A flow field simulation and neural network optimization method for liquid phase preparation of metal powders

Abstract

This invention relates to the field of liquid-phase metal powder preparation and intelligent process optimization, and particularly to a flow field simulation and neural network optimization method for the liquid-phase metal powder preparation process. The method involves establishing a reactor flow field mass transfer simulation model to obtain flow field mass transfer characteristics under different operating conditions; constructing a training dataset using the flow field mass transfer characteristics and experimental powder morphology indicators; and establishing a mapping relationship between process parameters, flow field characteristics, and powder morphology using a neural network. Recommended parameter combinations are output based on target morphology requirements, and verified and corrected through simulation recalculation and / or actual preparation experiments. This invention reduces the reliance on experience-based trial and error in traditional process scale-up and parameter selection, improves the accuracy of controlling metal powder particle size, sphericity, and dispersibility, enhances batch stability, and provides a basis for intelligent design and industrial scale-up of the liquid-phase metal powder preparation process.

Description

Technical Field

[0001] This invention relates to the field of liquid phase preparation of metal powder and intelligent process optimization technology, specifically a flow field simulation and neural network optimization method for the liquid phase preparation process of metal powder. Background Technology

[0002] Metal powders are essential raw materials for conductive pastes, electrical contact materials, electronic packaging, powder metallurgy, additive manufacturing, electromagnetic functional materials, and catalytic materials. The particle size distribution, sphericity, dispersibility, and agglomeration of silver powder, copper powder, nickel powder, cobalt powder, tin powder, zinc powder, iron powder, and their alloy powders directly affect the powder's filling performance, conductive network formation ability, sintering behavior, and subsequent application stability.

[0003] Liquid-phase preparation methods have become an important approach for metal powder preparation due to their mild reaction conditions, good equipment adaptability, and wide range of adjustable particle size and morphology. In liquid-phase reduction, displacement reactions, chemical precipitation, or composite reduction processes, the reduction of metal ions, crystal nucleation, grain growth, particle collision, and agglomeration all occur within a stirred reactor. The velocity field, circulation flow, turbulent kinetic energy distribution, low-velocity retention zone, local concentration peaks in the feeding zone, and concentration gradients within the reactor directly affect local supersaturation, reduction rate, and the microenvironment for particle growth, thus determining the final powder morphology.

[0004] During the process of scaling up from laboratory pilot-scale to pilot-scale or industrial production, even if the material concentration, reaction time, and feed ratio remain constant, changes in parameters such as the reactor inner diameter, liquid level, impeller diameter, impeller height from the bottom, stirring speed, and feed position can lead to significant differences in the three-dimensional flow field and mass transfer state within the reactor. Traditional scale-up methods that rely on trial and error typically require extensive experimental verification, resulting in problems such as long R&D cycles, high raw material consumption, and insufficient batch stability.

[0005] Computational fluid dynamics (CFD) simulations can obtain information on the flow field and mass transfer of components inside a reactor, but relying solely on simulation results is still insufficient to directly determine powder particle size, sphericity, and agglomeration state. Especially in different metal powder systems, there are inherent differences in reaction kinetics, solution viscosity, feeding methods, and particle growth processes, making it difficult to formulate universal parameter optimization criteria based on a single simulation index.

[0006] Neural network models possess the ability to handle multivariate nonlinear mapping relationships, and can correlate reactor structural parameters, stirring process parameters, flow field mass transfer characteristics, and experimental morphological indicators. By combining flow field mass transfer simulation with neural network prediction, and performing feedback corrections through experimental or simulation recalculations, an intelligent parameter optimization method for the liquid-phase preparation process of metal powders can be established, providing a basis for morphological control and large-scale preparation of silver powder, copper powder, and other metal powders. Summary of the Invention

[0007] The purpose of this invention is to provide a flow field simulation and neural network optimization method for the liquid phase preparation process of metal powder, so as to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution: A flow field simulation and neural network optimization method for the liquid phase preparation process of metal powder includes the following steps: S1. Establish a flow field mass transfer simulation model for a metal powder liquid phase preparation reactor. Input the reactor structural parameters, stirring process parameters, feeding parameters and liquid phase physical property parameters, and obtain the flow field mass transfer characteristics under different working conditions through numerical calculation. S2. The flow field mass transfer characteristics obtained in step S1 and the morphology index of the metal powder actually prepared under the corresponding working conditions are used to form a training dataset. A neural network model is used to establish the mapping relationship between the reactor structural parameters, stirring process parameters, feeding parameters, flow field mass transfer characteristics and powder morphology index. S3. Based on the target powder morphology requirements, the trained neural network model outputs a recommended parameter combination, and the recommended parameter combination is verified and corrected through simulation recalculation and / or actual preparation experiments until the process parameters that meet the target powder morphology requirements are obtained.

[0009] Furthermore, the metal powder is one or more of silver powder, copper powder, nickel powder, cobalt powder, tin powder, zinc powder, iron powder and their alloy powders, preferably silver powder or copper powder; the powder morphology indicators include at least one of particle size distribution, average particle size, sphericity, dispersibility and agglomeration degree.

[0010] Further, in step S1, the flow field mass transfer simulation model includes a vessel body, a liquid phase region, a stirring shaft, a stirring paddle, a feeding inlet, a rotating region, and a stationary region; the reactor structural parameters include at least one of the following: reactor inner diameter, vessel height, liquid level height, stirring shaft diameter, stirring paddle type, stirring paddle diameter, paddle width, paddle installation height, and baffle structure; the stirring process parameters include at least one of the following: stirring speed, stirring time, stirring direction, and stirring program; the feeding parameters include at least one of the following: feeding position, feeding speed, feeding time, feeding concentration, and feeding method; the liquid phase physical property parameters include at least one of the following: liquid phase density, liquid phase viscosity, solute diffusion coefficient, and reaction system temperature.

[0011] Further, in step S1, the flow field mass transfer characteristics include one or more of the following: velocity field characteristics, turbulent kinetic energy characteristics, circulating flow characteristics, low-velocity stagnation zone ratio, concentration peak near the feeding area, concentration gradient, and mixing uniformity index.

[0012] Further, in step S2, the neural network model uses the reactor structural parameters, stirring process parameters, feeding parameters, and flow field mass transfer characteristics as input layer nodes, and the powder particle size distribution, average particle size, sphericity, agglomeration degree, and dispersion evaluation values ​​as output layer nodes; the neural network model is a multilayer feedforward neural network, trained with historical simulation data and experimental characterization data, and updated with an error backpropagation algorithm.

[0013] Further, in step S3, the feedback correction specifically includes: inputting the recommended parameter combination output by the neural network model into the flow field mass transfer simulation model for recalculation to obtain the recalculation result; comparing the recalculation result and / or the morphology index of the powder actually prepared based on the recommended parameter combination with the target value; when the deviation between the predicted value and the verification value exceeds a preset threshold, supplementing the verification result into the training dataset, and retraining the neural network model, iterating until the reactor structure parameters, stirring process parameters, and feeding parameters that meet the morphology requirements of the target powder are output.

[0014] As can be seen from the technical solution provided by the present invention above, the flow field simulation and neural network optimization method for the liquid phase preparation process of metal powder provided by the present invention has the following beneficial effects: This invention combines the simulation results of mass transfer in the reactor flow field with experimental powder morphology data to establish a quantitative correspondence between equipment parameters, stirring parameters, mass transfer characteristics of the flow field and powder morphology, thus avoiding the limitations of relying solely on empirical judgment. This invention learns complex nonlinear relationships through a neural network model, and can output recommended parameter combinations based on target particle size, sphericity, dispersion and aggregation degree, reducing multiple rounds of trial and error experiments and shortening the process development cycle; This invention is applicable to the liquid phase preparation process of silver powder, copper powder and various metal or alloy powders. It can provide a basis for parameter selection from laboratory small-scale to pilot-scale or industrial scale-up, and improve the stability of powder morphology and batch consistency. This invention uses simulation recalculation and / or actual preparation experiments for feedback correction, enabling the neural network model to be continuously iterated and updated, gradually improving the reliability and applicability of the recommended parameters. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the flow field simulation and neural network optimization method for the liquid phase preparation process of metal powder according to the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0017] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific embodiments.

[0018] like Figure 1 As shown in the figure, this invention provides a flow field simulation and neural network optimization method for the liquid phase preparation process of metal powder, including the following steps: Step S1: Establish a flow field mass transfer simulation model and obtain flow field mass transfer characteristics: S1-1: Establishing the Geometric Model of the Reactor: Based on the actual structure of the reactor used for liquid phase preparation of metal powder, construct a three-dimensional geometric model including the reactor body, liquid phase region, stirring shaft, stirring paddle, feed inlet, rotating region, and stationary region; among which, the reactor structural parameters include the reactor inner diameter, reactor height, liquid level height, stirring shaft diameter, stirring paddle type, stirring paddle diameter, paddle width, paddle installation height, and baffle structure, etc. S1-2: Set process parameters and physical property parameters: Stirring process parameters include stirring speed, stirring time, stirring direction and stirring program; feeding parameters include feeding location, feeding speed, feeding time, feeding concentration and feeding method; liquid phase physical property parameters include liquid phase density, liquid phase viscosity, solute diffusion coefficient and reaction system temperature; S1-3: Mesh generation and model solution: Refine the mesh in the region near the agitator, the region near the feed inlet, the interface between the rotating and stationary regions, and the wall region; Set up the incompressible fluid model, turbulent model, and component transport model; Define the coupling method between the rotating and stationary regions; Set boundary conditions and solution control parameters, and perform steady-state or transient numerical calculations. S1-4: Extract flow field mass transfer characteristics: Post-process the simulation results and extract one or more of the following characteristics: average velocity, maximum velocity, local velocity in the feeding zone, turbulent kinetic energy and its distribution, circulating flow intensity, proportion of low-speed stagnation zone, peak concentration near the feeding zone, concentration gradient, mixing time, and mixing uniformity index. By changing the reactor structural parameters, stirring process parameters, and feeding parameters, S1-1 to S1-4 were repeatedly executed to obtain a flow field mass transfer characteristic dataset covering different operating conditions. Step S2: Construct the training dataset and build the neural network model: S2-1: Constructing the training dataset: The reactor structural parameters, stirring process parameters, feeding parameters, and their corresponding flow field mass transfer characteristics under different operating conditions are used as input data. The average particle size, particle size distribution, sphericity, dispersibility, and agglomeration degree of the metal powder actually prepared under the corresponding operating conditions are used as output labels to form the training dataset. Among them, the average particle size and particle size distribution are measured by a laser particle size analyzer, and the sphericity, dispersibility, and agglomeration degree are obtained by scanning electron microscopy (SEM) image analysis. S2-2: Data preprocessing: Normalize the input parameters and output indicators, remove outlier data points, and divide the dataset into training set, validation set and test set according to a preset ratio; S2-3: Establishing a Neural Network Model: A multi-layer feedforward neural network structure is adopted. The input layer nodes correspond to the reactor structural parameters, stirring process parameters, feeding parameters, and flow field mass transfer characteristics. The output layer nodes correspond to the powder particle size distribution, average particle size, sphericity, agglomeration degree, and dispersion evaluation values. The number of hidden layers and nodes is determined according to the data scale and problem complexity. The activation function can be ReLU, Sigmoid, or Tanh function. S2-4: Training the neural network model: Using the mean squared error (MSE) or mean absolute error (MAE) between the predicted morphology index and the experimental morphology index as the loss function, the model weights are updated using the backpropagation algorithm; when the validation set loss no longer decreases or the preset convergence condition is met, training is stopped, and a neural network model for predicting powder morphology is obtained. Step S3: Output recommended parameter combinations and verify and provide feedback for correction: S3-1: Input target powder morphology requirements: Based on actual preparation needs, set the target average particle size, target particle size distribution range, target sphericity, target dispersibility, and allowable agglomeration degree; S3-2: Output recommended parameter combination: Input multiple sets of candidate process parameters into the trained neural network model for prediction, and select the recommended parameter combination whose predicted morphology index is closest to the target powder morphology requirement; the recommended parameter combination includes one or more of the following: reactor inner diameter, stirring blade diameter, stirring blade height from bottom, stirring speed, feeding position and feeding speed. S3-3: Simulation Recalculation and Experimental Verification: Re-enter the recommended parameter combination into the flow field mass transfer simulation model established in step S1 for recalculation, and determine whether the flow field mass transfer characteristics meet the process requirements of uniform mixing, reduction of low-speed retention zone and control of peak concentration in the feeding zone; if necessary, conduct actual preparation experiments according to the recommended parameter combination, and characterize the morphology of the obtained powder to obtain actual morphology indicators. S3-4: Feedback Correction: Compare the simulation results and / or the actual powder morphology indicators with the target values; when the deviation between the predicted value and the verification value exceeds the preset threshold, supplement the verification results to the training dataset, and re-execute step S2-4 to update the neural network model weights; iteratively execute S3-2 to S3-4 until the reactor structure parameters, stirring process parameters and feeding parameters that meet the target powder morphology requirements are output.

[0019] Example 2 This embodiment is based on Example 1 and is specifically applied to the preparation of silver powder by liquid-phase reduction method.

[0020] Using silver nitrate solution as the silver source, formaldehyde or ascorbic acid as the reducing agent, and polyvinylpyrrolidone (PVP) or gelatin as the dispersant, a liquid-phase reduction reaction was carried out in a stirred reactor. Based on the target silver powder particle size, sphericity, and dispersibility requirements, different combinations of parameters such as reactor inner diameter, impeller diameter to reactor diameter ratio, stirring speed, and feeding rate were set to simulate the flow field mass transfer.

[0021] The flow field mass transfer characteristics obtained from simulation, such as average velocity, turbulent kinetic energy, circulating flow intensity, proportion of low-velocity retention zone, peak concentration in the feeding zone, and mixing time, are combined with the particle size distribution, sphericity, dispersibility, and agglomeration degree of experimentally prepared silver powder under corresponding operating conditions to form a training dataset. After training the neural network model, the model outputs the following recommended parameter combination when the target silver powder morphology requirements are input: reactor inner diameter 200 mm, impeller diameter 80 mm, impeller height from bottom 40 mm, stirring speed 450 rpm, feeding position 30 mm from liquid surface, and feeding rate 15 mL / min.

[0022] The recommended parameter combination was input into the flow field mass transfer simulation model for recalculation. After confirming that the flow and mass transfer state inside the reactor met the process requirements, three batches of actual preparation experiments were conducted according to this parameter combination. The D50 values ​​of the silver powder obtained in the three batches were 0.98 μm, 1.03 μm, and 1.01 μm, respectively, and the sphericity values ​​were 0.91, 0.89, and 0.90, respectively. The dispersibility of all batches was rated as good, and the consistency between batches met the requirements.

[0023] Example 3 This embodiment is based on Example 1 and is specifically applied to the preparation of copper powder by liquid-phase reduction.

[0024] Copper sulfate solution was used as the copper source, and sodium hypophosphite or sodium borohydride was used as the reducing agent in a stirred reactor for liquid-phase reduction. Based on the target copper powder particle size and dispersibility requirements, different combinations of process parameters were set for flow field mass transfer simulation and experimental preparation. The same method as in Example 2 was used to establish a training dataset, train a neural network model, and output recommended parameter combinations. Simulation recalculation and three batches of actual preparation experiments verified that the obtained copper powder's particle size distribution, sphericity, and dispersibility all met the target requirements.

[0025] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. For those skilled in the art, appropriate adjustments or substitutions can be made to the reactor structure, agitator type, number and structure of neural network layers, number of input features, and powder morphology evaluation indicators without departing from the principles and spirit of the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0026] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A flow field simulation and neural network optimization method for the liquid phase preparation process of metal powder, characterized in that: include: S1. Establish a flow field mass transfer simulation model for a metal powder liquid phase preparation reactor. Input the reactor structural parameters, stirring process parameters, feeding parameters and liquid phase physical property parameters, and obtain the flow field mass transfer characteristics under different working conditions through numerical calculation. S2. The flow field mass transfer characteristics obtained in step S1 and the morphology index of the metal powder actually prepared under the corresponding working conditions are used to form a training dataset. A neural network model is used to establish the mapping relationship between the reactor structural parameters, stirring process parameters, feeding parameters, flow field mass transfer characteristics and powder morphology index. S3. Based on the target powder morphology requirements, the recommended parameter combination is output using the trained neural network model, and the recommended parameter combination is verified and corrected through simulation recalculation and / or actual preparation experiments until the process parameters that meet the target powder morphology requirements are obtained.

2. The flow field simulation and neural network optimization method for the liquid phase preparation process of metal powder according to claim 1, characterized in that: The metal powder is one or more of silver powder, copper powder, nickel powder, cobalt powder, tin powder, zinc powder, iron powder and their alloy powders; the powder morphology indicators include at least one of particle size distribution, average particle size, sphericity, dispersibility and agglomeration degree.

3. The flow field simulation and neural network optimization method for the liquid phase preparation process of metal powder according to claim 2, characterized in that: The metal powder is silver powder or copper powder; the liquid phase preparation method of the metal powder is liquid phase chemical reduction method.

4. The flow field simulation and neural network optimization method for the liquid phase preparation process of metal powder according to claim 1, characterized in that: In step S1, the flow field mass transfer simulation model includes a vessel, a liquid phase region, a stirring shaft, a stirring paddle, a feeding inlet, a rotating region, and a stationary region; the reactor structural parameters include at least one of the following: reactor inner diameter, vessel height, liquid level height, stirring shaft diameter, stirring paddle type, stirring paddle diameter, paddle width, paddle installation height, and baffle structure; the stirring process parameters include at least one of the following: stirring speed, stirring time, stirring direction, and stirring program; the feeding parameters include at least one of the following: feeding position, feeding speed, feeding time, feeding concentration, and feeding method; the liquid phase physical property parameters include at least one of the following: liquid phase density, liquid phase viscosity, solute diffusion coefficient, and reaction system temperature.

5. The flow field simulation and neural network optimization method for the liquid phase preparation process of metal powder according to claim 1, characterized in that: In step S1, the flow field mass transfer characteristics include one or more of the following: velocity field characteristics, turbulent kinetic energy characteristics, circulating flow characteristics, low-velocity retention zone ratio, concentration peak near the feeding area, concentration gradient, and mixing uniformity index.

6. The flow field simulation and neural network optimization method for the liquid phase preparation process of metal powder according to claim 1, characterized in that: In step S2, the neural network model uses the reactor structural parameters, stirring process parameters, feeding parameters, and flow field mass transfer characteristics as input layer nodes, and the powder particle size distribution, average particle size, sphericity, agglomeration degree, and dispersion evaluation values ​​as output layer nodes. The neural network model is a multilayer feedforward neural network, which is trained using historical simulation data and experimental characterization data, and the model weights are updated using an error backpropagation algorithm.

7. The flow field simulation and neural network optimization method for the liquid phase preparation process of metal powder according to claim 1, characterized in that: In step S3, the feedback correction specifically includes: inputting the recommended parameter combination output by the neural network model into the flow field mass transfer simulation model for recalculation to obtain the recalculation result; comparing the recalculation result and / or the morphology index of the powder actually prepared based on the recommended parameter combination with the target value; when the deviation between the predicted value and the verification value exceeds a preset threshold, supplementing the verification result into the training dataset, and retraining the neural network model until the output of the reactor structure parameters, stirring process parameters, and feeding parameter combination that meet the morphology requirements of the target powder is reached.

8. The flow field simulation and neural network optimization method for the liquid phase preparation process of metal powder according to claim 1, characterized in that: In step S1, the establishment of the flow field mass transfer simulation model includes: refining the mesh of the region near the stirring impeller, the region near the feed inlet, the interface between the rotating region and the stationary region, and the wall region; setting the incompressible fluid model, the turbulence model, the component transport model, and the coupling boundary conditions between the rotating region and the stationary region.

9. The flow field simulation and neural network optimization method for the liquid phase preparation process of metal powder according to claim 1, characterized in that: In step S2, before constructing the training dataset, the input parameters and output metrics are normalized and preprocessed to remove outlier data points, and the dataset is divided into training set, validation set and test set according to a preset ratio.

10. The flow field simulation and neural network optimization method for the liquid phase preparation process of metal powder according to claim 1, characterized in that: In step S3, the recommended parameter combination includes at least two of the following parameters: reactor inner diameter, impeller diameter, impeller height from bottom, stirring speed, feeding position, and feeding speed.

Images