Optimal control method of ai for process parameters of dry granulation and robot system

Abstract

The present application relates to the technical field of granulation process parameter regulation, in particular to an AI optimal regulation method and a robot system for dry granulation process parameters. The regulation method comprises: S1, parameter identification: identifying key process parameters in the dry granulation process; S2, parameter acquisition: acquiring key process parameters and quality attribute parameters in the dry granulation process, and forming a multi-dimensional data lake; S3, constructing a mapping model; S4, parameter optimization: according to the mapping model, calculating the optimization interval of the key process parameters corresponding to the required quality attribute parameters; S5, regulation decision and action: acquiring real-time key process parameters, judging whether each parameter in the key process parameters is in the corresponding optimization interval, and if not, adjusting the key process parameters. The present application realizes automatic tuning of the control system parameters based on AI, and overcomes the problems of traditional production process parameter adjustment, such as untimely adjustment, low process quality control level, low efficiency, high dependence on artificial experience, etc.

Description

Technical Field

[0001] This invention relates to the field of parameter control technology in granulation process, specifically to an AI-based optimal control method and robotic system for parameters in dry granulation process. Background Technology

[0002] Dry granulation is a crucial step in solid dosage form production, and its granulation quality directly affects subsequent tableting, filling, and dissolution of the finished product. Excessively large particle size can negatively impact tablet hardness and dissolution, while excessively small particle size leads to poor flowability and static electricity buildup. High moisture content can cause clumping, while low moisture content can cause granules to become brittle. Therefore, monitoring and controlling key quality attributes such as particle size distribution and moisture content during the dry granulation process helps improve batch-to-batch consistency of drug quality.

[0003] Currently, the production process relies heavily on human experience, and the industry still primarily depends on a closed loop of "manual sampling - laboratory testing - experience-based interpretation - manual parameter adjustment." During granulation, manual sampling and visual inspection or offline endpoint testing are mainly used, with operators adjusting process parameters based on experience. This approach has significant bottlenecks: firstly, manual sampling and laboratory testing are time-consuming, leading to delays in process parameter adjustments and hindering production; secondly, operators have varying levels of experience, and experience-based judgments lack quantifiable indicators, potentially causing quality deviations.

[0004] Furthermore, due to the real-time nature of measurement and control, manual judgment cannot respond in real time to fluctuations in material characteristics or variations in process parameters. The control logic becomes rigid, and manual adjustment or experience-based setting of process parameters such as environmental parameters, airflow, air temperature, material temperature, air outlet temperature, and atomization pressure involves the coupling of multiple variables. Commonly used single-loop or fixed PID algorithms cannot adapt to changes in quality attributes caused by the interaction of parameters, resulting in quality fluctuations and significant batch-to-batch differences. Summary of the Invention

[0005] To address the aforementioned technical problems in existing technologies, this invention provides an AI-based optimal control method and robotic system for dry granulation process parameters. By monitoring process parameters and quality attributes online, real-time feedback control of process parameters is achieved, replacing traditional methods of manual sampling and experience-based adjustment of process parameters. This overcomes problems such as untimely process parameter adjustments, low levels of process quality control, low efficiency, and high reliance on human experience, thereby achieving precise control of pharmaceutical process quality and improving quality consistency.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows:

[0007] In a first aspect, the present invention provides a method for optimal control of AI parameters in a dry granulation process, comprising:

[0008] S1. Parameter identification: Identify key process parameters in the dry granulation process;

[0009] S2. Parameter Acquisition: Acquire key process parameters and quality attribute parameters in the dry granulation process. Establish a correspondence between key process parameters and quality attribute parameters according to time sequence, and combine multiple sets of key process parameters and quality attribute parameters into a multi-dimensional data lake.

[0010] S3. Construct a mapping model: Establish a mapping model between key process parameters and quality attribute parameters through convolutional neural network modeling;

[0011] S4. Parameter optimization: Based on the mapping model, calculate the optimization range of the key process parameters corresponding to the required quality attribute parameters;

[0012] S5. Control Decision and Action: Obtain real-time key process parameters. Based on the optimization range of the key process parameters obtained in S4, determine whether each parameter in the key process parameters is within its corresponding optimization range. If not, adjust the key process parameters until each parameter in the key process parameters is within its corresponding optimization range.

[0013] Furthermore, in step S2, multiple sets of key process parameters and quality attribute parameters are combined to form a multi-dimensional data lake, specifically including:

[0014] The key process parameters and quality attribute parameters are synchronized, outlier correction is performed, and missing value imputation is performed; then all data are aligned and merged to obtain a multi-dimensional data lake.

[0015] Furthermore, in step S3, a mapping model between key process parameters and quality attribute parameters is constructed using a convolutional neural network. Specific methods include:

[0016] S301, Input data representation:

[0017] The key process parameters are represented as a vector. :

[0018]

[0019] in, Indicates the first Key process parameters;

[0020] S302, Convolutional Layer Output:

[0021] After convolution, the feature map is output. Feature map Characterizing the relationship between key process parameters and quality attribute parameters:

[0022]

[0023] in, Indicates the key process parameters input. It is a convolution kernel. It is an activation function; Represents the spatial coordinates of the feature map;

[0024] S303, Output Mapping Relationship:

[0025] The feature map is processed through the fully connected layer of a convolutional neural network. Mapped to quality attribute parameters; the formula for outputting quality attribute parameters is:

[0026]

[0027] in, Indicates the first j One quality attribute parameter; Represents the weight matrix. This is a bias term.

[0028] Furthermore, key process parameters include atomization pressure, ambient temperature and humidity, relative density of incoming material, inlet air temperature, inlet air flow rate, peristaltic pump delivery frequency of extract, material temperature, and outlet air temperature.

[0029] Furthermore, quality attribute parameters include particle size distribution and particle moisture content.

[0030] Furthermore, activation function This is the ReLU function.

[0031] Furthermore, in step S4, based on the mapping model, a non-dominated sorting genetic algorithm is used to calculate the optimization range of the key process parameters corresponding to the required quality attribute parameters; specifically, this includes:

[0032] S401, Initialization:

[0033] Randomly generate parent population Each individual is a vector of key process parameters;

[0034] S402, Fast Non-Dominated Sort and Crowding Calculation:

[0035] right Performing a fast nondominated sort yields several nondominated fronts. ;

[0036] Individual crowding distances are calculated within each front to maintain diversity;

[0037] S403. Selection, Crossover, and Mutation:

[0038] The parent pair was selected using a tournament selection method;

[0039] Generate offspring by performing crossover operators ;

[0040] Apply the mutation operator to the offspring;

[0041] S404, Merging and Generation of New Generation:

[0042] merge ;

[0043] right Perform a non-dominated sort again, selecting the top M individuals based on the frontier and crowding degree. ;

[0044] S405, Termination and Output:

[0045] When the maximum algebra is reached When the population converges, output the first non-dominated frontier. The Pareto optimal solution set consisting of all individuals in the set, and its corresponding parameter range.

[0046] Furthermore, the AI-based optimal control method for parameters in the dry granulation process also includes:

[0047] S6. Model Update: Establish a correspondence between the key process parameters adjusted in S5 and the adjusted real-time quality attribute parameters, and update the mapping model accordingly.

[0048] Furthermore, in step S2, the quality attribute parameters of the dry granulation process are obtained, specifically through the following method:

[0049] Near-infrared spectral cameras are used to acquire spectral images of particle samples, and white light cameras are used to acquire white light images of particle samples. Based on the spectral and white light images, the particle size distribution and particle moisture content of the particle samples are predicted.

[0050] Furthermore, based on spectral and white light images, the particle size distribution and moisture content of the particle samples are predicted, specifically including:

[0051] Particle size distribution and particle moisture content are predicted based on spectral images.

[0052] The specific formula for predicting particle size distribution is as follows:

[0053]

[0054] in, The particle size distribution is predicted from spectral images. These are the spectral features of near-infrared spectral images. It is a function of the regression model that predicts particle size distribution based on spectral characteristics;

[0055] The specific formula for predicting particle moisture content is as follows:

[0056]

[0057] in, It is particulate moisture. These are the spectral features of near-infrared spectral images. It is a function of the regression model that predicts particle moisture based on spectral characteristics;

[0058] The formula for predicting particle size distribution based on white light images is as follows:

[0059]

[0060] in, The particle size distribution is predicted from white light images. These are the geometric features of a white light image. It is a function of the regression model that predicts particle size distribution based on the geometric features of white light images;

[0061] The final particle size distribution is as follows:

[0062]

[0063] in, , These are the weighting coefficients. .

[0064] Furthermore, the gradient descent method is used to determine the weight coefficients. , Define the objective function

[0065]

[0066] Where N represents the total sample size. This represents the true particle size distribution of the i-th sample. This represents the particle size distribution of the i-th sample predicted from the spectral image. This represents the particle size distribution of the i-th sample predicted from the white light image;

[0067] Minimize using gradient descent To obtain the weighting coefficients , .

[0068] Furthermore, the update rule for gradient descent is as follows:

[0069] ;

[0070]

[0071] in, , The updated weighting coefficients, It is the learning rate.

[0072] Secondly, the present invention also provides an AI-based optimal control robot system for dry granulation process parameters, comprising:

[0073] The multimodal intelligent sensing module is used to acquire key process parameters and quality attribute parameters in the dry granulation process and display the status of the dry granulation process in a machine vision manner.

[0074] The data governance module is used to establish a time-series correspondence between key process parameters and quality attribute parameters, and to combine multiple sets of key process parameters and quality attribute parameters into a multi-dimensional data lake.

[0075] The key process parameter identification module is used to collect relevant data from experimental batches and identify the key process parameters that cause quality fluctuations through a causal discovery method based on Bayesian network learning.

[0076] The process mechanism modeling module, based on experimental batch data, uses a CNN algorithm to construct a nonlinear mapping model between key process parameters and quality attribute parameters.

[0077] The parameter optimization module defines the search range for each key process parameter, takes the expected value of the control range of the quality attribute parameter as the optimization target, and uses the optimization algorithm to calculate the optimization space of the key process parameters.

[0078] The humanoid intelligent control decision module is used to optimize the range obtained by the parameter optimization module and to update the mapping model.

[0079] The control and execution module is used to convert the control variables corresponding to the feasible range of key process parameters into control values, which are then input into the PLC parameter online control system to execute production process control.

[0080] Furthermore, the multimodal intelligent sensing module includes:

[0081] The process quality inspection unit is used to acquire quality attribute parameters;

[0082] The process parameter acquisition unit is used to acquire key process parameters;

[0083] The granulation process state calculation vision unit is used to convert the acquired key process parameters into a multi-dimensional state diagram and display the feasible range and real-time value of the key process parameters in real time.

[0084] Furthermore, the process parameter acquisition unit includes: an atomizing pressure sensor, an ambient temperature and humidity sensor, a densitometer, an infusion frequency acquisition device, and a PLC parameter acquisition subunit. The infusion frequency acquisition device is used to acquire the infusion frequency of the extract peristaltic pump, and the PLC parameter acquisition subunit is used to acquire the inlet air temperature, inlet air flow rate, material temperature, and outlet air temperature.

[0085] Furthermore, the robotic system for controlling parameters in the dry granulation process also includes:

[0086] A mobile material handling module is used to acquire particle samples, and the process quality detection unit uses the particle samples acquired by the mobile material handling module to obtain quality attribute parameters.

[0087] Furthermore, the robot system for controlling parameters in the dry granulation process also includes:

[0088] The remote monitoring module is used to monitor the operating status, key process parameters, quality index detection and prediction data of the dry granulation process parameter control robot system in real time. When the quality attribute parameters exceed the control limits, an alarm signal is issued; when the key process parameters deviate from the feasible range, a process quality warning signal is issued.

[0089] Compared with the prior art, the present invention has the following beneficial effects:

[0090] The dry granulation process parameter AI-based optimal control method and robotic system provided by this invention establishes an accurate mapping model between key process parameters and quality attribute parameters by accurately measuring quality attribute parameters. Then, based on the mapping model and the required quality attribute parameters, the optimization range of the key process parameters is obtained. Subsequently, the key process parameters can be controlled according to real-time key process parameters during production. This eliminates the need to measure actual quality attribute parameters, achieving automated optimization of control system parameters based on AI. It overcomes the problems of untimely parameter adjustments, low process quality control levels, low efficiency, and high reliance on manual experience in traditional production processes. Simultaneously, it avoids sample waste and high testing costs caused by frequent sampling.

[0091] This invention optimizes key process parameters and improves the quality of the pharmaceutical process through AI-based optimal control methods. Attached Figure Description

[0092] Figure 1 The flowchart illustrates the AI-based optimal control method for dry granulation process parameters provided by this invention.

[0093] Figure 2 The structural block diagram of the robot system for optimal control of dry granulation process parameters using AI provided by this invention. Detailed Implementation

[0094] The technical solution of the present invention will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are not all embodiments of the present invention. All other embodiments obtained by those skilled in the art without creative effort are within the protection scope of the present invention.

[0095] It should be noted that, unless otherwise specifically stated, the relative arrangement and numerical expressions of the components and steps described in these embodiments should not be construed as limiting the scope of the invention.

[0096] The following description of exemplary embodiments is merely illustrative and is not intended to limit the invention or its application or use in any way. Techniques, methods, and apparatus known to those skilled in the art may not be discussed in detail herein, but where applicable, such techniques, methods, and apparatus should be considered part of this specification.

[0097] Example 1

[0098] This embodiment provides an AI-based optimal control method for parameters in the dry granulation process, such as... Figure 1 As shown, it includes:

[0099] S1. Parameter identification: Identify key process parameters in the dry granulation process;

[0100] Process parameters in dry granulation, including atomization pressure and ambient temperature and humidity, can affect the final quality attributes, but the degree of influence varies. Including every process parameter in the mapping model could lead to excessive data volume, negatively impacting the model's accuracy. Therefore, it is necessary to screen key process parameters to identify those that significantly influence quality attributes.

[0101] In this embodiment, key process parameters can be determined using a causal relationship algorithm, or they can be directly judged based on human experience. Ultimately, the key processes determined in this embodiment are: atomization pressure, ambient temperature and humidity, relative density of incoming material, inlet air temperature, inlet air flow rate, peristaltic pump delivery frequency of extract, material temperature, and outlet air temperature.

[0102] The quality attribute parameters in this embodiment include particle size distribution and particle moisture content.

[0103] S2. Parameter Acquisition: Acquire key process parameters and quality attribute parameters in the dry granulation process. Establish a correspondence between key process parameters and quality attribute parameters according to time sequence, and combine multiple sets of key process parameters and quality attribute parameters into a multi-dimensional data lake.

[0104] The key process parameters and quality attribute parameters are synchronized, outlier correction is performed, and missing value imputation is performed; then all data are aligned and merged to obtain a multi-dimensional data lake.

[0105] Different key process parameters are collected using different sensors, which have different acquisition frequencies, formats, and qualities. Therefore, the data collected from different sensors, as well as the quality attribute parameters, are processed by unifying the format and aligning the data.

[0106] S3. Constructing a mapping model: Using convolutional neural networks, establish a mapping model between key process parameters and quality attribute parameters; specific methods include:

[0107] S301, Input data representation:

[0108] The key process parameters are represented as a vector. :

[0109]

[0110] in, Indicates the first Key process parameters;

[0111] Specifically:

[0112]

[0113] in, Indicates atomization pressure, Indicates ambient temperature. Indicates ambient humidity. Indicates the relative density of the incoming material. Indicates the intake air temperature. Indicates the air intake flow rate. This indicates the infusion frequency of the extract peristaltic pump. Indicates the temperature of the material. This represents the outlet air temperature. Each key process parameter can be a scalar or a series of values ​​with a time-series dimension.

[0114] S302, Convolutional Layer Output:

[0115] After convolution, the feature map is output. Feature map It includes multiple extracted features that can characterize the relationship between key process parameters and quality attribute parameters:

[0116]

[0117] in, Indicates the key process parameters input. It is a convolution kernel. It is an activation function; activation function Represents the spatial coordinates of the feature map;

[0118] S303, Output Mapping Relationship:

[0119] The feature map is processed through the fully connected layer of a convolutional neural network. Mapped to quality attribute parameters; the formula for outputting quality attribute parameters is:

[0120]

[0121] in, Indicates particle size distribution. Indicates the moisture content of the particles; Represents the weight matrix. This is a bias term.

[0122] S4. Parameter Optimization: Based on the mapping model, use a non-dominated sorting genetic algorithm to calculate the optimization range of the key process parameters corresponding to the required quality attribute parameters; specifically including:

[0123] S401, Initialization:

[0124] Randomly generate parent population Each individual is a vector of key process parameters;

[0125] S402, Fast Non-Dominated Sort and Crowding Calculation:

[0126] right Performing a fast nondominated sort yields several nondominated fronts. ;

[0127] Individual crowding distances are calculated within each front to maintain diversity;

[0128] S403. Selection, Crossover, and Mutation:

[0129] The parent pair was selected using a tournament selection method;

[0130] Generate offspring by performing crossover operators ;

[0131] Apply the mutation operator to the offspring;

[0132] S404, Merging and Generation of New Generation:

[0133] merge ;

[0134] right Perform a non-dominated sort again, selecting the top M individuals based on the frontier and crowding degree. ;

[0135] S405, Termination and Output:

[0136] When the maximum algebra is reached When the population converges, output the first non-dominated frontier. The Pareto optimal solution set consisting of all individuals in the set, and its corresponding parameter range.

[0137] S5. Control Decision and Action: Obtain real-time key process parameters. Based on the optimization range of the key process parameters obtained in S4, determine whether each parameter in the key process parameters is within its corresponding optimization range. If not, adjust the key process parameters until each parameter in the key process parameters is within its corresponding optimization range.

[0138] S6. Model Update: Establish a correspondence between the key process parameters adjusted in S5 and the adjusted real-time quality attribute parameters, and update the mapping model accordingly.

[0139] In step S2, the quality attribute parameters in the dry granulation process are obtained using the following method:

[0140] The spectral images of particle samples are acquired using a near-infrared spectral camera, and white light images of the particle samples are acquired using a white light camera. Based on the spectral and white light images, the particle size distribution and moisture content of the particle samples are predicted. Specifically, this includes:

[0141] Particle size distribution and particle moisture content are predicted based on spectral images.

[0142] The specific formula for predicting particle size distribution is as follows:

[0143]

[0144] in, The particle size distribution is predicted from spectral images. These are the spectral features of near-infrared spectral images. It is a function of the regression model that predicts particle size distribution based on spectral characteristics;

[0145] The specific formula for predicting particle moisture content is as follows:

[0146]

[0147] in, It is particulate moisture. These are the spectral features of near-infrared spectral images. It is a function of the regression model that predicts particle moisture based on spectral characteristics;

[0148] The formula for predicting particle size distribution based on white light images is as follows:

[0149]

[0150] in, The particle size distribution is predicted from white light images. These are the geometric features of a white light image. It is a function of the regression model that predicts particle size distribution based on the geometric features of white light images;

[0151] The final particle size distribution is as follows:

[0152]

[0153] in, , These are the weighting coefficients. .

[0154] Determine the weight coefficients using gradient descent. , Define the objective function

[0155]

[0156] Where N represents the total sample size. This represents the true particle size distribution of the i-th sample. This represents the particle size distribution of the i-th sample predicted from the spectral image. This represents the particle size distribution of the i-th sample predicted from the white light image;

[0157] Minimize using gradient descent To obtain the weighting coefficients , .

[0158] The update rule for gradient descent is:

[0159] ;

[0160]

[0161] in, , The updated weighting coefficients, It is the learning rate, used to control the size of the step.

[0162] Example 2

[0163] This embodiment provides an AI-optimized robot system for controlling parameters in the dry granulation process, such as... Figure 2As shown, it includes a mobile material handling module, a multimodal intelligent sensing module, a data governance module, a key process parameter identification module, a process mechanism modeling module, a parameter optimization module, a humanoid intelligent control decision-making module, a control and execution module, and a remote monitoring module.

[0164] The mobile material handling module is mainly divided into two parts: movement and material handling. The movement part consists of a mobile robot that uses laser SLAM navigation to autonomously navigate to the vicinity of the pellet mill's sampling port. The material handling part consists of a material handling hopper clamped at the end of a robotic arm. During sampling, the robot autonomously navigates to the vicinity of the pellet mill's sample outlet, the robotic arm automatically opens the pellet mill's sampling port, and the sample from the pellet mill automatically falls into the material handling hopper.

[0165] The multimodal intelligent sensing module includes a process quality detection unit, a process parameter acquisition unit, and a granulation process status calculation and vision unit.

[0166] The process quality monitoring unit employs a combination of a near-infrared hyperspectral camera and a white light camera to acquire hyperspectral and white light images of samples during granulation via coaxial illumination. These data are used to predict particle size distribution and moisture content profiles for real-time quality monitoring.

[0167] The process parameter acquisition unit includes multiple sensors for real-time acquisition of key process parameters during granulation, such as atomization pressure, ambient temperature and humidity, relative density of incoming material, inlet air temperature, inlet air flow rate, peristaltic pump delivery frequency of extract, material temperature, and outlet air temperature. This sensor data provides raw data support for subsequent data processing and analysis.

[0168] The process parameter acquisition unit includes: an atomizing pressure sensor, an ambient temperature and humidity sensor, an incoming material relative density meter, a liquid infusion frequency acquisition device, and a PLC parameter acquisition subunit. The liquid infusion frequency acquisition device is used to acquire the liquid infusion frequency of the extract peristaltic pump, and the PLC parameter acquisition subunit is used to acquire the inlet air temperature, inlet air flow rate, material temperature, and outlet air temperature.

[0169] The granulation process status calculation vision unit converts key process parameters such as inlet air temperature, inlet air flow rate, extract peristaltic pump delivery frequency, and atomization pressure into a high-dimensional status diagram, and displays the feasible range and real-time values ​​of controllable variables. This imaging method can help operators monitor the status of the granulation process in real time.

[0170] The data governance module manages multimodal data through data aggregation, preprocessing, integration, and fusion. This module can process cross-modal data from different data sources (such as sensors and robots), performing operations such as synchronization, outlier correction, and feature alignment to form a multi-granularity data lake. The data in the data lake includes batch-level and particle-level data, providing a foundation for subsequent model training and optimization.

[0171] During the data aggregation phase, raw data streams are first obtained from different data sources. These data sources typically have different collection frequencies, formats, and qualities.

[0172] During the data preprocessing stage, the raw data is synchronized, outlier correction is performed, and missing values ​​are filled.

[0173] During the data integration and fusion phase, data streams from different modalities are aligned and merged to obtain a unified representation.

[0174] The key process parameter identification module collects relevant data from experimental batches and identifies the key process parameters that cause quality fluctuations through a causal discovery method based on Bayesian network learning, thereby determining the key process parameters in the dry granulation process.

[0175] The process mechanism modeling module, based on relevant experimental batch data, uses a convolutional neural network (CNN) algorithm to perform nonlinear mapping on multimodal key process parameters such as inlet air temperature, inlet air flow rate, material temperature, outlet air temperature, infusion frequency, and ambient temperature and humidity, in order to determine their relationship with quality attribute parameters (particle moisture and particle size distribution) and establish a mapping model.

[0176] The parameter optimization module takes particle moisture and particle size distribution as optimization objectives and uses multi-objective optimization algorithms such as particle swarm optimization, simulated annealing, or non-dominated sorting genetic algorithm to calculate the optimization range of each key process parameter.

[0177] The humanoid intelligent control decision-making module learns PID curves through full-process (time series) simulation, forming a cluster of typical patterns (i.e., patterns corresponding to predicted quality indicators such as particle moisture and particle size distribution) curves for key process parameters such as inlet air temperature, inlet air flow, extract peristaltic pump delivery frequency, and atomization pressure.

[0178] After implementation, the system collects actual production batch time-series data for key process parameters and quality attribute parameters. Specifically, the mobile sampling module sequentially samples at set time intervals and uses a multimodal intelligent sensing module to detect quality attribute parameters such as moisture and particle size of the samples as network output values. Meanwhile, key process parameters such as inlet air temperature and inlet air flow rate before the sampling point within the batch are used as network input values. The system continuously learns, automatically iterates and updates the mapping model, and continuously optimizes the feasible range of key process parameters.

[0179] With the continuous accumulation of production and testing data, a series of process quality pattern curves are obtained through autonomous learning, enabling control decisions to be made in a human-like manner. Specifically, control decision rules are first obtained through reasoning methods such as automatic optimization using time series similarity measurement algorithms, i.e., a control decision model is established. Then, based on the actual variation of key process parameters, the appropriate mode is determined, and the corresponding operation variable values ​​for the key process parameters are automatically set. Under given input conditions, the appropriate mode of parameter control PID curve is automatically generated, and feedback control is further implemented based on the actual production situation.

[0180] The control and execution module converts the control variables corresponding to the feasible range of key process parameters such as atomization pressure, ambient temperature and humidity, relative density of incoming material, inlet air temperature, inlet air flow rate, liquid delivery frequency of the extract peristaltic pump, material temperature, and outlet air temperature into operable control values, which are then input into the PLC parameter online control system to execute production process control.

[0181] Specifically, the parameter optimization module uses algorithms to derive optimal key process parameters, such as inlet air temperature, inlet air flow rate, extract peristaltic pump infusion frequency, and nebulization pressure. These parameters are crucial for ensuring the ideal quality of drug particles. In the control and execution module, these optimal key process parameters are input into the PLC in real time to achieve online feedback control. The PLC receives the key process parameters output by the optimization module and converts them into specific control commands. These control commands are applied to the dry granulation process to ensure that the key process parameters are always maintained within the optimized range.

[0182] Control commands are generated and written to the PLC via OPC UA. Feedforward correction is triggered by the density fluctuation of the incoming extract through the feedforward feedback coordination mechanism, and feedback iteration is triggered by the real-time hyperspectral deviation, thereby ensuring that the system can adjust parameters in a timely manner to avoid quality fluctuations caused by raw material density fluctuations. The control strategy is corrected and optimized through real-time feedback.

[0183] The remote monitoring module can monitor the robot system's operating status, detection data, prediction data, and control parameters in real time. Through a remote communication interface, operators can access and monitor this data from anywhere. When key quality attributes such as particle moisture and particle size distribution exceed set control limits, the system automatically issues an alarm; or when key process parameters deviate from the feasible range, the system automatically issues a process quality warning signal and prompts intervention. Through this module, any abnormalities or deviations in the production process can be quickly detected, and necessary adjustments can be made to ensure efficient and high-quality operation of the production process.

[0184] The above specific 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 examples, 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 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. A method for optimal control of AI parameters in dry granulation process, characterized in that, include: S1. Parameter identification: Identify key process parameters in the dry granulation process; S2. Parameter Acquisition: Acquire key process parameters and quality attribute parameters in the dry granulation process. Establish a correspondence between key process parameters and quality attribute parameters according to time sequence, and combine multiple sets of key process parameters and quality attribute parameters into a multi-dimensional data lake. S3. Construct a mapping model: Establish a mapping model between key process parameters and quality attribute parameters through convolutional neural network modeling; S4. Parameter optimization: Based on the mapping model, calculate the optimization range of the key process parameters corresponding to the required quality attribute parameters; S5. Control Decision and Action: Obtain real-time key process parameters. Based on the optimization range of the key process parameters obtained in S4, determine whether each parameter in the key process parameters is within the corresponding optimization range. If not, adjust the key process parameters until each parameter in the key process parameters is within the corresponding optimization range. Step S2 involves obtaining the quality attribute parameters during the dry granulation process. Specific methods include: Spectral images of particle samples are acquired using a near-infrared spectral camera, and white light images of particle samples are acquired using a white light camera; the particle size distribution of the particle samples is predicted based on the spectral and white light images. Specifically, it includes: Based on spectral images, particle size distribution is predicted. The specific formula for particle size distribution prediction is as follows: in, The particle size distribution is predicted from spectral images. These are the spectral features of near-infrared spectral images. It is a function of the regression model that predicts particle size distribution based on spectral characteristics; The formula for predicting particle size distribution based on white light images is as follows: in, The particle size distribution is predicted from white light images. These are the geometric features of a white light image. It is a function of the regression model that predicts particle size distribution based on the geometric features of white light images; The final particle size distribution is as follows: in, , These are the weighting coefficients. ; Determine the weight coefficients using gradient descent. , Define the objective function: Where N represents the total sample size. This represents the true particle size distribution of the i-th sample. This represents the particle size distribution of the i-th sample predicted from the spectral image. This represents the particle size distribution of the i-th sample predicted from the white light image; Minimize using gradient descent To obtain the weighting coefficients , ; The update rule for gradient descent is: ； in, , The updated weighting coefficients, It is the learning rate.

2. The method for optimal control of AI parameters in the dry granulation process according to claim 1, characterized in that, Step S2 involves assembling multiple sets of key process parameters and quality attribute parameters into a multi-dimensional data lake, specifically including: The key process parameters and quality attribute parameters are synchronized, outlier correction is performed, and missing value imputation is performed; then all data are aligned and merged to obtain a multi-dimensional data lake.

3. The method for optimal control of AI parameters in the dry granulation process according to claim 1, characterized in that, Step S3 utilizes a convolutional neural network to construct a mapping model between key process parameters and quality attribute parameters. Specific methods include: S301, Input data representation: The key process parameters are represented as a vector. : in, Indicates the first Key process parameters; S302, Convolutional Layer Output: After convolution, the feature map is output. Feature map Characterizing the relationship between key process parameters and quality attribute parameters: in, Indicates the key process parameters input. It is a convolution kernel. It is an activation function; Represents the spatial coordinates of the feature map; S303, Output Mapping Relationship: The feature map is processed through the fully connected layer of a convolutional neural network. Mapped to quality attribute parameters; the formula for outputting quality attribute parameters is: in, Indicates the first j One quality attribute parameter; Represents the weight matrix. This is a bias term.

4. The method for optimal control of AI parameters in the dry granulation process according to claim 3, characterized in that, Key process parameters include atomization pressure, ambient temperature and humidity, relative density of incoming material, inlet air temperature, inlet air flow rate, peristaltic pump delivery frequency of extract, material temperature, and outlet air temperature. Quality attribute parameters include particle size distribution and particle moisture content.

5. The method for optimal control of AI parameters in the dry granulation process according to claim 4, characterized in that, In step S4, based on the mapping model, a non-dominated sorting genetic algorithm is used to calculate the optimization range of the key process parameters corresponding to the required quality attribute parameters; specifically, this includes: S401, Initialization: Randomly generate parent population Each individual is a vector of key process parameters; S402, Fast Non-Dominated Sort and Crowding Calculation: right Performing a fast nondominated sort yields several nondominated fronts. ; Individual crowding distances are calculated within each front to maintain diversity; S403. Selection, Crossover, and Mutation: The parent pair was selected using a tournament selection method; Generate offspring by performing crossover operators ; Apply the mutation operator to the offspring; S404, Merging and Generation of New Generation: merge ; right Perform a non-dominated sort again, selecting the top M individuals based on the frontier and crowding degree. ; S405, Termination and Output: When the maximum algebra is reached When the population converges, output the first non-dominated frontier. The Pareto optimal solution set consisting of all individuals in the set, and its corresponding parameter range.

6. The method for optimal control of AI parameters in the dry granulation process according to claim 1, characterized in that, The AI-based optimal control method for parameters in dry granulation also includes: S6. Model Update: Establish a correspondence between the key process parameters adjusted in S5 and the adjusted real-time quality attribute parameters, and update the mapping model accordingly.

7. The method for optimal control of AI parameters in the dry granulation process according to claim 4, characterized in that, Step S2 involves obtaining quality attribute parameters during the dry granulation process. The specific method further includes: White light images of particulate samples are acquired using a white light camera, and the moisture content of the particulate samples is predicted based on the white light images.

8. The method for optimal control of AI parameters in the dry granulation process according to claim 7, characterized in that, Predicting particle size distribution and particle moisture content in particulate samples based on spectral and white light images, specifically including: Predicting particle moisture based on spectral images The specific formula for predicting particle moisture content is as follows: in, It is particulate moisture. These are the spectral features of near-infrared spectral images. It is a function of the regression model that predicts particle moisture based on spectral characteristics.

9. A robot system for optimal control of dry granulation process parameters using AI, used to execute the optimal control method for dry granulation process parameters using AI as described in any one of claims 1-8, characterized in that, include: The multimodal intelligent sensing module is used to acquire key process parameters and quality attribute parameters in the dry granulation process and display the status of the dry granulation process in a machine vision manner. The data governance module is used to establish a time-series correspondence between key process parameters and quality attribute parameters, and to combine multiple sets of key process parameters and quality attribute parameters into a multi-dimensional data lake. The key process parameter identification module is used to collect relevant data from experimental batches and identify the key process parameters that cause quality fluctuations through a causal discovery method based on Bayesian network learning. The process mechanism modeling module, based on experimental batch data, uses a CNN algorithm to construct a nonlinear mapping model between key process parameters and quality attribute parameters. The parameter optimization module defines the search range for each key process parameter, takes the expected value of the control range of the quality attribute parameter as the optimization target, and uses the optimization algorithm to calculate the optimization space of the key process parameters. The humanoid intelligent control decision module is used to optimize the range obtained by the parameter optimization module and to update the mapping model. The control and execution module is used to convert the control variables corresponding to the feasible range of key process parameters into control values, which are then input into the PLC parameter online control system to execute production process control.

10. The AI-based optimal control robot system for dry granulation process parameters according to claim 9, characterized in that, The multimodal intelligent sensing module includes: The process quality inspection unit is used to acquire quality attribute parameters; The process parameter acquisition unit is used to acquire key process parameters; The granulation process state calculation vision unit is used to convert the acquired key process parameters into a multi-dimensional state diagram and display the feasible range and real-time value of the key process parameters in real time.

11. The AI-based optimal control robot system for dry granulation process parameters according to claim 10, characterized in that, The dry granulation process parameter control robot system also includes: A mobile material handling module is used to acquire particle samples, and the process quality detection unit uses the particle samples acquired by the mobile material handling module to obtain quality attribute parameters.

12. The AI-based optimal control robot system for dry granulation process parameters according to claim 9, characterized in that, The dry granulation process parameter control robot system also includes: The remote monitoring module is used to monitor the operating status, key process parameters, quality index detection and prediction data of the dry granulation process parameter control robot system in real time. When the quality attribute parameters exceed the control limits, an alarm signal is issued; when the key process parameters deviate from the feasible range, a process quality warning signal is issued.

Images