An AI-based processing process quality control system for gordon euryale seeds

Abstract

The application discloses an AI-based processing process quality control system for gordon euryale seeds, belongs to the technical field of food processing and quality control, and comprises a quality perception quantification module, which is used for performing real-time nondestructive detection on material units entering a processing flow, analyzing and quantifying original spectrum data collected into an initial starch crystallinity index through a pre-established chemometrics model, and a quality dynamic evolution prediction module, which is used for predicting a quality evolution track of the material units in future processing time through a dynamic evolution prediction model based on the initial starch crystallinity index, a current quality state and real-time process parameters, so that different processing paths can be planned for materials with different characteristics, and disturbance caused by raw material heterogeneity can be effectively overcome.

Description

Technical Field

[0001] This invention relates to the field of food processing and quality control technology, specifically to an AI-based quality control system for the processing of water chestnuts. Background Technology

[0002] The main challenge facing the deep processing industry of water chestnut is the consistent control of the final product quality. Traditional processing uses fixed process parameters, which cannot adapt to the significant quality heterogeneity within raw material batches caused by factors such as breeding and origin, especially the difference in starch crystallinity. This difference directly affects the final textural properties of the product. Existing technologies lack the ability to identify raw material differences online and adjust the process in real time, resulting in unstable product quality. Therefore, there is an urgent need in this field for an intelligent quality control method to solve this problem.

[0003] The information disclosed in the background section above is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0004] The purpose of this invention is to provide an AI-based quality control system for the processing of water chestnuts, in order to solve the problems mentioned in the background art.

[0005] The technical solution of the present invention includes: a quality perception and quantification module, used to perform real-time non-destructive testing on material units entering the processing flow, and to analyze and quantify the collected raw spectral data into an initial starch crystallinity index through a pre-established chemometric model;

[0006] The quality dynamic evolution prediction module, based on the initial starch crystallinity index, current quality status and real-time process parameters, predicts the quality evolution trajectory of the material unit in the future processing time through a dynamic evolution prediction model.

[0007] The process parameter dynamic optimization module is used to compare the quality evolution trajectory with the preset ideal target quality trajectory, and calculate the optimal combination of process parameters for the next moment through the model predictive control optimization framework.

[0008] The closed-loop feedback control execution module is used to send the optimal combination of process parameters to the execution mechanism of the processing equipment to complete the real-time control of the processing process. Based on the quality status of the material unit after the control, a new round of quality evolution trajectory prediction, optimization and control is carried out to form a closed-loop feedback.

[0009] Preferably, the quality perception and quantification module includes a near-infrared spectroscopy detection system deployed on the material conveying line, which is used to collect the raw spectral data of each material unit.

[0010] Preferably, the chemometric model includes: collecting gorgon fruit samples with different starch crystallinity; determining the crystallinity value of the gorgon fruit samples using X-ray diffraction and simultaneously acquiring their near-infrared spectra; and training the model using a partial least squares regression algorithm with the near-infrared spectra as input and the crystallinity values ​​as output.

[0011] Preferably, the dynamic evolution prediction model consists of a discrete-time evolution equation of quality state and a constitutive model of instantaneous change rate of quality; the discrete-time evolution equation of quality state is used to iteratively calculate the quality state at the next moment based on the current quality state and the instantaneous change rate of quality.

[0012] Preferably, the constitutive model of instantaneous quality change rate is used to construct the functional relationship between the instantaneous quality change rate and the real-time process parameters, the current quality state, and the initial starch crystallinity index; the initial starch crystallinity index is introduced as a raw material quality correction term to characterize the characteristic that the higher the initial crystallinity of the raw material, the slower the rate of change of raw material quality.

[0013] Preferably, the model predictive control optimization framework calculates the optimal combination of process parameters by solving a cost function; the cost function includes a tracking error term for minimizing the deviation between the quality evolution trajectory and the ideal target quality trajectory, and a control cost term for penalizing changes in process parameters.

[0014] Preferably, the control cost term is a weighted control increment penalty term. The weighted control increment penalty term normalizes the change in control quantity using the normal operating range of each process parameter, and adjusts the penalty intensity for different control quantity changes through a weight matrix.

[0015] Preferably, the closed-loop feedback control execution module sends the optimal combination of process parameters to the programmable logic controller (PLC), which then drives the heater, valve, or motor.

[0016] Preferably, the current quality status is obtained through real-time measurement by online sensors deployed on the processing equipment, or, in the absence of online sensors, through a model-based state estimation method, combined with the predicted value from the previous control cycle and model correction.

[0017] This invention provides an AI-based quality control system for the processing of water chestnuts, which has the following improvements and advantages compared to existing technologies:

[0018] 1. The revolutionary advancement of this invention lies in its refinement of particle size control from macroscopic batches to microscopic material units. Through a quality-sensing quantification module, the system no longer blindly processes all raw materials. Instead, it utilizes a near-infrared spectroscopy detection system and a pre-established chemometric model to assign a quantitative initial starch crystallinity index to each material unit entering the processing flow. This is equivalent to issuing a unique quality ID to each material unit, enabling the system to proactively and accurately predict its processing characteristics. This is fundamentally different from the passive quality control method of existing technologies that only conducts sampling inspections after processing is completed;

[0019] 2. A core contribution of this invention is the creation of a quality dynamic evolution prediction module. The dynamic evolution prediction model within this module, especially its constitutive model of instantaneous quality change rate, has profound practical significance. At the same time, the system can actively predict rather than passively respond, and can proactively plan different processing paths for materials with different characteristics, thereby effectively overcoming the disturbances caused by the heterogeneity of raw materials.

[0020] 3. This invention does not employ simple proportional-integral-derivative control, but instead introduces a more advanced model predictive control optimization framework through a process parameter dynamic optimization module. This replaces the traditional repeated trial and error and adjustment that relies on human experience, transforming it into an optimal path-finding process that can proactively plan a series of future control actions, making process control more precise, stable, and efficient.

[0021] 4. The calculated optimal process parameters are sent to the underlying actuator through the closed-loop feedback control execution module, and the adjusted material quality status is used as the starting point of a new cycle. This forms a complete cyber-physical fusion closed loop. Compared with the open-loop and static control of the prior art, the closed-loop system of the present invention can suppress various unmodeled disturbances in real time and continuously stabilize the processing process on the optimal path. Attached Figure Description

[0022] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0023] Figure 1 This is a flowchart of the system of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0025] Example 1:

[0026] Please see Figure 1This invention provides a quality control system for the processing of water chestnut based on AI, including: a quality perception and quantification module, used to perform real-time non-destructive testing on material units entering the processing flow, and to analyze and quantify the collected raw spectral data into an initial starch crystallinity index through a pre-established chemometric model;

[0027] The quality dynamic evolution prediction module, based on the initial starch crystallinity index, current quality status, and real-time process parameters, predicts the quality evolution trajectory of the material unit within the future processing time through a dynamic evolution prediction model.

[0028] The process parameter dynamic optimization module is used to compare the quality evolution trajectory with the preset ideal target quality trajectory, and calculate the optimal combination of process parameters for the next moment through the model predictive control optimization framework.

[0029] The closed-loop feedback control execution module is used to send the optimal combination of process parameters to the actuator of the processing equipment to complete the real-time control of the processing process. Based on the quality status of the material unit after the control, a new round of quality evolution trajectory prediction, optimization and control is carried out to form a closed-loop feedback.

[0030] This invention provides an AI-based quality control system for the processing of water chestnuts, aiming to overcome the technical challenge of unstable finished product quality caused by the heterogeneity of raw material quality. The system, through a highly integrated cyber-physical architecture of perception, prediction, decision-making, and execution, transforms the traditional batch control paradigm into precise unit control and the lagging passive detection paradigm into a proactive prediction paradigm. The system's operation begins with a quality perception and quantification module, which assigns a quantifiable initial quality identity to each individual water chestnut material unit. Sequentially, a quality dynamic evolution prediction module, based on this identity and... Real-time operating conditions are used to construct mathematical models to deduce the intrinsic quality changes of materials during processing. Based on this, the dynamic optimization module for process parameters calculates the optimal process intervention measures using advanced control theory based on the deviation between the predicted evolution trajectory and the ideal target. The closed-loop feedback control execution module accurately sends this optimal decision to the physical equipment and continuously iterates this process, thus forming an adaptive closed-loop control system. This system effectively overcomes the interference of raw material differences on the quality of the final product, ensuring that products such as gorgon fruit cake and gorgon fruit powder achieve a high degree of consistency and excellent taste and texture.

[0031] Example 2

[0032] The quality perception and quantification module includes a near-infrared spectroscopy detection system deployed on the material conveyor line, which is used to collect the raw spectral data of each material unit;

[0033] The chemometric model includes: collecting gorgon fruit samples with different starch crystallinities; determining the crystallinity values ​​of the gorgon fruit samples using X-ray diffraction and simultaneously acquiring their near-infrared spectra; and training the model using a partial least squares regression algorithm with near-infrared spectra as input and crystallinity values ​​as output.

[0034] In this embodiment, the core component of the quality perception and quantification module is a near-infrared spectroscopy (NIRS) detection system deployed on the gorgon fruit slurry conveying pipeline. This system performs non-invasive, high-throughput real-time scanning on each material unit flowing through it, capturing its unique raw spectral data. The fundamental purpose of this measure is to provide the most essential initial state information of each material unit for subsequent precise control.

[0035] To convert high-dimensional spectral data Transforming the data into a single indicator with clear physical meaning, the quality perception quantification module utilizes a pre-built chemometric model. The model was established with rigorous scientific procedures: researchers collected a large number of water chestnut samples covering different origins, varieties, and batches, and used X-ray diffraction (XRD), the recognized gold standard method, to accurately determine the crystallinity of their starch. Simultaneously, NIRS spectral data of the corresponding samples were collected. Based on this, multiple sets of spectral data were used as input variables X, and the crystallinity values ​​determined by XRD were used as output variables Y. Partial least squares regression (PLS-R) was employed for fitting and training, ultimately obtaining a calibration model capable of accurately retrieving crystallinity from spectral data. ;

[0036] In actual operation of the system, this module uses the following formula to parse the real-time acquired spectral data into an initial starch crystallinity index. :

[0037]

[0038] in, As a dimensionless relative index, the value directly reflects the degree of orderliness of the starch structure in the material unit, thus characterizing the ease or difficulty of its processing. It becomes the first key link connecting the heterogeneity of raw materials with adaptive control of the process. This method enables the system to see through the appearance of raw materials and understand the differences in their internal quality, providing a solid data foundation for achieving quality control from the source.

[0039] Example 3

[0040] The dynamic evolution prediction model consists of a discrete-time evolution equation of quality state and a constitutive model of instantaneous change rate of quality. The discrete-time evolution equation of quality state is used to iteratively calculate the quality state at the next moment based on the current quality state and the instantaneous change rate of quality.

[0041] A constitutive model of instantaneous quality change rate is used to construct the functional relationship between the instantaneous quality change rate and real-time process parameters, current quality status, and initial starch crystallinity index. The initial starch crystallinity index is introduced as a raw material quality correction term to characterize the characteristic that the higher the initial crystallinity of the raw material, the slower the rate of change in raw material quality.

[0042] In this embodiment, the quality dynamic evolution prediction module constructs a dynamic evolution prediction model that can accurately depict the quality evolution trajectory of water chestnut material during processing. A key technical feature is that the model is not a single black box model, but consists of two mutually coupled sub-models with clear physical connotations: a macroscopic discrete-time evolution equation of quality state and a microscopic constitutive model of instantaneous change rate of quality. The two work together to completely describe the dynamic process from raw materials to finished products.

[0043] The first component is the discrete-time evolution equation of the quality state; this equation originates from the forward Euler method for solving ordinary differential equations in numerical analysis. The technical motivation is to cleverly decompose the complex, nonlinear, continuous quality evolution process into a series of equations with extremely small time steps. The linear change within the process allows the computer to predict and simulate the long-term evolution trajectory of material quality throughout the entire processing cycle through simple iterative calculations. This equation forms the iterative framework for quality prediction. In each control cycle, it uses the current state and the instantaneous rate of change to recursively deduce the state at the next moment, advancing layer by layer. This provides predicted values ​​of future states for model predictive control and is the foundation for achieving forward-looking control. The expression is:

[0044]

[0045] in, The model predicts the material quality state at the next moment, which is the output of this equation. The current material quality status is used as the input to the model. This is the preset time step for system control and calculation. It is a constant, and its selection requires comprehensive consideration of the system's computing power and the dynamic characteristics of the process. It should generally be smaller than the main time constant of the controlled process to ensure the stability of numerical integration and the timeliness of control. The specific value can be determined through simulation or field experiments, and its dimension is time. ; It is the instantaneous rate of change of quality at the current moment, and its dimension is the reciprocal of time. It is not an independent parameter, but a bridge connecting macroscopic and microscopic models. Its value is calculated by the constitutive model described below.

[0046] This constitutive model is one of the core innovations of this invention; it is a constitutive model of the instantaneous rate of change of quality. This model profoundly integrates scientific principles from multiple fields, drawing on the Arrhenius equation, rheological theory, and first-order reaction rate models in chemical reaction kinetics. The technical motivation is to create a compact mathematical expression with clear physical meaning that can simultaneously capture the combined influence of thermo-coupling effects, the current state of the material, and, most importantly, the initial characteristics of the raw materials on the processing rate. The model receives real-time process parameters from sensors and raw material characteristic parameters from Example 2, calculating the instantaneous rate of change driving quality evolution. It is precisely because of the existence of the raw material quality correction term that the entire dynamic evolution prediction model possesses the core capability to respond to raw material heterogeneity. The prediction results are no longer uniform but rather tailored to the personalized evolution trajectory of each material unit. The expression is:

[0047]

[0048] Among them, real-time process parameter vector It is the control input of the model, including the effective processing temperature. The dimension is temperature. and equivalent mechanical force The dimension can be pressure. or shear rate ;

[0049] For example, in a cooking tank with an agitator, the equivalent mechanical force This can be mainly reflected in the average shear rate. This value can be estimated using the following industry-recognized formula:

[0050]

[0051] in, It refers to the mixer speed, measured in revolutions per second (rpm). It is a dimensionless constant related to the geometry of the agitator, such as diameter and blade type. This constant can be obtained by consulting relevant chemical engineering handbooks or through fluid dynamics simulation.

[0052] It is an experimentally calibrated inherent property of the material, representing the critical reference temperature at which significant changes in starch quality begin to occur, with the dimension of temperature. ;in, It is the temperature sensitivity coefficient, and its dimensions are... ; It is the mechanical force sensitivity coefficient; when the equivalent mechanical force In terms of pressure, dimensions When measuring, The dimensions are ;when In terms of shear rate, dimensions When measuring, These are dimensionless coefficients; their values ​​are obtained through fitting and training on a large amount of historical processed data. It is a dimensionless state saturation term, which ensures that when the material quality approaches the ideal state (i.e., When the rate of change of a material is zero, its rate of change naturally slows down and tends to zero, which is entirely in line with the laws of physics; the crucial innovation lies in the dimensionless raw material quality correction term. This item refers to the initial starch crystallinity index obtained in Example 2. As a key variable introduced into the dynamic model, It is a dimensionless undetermined coefficient obtained through model training; the introduction of this correction term mathematically accurately describes the core physicochemical characteristic that the higher the initial crystallinity of the gorgon fruit raw material, the more difficult it is for its molecular chain to deconstruct and rearrange, and therefore the slower its quality change rate.

[0053] To enable those skilled in the art to better understand and implement the present invention, a set of exemplary parameter values ​​obtained through experimental calibration in the context of gorgon fruit powder processing are provided herein; these values ​​can serve as an initial reference for implementing the present invention: critical reference temperature : Temperature sensitivity coefficient : Mechanical force sensitivity coefficient When equivalent mechanical force In terms of shear rate (unit) When measured by ) It is a dimensionless coefficient, and its value can be taken as... Raw material quality correction factor : (dimensionless);

[0054] It should be emphasized that the optimal parameter values ​​may vary depending on the specific equipment and raw material batch, and those skilled in the art can make fine adjustments based on the above reference values.

[0055] Example 4

[0056] The model predictive control optimization framework calculates the optimal combination of process parameters by solving a cost function. The cost function includes a tracking error term to minimize the deviation between the quality evolution trajectory and the ideal target quality trajectory, and a control cost term to penalize changes in process parameters.

[0057] The control cost term is a weighted control increment penalty term. The weighted control increment penalty term normalizes the change in control quantity using the normal operating range of each process parameter, and adjusts the penalty intensity for different control quantity changes through a weight matrix.

[0058] In this embodiment, the function of the process parameter dynamic optimization module is to find the optimal path to the ideal quality target for the system based on the quality evolution trajectory predicted in Embodiment 3. To achieve this, the module adopts Model Predictive Control (MPC) optimization framework. In each control cycle... This framework does not simply calculate the optimal action for the next instant, but rather solves a carefully designed cost function. To plan for a finite time domain in the future A complete set of optimal process parameters;

[0059] The design of this cost function aims to achieve a well-considered control strategy. It not only ensures that the future quality trajectory closely follows the preset ideal target, but also considers process stability and safety, effectively suppressing drastic fluctuations in process parameters to avoid impacting equipment or causing product quality oscillations. In each control cycle, the optimization solver searches for a set of future control sequences to minimize this total cost—essentially a dynamic trade-off process. This guides the system to find a realistic control strategy that effectively corrects quality deviations while maintaining smooth changes in process parameters. This directly ensures the stability, efficiency, and safety of the entire water chestnut processing process. The model predicts the control cost function as follows:

[0060]

[0061] in, It is the dimensionless total cost to be minimized; the first term of the cost function This is the tracking error term; here, Based on the current moment All information, obtained by repeatedly calling the prediction model of Example 3, is the future... The predicted quality status of the step; and This is the ideal quality target value predefined by process experts at the same future moment based on historical data from the production of gold batches or the texture design during the product development phase; the core function of this is to drive the future output of the system to infinitely approach the desired target; the second term of the cost function This is a weighted control increment penalty term, i.e., a control cost term; here, Represents the future The change in the vector of process parameters. A clever design lies in the fact that this change is within the normal operating range of each process parameter. Normalization was performed; The source is the physical specifications and safety regulations of the processing equipment; this normalization allows changes in control quantities of different physical dimensions and orders of magnitude to be compared and weighted on the same scale; furthermore... It is a diagonal weight matrix used to adjust the penalty intensity for different control variable changes. Its specific value is a tunable hyperparameter derived from on-site process debugging, aiming to balance the speed of control response and the stability of process operation. The length of the prediction time domain needs to be chosen by balancing prediction accuracy and computational burden. Its value is usually determined by simulation or field testing so that the total prediction time can cover a major time constant of material quality changes.

[0062] Example 5

[0063] The closed-loop feedback control execution module sends the optimal combination of process parameters to the programmable logic controller (PLC), which then drives the heater, valve, or motor.

[0064] In this embodiment, the function of the closed-loop feedback control execution module is to transform decisions in the information world into actions in the physical world; in the process parameter dynamic optimization module of embodiment 4, the optimal process parameter sequence for the next N steps is obtained by solving the cost function. Subsequently, the closed-loop feedback control execution module strictly follows the basic principles of model predictive control, extracting and executing only the first element in the sequence, namely the optimal combination of process parameters at the current moment. ;

[0065] Sequentially, this module will use this optimal instruction For example, optimal temperature and optimal shear rate These instructions are sent as digital signals to the programmable logic controller (PLC) at the factory's bottom layer. As the core control unit of industrial automation, the PLC immediately parses these instructions and precisely drives the connected physical actuators to respond. For example, the PLC adjusts the opening of the steam valve supplying the cooking tank to ensure the material temperature reaches the specified level. At the same time, adjust the motor speed of the stirring system to apply an equivalent to The mechanical force; this action completes the transformation from an optimal calculation result to a precise physical operation, constituting the execution link of the control loop; one control cycle After completion, the system will obtain the new material quality status and use it as a starting point to start a new round of prediction-optimization-execution cycle, thus forming a closed-loop feedback that is constantly being corrected and ensures that the processing process always runs on the optimal track.

[0066] Example 6

[0067] The current quality status is obtained through real-time measurements by online sensors deployed on the processing equipment, or, in the absence of online sensors, through a model-based state estimation method, combined with the predicted value from the previous control cycle and model correction.

[0068] In this embodiment, to achieve effective closed-loop feedback, the current quality status at the beginning of each control cycle is accurately determined. This is crucial; the present invention provides two highly flexible and adaptable technical approaches for this purpose.

[0069] The first and optimal approach is direct measurement. This involves deploying online sensors, such as online viscometers, rheometers, or specific optical probes, on core processing equipment like cooking tanks or aging pipelines to continuously monitor the physical or chemical properties of the material in real time. The measured values ​​are then calibrated and transformed into the current quality status using a calibration model. ;

[0070] To give the core indicator of "quality status" a clear physical meaning and measurability, this invention defines it as the degree of gelatinization of the gorgon fruit material, a dimensionless index between 0 and 1; the calculation method can be correlated with the measurement value of an online viscometer; for example, it can be normalized using the following formula:

[0071]

[0072] in, It is the current moment. The quality status; It is the material viscosity measured in real time by an online viscometer; It is the initial viscosity of the material when it enters the processing equipment; It is the material viscosity at the ideal processing endpoint, i.e., the gold batch product, determined experimentally; through this definition, This indicates that the material is in its initial, unprocessed state, while This means that the material has reached the ideal final quality state;

[0073] This method of acquisition It is the most authentic and timely, providing the highest quality feedback signal for subsequent prediction and control;

[0074] In scenarios involving the modification of existing production lines or cost constraints, it may be impossible to install online sensors. To address this, the present invention provides a second approach: a model-based state estimation method. In this mode, the system will use the previous control cycle, i.e., time [time value missing], [time value missing]. Regarding the current moment The predicted quality status is combined with a correction term based on actual process parameters and model deviation to calculate an optimal estimate of the current quality status.

[0075] This model-based state estimation method employs an observer structure with an intermittent correction term. During most control cycles without offline sampling, the system directly uses the model's own prediction as the optimal estimate of the current state. When new offline measurement data is obtained, the model state is corrected once to eliminate accumulated errors, resulting in a more accurate estimate of the current quality state. The following formula can be used to calculate:

[0076] 1. At times when no offline measurements are available :

[0077]

[0078] 2. At a specific moment when offline measurement values ​​are obtained. :

[0079]

[0080] in, This is an estimate of the current quality status; This is the model's prediction of the current quality state from the previous time step; correction only occurs when offline measurements are obtained, and the correction term consists of two parts: At a specific moment Corrected state estimate; It refers to a specific point in time. The true quality state of the material is obtained by offline sampling and testing in the laboratory, for example, by measuring its viscosity and calculating it according to the supplementary formula in point 1. The quality prediction model at the same time point as offline sampling The predicted quality value; It is a preset observer gain coefficient, for example, between 0.1 and 0.5, used to adjust the strength of the correction;

[0081] Although this method is indirect, it fully utilizes the predictive power of existing high-precision dynamic models and anchors and calibrates the model through periodic, precise offline measurements, achieving soft measurement even in the absence of hardware. Therefore, whether through direct physical sensing or indirect model estimation, this system can reliably obtain the current quality state necessary to initiate each closed-loop iteration. This greatly enhances the universality and robustness of the quality control system in different industrial environments.

[0082] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An AI-based processing process quality control system for gordon euryale seeds, characterized in that, The application relates to a quality-aware and closed-loop control system for starch processing, comprising: a quality-aware quantification module for real-time non-destructive detection of material units entering the processing flow, and for analyzing and quantifying the collected original spectrum data into an initial starch crystallinity index through a pre-established chemometrics model; a quality dynamic evolution prediction module for predicting the quality evolution trajectory of the material units in future processing time through a dynamic evolution prediction model based on the initial starch crystallinity index, the current quality state and real-time process parameters; a process parameter dynamic optimization module for comparing the quality evolution trajectory with a preset ideal target quality trajectory, and for calculating the optimal process parameter combination at the next moment through a model predictive control optimization framework; a closed-loop feedback control execution module for issuing the optimal process parameter combination to the execution mechanism of the processing equipment to complete real-time regulation and control of the processing process, and for performing a new round of quality evolution trajectory prediction, optimization and control based on the quality state of the regulated material units, thereby forming a closed-loop feedback. The dynamic evolution prediction model is composed of a quality state discrete time evolution equation and a quality instantaneous change rate constitutive model; the quality state discrete time evolution equation is used for iteratively calculating the quality state at the next moment based on the current quality state and the quality instantaneous change rate; the quality instantaneous change rate constitutive model is used for establishing a functional relationship between the quality instantaneous change rate and the real-time process parameters, the current quality state and the initial starch crystallinity index; the initial starch crystallinity index is introduced as a raw material quality correction term to represent the characteristic that the higher the initial crystallinity, the slower the raw material quality change rate.

2. The AI-based processing quality control system of gordon euryale according to claim 1, characterized in that, The quality-aware quantification module comprises a near-infrared spectrum detection system arranged on a material conveying line, and the near-infrared spectrum detection system is used for collecting the original spectrum data of each material unit. 3.The AI-based processing process quality control system of gordon euryale seed according to claim 2, characterized in that, The chemometrics model comprises the following steps: collecting Ginkgo seed samples with different starch crystallinity; using an X-ray diffraction method to measure the crystallinity value of the Ginkgo seed samples, and synchronously collecting the near-infrared spectrum thereof; and using a partial least squares regression algorithm to train, taking the near-infrared spectrum as input and taking the crystallinity value as output.

4. The AI-based processing quality control system of gordon euryale according to claim 1, characterized in that, The model predictive control optimization framework calculates the optimal process parameter combination by solving a cost function; the cost function comprises a tracking error term for minimizing the deviation between the quality evolution trajectory and the ideal target quality trajectory, and a control cost term for punishing the change of process parameters.

5. The AI-based processing quality control system of gordon euryale seed according to claim 4, characterized in that, The control cost term is a weighted control increment penalty term, which uses the normal operation range of each process parameter to normalize the change amount of the control quantity, and adjusts the punishment degree of different control quantity changes through a weight matrix.

6. The AI-based processing quality control system of Gegen fruit according to claim 1, characterized in that, The closed-loop feedback control execution module issues the optimal process parameter combination to a programmable logic controller, which drives a heater, a valve or a motor.

7. The AI-based processing quality control system of gordon euryale seed according to claim 1, characterized in that, The current quality state is obtained by real-time measurement through an online sensor arranged on the processing equipment, or is obtained through a model-based state estimation method combined with the predicted value of the last control period and the model after correction when there is no online sensor.

Images