Graphene-based functional nano composite electrode preparation system based on AI parameter optimization

Abstract

The invention relates to the technical field of new energy material and electrochemical device manufacturing, and discloses a graphene-based functional nano-composite electrode preparation system based on AI parameter optimization, which integrates physical modules such as intelligent material supply, synthesis modification, slurry forming, post-treatment and the like with online characterization equipment into a whole. And closed-loop control is realized through the central AI optimization unit. And the central AI unit establishes a prediction model of process parameters and electrode performance by utilizing a deep neural network, reversely optimizes the process parameters according to performance requirements by adopting a multi-target optimization algorithm, and drives a system to iteratively execute a'preparation-test-optimization 'cycle until a high-performance electrode meeting a target is obtained. According to the method, automatic, accurate and efficient reverse design from performance requirements to process parameters is achieved, the problems that a traditional preparation method depends on experience and is low in efficiency and poor in repeatability are solved, and the research and development efficiency and the product performance limit of the electrode material are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of new energy materials and electrochemical device manufacturing technology, and more specifically to a graphene-based functional nanocomposite electrode preparation system based on AI parameter optimization. Background Technology

[0002] Graphene-based nanocomposites are key materials for constructing high-performance electrochemical electrodes. Their traditional preparation process typically involves multiple discrete steps, including the composite of functional components with graphene, slurry preparation, coating and molding, and heat treatment. This process suffers from significant bottlenecks. 1) Process parameters (such as ratio, temperature, and time) are highly dependent on the experience of the experimenters. Finding the optimal combination through trial and error is extremely inefficient and it is difficult to obtain the global optimal solution. 2) Manual or semi-automatic operation leads to poor batch repeatability and large fluctuations in the microstructure (such as dispersion uniformity and defect density) and macroscopic properties (such as specific capacity and cycle life) of the material; 3) The mapping relationship from raw materials to final electrode performance is a complex "black box" affected by multivariate nonlinear coupling. Traditional methods are difficult to establish accurate quantitative models and cannot achieve performance-oriented precision preparation.

[0003] In existing technologies, although some automated equipment is applied to a single process (such as automatic coating) or offline experimental design (such as response surface methodology) is used for parameter optimization, none of them have formed a fully closed-loop intelligent system that integrates "preparation-characterization-decision-optimization" from raw materials to finished products.

[0004] Based on this, the present invention proposes a graphene-based functional nanocomposite electrode fabrication system based on AI parameter optimization to solve the above problems. Summary of the Invention

[0005] The technical problem this invention aims to solve is to provide an integrated, automated, and intelligent fabrication system and method to address the shortcomings of existing graphene-based electrode fabrication processes, which are characterized by dispersion, reliance on experience, poor controllability, and difficulty in accurately linking process and performance. This system can automatically and continuously optimize the entire process parameters according to target performance requirements using AI algorithms, and stably output high-performance electrode products.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A graphene-based functional nanocomposite electrode fabrication system based on AI parameter optimization is characterized in that the system is a physical system tightly integrated with a physical hardware layer, a data sensing layer, and an intelligent decision-making layer, comprising: Physical hardware execution layer: The intelligent feeding and pretreatment module includes multiple raw material storage tanks (for storing graphene oxide dispersions, metal salt precursor solutions, etc.) connected to precision metering pumps and mass flow controllers, as well as a premixing reactor with mechanical stirring and circulating water bath temperature control functions. All actuators receive control commands via an industrial fieldbus.

[0007] Integrated Synthesis and Modification Module: This is a modular platform that integrates at least a hydrothermal / solvothermal reaction unit (an autoclave with a heating mantle and pressure sensor) and an electrochemical deposition unit (including a three-electrode system and a potentiostat). The module is equipped with an automated material transfer device (such as a robotic arm or pneumatic conveyor line).

[0008] Adaptive slurry preparation and electrode forming module: includes a high-speed dispersion homogenizer with an online viscosity sensor and pH meter, and a slurry delivery system that can be linked with a precision coating machine or 3D printing equipment.

[0009] Intelligent post-processing module: mainly a programmable temperature-controlled tubular atmosphere furnace, whose temperature zone, heating program and atmosphere flow rate are programmable.

[0010] Online multi-dimensional characterization and performance evaluation module: a) Structural characterization unit: It integrates a Raman spectroscopy probe and a small scanning electron microscope (SEM) sample stage, and uses an automated sampling robot to obtain trace samples from the synthesized and post-processed materials for analysis.

[0011] b) Performance evaluation unit: Includes automatic punching, stacking and liquid injection devices, used to quickly assemble dry electrode sheets into coin cell analog batteries and send them to a multi-channel electrochemical workstation for standardized tests such as constant current charge-discharge and cyclic voltammetry.

[0012] Data perception and transmission layer: All the sensors in the aforementioned physical modules (temperature, pressure, flow rate, viscosity, spectral data, and electrochemical signals) upload real-time data to the central database via industrial Ethernet or OPC UA protocol through data acquisition cards or the device's built-in interfaces. Simultaneously, the status and control parameters of the actuators in each module are also recorded synchronously.

[0013] Intelligent Decision-Making Layer (Central AI Optimization Unit): This unit is a software system running on an industrial server, and it is the brain of this invention. Its technical components include: A unified spatiotemporal identifier for the process-performance database: A unique ID is established for each batch of preparation tasks, and the complete set of process parameters X (from control commands and sensor readings of each module) and the set of performance indicators Y (from structural and electrochemical data from the characterization and evaluation module) for that batch are stored synchronously in time sequence.

[0014] The process-performance dynamic prediction model employs a deep neural network (such as a multilayer perceptron (MLP) or a combination of a convolutional neural network (CNN) and a long short-term memory (LSTM) network) as the core predictor. The input layer nodes correspond to the dimension of the process parameter set X, and the output layer nodes correspond to the dimension of the performance index set Y. The model is trained under supervised supervision using historical {X, Y} data pairs from a database, and the weights are optimized through backpropagation, aiming to establish a high-precision nonlinear mapping from X to Y.

[0015] Multi-objective inverse optimization engine: Employs a non-dominated sorting genetic algorithm (NSGA-II) or a Bayesian optimization algorithm based on Pareto optimality. Users set objectives (e.g., "maximize specific capacity while minimizing the rate of internal resistance growth") through a user interface. The optimization engine uses the prediction model as its internal evaluation function and performs a global search within the feasible region of process parameters to find a new combination of process parameters X_opt that optimizes the Pareto front of the prediction performance.

[0016] The control command generation and distribution module parses the X_opt vector output by the optimization engine into a specific sequence of control commands that can be executed by each physical module (e.g., metering pump flow rate = 5.2 mL / min, reactor temperature = 158°C, constant potential = 0.8V vs. SCE, annealing program = [300°C, 120 min; 450°C, 60 min]), and distributes them to the corresponding controllers through the communication protocol.

[0017] The system's workflow forms a reinforcement learning closed loop of "perception-decision-execution-learning": the system performs a round of preparation and testing to obtain new data {X_i,Y_i} → updates the prediction model → the optimization engine calculates the next round of parameters X_{i+1} → issues the execution → obtains new data {X_{i+1}, Y_{i+1}}, and so on, until the electrode performance reaches the predetermined target or the optimization converges.

[0018] Another object of the present invention is to provide a method for preparing graphene-based functional nanocomposite electrodes using the above-described system, characterized by comprising the following steps: S1: Set the target electrode performance indicators and process parameter boundary conditions in the central AI optimization unit; S2: Based on initial parameters or historical optimal parameters, the system initiates the first round of fully automated preparation process, completes online characterization and performance evaluation, and obtains initial data pairs {X1,Y1}. S3: The central AI optimization unit uses accumulated data to train or update the process-performance prediction model; S4: Based on the updated model and preset objectives, the multi-objective optimization engine generates a set of optimized process parameters X_new; S5: Decompose X_new into control instructions to drive each physical module to execute a new round of preparation and testing; S6: Repeat steps S3-S5 to form a continuous optimization cycle. When the performance improvement is less than the set threshold or the absolute performance meets the requirements for N consecutive rounds (N≥3), the optimization is considered complete, and the final process solution and the corresponding high-performance electrode product are output.

[0019] The technical effects and advantages of this invention are as follows: This invention transforms the traditional trial-and-error process, which relies on experience, into an automated global optimization driven by AI models. This can shorten the development cycle of new material formulations and processes from months or years to weeks.

[0020] This invention enables the stable fabrication of electrodes with performance close to the theoretical limit by precisely executing the optimal parameters output by AI, and significantly reduces batch-to-batch fluctuations in key performance characteristics (e.g., controlling performance deviation from >10% to <3%).

[0021] The system proposed in this invention automatically builds and continuously improves a "process-performance" database and prediction model during operation, forming a reusable digital process knowledge asset, which significantly reduces the dependence on the experience of core personnel.

[0022] In this invention, the same hardware system can be reconfigured through software to quickly adapt and develop different types of electrode products (such as high-power and high-energy types), thus realizing flexible R&D and production.

[0023] This invention can significantly reduce the number of experimental rounds, saving expensive raw materials and energy. It locks in the optimal process in one go, avoiding quality fluctuations and waste in large-scale production, thereby reducing overall costs. Attached Figure Description

[0024] The invention will now be further described with reference to the accompanying drawings.

[0025] Figure 1 This is a system block diagram of the graphene-based functional nanocomposite electrode fabrication system based on AI parameter optimization of the present invention; Figure 2 This is a flowchart of the method for preparing graphene-based functional nanocomposite electrodes based on AI parameter optimization according to the present invention. Detailed Implementation

[0026] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. These embodiments are only used to explain the present invention and are not intended to limit the scope of protection of the present invention.

[0027] Please see Figure 1As shown, a graphene-based functional nanocomposite electrode fabrication system based on AI parameter optimization includes a physical execution subsystem, a data sensing subsystem, and a central AI optimization subsystem. The physical execution subsystem includes components connected via material transfer devices according to the process flow: The intelligent feeding and pre-processing module is used for the precise metering, conveying and pre-mixing of raw materials; An integrated synthesis and modification module is used to realize the composite of graphene and functional nanomaterials, integrating at least a hydrothermal reaction unit and an electrochemical deposition unit; An adaptive slurry preparation and electrode forming module is used to prepare composite powder into a slurry and then coat it. The intelligent post-processing module is used for heat treatment of the formed electrodes; An online multi-dimensional characterization and performance evaluation module is used for structural characterization and rapid electrochemical testing of intermediate products and final electrodes; The data sensing subsystem is used to collect process parameter data and characterization performance data of each physical module in real time and upload them to the database; The central AI optimization subsystem includes: A process-performance database is used to store sets of process parameters and performance indicators that are spatially and temporally correlated. The machine learning-based process-performance prediction model is trained using historical data in a database to predict performance indicators based on input process parameters. The multi-objective optimization engine is used to call the prediction model based on the electrode performance target set by the user, search within the process parameter space, and output the optimized process parameter scheme. The control interface is used to decompose the optimized process parameter scheme and issue it as control commands to each module in the physical execution subsystem.

[0028] In one embodiment, the physical execution subsystem is implemented as follows: Intelligent feeding and pretreatment module: Precision injection pumps from companies such as Metrohm (Switzerland) can be used as metering pumps, combined with an Omega brand mass flow controller, and connected to a Siemens S7-1200 PLC via Modbus RTU protocol. The premixing reactor can be an IKA brand jacketed reactor, with the stirring motor and circulating water bath (such as Julabo brand) connected to the same PLC. Material piping is made of PFA material.

[0029] Integrated synthesis and modification module: The hydrothermal reaction unit can utilize a 100mL PTFE-lined high-pressure reactor from Xi'an Yibei, equipped with a Heidolph electric heating mantle and a WIKA pressure sensor. The electrochemical deposition unit employs a Gamry Reference 600+ potentiostat, with the three-electrode system placed in a custom-designed electrolytic cell. The two units are connected via an Epson six-axis robotic arm for vessel transfer.

[0030] Adaptive slurry preparation and electrode forming module: The dispersion homogenizer adopts the German IKA Ultra-Turrax series, equipped with a Brookfield online viscometer. The coating machine adopts the Shenzhen Xinjiatuo automatic blade coating machine. The two are coordinated by PLC, and the slurry is delivered by a corrosion-resistant peristaltic pump.

[0031] Intelligent post-processing module: It adopts Hefei Kejing's OTF-1200X tube furnace, equipped with Eurotherm brand multi-segment programmable temperature controller and high-purity argon gas path.

[0032] Online multi-dimensional characterization and performance evaluation module: The Raman spectrometer used is the portable i-Raman Pro from BitaTek, with the probe fixed above the sampling point via optical fiber. The SEM sample stage is connected to the sample chamber of a Phenom desktop scanning electron microscope. Electrochemical testing uses a Shanghai Chenhua CHI760E multi-channel workstation, and the button cell automated assembly line is designed and customized according to the standards of MTI Corporation of South Korea.

[0033] In one embodiment, the data sensing subsystem is implemented as follows: all the signals from the aforementioned sensors (flow rate, temperature, pressure, viscosity, spectral signals, current and voltage signals) are connected to an industrial computer with configuration software (such as KingSCADA) installed, either through their respective transmitters or directly through the device communication port (such as RS-485 or Ethernet). This computer acts as a data acquisition server, packaging the data into a unified JSON format and sending it to the central AI server via the TCP / IP protocol.

[0034] In one embodiment, the central AI optimization subsystem is implemented as follows: This subsystem is deployed on a server running Ubuntu and a Python environment. The database uses PostgreSQL, with three core tables: `batch_info` (batch information), `process_parameters` (process parameters), and `performance_metrics` (performance metrics), linked by `batch_id`. The prediction model uses the PyTorch framework to build a four-layer fully connected neural network (input layer, two hidden layers each with 256 nodes, and an output layer). The optimization engine is implemented using the NSGA-II algorithm from the pymoo library. The control command generation module is a Python script that parses the optimized output parameter vector into specific command strings (such as SCPI commands and Modbus write commands) for different device controllers, and sends them out via the corresponding serial port or Ethernet port.

[0035] The process-performance prediction model based on machine learning is a deep neural network model, in which the number of nodes in the input layer and the output layer correspond to the dimensions of the process parameter set and the performance index set, respectively.

[0036] In practice, the number of input layer nodes is determined based on the actual number of process parameters collected. For example, if 15 key parameters are selected (such as concentration, temperature, and time), the input layer will have 15 nodes. The number of output layer nodes is determined based on the number of performance indicators of interest. For example, if 5 indicators are selected (such as specific capacity, first-time efficiency, and retention rate after 100 cycles), the output layer will have 5 nodes. Both hidden layers use the ReLU activation function, and the output layer uses a linear activation function. During training, historical data from the database is used, divided into training and validation sets in a 7:3 ratio. The Adam optimizer is used, with mean squared error (MSE) as the loss function. Training stops when the validation set loss no longer decreases for 10 consecutive epochs. The model is saved as a .pt file for the optimization engine to use.

[0037] The multi-objective optimization engine uses either the Non-Dominated Sorting Genetic Algorithm (NSGA-II) or the Bayesian optimization algorithm.

[0038] In one embodiment, taking NSGA-II as an example, the pymoo library is used, the population size is set to 40, simulated binary crossover (SBX) is used with a probability of 0.9 and a distribution exponent of 15, and multinomial mutation is used with a probability of 0.1 and a distribution exponent of 20. The maximum number of iterations is set to 50. The objective function is defined as: f2 = Predicted internal resistance growth rate (minimize internal resistance growth). Constraints are set according to equipment limits (e.g., temperature range 0-200°C). The optimization engine calls the pre-trained DNN model to evaluate the objective function value for each individual.

[0039] The online multidimensional characterization and performance evaluation module includes an automatic sampling device, a Raman spectrometer, a scanning electron microscope sample stage, and an automated assembly line and multi-channel electrochemical workstation for rapid assembly and testing of coin cell analog batteries.

[0040] In practice, the automated sampling device uses a miniature triaxial linear module (such as the HIWIN brand) to drive a quartz capillary sampling needle. Under PLC control, it extracts approximately 10 microliters of sample droplets or trace amounts of powder from the sampling valves at the outlets of the synthesis reactor and post-processing furnace, respectively, and adds them to the slide of the Raman spectrometer and the sample post of the SEM. The Raman spectrometer automatically acquires spectra and calculates the intensity ratio (ID / IG) of the D peak to the G peak under 532nm laser light with an integration time of 2 seconds. The SEM automatically captures backscattered electron images of five fields of view in low-pressure mode (0.1mbar), and estimates the average particle size and distribution uniformity using built-in image analysis software. The automated assembly line for coin cells (CR2032) completes electrode cutting, separator placement, lithium sheet placement, quantitative electrolyte addition, and battery casing encapsulation within an argon-filled glove box. The assembled batteries were then moved into a multi-channel test rack outside the glove box, where the electrochemical workstation automatically executed a predetermined test protocol (such as 0.1C constant current charge and discharge for 3 cycles).

[0041] The data sensing subsystem enables data communication between various devices and the central AI optimization subsystem via industrial Ethernet or OPC UA protocol.

[0042] In implementation, for smart devices supporting Ethernet (such as potentiostats, electrochemical workstations, and some temperature controllers), they are directly connected to the local area network, configured with fixed IP addresses, and communicate using the SDK provided by the device manufacturer or a custom TCP-based protocol. For traditional devices that only support serial ports (such as some metering pumps and flow meters), a MOXA-branded serial port server is used to convert them into Ethernet signals. On the data acquisition server, an OPC UA server (e.g., developed using the opcua-asyncio library) is deployed to uniformly map data from different protocols into the OPC UA information model. The central AI optimization subsystem acts as an OPC UA client, subscribing to the required data nodes to achieve real-time, unified data acquisition.

[0043] Please see Figure 2 As shown, a method for preparing graphene-based functional nanocomposite electrodes based on AI parameter optimization is presented, the method comprising the following steps: S1: Set the target electrode performance indicators and process parameter constraints; S2: Perform the first preparation and testing cycle to obtain initial process-performance data pairs; S3: Utilize accumulated data to update the process-performance prediction model in the central AI optimization subsystem; S4: Based on the updated model and target indicators, a new set of optimized process parameters is generated through a multi-objective optimization engine; S5: Execute the new preparation and testing cycle; S6: Repeat steps S3 to S5 until the electrode performance reaches the predetermined optimization target or the optimization process converges.

[0044] In one specific embodiment, the implementation steps are as follows: S1: In the web graphical interface of the central AI optimization subsystem (e.g., developed using the Flask framework), input the target: initial discharge specific capacity > 350 mAh / g, capacity retention rate after 100 cycles > 95%. Set the process boundaries: hydrothermal temperature: 120-180°C, precursor concentration: 0.1-0.5M.

[0045] S2: The system selects a set of benchmark parameters close to the median from the database to start. The robotic arm loads the mixed precursor solution into the autoclave and reacts according to the set program. After completion, the slurry is coated, dried, and annealed to form electrode sheets. Online Raman, SEM, and coin cell tests are then performed, and the data is automatically stored in the database, generating a record for batch B001.

[0046] S3: The central AI server detects new data B001 and automatically triggers the incremental training process of the model. It uses all data (including B001) to fine-tune the existing DNN model for 10 epochs and updates the model weights.

[0047] S4: The optimization engine loads the new model and runs the NSGA-II algorithm for 50 generations with the target set in S1. It selects the optimal solution that balances specific capacity and retention rate from the Pareto optimal solution set and outputs a new set of process parameters P_new.

[0048] S5: The control instruction generation module converts P_new into specific instructions: {"pump_1_flow":4.7,"reactor_temp":165,...}, and writes them to the controllers of each device via OPCUA to start the preparation of batch B002.

[0049] S6: Repeat S2-S5. When the system detects that the measured specific capacity of three consecutive batches (e.g., B003, B004, B005) is all within 355±2 mAh / g and the retention rate is all within 96±0.5%, the performance is considered to have stabilized and converged. The system locks the process parameters of B004 as the optimal solution and can indicate on the interface that the optimization is complete.

[0050] The convergence criterion in step S6 is: the improvement of the core performance indicators of the electrodes prepared in M ​​consecutive rounds (M≥3) is less than the preset threshold.

[0051] In the software implementation, the following steps are taken: After each round of testing, the system extracts the core performance metrics (such as specific capacity Y_cap) from the database for the most recent M rounds (e.g., M=3). The mean μ and standard deviation σ of these M data points are calculated. A threshold δ is set (e.g., δ=1%ofμ). The judgment condition is: σ<δ and |Y_cap_current-μ|<δ. If the above conditions are met consecutively, a "convergence" flag is triggered, the optimization loop stops, and a report is output. This logic is implemented in Python and runs as a standalone monitoring service in the background.

[0052] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A graphene-based functional nanocomposite electrode fabrication system based on AI parameter optimization, characterized in that, It includes a physical execution subsystem, a data perception subsystem, and a central AI optimization subsystem; The physical execution subsystem includes components connected via material transfer devices according to the process flow: The intelligent feeding and pre-processing module is used for the precise metering, conveying and pre-mixing of raw materials; An integrated synthesis and modification module is used to realize the composite of graphene and functional nanomaterials, integrating at least a hydrothermal reaction unit and an electrochemical deposition unit; An adaptive slurry preparation and electrode forming module is used to prepare composite powder into a slurry and then coat it. The intelligent post-processing module is used for heat treatment of the formed electrodes; An online multi-dimensional characterization and performance evaluation module is used for structural characterization and rapid electrochemical testing of intermediate products and final electrodes; The data sensing subsystem is used to collect process parameter data and characterization performance data of each physical module in real time and upload them to the database. The central AI optimization subsystem includes: A process-performance database is used to store sets of process parameters and performance indicators that are spatially and temporally correlated. The machine learning-based process-performance prediction model is trained using historical data in the database to predict performance indicators based on input process parameters. The multi-objective optimization engine is used to call the prediction model according to the electrode performance target set by the user, search in the process parameter space, and output the optimized process parameter scheme. The control interface is used to decompose the optimized process parameter scheme and issue it as control commands to each module in the physical execution subsystem.

2. The system according to claim 1, characterized in that, The machine learning-based process-performance prediction model is a deep neural network model, with the number of nodes in its input and output layers corresponding to the dimensions of the process parameter set and performance index set, respectively.

3. The system according to claim 1, characterized in that, The multi-objective optimization engine employs either the Non-Dominated Sorting Genetic Algorithm (NSGA-II) or the Bayesian optimization algorithm.

4. The system according to claim 1, characterized in that, The online multidimensional characterization and performance evaluation module includes an automatic sampling device, a Raman spectrometer, a scanning electron microscope sample stage, and an automated assembly line and multi-channel electrochemical workstation for rapid assembly and testing of coin cell analog batteries.

5. The system according to claim 1, characterized in that, The data sensing subsystem enables data communication between each device and the central AI optimization subsystem via industrial Ethernet or OPCUA protocol.

6. A method for preparing graphene-based functional nanocomposite electrodes based on AI parameter optimization, using the system described in any one of claims 1 to 5, characterized in that, The method includes the following steps: S1: Set the target electrode performance indicators and process parameter constraints; S2: Perform the first preparation and testing cycle to obtain initial process-performance data pairs; S3: Utilize accumulated data to update the process-performance prediction model in the central AI optimization subsystem; S4: Based on the updated model and target indicators, a new set of optimized process parameters is generated through a multi-objective optimization engine; S5: Execute the new preparation and testing cycle; S6: Repeat steps S3 to S5 until the electrode performance reaches the predetermined optimization target or the optimization process converges.

7. The method according to claim 6, characterized in that, The convergence criterion in step S6 is that the improvement in the core performance indicators of the electrodes prepared in M ​​consecutive rounds (M≥3) is less than the preset threshold.

Images