Agent automatic preparation and dosage control system and method based on artificial intelligence technology

Abstract

The invention discloses an automatic medicament preparation and dosage control system based on an artificial intelligence technology and a method thereof, relates to the technical field of industrial medicament preparation, and solves the problems that although the flow of medicaments can be controlled to a certain extent in the prior art, most medicaments lack intelligent prediction capability and cannot effectively process complex nonlinear working conditions. Comprising a sensor module, an intelligent controller and an actuator module, the sensor module comprises a spectrum analyzer and a weighing module and is used for monitoring components and weight of a medicament in real time, the intelligent controller processes data of the sensor module and predicts medicament dispensing parameters, the actuator module executes medicament metering, water adding and stirring operations, the sensor module monitors a medicament dispensing result in real time and feeds back the medicament dispensing result, and the intelligent controller corrects errors. Through an integrated high-precision sensor, an advanced intelligent control algorithm and a highly-automatic execution mechanism, comprehensive automatic, precise and intelligent management of the medicament preparation process is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial pharmaceutical preparation, in particular to a pharmaceutical automatic preparation and dosage control system based on artificial intelligence technology and a method thereof. BACKGROUND

[0002] In the current technical field of industrial pharmaceutical preparation, traditional pharmaceutical preparation methods generally have various defects, which limit the improvement of production efficiency and the guarantee of product quality, as follows: Large human error: In the traditional pharmaceutical preparation process, the operator needs to rely on personal experience to weigh and mix the pharmaceuticals. This subjective operation method is prone to introduce large errors. For example, the typical concentration deviation can reach ±5%, which directly affects the quality and efficacy of the final product; Low efficiency: In the traditional pharmaceutical preparation process, the weighing, mixing and stirring of pharmaceuticals are usually completed manually. These operations not only consume time, but also are difficult to meet the requirements of high efficiency and rapid response under large-scale continuous production demand; Safety risk: When handling pharmaceuticals with strong corrosive and toxic properties (such as polyacrylamide PAM), manual operation has a high safety risk, which may threaten the health of the operator; Poor adaptability: The traditional pharmaceutical preparation method often uses a fixed mixing mode. This mode cannot adapt to fluctuations in raw material properties or changes in production processes, resulting in pharmaceuticals that may not meet actual production needs.

[0003] The prior art has solved some problems to some extent, but still has limitations. Although it can control the flow of pharmaceuticals to some extent, it lacks intelligent prediction capability and cannot effectively handle complex nonlinear working conditions, so its effect is limited in actual application. In view of the above defects and limitations, the present application aims to provide an automatic pharmaceutical preparation system based on artificial intelligence, which can realize the automation, precision and intelligence of pharmaceutical preparation, thereby improving production efficiency, ensuring product quality, reducing safety risks, and enhancing the adaptability of the system. SUMMARY

[0004] The purpose of the present application is to provide a pharmaceutical automatic preparation and dosage control system based on artificial intelligence technology and a method thereof, which realizes the comprehensive automation, precision and intelligent management of the pharmaceutical preparation process through integrated high-precision sensors, advanced intelligent control algorithms and highly automated actuators.

[0005] The above technical purposes of the present application are achieved by the following technical solutions: A pharmaceutical automatic preparation and dosage control system based on artificial intelligence technology, comprising a sensor module, an intelligent controller and an actuator module. The sensor module includes a spectrum analyzer and a weighing module to monitor the composition and weight of the medicament in real time, the intelligent controller processes the data of the sensor module and predicts the dispensing parameters, the actuator module performs the metering, water adding and stirring operations of the medicament, the sensor module monitors the dispensing results in real time and feeds back, and the intelligent controller corrects the errors.

[0006] Further, the sensor module further includes a conductivity sensor, the spectrum analyzer detects the composition of the medicament in real time, obtains spectrum data, and fuses the spectrum data with the data of the conductivity sensor to correct the concentration of the medicament.

[0007] Further, the intelligent controller adopts a hybrid model of convolutional neural network and long short-term memory network, the convolutional neural network extracts time series features from the sensor module data, captures key information in the data through multi-layer convolution and pooling operations, and the long short-term memory network further processes and memorizes these features, and predicts the optimal medicament concentration and water adding amount by using the time series processing capability.

[0008] Further, the hybrid model further includes a fusion layer to weight the outputs of the convolutional neural network and the long short-term memory network through an attention mechanism to generate the final dispensing instructions.

[0009] Further, the actuator module includes a metering and medicament adding system and a mixing control unit, the metering and medicament adding system includes a powder adding unit and a liquid metering unit, and the mixing control unit includes a stirrer and a temperature control module.

[0010] Further, when the weighing module detects that the powder adding amount reaches 90% of the target value, the actuator module starts the stirrer in advance.

[0011] Further, when the spectrum analysis shows that the dissolution is insufficient, the actuator module increases the stirring speed of the stirrer.

[0012] Further, a database is further included to quickly adapt and optimize the preparation strategy of new medicaments through transfer learning to quickly adapt to new medicament types.

[0013] An automatic medicament preparation and dosage control method based on artificial intelligence technology, comprising the following steps, The intelligent controller loads the historical process parameters from the database, and the sensor module and the actuator module perform self-checking to ensure that the components work normally; The sensor module starts to collect real-time data; The intelligent controller receives the sensor data and predicts the optimal medicament concentration and water adding amount through a hybrid model of convolutional neural network and long short-term memory network; The actuator module automatically completes the preparation of the medicament according to the predicted parameters of the intelligent controller; In the preparation process, the weighing module and the spectrum analyzer monitor the dispensing result in real time, and the intelligent controller corrects the error in time through a closed-loop feedback mechanism to ensure the final dispensing accuracy.

[0014] Further, the closed-loop feedback mechanism includes that the intelligent controller drives the metering pump to add the medicament according to the prediction, the weighing module feeds back the powder weight in real time, and if the deviation exceeds a preset value, the feeding speed is adjusted through PID control.

[0015] In summary, the present application has the following beneficial effects: Multi-modal perception fusion is adopted, single sensor limitations are broken through through spectrum + conductivity + weighing data cross-validation, hybrid AI architecture, i.e. CNN + LSTM + Attention realizes joint modeling of space-time features, edge-cloud collaboration, through edge real-time control, through cloud large-scale simulation and optimization, self-evolution system is adopted, through incremental learning + digital twin driving continuous performance improvement, can be extended to vaccine adjuvant preparation in the pharmaceutical industry, catalyst preparation in the chemical industry and other scenes, through changing sensor types (such as pH meter, viscometer) and adjusting model parameters, quickly adapt to new requirements; The medicament automatic preparation and dosage control system of the present application not only improves the production efficiency of the pharmaceutical, water treatment, chemical and other industries, but also greatly enhances the safety of operation; especially in dangerous or complex industrial environments, the system can realize unmanned operation, greatly reducing the risk of human operation, bringing unprecedented technological innovation and production safety protection to the field of industrial medicament preparation; The automatic medicament preparation system of the present application provides an efficient, reliable and easy-to-operate medicament preparation solution for the above-mentioned industries with its unique innovative design and great application potential. Through the application of the system, the automation level of the production process can be significantly improved, the errors caused by human operation can be greatly reduced, the stability of product quality can be effectively enhanced, and significant improvements can be made in production safety and economic benefits, injecting new vitality and power into the development of related industries. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 is a principle diagram of a medicament automatic preparation and dosage control system based on artificial intelligence technology of the present application; Figure 2 is a structure principle diagram of a medicament automatic preparation and dosage control system based on artificial intelligence technology of the present application; Figure 3 is a schematic diagram of an actuator part in a medicament automatic preparation and dosage control system based on artificial intelligence technology of the present application; Figure 4It is a flowchart of a medicine automatic preparation and dosage control method based on artificial intelligence technology. DETAILED DESCRIPTION

[0017] The specific embodiments of the present application will be further described below with reference to the accompanying drawings, and the embodiments do not constitute a limitation on the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0018] A medicine automatic preparation and dosage control system based on artificial intelligence technology, as shown in Figure 1 focuses on the field of industrial medicine preparation technology, and focuses on the research and development and actual application of automatic medicine dispensing system. It is designed for the industrial-grade accurate preparation of key medicines such as sodium carbonate, polyacrylamide (PAM), magnesium agent, etc. It is suitable for various industrial application scenarios such as pharmaceutical, water treatment, chemical industry, etc. It is mainly composed of four modules of intelligent controller, multi-modal sensor network, high-precision actuator and data management platform. It adopts a closed-loop architecture of "perception-decision-execution-optimization", combines advanced sensor technology, intelligent control algorithm and automatic execution mechanism, and realizes the full-process automation, precision and intelligence of medicine preparation.

[0019] As shown in Figure 1 and Figure 2 , it specifically includes the following components: 1. Sensor module (multi-modal sensor network): including high-precision spectrum analyzer and weighing module, for real-time monitoring of medicine ingredients and weight; The spectrum analyzer can detect the composition of the medicine in real time, obtain spectral data, and fuse the spectral data with conductivity data to correct the medicine concentration and ensure the accuracy of the medicine dispensing. It can also provide high-resolution spectral analysis to identify and quantify specific components in the medicine; The weighing module can use pressure sensors and other devices, and use dynamic filtering algorithms such as Kalman filtering to improve the stability and accuracy of the weighing data; the precision reaches ±0.05g, ensuring the accuracy of medicine preparation.

[0020] Specifically, the sensor module includes: High-spectrum analyzer, the present embodiment uses OceanInsight Flame-S-VIS-NIR to analyze the medicine ingredients (such as PAM hydrolysis degree) in real time; Precise high-precision pressure sensor, the present embodiment uses Sartorius Quintix5102-1S (±0.05g) to dynamically monitor the quality of medicine powder; Mass flow meter, in this embodiment, a model parameter of KROHNE OPTIMASS 7400 (precision ±0.05%) is used to measure the liquid flow; Conductivity sensor, in this embodiment, a model parameter of METTLER-TOLEDO InPro8630 (±0.01 mS / cm) is used to verify the solution concentration; Torque sensor, in this embodiment, a model parameter of HONEYWELL TBF series (range 0-10 N·m) is used to monitor the stirring resistance and prevent motor overload.

[0021] Among them, the sensor module adopts the following data fusion mechanism, Time synchronization: align multi-source data timestamps based on PTP protocol (IEEE1588); Anomaly detection: real-time elimination of abnormal sampling points using Isolation Forest algorithm.

[0022] In this embodiment, during the construction of the sensor network, In terms of concentration monitoring, a METTLER-TOLEDO InPro8630 conductivity sensor is installed at the outlet of the dispensing tank to monitor the solution conductivity in real time (precision ±0.01 mS / cm), and the data is transmitted to the controller through ModbusTCP. An OceanInsight Flame-S-VIS-NIR spectrometer is also added to scan the solution absorption spectrum (wavelength 350-1000 nm) every 5 seconds to identify abnormal drug components (such as impurities mixed in).

[0023] In terms of flow and weight monitoring, a KROHNE OPTIMASS 7400 mass flow meter is installed on the water inlet pipe with a measurement accuracy of ±0.05% and a data sampling rate of 1 kHz. A Sartorius Quintix5102-1S high-precision pressure sensor is integrated into the powder feeding port to dynamically calibrate the metering pump feeding speed, ensuring an accuracy of ±0.05g.

[0024] In summary, this embodiment uses a multi-modal data fusion technology, which is an innovative data fusion framework that integrates various sensor data, including but not limited to pressure sensors, flow meters, pH sensors, and turbidity sensors, as well as process parameters from historical databases. Through high-dimensional data processing and machine learning algorithms such as principal component analysis (PCA) and clustering analysis, the system can identify the internal relationships between data and dynamically optimize the dispensing strategy, thereby improving the efficiency and adaptability of drug preparation; 2. Intelligent controller (AI intelligent computer), As the core processing unit of the system, an intelligent control driven by deep learning is adopted, including a hybrid model integrating an improved convolutional neural network (CNN) and a long short-term memory network (LSTM) for processing data of the sensor module and predicting dispensing parameters, and the hybrid model is used to realize intelligent decision-making of the medicine preparation; Specifically, the CNN layer is responsible for extracting time series features from the water inlet sensor data, and the LSTM layer further processes these features to predict the optimal medicine concentration and water addition amount. The intelligent controller continuously learns to optimize the medicine dispensing strategy and improve the prediction accuracy; the training of the model can use the Adam optimizer and the cross-entropy loss function to improve the learning efficiency and prediction accuracy; The intelligent controller can further realize the control of the medicine preparation precision through a closed-loop feedback correction mechanism, and its precision reaches ±0.1% in the embodiment. The closed-loop feedback correction mechanism includes that the real-time online weighing system uses a high-precision pressure sensor (precision ±0.01g) combined with a microprocessor-controlled weighing module to ensure the accuracy of medicine weighing; the spectral analysis module uses near-infrared (NIR) spectroscopy technology to monitor the composition and concentration of the medicine in real time, and uses support vector machine (SVM) and other classification algorithms to quickly evaluate the quality of the medicine; and the PID control algorithm is used to adjust the action of the actuator in real time to correct any detected deviation, ensuring that the medicine dispensing error is controlled within 0.1%.

[0025] Specifically, the hardware configuration of the intelligent controller includes a main control unit, a coprocessor, and an extended I / O module, The main control unit includes an NVIDIA Jetson AGX Orin embedded AI module (64GB RAM, 32TOPS computing power), which is equipped with an Ubuntu 20.04LTS operating system and a built-in TensorRT 8.6 acceleration engine to support real-time inference (delay ≤50ms); The coprocessor includes an STM32H743 microcontroller responsible for real-time I / O signal processing (response delay ≤1ms), and the extended I / O module is equipped with RS485 and EtherCAT interfaces to connect sensors and actuators.

[0026] The software architecture of the intelligent controller includes an algorithm layer and a control layer, The algorithm layer adopts a CNN-LSTM hybrid prediction model based on the PyTorch framework, with an input dimension of (time step x sensor channel number) and an output of the drug concentration and water addition instruction; the hybrid prediction model specifically includes a convolutional neural network (CNN), a long short-term memory network (LSTM), and a fusion layer, the convolutional neural network (CNN) processes the drug component features (wavelength 350-1000 nm) input by the spectral sensor; the long short-term memory network (LSTM) models the time series data such as flow, pressure, and weight to predict the dynamic proportioning demand; and the fusion layer weights the CNN and LSTM outputs through the attention mechanism (Attention) to generate the final dispensing instruction.

[0027] The algorithm layer can further include an adaptive optimization module, which updates the model parameters (learning rate = 1e-5) by online learning every day to adapt to the batch differences of raw materials, and supports rapid adaptation to new drug types (such as migration from PAM to cationic polyacrylamide) through a pre-trained model library by transfer learning; The adaptive optimization mechanism uses incremental learning, automatically annotates through daily new data, prevents catastrophic forgetting through elastic weight consolidation (EWC), and sets the model version management to retain the last 5 versions, supporting one-key rollback; digital twin simulation can also be used to build a dispensing process digital twin body through ANSYS TwinBuilder, to pre-visualize extreme conditions (such as flow mutation), and in some embodiments, the simulation results can be fed back to the AI model to enhance robustness; The control layer manages the actuator instruction queue through ROS (Robot Operating System), and priority sorting ensures instantaneous response of critical operations (such as emergency stop).

[0028] 3. Actuator module (high-precision actuator): As shown in Figure 3 , the actuator module includes metering pumps, solenoid valves, mixers, and other automated devices for performing drug metering, water addition, and mixing operations; the metering pump precisely controls the flow of the drug, the solenoid valve controls the switching of the medium, and the mixer ensures uniform mixing of the drug.

[0029] The actuator module supports pulse width modulation (PWM) technology to accurately control the actuator's actions; it also includes a torque feedback mechanism to prevent the mixer from overloading through torque feedback, ensuring the uniformity and safety of the mixing process; it can effectively prevent mixing overload and ensure stable operation of the system.

[0030] Specifically, in this embodiment, the actuator module includes the following key execution units: The metering dosing system includes a powder dosing unit and a liquid metering unit. The powder dosing unit adopts a Sommer SC1000 loss-in-weight feeder (precision ±0.1%) and is controlled by a PID closed loop; the liquid metering unit adopts a ProMinent DME12-24V electromagnetic diaphragm pump (flow rate 0.1-10 L / h, repeatability ±0.5%). The mixing control unit includes a stirrer and a temperature control module. The stirrer adopts an IKAR W20 digital (rotational speed 0-2000 rpm) and is equipped with an adaptive PID algorithm to adjust the torque; the temperature control module adopts a Peltier semiconductor heating / cooling device (temperature control range 5-80℃, precision ±0.5℃).

[0031] The linkage logic is as follows: when the high-precision pressure sensor detects that the powder dosing amount reaches 90% of the target value, the stirrer is started in advance (pre-mixing mode); when spectral analysis shows that the dissolution is insufficient, the stirring speed is automatically increased (gradient increase of 10% per time).

[0032] Specifically, during the linkage debugging process, In the dosing system, the powder dosing part: Sommer SC1000 loss-in-weight feeder, dynamically adjusts the screw speed through the PID algorithm, matches the AI predicted dosing rate, the liquid metering part: ProMinent DME electromagnetic diaphragm pump, receives PWM signal to control flow rate (0.1-10 L / h adjustable), repeatability ±0.5%.

[0033] In the mixing control unit, the IKAR W20 stirrer is equipped with torque feedback, the initial rotational speed is set to 500 rpm, and when the viscosity rises (torque > 5 N·m) is detected, it is automatically increased to 800 rpm; the Peltier temperature control module is embedded in the interlayer of the dosing tank, and the heating / cooling power is adjusted according to the target temperature (such as 25℃±0.5℃).

[0034] 4. Database (data management platform), The database is used to store information such as drug formula, historical data, calibration curve, etc. The historical data includes historical process parameters and sensor data, and through transfer learning, it quickly adapts and optimizes the preparation strategy of new drug types, so that the system can quickly adapt to new drug types without the need to re-adjust a large number of parameters; also stores and updates the standard formula and operating parameters of multiple drugs; and provides data analysis and reporting to support continuous improvement and optimization of the system.

[0035] Specifically, the database framework includes a real-time library and a historical library. The real-time library uses InfluxDB to store sensor raw data (sampling frequency 1 kHz, retention 30 days), and the historical library uses MySQL cluster to archive formula records and operation logs (capacity ≥1PB).

[0036] 5. User interface, The user interface is used for system operation and monitoring, providing a platform for user interaction with the system, which provides an intuitive operation interface, allowing the operator to monitor the real-time status of the system, adjust the system settings, and view historical data and system reports; The human-computer interaction interface specifically includes a Web end and a mobile end. The Web end is built with a React framework, supporting 3D process simulation and batch import of formulas (Excel / CSV). The mobile end uses an Android / iOS App to push real-time alerts (such as concentration deviation > 0.2%).

[0037] The innovation of this embodiment lies in its multi-modal data fusion technology, deep learning driven control strategy, and closed-loop feedback mechanism. The integrated application of these technologies not only significantly improves the production efficiency of industries such as pharmaceuticals, water treatment, and chemical engineering, but also greatly enhances the safety of operations, especially in dangerous or complex working conditions, achieving unmanned operation of pharmaceutical preparation and reducing the risk of human error.

[0038] As shown in Figure 4 the dynamic pharmaceutical preparation workflow includes, System initialization: The intelligent controller loads historical process parameters from the database, and the sensor module and actuator module perform self-checking to ensure that all components are working properly. Data collection: The sensor module begins collecting real-time data, including key parameters such as pharmaceutical ingredients, concentration, flow rate, pH value, etc. Data processing and prediction: The intelligent controller receives sensor data and processes it through a CNN+LSTM hybrid model to predict the optimal pharmaceutical concentration and water addition, providing guidance for execution of the preparation.

[0039] Execution of preparation: The actuator module automatically performs precise measurement of pharmaceuticals, water addition, stirring, and other operations based on the predicted parameters from the intelligent controller, completing the preparation of pharmaceuticals.

[0040] Closed-loop feedback: During the preparation process, the weighing module and spectral analyzer monitor the preparation results in real time. Through the closed-loop feedback mechanism, the intelligent controller can promptly correct any errors to ensure that the final preparation accuracy is within ±0.1%.

[0041] Specifically, the system operation flow example (taking PAM solution preparation as an example), Step S10, formula analysis and initialization, including, Step S11, the user inputs target parameters (concentration, volume, temperature) through the Web interface, such as target concentration (0.1%), volume (500L), and temperature (25℃). The system calls the database to match the optimal historical formula. Step S12, initialize the device (device self-test), zero the metering pump, preheat the stirrer, baseline calibration of the spectrometer, automatically verify the sensor zero point, zero the metering pump, and start the stirrer under no load.

[0042] Step S20, multi-modal data acquisition, Real-time acquisition of spectral, weight, flow, conductivity, torque data, upload to edge server through 5G industrial gateway; and data preprocessing, using Kalman filter denoising, Min-Max normalization, time window alignment (1 second granularity); Among them, in the real-time data stream, sensor data is collected synchronously through EtherCAT bus, and the timestamp alignment accuracy is ≤1ms; spectral data is preprocessed by GPU acceleration (baseline correction, denoising) to generate 128-dimensional feature vectors; and set up an abnormal treatment: use the Isolation Forest algorithm to detect outliers, such as conductivity mutation exceeding 3σ range, trigger data rejection and alarm.

[0043] Step S30, data processing and prediction, AI decision and instruction generation, including through hybrid model inference, using CNN+LSTM+Attention for fusion inference, extracting spectral features, time series prediction, and concentration prediction; in some embodiments, the optimal solution can be solved by multi-objective optimization, with the lowest cost (amount of reagent) and the smallest error (concentration deviation <0.1%) as the target, and the constraint conditions including cost (reagent unit price) and quality (concentration error <0.1%), using multi-objective genetic algorithm (NSGA-II) to solve the Pareto optimal solution.

[0044] Step S40, closed-loop execution and feedback, including, Step S41, actuator control, including, the intelligent controller generates a PWM signal to drive the metering pump, the metering pump adds reagent according to the predicted flow curve, and the pulse width corresponds to the flow demand (such as 10L / h corresponding to duty cycle 30%); the high-precision pressure sensor feeds back the powder weight in real time, and if the deviation exceeds ±0.1%, the PID controller is triggered to adjust the speed of the feeder; the stirrer dynamically adjusts the speed according to the torque feedback (PID parameters: Kp=0.8, Ki=0.2, Kd=0.1); Specifically including, Powder addition: the AI model calculates the PAM powder demand as 5.0kg, the feeder adds at a rate of 2.5kg / h, and the high-precision pressure sensor calibrates in real time; the spectrometer monitors the dissolution state, and when undissolved particles are detected, the stirring speed is increased to 1000rpm; Water control: the mass flowmeter injects 495L of pure water according to the AI instruction, and the flow fluctuation is suppressed by the PID controller (overshoot <1%); Mixed control: The temperature control module maintains the solution temperature at 25±0.3℃, and the conductivity sensor verifies that the concentration meets the standard (0.099%-0.101%).

[0045] Step S42, feedback correction, including, Batch Summary: The system generates a report recording the actual dosage (4.98 kg), time taken (7 minutes and 30 seconds), and energy consumption (1.2 kWh). After each batch, the residual between the measured and predicted conductivity values ​​is input into the online learning module to update the model weights (learning rate = 1e-5). If the deviation between the measured and predicted conductivity values ​​is ≥0.1%, the incremental learning module is triggered to update the model. The incremental learning module uses the data from this batch to fine-tune the LSTM parameters and improve the prediction accuracy for the next batch. If there are spectral anomalies (such as impurity peaks), the process is immediately paused and a self-cleaning procedure is initiated.

[0046] In other embodiments, security and fault tolerance mechanisms are also provided, including, Emergency stop protection: When pressure exceeds the limit (>1MPa) or temperature is abnormal (>50℃), the actuator power is immediately cut off and the pressure relief valve is activated; Redundancy design: Dual metering pumps are connected in parallel, and the backup pump takes over the task within 0.5 seconds when the main pump fails; The database is backed up off-site daily to prevent data loss.

[0047] In this embodiment, the system is implemented; 1) Breakthrough in precision: Weighing accuracy ±0.05g, concentration control error ≤0.1%, far exceeding traditional manual (±5%) and PLC control (±1%); spectral component identification accuracy ≥98% (compared to HPLC standard).

[0048] 2) Efficiency improvement: Full-process automation: Drug preparation speed increased by 300% (e.g., PAM solution preparation time reduced from 30 minutes to 8 minutes); supports continuous operation 24 / 7, with a fault self-diagnosis rate of ≥95%.

[0049] 3) Safe and reliable: Fully unmanned operation in hazardous scenarios (strong acids, highly toxic agents); multiple redundancy design (dual metering pumps, backup power supply) ensures system availability ≥99.9%.

[0050] The technical solution of this invention is applicable to multiple industrial fields, including but not limited to: Pharmaceutical industry: precise formulation of raw materials and excipients in the production process of pharmaceuticals, ensuring the stability of drug quality and therapeutic effect; water treatment industry: precise formulation of water treatment agents such as flocculants, disinfectants, pH regulators, etc. to optimize water treatment effect and improve water quality safety; chemical industry: formulation of chemicals in the synthesis, formulation and processing of chemicals to improve product uniformity and production efficiency; environmental protection industry: automatic formulation of agents in wastewater treatment, waste gas treatment and other environmental protection projects to reduce pollutant emissions and improve treatment efficiency; food and beverage industry: precise formulation of food additives and beverage formulations to ensure food safety and consistency of product standards.

[0051] The technical solution of the embodiment has wide applicability and can play a role in many important industrial fields. The present application also includes the following protection range: intelligent formulation method of other industrial agents such as flocculants and corrosion inhibitors; application of extended AI models such as reinforcement learning and federated learning in agent formulation; and any equivalent improvement based on the principles of the present application, including but not limited to improvements in system structure, algorithm, sensor, actuator, etc.

[0052] The above is only a preferred embodiment of the present application and is not intended to limit the present application. Those skilled in the art can make various modifications or equivalent replacements to the present application within the spirit and protection scope of the present application. Such modifications or equivalent replacements shall also be considered to fall within the protection scope of the technical solution of the present application.

Claims

1. An automatic medicine preparation and dosage control system based on artificial intelligence technology, characterized in that: The system comprises a sensor module, an intelligent controller, and an actuator module. The sensor module comprises a spectrometer and a weighing module, which monitor the composition and weight of the medicament in real time. The intelligent controller processes the data from the sensor module and predicts the dispensing parameters. The actuator module performs the metering, water adding, and stirring operations of the medicament. The sensor module monitors the dispensing results in real time and feeds back the information. The intelligent controller corrects the errors.

2. The medicine automatic preparation and dosage control system based on artificial intelligence technology according to claim 1, characterized in that: The sensor module further comprises a conductivity sensor. The spectrometer detects the composition of the medicament in real time, obtains spectral data, and fuses the spectral data with the data from the conductivity sensor to correct the concentration of the medicament.

3. The medicine automatic preparation and dosage control system based on artificial intelligence technology according to claim 1 or 2, characterized in that: The intelligent controller adopts a hybrid model of convolutional neural network and long short-term memory network. The convolutional neural network extracts time series features from the data of the sensor module, captures the key information in the data through multi-layer convolution and pooling operations, and the long short-term memory network further processes and memorizes these features, and predicts the optimal medicament concentration and water adding amount by using its time series processing capability.

4. The medicine automatic preparation and dosage control system based on artificial intelligence technology according to claim 3, characterized in that: The hybrid model further comprises a fusion layer, which weights the outputs of the convolutional neural network and the long short-term memory network through an attention mechanism to generate the final dispensing instructions.

5. The medicine automatic preparation and dosage control system based on artificial intelligence technology according to claim 1, characterized in that: The actuator module comprises a metering and adding system and a mixing control unit. The metering and adding system comprises a powder adding unit and a liquid metering unit. The mixing control unit comprises a stirrer and a temperature control module.

6. The medicine automatic preparation and dosage control system based on artificial intelligence technology according to claim 5, characterized in that: When the weighing module detects that the powder adding amount reaches 90% of the target value, the actuator module starts the stirrer in advance.

7. The medicine automatic preparation and dosage control system based on artificial intelligence technology according to claim 5 or 6, characterized in that: When the spectrometer shows that the dissolution is insufficient, the actuator module increases the stirring speed of the stirrer. 8.The medicine automatic preparation and dosage control system based on artificial intelligence technology according to claim 1, characterized in that: The system further comprises a database, which quickly adapts and optimizes the dispensing strategy for new medicament types through transfer learning.

9. An automatic medicine preparation and dosage control method based on artificial intelligence technology, characterized in that: The system comprises the following steps, The intelligent controller loads the historical process parameters from the database, and the sensor module and the actuator module perform self-checking to ensure that the components are working properly. The sensor module starts collecting real-time data. The intelligent controller receives the sensor data and predicts the optimal medicament concentration and water adding amount through the hybrid model of convolutional neural network and long short-term memory network. The actuator module automatically completes the dispensing of the medicament according to the predicted parameters of the intelligent controller. During the dispensing process, the weighing module and the spectrometer monitor the dispensing results in real time. The intelligent controller corrects the errors in a timely manner through a closed-loop feedback mechanism to ensure the final dispensing accuracy. 10.The method of claim 9, wherein the method further comprises: determining a target volume of the medicine based on the volume of the medicine and the volume of the diluent; and controlling the amount of the medicine to be added to the diluent based on the target volume of the medicine. The closed-loop feedback mechanism comprises that the intelligent controller drives the metering pump to add the medicament according to the prediction. The weighing module feeds back the weight of the powder in real time. If the deviation exceeds the preset value, the feeding speed is adjusted through PID control.

Images