Multi-channel rotational flow spray gun optimization method and system based on graph convolution

Abstract

The invention relates to the technical field of immersed injection spray guns, and discloses a multi-channel rotational flow spray gun optimization method and system based on graph convolution, and the method comprises the steps: combining computational fluid dynamics and a graph convolution neural network technology, and rapidly predicting a velocity field and a pressure field of a spray gun through rotational flow spray gun grid coordinates and rotational flow blade boundary information; the design period is obviously shortened, the test cost is reduced, and the research and development efficiency is improved; the graph convolutional neural network model subjected to repeated training and verification has high generalization ability and high accuracy, and it is guaranteed that the spray gun can still provide reliable prediction results under different working conditions and variable industrial conditions. The system comprises a data acquisition and preprocessing module, a graph convolutional neural network model module and a rotational flow spray gun structure optimization module. The flow characteristics and local physical characteristics of the spray gun are further analyzed, efficient and stable performance can be kept in actual working conditions, the service life is prolonged, and the metallurgical reaction efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of submerged blowing spray guns, and in particular to a multi-channel swirl spray gun optimization method and system based on graph convolution. Background Art

[0002] The top-submerged lance (TSL) process is a mature and efficient non-ferrous metal intensified smelting technology. This process combines the top-blown molten pool smelting process and the submerged injection technology. Air or oxygen-enriched air and fuel are injected into the high-temperature molten pool through a specially designed vertically mounted lance. It was developed by Dr. John Floyd in the 1970s. In this process, the main functions of the gas include: strengthening stirring, promoting full mixing and homogenization of the materials in the furnace, and improving the efficiency of heat and material transfer; sending solid additives (such as solid fuel, slag modifier / stabilizer) into the molten pool; under high temperature conditions, oxygen or other reaction gases react with impurities in the ore to form slag and metal separation, effectively extracting the target metal. Due to its superior efficiency and environmental performance, it has significant advantages in the field of non-ferrous metal refining with complex ores and high impurities. The TSL process has become an important means of modern intensified smelting and is widely used in the extraction of non-ferrous metals such as copper, tin, lead and zinc. During the smelting process, the submerged top-blowing lance equipped with a cyclone converts the axial jet into a chaotic cyclone and then injects it into the high-temperature molten pool, generating a strong turbulent stirring effect, increasing the gas-liquid contact area, and greatly improving the smelting reaction rate. However, in actual application, the erosion of the lance wall by the cyclone will cause the lance to bend or the nozzle to burn, greatly limiting the service life of the submerged top-blowing lance. The TSL molten pool smelting system involves complex cyclone-axial coaxial jet motion, jet-particle mixing, and gas-slag multiphase stirring behavior. A comprehensive and in-depth understanding of the complex cyclone fluid dynamics in the TSL furnace will help the operation, optimization, and performance improvement of the top-blown molten pool smelting system.

[0003] At present, the research on TSL process mainly focuses on the gas-slag multiphase stirring characteristics in the molten pool, while the research on the chaotic motion and attenuation characteristics of the swirl in the top-blowing lance is rarely reported. Among them, the cyclone structure is an important factor affecting the intensity of the chaotic motion of the swirl and its flow morphology. Improving the swirl performance in the industrial reactor by optimizing the cyclone structure is helpful to improve the operating efficiency of industrial equipment. In recent years, domestic and foreign scholars have mainly studied the influence of spiral cyclone geometric parameters (such as blade length, blade rotation angle, blade installation height and blade thickness, etc.) on swirl disturbance, swirl intensity attenuation, and effective influence distance of the cyclone through experiments. However, due to the complex high temperature distribution state, harsh operating conditions and expensive experimental costs in industrial-scale reactors, it is difficult for experimental methods to deeply understand the swirl turbulence characteristics in the equipment. As a method that can complement experiments, computational fluid dynamics (CFD) can study and explain various physical phenomena occurring in the reactor under harsh operating conditions by visualizing the multiphase interaction, multi-scale motion characteristics and multi-physical processes in the reactor, and conduct in-depth exploration of the coupling control mechanism of the process. With the update of computing hardware and the development of numerical algorithms, CFD has become a powerful tool for exploring the nonlinear characteristics of swirl in industrial applications. In the field of deep learning, graph convolutional neural networks (GCNNs) can process graph structure data with complex geometric shapes without converting them into regular grid or image formats. Therefore, GCNN can directly process unstructured grid data commonly used in CFD simulations, such as triangular or tetrahedral grids. And its basic idea is to update node features by aggregating node neighbor information (each node and edge has an initial feature vector, representing its attributes. For example, in the present invention, node features include node coordinates, velocity and pressure), which can effectively capture local structural information. Therefore, the present invention predicts the velocity field and pressure field of swirl spray guns with different geometric structures through the combination of CFD and GCNN, so as to deeply understand and analyze the flow characteristics, local physical characteristics and resistance parameters of the swirl spray gun.

[0004] Prior art 1, Chinese patent, application number: CN201910751432.9 discloses a Laval jet submerged top blowing spray gun, which relates to the field of non-ferrous metal smelting technology, including a hollow gun body, the gun body is cylindrical, one end of the gun body is a closed end, and the other end is an open end, characterized in that the inner cavity of the gun body is divided into a mixing section, a conveying section, a swirl section and a turbulent section from the closed end to the open end, a fuel pipe coaxially penetrates into the gun body from the center of the closed end and extends into the turbulent section, a pressure pipe is arranged on the outer sleeve of the fuel pipe in the gun body, an end pressure sampler is installed outside the gun body at the closed end, the end pressure sampler is inserted into the gun body and connected to the pressure pipe; at least two fluid interfaces are opened on the side wall of the gun body in the mixing section; at least one Laval injection device is installed in the gun body, and the Laval injection device is located between any two adjacent fluid interfaces; a swirler is installed in the gun body in the swirl section. Although the structure is simple, the fluid is mixed evenly, it is safe and reliable, and has a long service life, it does not take into account that the flow characteristics of the spray gun under different geometric structures are different, thus resulting in a short service life of the spray gun.

[0005] Prior art 2, Chinese patent, application number: CN202110223827.9 discloses a submerged top-blown converter, including a top-blown converter body, a material inlet and outlet are provided on the top of the top-blown converter body, and a submerged top-blown lance is inserted into the material inlet and outlet. The furnace has high reaction efficiency. By using the submerged top-blown molten pool for smelting, materials can be continuously added to the furnace, reducing the shutdown time for charging and improving the smelting operation rate. Since the submerged top-blown lance can be directly inserted into the molten pool, and a cyclone is provided in the submerged top-blown lance, the submerged top-blown lance stirs the molten pool violently, the materials are mixed more evenly, and the reaction efficiency in the furnace is improved. The top-blown converter is provided with a detection rod, which can sample and analyze the melt at any time, reduce the number of shutdown sampling times, keep the furnace temperature constant, and extend the furnace life. The support of the top-blown converter adopts a combination structure of a support ring and a roller, and adopts a large and small gear transmission. Although the transmission is more stable and overcomes the vibration impact of the furnace body, the use and life of the lance are not taken into account.

[0006] Prior art three, Chinese patent, application number: CN202010588955.9 discloses a modification method of a top-blown smelting facility, a continuous copper smelting facility and a method. The modification method of the top-blown smelting facility includes: canceling the converter or converting furnace of the top-blown converting facility; reducing the furnace height of the top-blown smelting furnace of the top-blown smelting facility; canceling the single submerged top-blown main lance of the top-blown smelting furnace; setting at least three top-blown smelting lances spaced from each other on the top of the top-blown smelting furnace; adding a copper matte discharge port to the top-blown smelting furnace; canceling the electrodes and electrode supporting devices of the sedimentation electric furnace of the top-blown smelting facility, and setting at least three top-blown converting lances spaced from each other on the top of the sedimentation electric furnace to transform the sedimentation electric furnace into a top-blown converting furnace; connecting the copper matte discharge port of the top-blown smelting furnace with the copper matte inlet of the top-blown converting furnace through a flow channel. Although the transformation method of the top-blowing smelting facility can make full use of the existing top-blowing smelting facilities, it can also be beneficial to the output of high-grade copper matte, create conditions for continuous blowing, and reduce the investment cost of continuous copper smelting. However, the flow characteristics, local physical characteristics and resistance of the swirl spray gun are not taken into consideration and calculated, resulting in a short spray gun life.

[0007] At present, the prior art 1, the prior art 2 and the prior art 3 do not take into account that the flow characteristics of the spray gun under different geometric structures are different, and do not consider and calculate the flow characteristics, local physical characteristics and resistance of the swirl spray gun, thus resulting in the problem of too short spray gun life. In order to solve the above problems, the present invention provides a multi-channel swirl spray gun optimization method and system based on graph convolution. Summary of the invention

[0008] The main purpose of the present invention is to provide a multi-channel swirl spray gun optimization method based on graph convolution to solve the problem that the flow characteristics of the spray gun under different geometric structures are different in the prior art, and the flow characteristics, local physical characteristics and resistance of the swirl spray gun are not considered and calculated, thus resulting in a short spray gun life.

[0009] To achieve the above object, the present invention provides the following technical solutions:

[0010] A multi-channel swirl spray gun optimization method based on graph convolution, the multi-channel swirl spray gun optimization method based on graph convolution comprising:

[0011] Use CFD software to simulate and calculate the swirl spray gun under different geometric structures, obtain the three velocity fields and pressure fields of the swirl spray gun, a total of four three-dimensional flow field data, obtain a data set, convert the three-dimensional data into a graph structure, and process the obtained data to reconstruct a new data set;

[0012] Design the graph convolutional neural network model structure and train the graph convolutional neural network model by minimizing the mean absolute error loss function; the last 1x1 convolution layer converts the node features into predicted velocity and pressure fields; use the Adam optimizer to update the parameters and adjust the hyperparameters according to the validation set during the training process; compare the predicted flow field with the CFD simulation results;

[0013] According to the velocity components and pressures in three directions of each node output by the graph convolutional neural network model, the velocity and pressure fields of the entire flow field are reconstructed; by drawing data images and calculating the pressure and shear stress on the surface of the swirl blades based on the data images, the distribution of velocity and pressure and the values ​​of drag and lift are obtained; based on a comprehensive and detailed analysis of the flow characteristics and local physical characteristics of the swirl spray gun, various parameters of the swirl spray gun are continuously adjusted and optimized, and the adjusted parameters are input into the CFD software simulation. The simulation results and experiments verify whether the optimization is qualified. If not, adjustments will continue.

[0014] As a further improvement of the present invention, the process of reconstructing a new data set includes:

[0015] The key areas are encrypted, including the boundary of the swirl blade and the outlet of the spray gun; CFD software is used to simulate and calculate the swirl spray gun under different geometric structures; after calculation, four three-dimensional flow field data of the swirl spray gun, three velocity fields and one pressure field, are obtained;

[0016] Convert the three-dimensional data into a graph structure and project the pressure and velocity data onto the grid by linear interpolation. Discretize each shape and the flow field area around it into a triangular grid, with each node containing coordinates, velocity and pressure information. Normalize the values ​​of the velocity field and pressure field.

[0017] Each node in the grid is considered as a node in the graph structure, and the edge is defined by the spatial distance. The boundary conditions of the swirl blades are processed by SDF, and the data is assigned to each node as a feature.

[0018] The reconstructed dataset contains 80% training set, 10% validation set and 10% test set, and the reconstructed dataset is fed into the convolutional network model.

[0019] As a further improvement of the present invention, the process of discretizing into a triangular network includes:

[0020] Collect the point set of triangulation, select an initial triangle, the three vertices must be three points in the point set, and these three points cannot be collinear; the point set includes the key points at the boundary of the swirl blade and the outlet of the spray gun;

[0021] Insert the rest of the points in the point set into the current triangulated network; each time a new point is inserted, check whether the new point will cause some triangular units to no longer conform to the empty circle characteristic; if not, delete the non-conforming triangular units to form a hole;

[0022] Connect the new point to each vertex that makes up the hole to generate a new edge; update the triangulated network based on the new edge, and repeat the above process until all points are inserted into the mesh.

[0023] As a further improvement of the present invention, the process of comparing the predicted flow field with the CFD simulation result includes:

[0024] Design a graph convolutional neural network model, which consists of multiple graph convolution blocks. Each convolution block contains two-step graph convolution layers and two-step smoothing layers. The last 1x1 convolution layer converts the node features into predicted velocity and pressure fields.

[0025] The graph convolutional neural network model is trained by minimizing the mean absolute error loss function, using the Adam optimizer to update parameters, and adjusting hyperparameters based on the results of the validation set during training;

[0026] Calculate the absolute error between the predicted and true values; compare the predicted flow field with the CFD simulation results.

[0027] As a further improvement of the present invention, the process of designing a graph convolutional neural network model includes:

[0028] The graph convolutional neural network model consists of multiple graph convolution blocks, each of which contains two-step graph convolution layers and two-step smoothing layers. New edge features are calculated through edge convolution and fused to the nodes.

[0029] New node features are calculated using the updated node features and edge information; the coordinates of the input layer are directly connected to each smoothing layer to provide spatial information for edge convolution;

[0030] Use a 1x1 convolutional layer to convert node features into predicted flow and pressure fields;

[0031] The 1x1 convolutional layer is equivalent to a fully connected layer.

[0032] As a further improvement of the present invention, the process of adjusting and optimizing various parameters of the swirl spray gun includes:

[0033] According to the velocity components and pressure in three directions of each node output by the graph convolutional neural network model, the velocity field and pressure field of the entire flow field are reconstructed;

[0034] By drawing streamlines, velocity vector diagrams and pressure distribution data images, the distribution of velocity and pressure is obtained; according to the drawn pressure distribution diagram, the area on the surface of the swirl blade is selected for integral calculation of the pressure and shear stress on the surface of the swirl blade, and the drag and lift values ​​are calculated based on the integral results of the pressure and shear stress; based on a comprehensive and detailed analysis of the flow characteristics and local physical characteristics of the swirl spray gun, various parameters of the swirl spray gun are continuously adjusted;

[0035] Analyze the calculation results and verify through CFD software simulation and experimental verification whether the various parameters of the adjusted swirl spray gun are qualified. If not, continue to adjust the swirl spray gun parameters.

[0036] As a further improvement of the present invention, the process of obtaining the drag and lift values ​​includes:

[0037] Obtain the flow field velocity data and pressure data in the graph convolutional neural network model, and determine the vertex data for drawing the flow field velocity and pressure according to each data; draw a data image of the distribution of the flow field velocity and pressure according to the vertex data;

[0038] Identify the drawn pressure distribution diagram, select the area on the swirl blade surface for integral calculation, and obtain the pressure and shear stress on the blade surface; calculate the drag and lift values ​​of the swirl spray gun based on the integral structure of pressure and shear stress; calculate the pressure gradient force and vortex data to obtain the local flow characteristics and complex fluid mechanics behavior of the swirl blade area;

[0039] The calculation results are analyzed, and the flow characteristics and local physical characteristics of the swirl spray gun are analyzed through numerical simulation and experimental data; among them, the flow characteristics include vortex and separation phenomenon, and the local physical characteristics include pressure center and velocity peak.

[0040] As a further improvement of the present invention, the process of local flow characteristics and complex fluid dynamics behavior of the region includes:

[0041] According to the drawn pressure distribution diagram, the surface area of ​​the swirl blade is identified and selected, and the pressure distribution is integrated to obtain the total pressure and shear stress on the blade surface;

[0042] According to the pressure value of the pressure sensor in the swirl blade surface area, the pressure value of the sensor is integrated to obtain the pressure on the blade surface; the pressure on the blade surface is expressed as:

[0043] P total =∫ A P(x,y)dA

[0044] Where P(x,y) is the pressure distribution function, and A is the area of ​​the blade surface;

[0045] Shear stress is realized by solving the moment and swirl blade surface; first derive the force F in the x direction of abcd t , using the equation Get the expressions for forces F1 and F2; substitute the expressions into the equation:

[0046] Tt dx=F1-F2

[0047] Get Moment The ratio of dx is equal to M z Multiply by S d A y The integral of V y The shear stress on the blade surface is finally obtained by being parallel to the y-axis and positively or negatively with the y-axis; the shear stress on the blade surface is:

[0048]

[0049] Among them, M z is the moment about the z-axis, I z is the moment of inertia of the section, y and z are the coordinates on the section;

[0050] According to the integral results of pressure and shear stress, the drag and lift values ​​of the swirl spray gun are calculated;

[0051]

[0052] d p =f(V axial ,G·P)

[0053]

[0054] Where a is the surface tension coefficient;

[0055] The pressure gradient force is calculated based on the above results; the pressure gradient force calculation formula is:

[0056]

[0057] where ρ is the fluid density, is the pressure gradient derived from the above pressures;

[0058] Lift and drag are calculated using the integral of pressure and shear stress, where the formula for lift, L, is:

[0059] L = ∫ A (P top -P tottom )dA

[0060] The formula for resistance D is:

[0061] D = ∫A (τ top +τ bottom )dA

[0062] Among them, the above-mentioned P top and P tottom are the pressure on the upper and lower surfaces, τ top and τ bottom are the shear stresses on the upper and lower surfaces, respectively;

[0063] The calculation formula of vorticity Γ is:

[0064] Γ=∫ A (ω x dy-ω y dx)dA

[0065] Among them, ω x and ω y are the components of vorticity in the x and y directions, respectively;

[0066] By analyzing the distribution of pressure gradient force and vorticity, it is possible to determine whether the flow is separating, reattaching or vortex formation occurs.

[0067] As a further improvement of the present invention, the process of analyzing the flow characteristics and local physical characteristics of the swirl spray gun includes:

[0068] By drawing streamlines, velocity vector diagrams and pressure distribution data images, flow separation, reattachment and vortex formation phenomena can be identified; by integrating and calculating the pressure and shear stress on the surface of the swirl blade, the drag and lift values ​​can be obtained; by calculating the pressure gradient force and vortex data through the drag and lift values, the local flow characteristics and complex fluid mechanics behavior of the swirl blade area can be obtained;

[0069] The obtained results are output to CFD software for simulation, and the simulation results are compared with the experimental data. A threshold is set. If the value is greater than the threshold, the numerical simulation result is considered unqualified and the design parameters of the swirl spray gun are adjusted.

[0070] The adjusted parameters are simulated and analyzed again, and the design parameters of the swirl spray gun are iterated until they are less than the threshold value.

[0071] To achieve the above object, the present invention provides the following technical solutions:

[0072] A multi-channel swirl spray gun optimization system based on graph convolution, which is applied to the multi-channel swirl chaotic metallurgical spray gun design and optimization method based on deep learning. The multi-channel swirl spray gun optimization and optimization system based on graph convolution includes:

[0073] The data acquisition and preprocessing module is responsible for calculating and extracting the required data from the CFD software, projecting the pressure and velocity data onto the grid through linear interpolation, discretizing each shape and the surrounding flow field area into a triangular grid, and each node contains coordinate, velocity and pressure information; normalizing the values ​​of the velocity field and pressure field, reconstructing the data set, and constructing the data set into 80% training set, 10% validation set and 10% test set; inputting the reconstructed data set into the graph convolutional neural network model module;

[0074] The graph convolutional neural network model module is responsible for designing the convolutional network neural structure, which consists of multiple convolution blocks. Each convolution block contains a two-step graph convolution layer and a two-step smoothing layer. The last 1x1 convolution layer converts the node features into predicted velocity and pressure fields. The graph convolutional neural network is trained by minimizing the mean absolute error loss function, updating the parameters through the Adam optimizer, and adjusting the hyperparameters according to the performance of the validation set during the training process. The generalization ability of the graph convolutional neural network model is evaluated using the test set, and the MAE between the predicted value and the true value is calculated. The accuracy of the prediction results of the graph convolutional neural network is obtained by comparing the predicted flow field with the CFD simulation results. The final results are output to the swirl spray gun structure optimization module.

[0075] The swirl spray gun structure optimization module is responsible for drawing data images such as streamlines, velocity vector diagrams and pressure distribution diagrams, identifying flow, reattachment and swirl formation phenomena, and comprehensively analyzing the flow characteristics of the swirl spray gun. The pressure and shear stress on the surface of the swirl blade are calculated by integration to obtain the drag and lift values, and the pressure gradient force and vortex data are calculated to analyze the local flow characteristics of the swirl spray gun; various parameters of the swirl spray gun are adjusted according to the analyzed characteristics and data, and the adjusted various parameters of the swirl spray gun are input into the CFD software for simulation, and experimental verification is carried out based on the adjusted various parameters of the swirl spray gun, and the qualification is checked based on the simulation results and experimental verification; if it is unqualified, the various parameters of the swirl spray gun are readjusted and the simulation and experiment are repeated.

[0076] The present invention combines computational fluid dynamics (CFD) and graph convolutional neural network (GCNN) technology, and can quickly predict the velocity field and pressure field of the spray gun through the grid coordinates of the swirl spray gun and the boundary information of the swirl blades. Compared with the traditional experimental method, it significantly shortens the design cycle, reduces the experimental cost, and improves the research and development efficiency. The repeatedly trained and verified GCNN model has a strong generalization ability and high accuracy, ensuring that the spray gun can still provide reliable prediction results under different working conditions and variable industrial conditions. Further analysis of the flow characteristics and local physical characteristics of the spray gun helps to understand its complex fluid mechanics behavior, ensure that it can maintain efficient and stable performance in actual working conditions, extend its service life, and improve metallurgical reaction efficiency. It reduces the frequency of equipment replacement and resource consumption from the source, while reducing energy consumption and pollution emissions in the process, helping to achieve green and sustainable development of industrial production. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 It is a schematic diagram of the steps of an embodiment of a multi-channel swirl spray gun optimization method based on graph convolution of the present invention;

[0078] Figure 2 A flowchart of the steps of reconstructing a new data set in one embodiment of the multi-channel swirl spray gun optimization method based on graph convolution of the present invention;

[0079] Figure 3 A flowchart of the steps of discretizing an embodiment of a multi-channel swirl spray gun optimization method based on graph convolution of the present invention into a triangular network;

[0080] Figure 4 A flowchart of an embodiment of a multi-channel swirl spray gun optimization method based on graph convolution of the present invention by comparing the predicted flow field with the CFD simulation result;

[0081] Figure 5 It is a principle diagram of the multi-channel swirl spray gun optimization method based on graph convolution of the present invention;

[0082] Figure 6 A flowchart of the steps of designing a graph convolutional neural network model for an embodiment of a multi-channel swirl spray gun optimization method based on graph convolution of the present invention;

[0083] Figure 7 A flowchart of the steps for adjusting and optimizing various parameters of a swirl spray gun in one embodiment of a multi-channel swirl spray gun optimization method based on graph convolution of the present invention;

[0084] Figure 8 It is a principle diagram for adjusting and optimizing various parameters of a swirl spray gun in a multi-channel swirl spray gun optimization method based on graph convolution of the present invention;

[0085] Fig. 9A flowchart of the steps of obtaining drag and lift values ​​in an embodiment of a multi-channel swirl spray gun optimization method based on graph convolution of the present invention;

[0086] Fig.10 A flowchart of the steps of local flow characteristics and complex fluid mechanics behavior of a domain in one embodiment of a multi-channel swirl spray gun optimization method based on graph convolution of the present invention;

[0087] Fig.11 A flowchart of the steps of analyzing the flow characteristics and local physical characteristics of a swirl spray gun in one embodiment of a multi-channel swirl spray gun optimization method based on graph convolution of the present invention;

[0088] Fig.12 It is a functional module schematic diagram of an embodiment of a multi-channel swirl spray gun optimization and optimization system based on graph convolution of the present invention;

[0089] Fig.13 It is a structural schematic diagram of an embodiment of an electronic device of the present invention;

[0090] Fig.14 It is a schematic structural diagram of an embodiment of the storage medium of the present invention. DETAILED DESCRIPTION

[0091] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0092] The terms "first", "second" and "third" in the present invention are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Therefore, the features defined as "first", "second" and "third" can explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "multiple" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined. All directional indications (such as up, down, left, right, front, back...) in the embodiments of the present invention are only used to explain the relative position relationship, movement, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication also changes accordingly. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally also includes steps or units that are not listed, or optionally also includes other steps or units inherent to these processes, methods, products or devices.

[0093] Reference to an "embodiment" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiment may be included in at least one embodiment of the present invention. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0094] like Figure 1 As shown, this embodiment provides an embodiment of a multi-channel swirl spray gun design and optimization method based on graph convolution. The multi-channel swirl spray gun optimization method based on graph convolution specifically includes the following steps:

[0095] Step S1: using CFD software to simulate and calculate the swirl spray gun under different geometric structures, obtain three velocity fields and pressure fields of the swirl spray gun, a total of four three-dimensional flow field data, obtain a data set, convert the three-dimensional data into a graph structure, and process the obtained data to reconstruct a new data set;

[0096] Step S2: Design the graph convolutional neural network model structure and train the graph convolutional neural network model by minimizing the mean absolute error loss function; the last 1x1 convolutional layer converts the node features into predicted velocity and pressure fields; use the Adam optimizer to update the parameters and adjust the hyperparameters according to the validation set during the training process; compare the predicted flow field with the CFD simulation results;

[0097] Step S3: Reconstruct the velocity and pressure field of the entire flow field according to the velocity components and pressures in three directions of each node output by the graph convolutional neural network model; obtain the distribution of velocity and pressure and the values ​​of drag and lift by drawing data images and calculating the pressure and shear stress on the surface of the swirl blades according to the data images; continuously adjust and optimize various parameters of the swirl spray gun based on a comprehensive and detailed analysis of the flow characteristics and local physical characteristics of the swirl spray gun, and input the adjusted parameters into the CFD software simulation, and verify whether the optimization is qualified with the simulation results and experiments. If not, continue to adjust.

[0098] Preferably, this embodiment combines computational fluid dynamics (CFD) and (GCNN) technology, and the velocity field and pressure field of the spray gun can be quickly predicted through the swirl spray gun grid coordinates and swirl blade boundary information. Compared with the traditional experimental method, it significantly shortens the design cycle, reduces the test cost, and improves the research and development efficiency. The repeatedly trained and verified GCNN model has a strong generalization ability and high accuracy, ensuring that the spray gun can still provide reliable prediction results under different working conditions and changing industrial conditions. Further analysis of the flow characteristics and local physical characteristics of the spray gun will help to understand its complex fluid mechanics behavior, ensure that it can maintain efficient and stable performance in actual working conditions, extend its service life, and improve the efficiency of metallurgical reactions. It reduces the frequency of equipment replacement and resource consumption from the source, while reducing energy consumption and pollution emissions in the process, helping to achieve green and sustainable development of industrial production.

[0099] In summary, the significance of this embodiment is that by combining computational fluid dynamics (CFD) and graph convolutional neural network (GCNN) technology, the flow characteristics of spray guns under different geometric structures can be predicted.

[0100] Further, if Figure 2 As shown, the process of reconstructing a new data set in step S1 specifically includes the following steps:

[0101] Step S11: Encrypt the key area, which includes the boundary of the swirl blade and the outlet of the spray gun; use CFD software to simulate and calculate the swirl spray gun under different geometric structures; after calculation, obtain three velocity fields and one pressure field of the swirl spray gun, a total of four three-dimensional flow field data;

[0102] Step S12: convert the three-dimensional data into a graph structure, project the pressure and velocity data onto the grid by linear interpolation; discretize each shape and the flow field area around it into a triangular grid, and each node contains coordinates, velocity and pressure information; and normalize the values ​​of the velocity field and the pressure field;

[0103] Each node in the grid is considered as a node in the graph structure, and the edge is defined by the spatial distance. The boundary conditions of the swirl blades are processed by SDF, and the data is assigned to each node as a feature.

[0104] Step S13: Reconstruct the data set, which includes 80% training set, 10% validation set and 10% test set, and input the reconstructed data set into the convolutional network model.

[0105] Preferably, this embodiment first uses CFD software to simulate and calculate the swirl spray gun under different geometric structures. Before the calculation, local encryption is also performed to ensure the mesh accuracy of key areas (such as the boundary of the swirl blade and the outlet of the spray gun). After calculation, three velocity fields (x, y, z) and one pressure field of the swirl spray gun are obtained, totaling four three-dimensional flow field data. After that, the three-dimensional data is converted into a graph structure, each node in the grid is regarded as a node in the graph structure, and the edge is defined by the spatial distance. The velocity, pressure and coordinate data need to be normalized, and the boundary conditions of the swirl blade are processed by SDF (Signdistance function), and then these data are assigned to each node as features. The input features of the final model include i) swirl blade boundary; ii) x coordinate of each node; iii) y coordinate of each node; iv) z coordinate of each node. The output includes i) pressure field; ii) x-direction velocity field; iii) y-direction velocity field; iv) z-direction velocity field. In the present invention, the architecture of the GCNN model consists of multiple graph convolution blocks, each of which contains two-step graph convolution layers and two-step smoothing layers. The graph convolution layer first calculates new edge features through edge convolution, then aggregates these features to the nodes, and finally calculates new node features using the updated node features and edge information. The smoothing layer is used to reduce the spatial variability of node features. The model also uses jump connections to directly connect the coordinate information of the input layer to each smoothing layer to provide spatial information for edge convolution. Finally, a 1×1 convolution layer (equivalent to a fully connected layer) is used to convert the node features into predicted velocity fields and pressure fields.

[0106] In summary, this embodiment comprehensively improves the processing accuracy of swirl gun flow field data and the prediction ability of the model by encrypting key areas, using CFD software for simulation, data conversion and graph structure construction, data set reconstruction and convolutional network model training, SDF processing and boundary condition assignment, and unstructured grid and adaptive nested grid methods.

[0107] Further, if Figure 3 As shown, the process of discretizing into a triangular network in step S12 specifically includes the following steps:

[0108] Step S121: Collect the triangulated point set, select an initial triangle, the three vertices must be three points in the point set, and the three points cannot be collinear; wherein the point set includes key points at the swirl blade boundary and the spray gun outlet;

[0109] Step S122: inserting the rest of the points in the point set into the current triangulated network; each time a new point is inserted, checking whether the new point will cause some triangular units to no longer meet the empty circle characteristic; if not, deleting the non-compliant triangular units to form a hole;

[0110] Step S123: Connect the new point to each vertex that makes up the hole to generate a new edge; update the triangulated network according to the new edge, and repeat the above process until all points are inserted into the mesh.

[0111] Preferably, this embodiment can generate a high-quality Delaunay triangulation network through point-by-point insertion and checking of empty circle characteristics, ensuring that the circumscribed circle of each triangle does not contain other points and the minimum angle is the largest. It can adapt to point sets of different shapes and sizes, and through dynamic updating and optimization, it can handle complex boundary conditions and irregular point sets to generate a triangulated network that meets the requirements. The point-by-point insertion method has high efficiency when processing large-scale point sets, and can generate high-quality triangulated networks in a shorter time. It is particularly important for application scenarios that require rapid generation of meshes, such as real-time simulation and dynamic analysis. It is not only applicable to static point sets, but also can adapt to changes in point sets through dynamic updating and optimization. For example, when the point set changes, the high quality of the mesh can be maintained by reinserting new points and updating the triangulated network.

[0112] In summary, this embodiment can generate a high-quality, adaptable, efficient and flexible triangulated network. This method has broad application prospects in the fields of geometric modeling, numerical calculation and dynamic analysis.

[0113] Further, if Figure 4 As shown, the process of comparing the predicted flow field with the CFD simulation result in step S2 specifically includes the following steps:

[0114] Step S21: Design a graph convolutional neural network model, which consists of multiple graph convolution blocks. Each convolution block contains two-step graph convolution layers and two-step smoothing layers. The last 1x1 convolution layer converts the node features into predicted velocity fields and pressure fields.

[0115] Step S22: The graph convolutional neural network model is trained by minimizing the mean absolute error loss function, using the Adam optimizer to update the parameters, and adjusting the hyperparameters according to the results of the validation set during the training process;

[0116] Among them, the loss function expression is:

[0117] L=λ1L MAE +λ2L MSE +λ3L SSIM +λ4L grad

[0118] The meaning of each part is as follows: MAE Represents the mean absolute error (Mean Absolute Error), which is used to measure the average absolute error between the predicted velocity field and pressure field and the true value; the calculation formula is:

[0119]

[0120] In the formula, y i represents the true value, represents the predicted value, and N represents the number of samples;

[0121] L MSE Represents the mean square error (Mean Squared Error), which is used to measure the square error between the predicted value and the true value. The calculation formula is:

[0122]

[0123] L SSIM It represents the structural similarity index, which is used to measure the similarity between the structural information of the predicted flow field and the real flow field;

[0124] L grad Represents the gradient loss, which is used to measure the difference between the gradient information of the predicted flow field and the gradient information of the real flow field. The calculation formula is:

[0125]

[0126] In the formula, represents the gradient of the real flow field, represents the gradient of the predicted flow field;

[0127] λ1, λ2, λ3, λ4 represent weight coefficients, which are used to balance the importance of different loss terms;

[0128] Step S23: Calculate the absolute error between the predicted value and the true value; compare the predicted flow field with the CFD simulation result.

[0129] Preferably, the architecture of the convolutional neural network model of this embodiment is composed of multiple graph convolution blocks, each of which contains a two-step graph convolution layer and a two-step smoothing layer. The graph convolution layer first calculates new edge features through edge convolution, then aggregates these features to the nodes, and finally calculates new node features using the updated node features and edge information. The smoothing layer is used to reduce the spatial variability of node features. The model also uses a jump connection to directly connect the coordinate information of the input layer to each smoothing layer to provide spatial information for edge convolution. Finally, a 1×1 convolution layer (equivalent to a fully connected layer) is used to convert the node features into predicted velocity fields and pressure fields. Training is performed by minimizing the mean absolute error (MAE) loss function, using the Adam optimizer to update parameters, and adjusting hyperparameters according to the performance of the validation set during training. After training, the generalization ability of the model is evaluated using the test set, and the MAE between the predicted value and the true value is calculated. In addition, by comparing the predicted flow field with the CFD simulation results, the prediction effect of the model is intuitively evaluated, thereby verifying the accuracy and robustness of the model under different shapes and conditions (for specific principles, refer to the attached Figure 5 ).

[0130] In summary, the present embodiment designs a graph convolutional neural network model, especially a structure including multiple graph convolution blocks, which can effectively capture and process complex data features. Each convolution block includes a two-step graph convolution layer and a two-step smoothing layer, which helps the model to retain local features and smooth global features when processing flow field data, thereby improving the accuracy of prediction. The Adam optimizer is an adaptive learning rate optimization algorithm that can dynamically adjust the learning rate of each parameter according to the first-order moment estimation and second-order moment estimation of the gradient. This adaptive adjustment can accelerate the convergence speed of the model, reduce oscillation during training, and improve training efficiency. The mean absolute error (MAE) is a commonly used loss function, which is particularly suitable for processing data sets with more outliers. By minimizing MAE, the model can better capture the nonlinear relationship in the data and improve the accuracy of prediction. During the training process, verification and adjustment of hyperparameters through the validation set can ensure that the performance of the model on different data sets is stable and has good generalization ability. It helps to avoid overfitting of the model and improve the robustness of the model. Comparing the predicted flow field with the CFD simulation results can intuitively evaluate the prediction effect of the model. This comparison helps to identify deficiencies in the model and further optimize the model.

[0131] Further, if Figure 6 As shown, the process of designing the graph convolutional neural network model in step S21 specifically includes the following steps:

[0132] Step S211: The graph convolutional neural network model is composed of multiple graph convolution blocks, each of which includes two-step graph convolution layers and two-step smoothing layers; new edge features are calculated through edge convolution, and the new edge features are fused to the nodes;

[0133] Step S212: Calculate new node features using the updated node features and edge information; directly connect the coordinates of the input layer to each smoothing layer to provide spatial information for edge convolution;

[0134] Step S213: Use a 1x1 convolutional layer to convert node features into predicted flow fields and pressure fields;

[0135] The 1x1 convolutional layer is equivalent to a fully connected layer.

[0136] Preferably, each graph convolution block in this embodiment includes a two-step graph convolution layer and a two-step smoothing layer. It is helpful to perform effective feature extraction and information propagation on graph structure data. By calculating new edge features through edge convolution and fusing these new edge features to the nodes, the expressiveness of node features can be enhanced, so that the model can better capture the complex relationships in the graph structure. Using updated node features and edge information to calculate new node features helps the model gradually converge to a more accurate feature representation. After directly connecting the coordinates of the input layer to each smoothing layer, spatial information is provided for edge convolution, which can enhance the sensitivity of the model to spatial position information, thereby improving the accuracy of prediction. Using a 1x1 convolution layer to convert node features into predicted flow fields and pressure fields is equivalent to a fully connected layer. The main function of the 1x1 convolution layer is to adjust the number of channels between network layers, control the complexity of the model, and can be used for dimensionality reduction and reduce the amount of computation. Through the 1x1 convolution layer, the model can effectively adjust the network structure while maintaining feature quality, avoid overfitting, and improve the generalization ability of the model.

[0137] Therefore, through the iterative update of graph convolution blocks, node features and edge information, as well as the dimensionality reduction and channel adjustment of 1x1 convolution layers, effective processing and feature extraction of complex graph structure data are achieved, thereby improving the prediction accuracy and generalization ability of the model.

[0138] Further, if Figure 7 As shown, the process of adjusting and optimizing various parameters of the swirl spray gun in step S3 specifically includes the following steps:

[0139] Step S31: Reconstruct the velocity field and pressure field of the entire flow field according to the velocity components and pressures in three directions of each node output by the graph convolutional neural network model;

[0140] Step S32: Obtain the distribution of velocity and pressure by drawing data images such as streamlines, velocity vector diagrams and pressure distribution diagrams; select an area on the surface of the swirl blade to integrate and calculate the pressure and shear stress on the surface of the swirl blade according to the drawn pressure distribution diagram, and calculate the drag and lift values ​​according to the integral results of the pressure and shear stress; continuously adjust various parameters of the swirl spray gun based on a comprehensive and detailed analysis of the flow characteristics and local physical characteristics of the swirl spray gun;

[0141] Step S33: Analyze the calculation results and verify through CFD software simulation and experimental verification whether the various parameters of the adjusted swirl spray gun are qualified. If not, continue to adjust the swirl spray gun parameters.

[0142] Preferably, in this embodiment, the graph convolutional neural network model outputs the velocity components and pressures in three directions of each node, and reconstructs the velocity field and pressure field of the entire flow field. Therefore, by drawing data images such as streamlines, velocity vector diagrams and pressure distribution diagrams, the distribution of velocity and pressure can be revealed, which helps to identify phenomena such as flow separation, reattachment, and vortex formation, and comprehensively analyze the flow characteristics of the spray gun. In addition to the overall flow field analysis, the local physical characteristics of the swirl blade surface can also be analyzed in detail. By integrally calculating the pressure and shear stress on the swirl blade surface, the drag and lift values ​​can be obtained. Further analysis can reveal the local flow characteristics and complex fluid mechanics behavior of the swirl blade area by calculating data such as pressure gradient force and vortex. By comprehensively and in detail analyzing the flow characteristics and local physical characteristics of the swirl spray gun, the optimization of the swirl spray gun structure can be scientifically guided. For example, by adjusting the shape and angle of the swirl blade, flow separation and vortex can be reduced or eliminated; by improving the structural parameters of the swirl blade, the tangential momentum can be strengthened, and the stirring efficiency of the spray gun can be improved; in areas with higher shear stress, more wear-resistant blade materials can be selected to reduce blade wear. These measures are designed to make the flow more uniform, reduce energy loss and material stress concentration, thereby improving the overall performance of the spray gun and extending its life. In addition, the spray gun design is continuously improved through numerical simulation and experimental verification to ensure that the optimization measures are effective in actual applications, and ultimately achieve the goal of performance improvement and life extension (for specific principles, please refer to the attached Figure 8 ).

[0143] In summary, this embodiment not only improves the efficiency and performance of the swirl spray gun design, but also significantly improves the accuracy and efficiency of prediction through deep learning technology, reducing the number of experiments and costs.

[0144] Further, if Fig. 9 As shown, the process of obtaining the drag and lift values ​​in step S32 specifically includes the following steps:

[0145] Step S321: Obtain flow field velocity data and pressure data in the graph convolutional neural network model, and determine vertex data for drawing flow field velocity and pressure according to each data; draw a data image of the distribution of flow field velocity and pressure according to the vertex data;

[0146] Step S322: Identify the drawn pressure distribution diagram, select the area on the surface of the swirl blade for integral calculation, and obtain the pressure and shear stress on the blade surface; calculate the resistance and lift values ​​of the swirl spray gun according to the integral structure of the pressure and shear stress; calculate the pressure gradient force and vortex data to obtain the local flow characteristics and complex fluid mechanics behavior of the swirl blade area;

[0147] Step S323: Analyze the calculation results, and analyze the flow characteristics and local physical characteristics of the swirl spray gun through numerical simulation and experimental data; wherein the flow characteristics include vortex, separation and other phenomena, and the local physical characteristics include pressure center, velocity peak and the like.

[0148] Preferably, the present embodiment obtains flow field velocity and pressure data through a graph convolutional neural network model, and draws the corresponding vertex data image to intuitively understand the distribution of the flow field. The drawn pressure distribution diagram is identified, and the area on the surface of the swirl blade is selected for integral calculation to obtain the pressure and shear stress on the blade surface. It is essential to calculate the drag and lift values ​​of the swirl spray gun. In addition, the pressure gradient force and vortex data are calculated to obtain the local flow characteristics and complex fluid mechanics behavior of the swirl blade area. The accuracy of the model can be verified and the design can be further optimized by comparing and analyzing numerical simulation and experimental data. Calculating vorticity and pressure gradient can reveal phenomena such as eddy currents and separations in the flow field, as well as local physical characteristics such as pressure center and velocity peak. It is very important to understand the flow characteristics and local physical characteristics of the swirl spray gun. The flow field prediction method based on graph convolutional neural network can predict the pressure coefficient within a few seconds, which is 3 orders of magnitude faster than CFD simulation. This embodiment can quickly evaluate the aerodynamic force, velocity and pressure field of the fluid flowing around the object, and provide fast feedback for design optimization. The effect of pressure gradient on flow is significant, especially in the field of airfoil separation flow control. By introducing an improved model of the pressure gradient, the location of the separation zone and the wall pressure can be predicted more accurately. This improved model can improve the accuracy of the calculation results and thus optimize the design.

[0149] In summary, this embodiment can effectively reveal the flow characteristics and local physical characteristics of the swirl spray gun, optimize the design and improve the computational efficiency by acquiring and analyzing the flow field velocity and pressure data, combining the comparison of numerical simulation and experimental data, and utilizing advanced technologies such as graph convolutional neural networks.

[0150] Further, if Fig.10As shown, the process of local flow characteristics and complex fluid mechanics behavior of the region in step S322 specifically includes the following steps:

[0151] Step S3221: According to the drawn pressure distribution diagram, identify and select the surface area of ​​the swirl blade, and perform integral calculation on the pressure distribution to obtain the total pressure and shear stress on the blade surface;

[0152] According to the pressure value of the pressure sensor in the swirl blade surface area, the pressure value of the sensor is integrated to obtain the pressure on the blade surface; the pressure on the blade surface is expressed as:

[0153] P total =∫ A P(x,y)dA

[0154] Where P(x,y) is the pressure distribution function, and A is the area of ​​the blade surface;

[0155] Shear stress is realized by solving the moment and swirl blade surface; first derive the force F in the x direction of abcd t , using the equation Get the expressions for forces F1 and F2; substitute the expressions into the equation:

[0156] Tt dx=F1-F2

[0157] Get Moment The ratio of dx is equal to M z Multiply by S d A y The integral of V y The shear stress on the blade surface is finally obtained by being parallel to the y-axis and positively or negatively with the y-axis; the shear stress on the blade surface is:

[0158]

[0159] Among them, M z is the moment about the z-axis, I z is the moment of inertia of the section, y and z are the coordinates on the section;

[0160] Step S3222: Calculate the drag and lift values ​​of the swirl spray gun according to the integration results of the pressure and shear stress;

[0161]

[0162] d p =f(V axial ,G·P)

[0163]

[0164]

[0165] Where a is the surface tension coefficient;

[0166] Step S3223: Calculate the pressure gradient force according to the above results; the pressure gradient force calculation formula is:

[0167]

[0168] where ρ is the fluid density, is the pressure gradient derived from the above pressures;

[0169] Lift and drag are calculated using the integral of pressure and shear stress, where the formula for lift, L, is:

[0170] L = ∫ A (P top -P tottom )dA

[0171] The formula for resistance D is:

[0172] D = ∫ A (τ top +τ bottom )dA

[0173] Among them, the above-mentioned P top and P tottom are the pressure on the upper and lower surfaces, τ top and τ bottom are the shear stresses on the upper and lower surfaces, respectively;

[0174] The calculation formula of vorticity Γ is:

[0175] Γ=∫ A (ω x dy-ω y dx)dA

[0176] Among them, ω x and ω y are the components of vorticity in the x and y directions, respectively;

[0177] By analyzing the distribution of pressure gradient force and vorticity, it can be determined whether the flow is undergoing separation, reattachment, or vortex formation.

[0178] Preferably, the present embodiment can obtain the total pressure on the blade surface by identifying and selecting the surface area of ​​the swirl blade and performing integral calculation on the pressure distribution. It is helpful to accurately calculate the stress of the blade under different pressures, thereby providing data support for subsequent structural design and optimization. The calculation of shear stress is achieved by solving the moment and swirl blade surface. The calculation formula of shear stress involves the moment of inertia of the section and the coordinates. The accurate calculation of the parameters is crucial to understanding the deformation and stress distribution of the blade under stress. The calculation of shear stress involves the expressions of forces F1 and F2, which are derived by equations, and finally the expression of shear stress is obtained. It ensures that the calculation of shear stress is based on accurate mechanical principles and mathematical models. The calculation of shear stress also involves the use of the moment of inertia of the section, which helps to more accurately evaluate the deformation and stress distribution of the blade under stress. According to the integral results of pressure and shear stress, the drag and lift values ​​of the swirl spray gun are calculated. It is helpful to evaluate the performance of the swirl spray gun under actual working conditions and provide a basis for optimized design. The introduction of the surface tension coefficient further considers the influence of the liquid surface tension on the drag and lift, which is particularly important in practical applications, especially in applications involving liquid flow and injection. Through the above steps, the accurate calculation of the surface pressure and shear stress of the swirl blade can be achieved, thereby improving the accuracy and reliability of the design. The calculation results can be directly applied to structural design and optimization, reducing experimental cost and time, and improving design efficiency. The calculation method of this embodiment can also be used to predict and analyze the performance of the swirl spray gun under different working conditions, providing theoretical support for practical applications.

[0179] In summary, the technical significance of this embodiment is that it provides a scientific basis for the design and optimization of the swirl spray gun through the accurate calculation of pressure distribution, shear stress, drag, lift and pressure gradient force, thereby improving the accuracy and reliability of the design.

[0180] Furthermore, if Fig.11 As shown, in step S33, the process of analyzing the flow characteristics and local physical characteristics of the swirl spray gun specifically includes the following steps:

[0181] Step S331: Identify flow separation, reattachment, vortex formation and other phenomena through the drawn streamlines, velocity vector diagrams, pressure distribution diagrams and other data images; obtain drag and lift values ​​by integrating and calculating the pressure and shear stress on the surface of the swirl blades; calculate the pressure gradient force and vorticity and other data through the drag and lift values ​​to obtain the local flow characteristics and complex fluid mechanics behavior of the swirl blade area;

[0182] Step S332: Output the obtained results to CFD software for simulation, compare the simulation results with the experimental data, set a threshold, and if it is greater than the threshold, the numerical simulation results are considered unqualified, and return to adjust the design parameters of the swirl spray gun;

[0183] Step S333: The adjusted parameters are simulated and analyzed again, and the swirl spray gun design parameters are iterated until they are less than the threshold value.

[0184] Preferably, the present embodiment can identify phenomena such as flow separation, reattachment, and vortex formation by drawing data images such as streamlines, velocity vector diagrams, and pressure distribution diagrams. It is essential to understand fluid dynamics behavior, especially in the swirl blade area, which directly affects the flow characteristics of the fluid and the performance of the equipment. By integrally calculating the pressure and shear stress on the surface of the swirl blade, the drag and lift values ​​can be obtained. The values ​​are key parameters for evaluating the performance of the swirl blade and directly affect the efficiency and energy consumption of the equipment. By calculating the pressure gradient force and vortex and other data by drag and lift numerical values, the local flow characteristics and complex fluid mechanics behavior of the swirl blade area can be obtained. It helps to deeply understand the flow law of the fluid in the swirl blade area, thereby providing a basis for optimizing the design. The obtained results are output to the CFD software for simulation, and the simulation results are compared with the experimental data, so that the error of the simulation results can be quantitatively evaluated. It helps to verify the accuracy of the simulation model and ensure the reliability of the numerical simulation results. If the accuracy of the simulation results meets the requirements, the virtual performance sample can be directly entered, otherwise the design parameters need to be adjusted. By iteratively adjusting the design parameters of the swirl spray gun to less than the threshold, the design parameters can be continuously optimized to improve the performance of the equipment. This iterative process helps find the optimal combination of design parameters to minimize energy consumption and cost while ensuring performance.

[0185] Therefore, the present invention improves the design accuracy and performance of the swirl blades, reduces energy consumption and costs. By comparing numerical simulation and experimental data, the accuracy and reliability of the design can be guaranteed, thereby obtaining better performance in practical applications. In addition, the application of CFD technology can also reduce experimental costs and cycles, and improve design efficiency.

[0186] Further, if Fig.12 As shown, the present embodiment also provides a multi-channel swirl chaotic metallurgical spray gun design system based on graph convolutional neural network. In the present embodiment, a multi-channel swirl chaotic metallurgical spray gun design system based on graph convolutional neural network is applied to a multi-channel swirl chaotic metallurgical spray gun design and optimization method based on graph convolutional neural network in the above-mentioned embodiment. The multi-channel swirl chaotic metallurgical spray gun design system based on graph convolutional neural network includes a data acquisition and preprocessing module 1, a graph convolutional neural network model module 2, and a swirl spray gun structure optimization module 3.

[0187] Among them, the data acquisition and preprocessing module 1 is responsible for calculating and extracting the required data from the CFD software, projecting the pressure and velocity data onto the grid through linear interpolation, discretizing each shape and the surrounding flow field area into a triangular grid, and each node contains coordinate, velocity and pressure information; normalizing the values ​​of the velocity field and pressure field, reconstructing the data set, and constructing the data set into 80% training set, 10% validation set and 10% test set; inputting the reconstructed data set into the graph convolutional neural network model module 2;

[0188] The graph convolutional neural network model module 2 is responsible for designing the convolutional network neural structure, which consists of multiple convolution blocks. Each convolution block contains a two-step graph convolution layer and a two-step smoothing layer. The last 1x1 convolution layer converts the node features into predicted velocity and pressure fields. The graph convolutional neural network is trained by minimizing the mean absolute error loss function, updating the parameters through the Adam optimizer, and adjusting the hyperparameters according to the performance of the validation set during the training process. The generalization ability of the graph convolutional neural network model is evaluated using the test set, and the MAE between the predicted value and the true value is calculated. The accuracy of the prediction results of the graph convolutional neural network is obtained by comparing the predicted flow field with the CFD simulation results. The final results are output to the swirl spray gun structure optimization module 3.

[0189] The swirl spray gun structure optimization module 3 is responsible for drawing data images such as streamlines, velocity vector diagrams and pressure distribution diagrams, identifying flow, reattachment and swirl formation phenomena, and comprehensively analyzing the flow characteristics of the swirl spray gun. The pressure and shear stress on the surface of the swirl blade are calculated by integration to obtain the drag and lift values, and the pressure gradient force and vortex data are calculated to analyze the local flow characteristics of the swirl spray gun; various parameters of the swirl spray gun are adjusted according to the analyzed characteristics and data, and the adjusted various parameters of the swirl spray gun are input into the CFD software for simulation, and experimental verification is carried out based on the adjusted various parameters of the swirl spray gun, and whether it is qualified is checked based on the simulation results and experimental verification; if it is unqualified, the various parameters of the swirl spray gun are readjusted and the simulation and experiment are repeated.

[0190] Preferably, this embodiment combines computational fluid dynamics (CFD) and graph convolutional neural network (GCNN) technology, and the velocity field and pressure field of the spray gun can be quickly predicted through the grid coordinates of the swirl spray gun and the boundary information of the swirl blades. Compared with the traditional experimental method, the design cycle is significantly shortened, the experimental cost is reduced, and the research and development efficiency is improved. The repeatedly trained and verified GCNN model has a strong generalization ability and a high accuracy, ensuring that the spray gun can still provide reliable prediction results under different working conditions and variable industrial conditions. Further analysis of the flow characteristics and local physical characteristics of the spray gun helps to understand its complex fluid mechanics behavior, ensure that it can maintain efficient and stable performance in actual working conditions, extend its service life, and improve metallurgical reaction efficiency. It reduces the frequency of equipment replacement and resource consumption from the source, while reducing energy consumption and pollution emissions in the process, helping to achieve green and sustainable development of industrial production.

[0191] Therefore, this embodiment combines computational fluid dynamics (CFD) and graph convolutional neural network (GCNN) technology to predict the flow characteristics of spray guns under different geometric structures.

[0192] like Fig.13 As shown, this embodiment provides an embodiment of an electronic device. In this embodiment, the electronic device 8 includes a processor 41 and a memory 42 coupled to the processor 41.

[0193] The memory 42 stores program instructions for implementing the multi-channel swirl spray gun optimization method based on graph convolution according to any one of the above embodiments.

[0194] The processor 41 is used to execute program instructions stored in the memory 42 to perform multi-channel swirl spray gun optimization based on graph convolution.

[0195] The processor 41 may also be referred to as a CPU (Central Processing Unit). The processor 41 may be an integrated circuit chip having signal processing capabilities. The processor 41 may also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0196] Further, Fig.14The schematic diagram of the structure of the storage medium of an embodiment of the present application is that the storage medium 5 of the embodiment of the present application stores program instructions 51 that can implement all the above methods, wherein the program instructions 91 can be stored in the above storage medium in the form of a software product, including several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) or a processor to execute all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, or terminal devices such as a computer, a server, a mobile phone, and a tablet.

[0197] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0198] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above integrated unit may be implemented in the form of hardware or in the form of software functional units. The above is only an implementation mode of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly used in other related technical fields, is also included in the patent protection scope of the present invention.

[0199] The specific implementation methods of the invention are described in detail above, but they are only examples, and the invention is not limited to the specific implementation methods described above. For those skilled in the art, any equivalent modification or substitution of the invention is also within the scope of the invention, therefore, the equalization, modification, improvement, etc. made without departing from the spirit and principle of the invention should be included in the scope of the invention.

Claims

1. A multi-channel swirl spray gun optimization method based on graph convolution, characterized in that: The graph convolution-based cyclone metallurgical spray gun design method comprises: Use CFD software to simulate and calculate the swirl spray gun under different geometric structures, obtain the three velocity fields and pressure fields of the swirl spray gun, a total of four three-dimensional flow field data, obtain a data set, convert the three-dimensional data into a graph structure, and process the obtained data to reconstruct a new data set; Design the graph convolutional neural network model structure and train the graph convolutional neural network model by minimizing the mean absolute error loss function; the last 1x1 convolutional layer converts the node features into predicted velocity and pressure fields; use the Adam optimizer to update the parameters and adjust the hyperparameters according to the validation set during the training process; compare the predicted flow field with the CFD simulation results; According to the velocity components and pressures in three directions of each node output by the graph convolutional neural network model, the velocity and pressure fields of the entire flow field are reconstructed; by drawing data images and calculating the pressure and shear stress on the surface of the swirl blades based on the data images, the distribution of velocity and pressure and the values ​​of drag and lift are obtained; based on a comprehensive and detailed analysis of the flow characteristics and local physical characteristics of the swirl spray gun, various parameters of the swirl spray gun are continuously adjusted and optimized, and the adjusted parameters are input into the CFD software simulation. The simulation results and experiments verify whether the optimization is qualified. If not, adjustments will continue.

2. The multi-channel swirl spray gun optimization method based on graph convolution according to claim 1 is characterized in that: The process of rebuilding a new dataset includes: The key areas are encrypted, including the boundary of the swirl blade and the outlet of the spray gun; CFD software is used to simulate and calculate the swirl spray gun under different geometric structures; after calculation, four three-dimensional flow field data of the swirl spray gun, three velocity fields and one pressure field, are obtained; Convert the three-dimensional data into a graph structure and project the pressure and velocity data onto the grid by linear interpolation. Discretize each shape and the flow field area around it into a triangular grid, with each node containing coordinates, velocity and pressure information. Normalize the values ​​of the velocity field and pressure field. Each node in the grid is considered as a node in the graph structure, and the edge is defined by the spatial distance. The boundary conditions of the swirl blades are processed by SDF, and the data is assigned to each node as a feature. The reconstructed dataset contains 80% training set, 10% validation set and 10% test set, and the reconstructed dataset is fed into the convolutional network model.

3. The multi-channel swirl spray gun optimization method based on graph convolution according to claim 2 is characterized in that: The process of discretizing into a triangular network includes: Collect the point set of triangulation, select an initial triangle, the three vertices must be three points in the point set, and these three points cannot be collinear; the point set includes the key points at the boundary of the swirl blade and the outlet of the spray gun; Insert the rest of the points in the point set into the current triangulated network; each time a new point is inserted, check whether the new point will cause some triangular units to no longer conform to the empty circle characteristic; if not, delete the non-conforming triangular units to form a hole; Connect the new point to each vertex that makes up the hole to generate a new edge; update the triangulated network based on the new edge, and repeat the above process until all points are inserted into the mesh.

4. The multi-channel swirl spray gun optimization method based on graph convolution according to claim 1, characterized in that: The process of comparing the predicted flow field with the CFD simulation results includes: Design a graph convolutional neural network model, which consists of multiple graph convolution blocks. Each convolution block contains two-step graph convolution layers and two-step smoothing layers. The last 1x1 convolution layer converts the node features into predicted velocity and pressure fields. The graph convolutional neural network model is trained by minimizing the mean absolute error loss function, using the Adam optimizer to update parameters, and adjusting hyperparameters based on the results of the validation set during training; Calculate the absolute error between the predicted and true values; compare the predicted flow field with the CFD simulation results.

5. The multi-channel swirl spray gun optimization method based on graph convolution according to claim 4 is characterized in that: The process of designing a graph convolutional neural network model includes: The graph convolutional neural network model consists of multiple graph convolution blocks, each of which contains two-step graph convolution layers and two-step smoothing layers. New edge features are calculated through edge convolution and fused to the nodes. New node features are calculated using the updated node features and edge information; the coordinates of the input layer are directly connected to each smoothing layer to provide spatial information for edge convolution; Use a 1x1 convolutional layer to convert node features into predicted flow and pressure fields; The 1x1 convolutional layer is equivalent to a fully connected layer.

6. The multi-channel swirl spray gun optimization method based on graph convolution according to claim 1, characterized in that: The process of adjusting and optimizing various parameters of the swirl spray gun includes: According to the velocity components and pressure in three directions of each node output by the graph convolutional neural network model, the velocity field and pressure field of the entire flow field are reconstructed; By drawing streamlines, velocity vector diagrams and pressure distribution data images, the distribution of velocity and pressure is obtained; according to the drawn pressure distribution diagram, the area on the surface of the swirl blade is selected for integral calculation of the pressure and shear stress on the surface of the swirl blade, and the drag and lift values ​​are calculated based on the integral results of the pressure and shear stress; based on a comprehensive and detailed analysis of the flow characteristics and local physical characteristics of the swirl spray gun, various parameters of the swirl spray gun are continuously adjusted; Analyze the calculation results and verify through CFD software simulation and experimental verification whether the various parameters of the adjusted swirl spray gun are qualified. If not, continue to adjust the swirl spray gun parameters.

7. The multi-channel swirl spray gun optimization method based on graph convolution according to claim 6, characterized in that: The process of obtaining drag and lift values ​​includes: Obtain the flow field velocity data and pressure data in the graph convolutional neural network model, and determine the vertex data for drawing the flow field velocity and pressure according to each data; draw a data image of the distribution of the flow field velocity and pressure according to the vertex data; Identify the drawn pressure distribution diagram, select the area on the swirl blade surface for integral calculation, and obtain the pressure and shear stress on the blade surface; calculate the drag and lift values ​​of the swirl spray gun based on the integral structure of pressure and shear stress; calculate the pressure gradient force and vortex data to obtain the local flow characteristics and complex fluid mechanics behavior of the swirl blade area; The calculation results are analyzed, and the flow characteristics and local physical characteristics of the swirl spray gun are analyzed through numerical simulation and experimental data; among them, the flow characteristics include vortex and separation phenomenon, and the local physical characteristics include pressure center and velocity peak.

8. The multi-channel swirl spray gun optimization method based on graph convolution according to claim 7, characterized in that: The local flow characteristics and complex fluid dynamics behavior of the region include: According to the drawn pressure distribution diagram, the surface area of ​​the swirl blade is identified and selected, and the pressure distribution is integrated to obtain the total pressure and shear stress on the blade surface; According to the pressure value of the pressure sensor in the swirl blade surface area, the pressure value of the sensor is integrated to obtain the pressure on the blade surface; the pressure on the blade surface is expressed as: P total =∫ A P(x,y)dA Where P(x,y) is the pressure distribution function, and A is the area of ​​the blade surface; Shear stress is realized by solving the moment and swirl blade surface; first derive the force F in the x direction of abcd t , using the equation Get the expressions for forces F1 and F2; substitute the expressions into the equation: Tt dx=F1-F2 Get Moment The ratio of dx is equal to M z Multiply by S d A y The integral of V y The shear stress on the blade surface is finally obtained by being parallel to the y-axis and positively or negatively with the y-axis; the shear stress on the blade surface is: Among them, M z is the moment about the z-axis, I z is the moment of inertia of the section, y and z are the coordinates on the section; According to the integral results of pressure and shear stress, the drag and lift values ​​of the swirl spray gun are calculated; d p =f(V axial ,G·P) Where a is the surface tension coefficient; The pressure gradient force is calculated based on the above results; the pressure gradient force calculation formula is: where ρ is the fluid density, is the pressure gradient derived from the above pressures; Lift and drag are calculated using the integral of pressure and shear stress, where the formula for lift, L, is: L=∫ A (P top -P tottom )dA The formula for resistance D is: D=∫ A (t top +t bottom )dA Among them, the above-mentioned P top and P tottom are the pressure on the upper and lower surfaces, τ top and τ bottom are the shear stresses on the upper and lower surfaces, respectively; The calculation formula of vorticity Γ is: Among them, ω x and ω y are the components of vorticity in the x and y directions, respectively; By analyzing the distribution of pressure gradient force and vorticity, it is possible to determine whether the flow is separating, reattaching or vortex formation occurs.

9. The multi-channel swirl spray gun optimization method based on graph convolution according to claim 6, characterized in that: The process of analyzing the flow characteristics and local physical characteristics of the swirl spray gun includes: By drawing streamlines, velocity vector diagrams and pressure distribution data images, flow separation, reattachment and vortex formation phenomena can be identified; by integrating and calculating the pressure and shear stress on the surface of the swirl blade, the drag and lift values ​​can be obtained; by calculating the pressure gradient force and vortex data through the drag and lift values, the local flow characteristics and complex fluid mechanics behavior of the swirl blade area can be obtained; The obtained results are output to CFD software for simulation, and the simulation results are compared with the experimental data. A threshold is set. If the value is greater than the threshold, the numerical simulation result is considered unqualified and the design parameters of the swirl spray gun are adjusted. The adjusted parameters are simulated and analyzed again, and the design parameters of the swirl spray gun are iterated until they are less than the threshold value.

10. A multi-channel swirl spray gun optimization system based on graph convolution, which is applied to a multi-channel swirl spray gun design and optimization method based on graph convolution as claimed in any one of claims 1 to 9, characterized in that: The multi-channel swirl spray gun optimization and optimization system based on graph convolution includes: The data acquisition and preprocessing module is responsible for calculating and extracting the required data from the CFD software, projecting the pressure and velocity data onto the grid through linear interpolation, discretizing each shape and the surrounding flow field area into a triangular grid, and each node contains coordinate, velocity and pressure information; normalizing the values ​​of the velocity field and pressure field, reconstructing the data set, and constructing the data set into 80% training set, 10% validation set and 10% test set; inputting the reconstructed data set into the graph convolutional neural network model module; The graph convolutional neural network model module is responsible for designing the convolutional network neural structure, which consists of multiple convolution blocks. Each convolution block contains two-step graph convolution layers and two-step smoothing layers. The last 1x1 convolution layer converts the node features into predicted velocity and pressure fields. The graph convolutional neural network is trained by minimizing the mean absolute error loss function, updating the parameters through the Adam optimizer, and adjusting the hyperparameters according to the performance of the validation set during the training process. The generalization ability of the graph convolutional neural network model is evaluated using the test set, and the MAE between the predicted value and the true value is calculated. The accuracy of the prediction results of the graph convolutional neural network is obtained by comparing the predicted flow field with the CFD simulation results. The final results are output to the swirl spray gun structure optimization module. The swirl spray gun structure optimization module is responsible for drawing streamlines, velocity vector diagrams and pressure distribution diagram data images, identifying flow, reattachment and swirl formation phenomena, and comprehensively analyzing the flow characteristics of the swirl spray gun. The pressure and shear stress on the surface of the swirl blade are calculated by integration to obtain the drag and lift values, and the pressure gradient force and vortex data are calculated to analyze the local flow characteristics of the swirl spray gun; various parameters of the swirl spray gun are adjusted according to the analyzed characteristics and data, and the adjusted various parameters of the swirl spray gun are input into the CFD software for simulation, and experimental verification is carried out based on the adjusted various parameters of the swirl spray gun, and whether it is qualified is checked according to the simulation results and experimental verification; if it is unqualified, the various parameters of the swirl spray gun are readjusted and the simulation and experiment are repeated.

Images