity virtual mapping of the physical drying system. It has a built-in reduced-order model of drying dynamics with multi-physics field coupling. The model prediction controller, which is communicatively connected to the multimodal sensing network and the digital twin, is used to perform feedforward initialization and online rolling optimization; The system is configured to perform the following steps: Step S1: Based on the initial quality data collected by the multimodal sensing network, simulation and optimization are performed in the digital twin to generate the optimal process parameter setting trajectory for the current batch of cardamom. Step S2: During the drying process, the model predictive controller executes in a fixed control cycle: acquiring the current system state, using the corrected model in the digital twin to predict the future time domain system behavior, and solving the constrained optimization problem to obtain the optimal control sequence in the future control time domain; Step S3: Implement the first control quantity in the optimal control sequence, and based on actuator feedback and system output feedback, correct the model parameters of the digital twin through the state estimator to achieve closed-loop adaptive control.

2. The system according to claim 1, characterized in that, The multimodal sensing network includes: The hyperspectral imaging unit is installed in the raw material pretreatment section to acquire hyperspectral image data of the cardamom surface. I h ( λ, x, y ); An online near-infrared spectroscopy unit, installed in the raw material pretreatment section, is used to sample and obtain the near-infrared absorption spectrum of cardamom. A nir ( λ ); An array-type microwave moisture sensor, arranged laterally above the conveyor belt in at least one critical temperature zone inside the dryer, is used for online measurement of the dielectric constant distribution of the material layer. ε n ( t And invert the water distribution along the width direction. M distribution ( n , t ); An electronic nose sensor array, arranged inside the dryer's exhaust duct, is used to monitor the response vector of volatile organic compounds in the exhaust gas. R gas ( t ).

3. The system according to claim 1 or 2, characterized in that, The digital twin of the cardamom drying process includes: Three-dimensional geometric and mesh model based on physical dryer structure; The multiphysics coupled computation kernel is used to solve the CFD control equations describing the hot air flow, heat and mass transfer in the dryer, the DEM equations describing the movement and accumulation of cardamom particles, and the thin-layer drying kinetic equations describing the diffusion of moisture inside the cardamom. The thin-layer drying kinetic equation includes a moisture diffusion equation: ； in D eff The effective moisture diffusion coefficient; and the drying rate equation: 。 4. The system according to claim 3, characterized in that, In step S1, converting the initial quality data into digital twin model parameters includes: Surface moisture distribution retrieved from hyperspectral data W s0 ( x , y (and near-infrared prediction of internal moisture) W i0 Calculate the average initial moisture content of the material: ； Where α is a preset weighting coefficient; Based on the sugar (S0) and starch (C0) content predicted by near-infrared spectroscopy, the effective moisture diffusion coefficient of the model is corrected: ； Where f1 is the preset association function.

5. The system according to claim 4, characterized in that, In step S1, generating the optimal setpoint trajectory in the digital twin specifically involves: constructing and solving for the future control input sequence. U For a finite-time optimization problem with decision variables, the objective function J is: ； in, W ( t f To predict the final moisture content, W target For the target moisture content, P total For total power consumption, t f Drying time, σ W ( k () represents the standard deviation of the predicted moisture content distribution. w 1 to w 4 These are the weighting coefficients. Q This is the weight matrix.

6. The system according to claim 5, characterized in that, The online rolling optimization in step S2 occurs in each control cycle. k Execution, including: State acquisition: Obtain the current measurement output vector from the multimodal sensing network. y m ( k ); State and parameter estimation: Using an extended Kalman filter, state estimation is performed based on the previous time step. Control input u ( k-1 ) and current measurements y m ( k Update the joint estimate of the current state. The augmented state vector x a Includes system status x and model parameters θ ; Optimization solution: Estimate based on the current state As initial conditions, based on the corrected model Solve the following optimization problem to obtain the optimal control sequence. U * ( k ): ； in, For the model's predicted output, r ( k + i The setpoint trajectory is from step S1. N p To predict the time domain, N c To control the time domain.

7. The system according to claim 6, characterized in that, Step S3 includes: Control command output: The optimal control sequence U * ( k The first control variable in ) U * ( k | k (This is) sent to the physical actuator; Feedback correction: in the next cycle k +1, using the new measurement value y m ( k +1), the model parameters are corrected through the update step of the extended Kalman filter: ； in, For prior state prediction, K ( k +1) is the Kalman gain. h ( ) represents the observation model; Model update: The estimated new parameters Update the prediction model of the digital twin.

8. The system according to claim 1, characterized in that, The system also includes a global collaborative scheduling module, which is configured to dynamically adjust the production cycle of the pretreatment unit based on the digital twin's prediction of the drying completion time of the current batch and the real-time status of the downstream cooling unit and temporary storage silo, so as to achieve optimal matching of overall production capacity and energy consumption.

9. The system according to claim 6, characterized in that, The optimization problem solved in the online rolling optimization must satisfy the following constraints: System dynamic constraints: ； Input constraints: ; Input rate of change constraint: ; Output soft constraints: .

10. The system according to claim 1, characterized in that, The adaptive drying control method controls the fluctuation range of the cardamom moisture content at the drying endpoint within a specified range. Within, and the drying uniformity index σ W Compared to fixed parameter control, it reduces costs by more than 30%.

Images