Micromixer automatic design and performance prediction method based on big data and machine learning

Abstract

The invention discloses a micro-mixer automatic design and performance prediction method based on big data and machine learning, and the method comprises the following steps: 1, selecting a random on-off micro-mixer; 2, a grid random on-off micro mixer design parameter library is built through joint simulation, and an original database is obtained; 3, building a convolutional neural network model, performing digital structure adjustment on the original database to obtain a simulation database, and training the convolutional neural network model by using the simulation database to predict an outlet mixing efficiency value Cout of the micro mixer; according to the method, the number of the micro-mixer design parameter libraries can be quickly increased, so that more micro-mixer designs with the mixing efficiency not less than 95% can be quickly obtained.

Description

Technical Field

[0001] The present invention relates to the field of microfluidic technology, and in particular to an automated design and performance prediction method for a micromixer based on big data and machine learning. Background Art

[0002] As one of the core components of microfluidic chips, the main function of a micromixer is to achieve fast, efficient and uniform mixing of two or more liquids at a microscale. The design of a micromixer plays a vital role in fields such as biochemical analysis, medical diagnosis, drug development and environmental monitoring. Researchers have designed and developed various forms of micromixers for different application scenarios, which are generally divided into two categories: active micromixers and passive micromixers. Relatively speaking, passive micromixers promote natural mixing between fluids through sophisticated internal structure design, and are favored for their simple structure, low cost and easy integration.

[0003] Existing micromixer designs are all based on a hypothetical high-efficiency mixing design, which is imported into microfluidic simulation software through 3D modeling for microfluidic simulation. If the mixing efficiency in the simulation results does not meet the requirements, it is necessary to return to the modeling software to change the shape, size, structure and other parameters of the micromixer, and perform microfluidic simulation again, and continuously iterate and optimize the design until the mixing efficiency of the micromixer meets the actual requirements. Its design process faces technical challenges such as long design cycle, low mixing efficiency, complex microfluidic dynamics and difficulty in adapting to diverse experimental needs. Summary of the invention

[0004] The purpose of this section is to summarize some aspects of embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the specification abstract and the invention title of this application to avoid blurring the purpose of this section, the specification abstract and the invention title, and such simplifications or omissions cannot be used to limit the scope of the present invention.

[0005] In view of the above and / or existing problems in the design of existing micro mixers, the present invention is proposed.

[0006] Therefore, the purpose of the present invention is to provide a method for automated design and performance prediction of a micromixer based on big data and machine learning, which solves the technical problems of long mixer design cycle and low efficiency in the prior art. The present invention can quickly expand the number of micromixer design parameter libraries, thereby quickly obtaining more micromixer designs with a mixing efficiency of not less than 95%.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions: a method for automatic design and performance prediction of a micro-mixer based on big data and machine learning, which comprises the following steps: 1. Select the random on-off micro mixer; 2. Joint simulation builds a grid random on-off micro-mixer design parameter library to obtain the original database; 3. Build a convolutional neural network model, adjust the digital structure of the original database, obtain a simulation database, and use the simulation database to train the convolutional neural network model to predict the outlet mixing efficiency value Cout of the micromixer.

[0008] As a preferred solution of the method for automated design and performance prediction of micromixers based on big data and machine learning in the present invention, wherein: in the step 1, a grid random micromixer design parameter library is established based on a 4×4 grid micromixer.

[0009] As a preferred solution of the micromixer automated design and performance prediction method based on big data and machine learning in the present invention, wherein: the cross-sectional size of the 4×4 grid random micromixer channel is , the length of the internal branch channel is 3 mm, the internal quadrilateral is a rhombus with an angle of 90°, the external boundary triangle is an isosceles right triangle, and the length of the external boundary branch channel is mm.

[0010] As a preferred solution of the method for automated design and performance prediction of micromixers based on big data and machine learning in the present invention, wherein: in the step 2, COMSOL Multiphysics and MALAB are jointly simulated for a part of the feature design of the micromixer population through the LiveLink for MATLAB interface. In the joint simulation tool, the MALAB program controls the random on and off of the grid micromixer, continuously generates new micromixer designs, and records channel information. The COMSOL Multiphysics software is responsible for microfluidic simulation of each newly generated micromixer and quantifying the mixing efficiency value of each micromixer outlet.

[0011] As a preferred solution of the method for automated design and performance prediction of a micromixer based on big data and machine learning in the present invention, a two-dimensional laminar steady-state microfluidic simulation model is established in COMSOL Multiphysics to determine simulation parameters and realize automated joint simulation, and a randomly selected micromixer is subjected to mesh densification processing, and the mesh size is adjusted to reach the critical value of the maximum number of Newton iterations of the steady-state solver by adjusting the parameters of the free triangular mesh.

[0012] As a preferred solution of the micro-mixer automated design and performance prediction method based on big data and machine learning in the present invention, wherein: during the joint simulation, the Reynolds number Re is 4.

[0013] As a preferred solution of the method for automatic design and performance prediction of a micro-mixer based on big data and machine learning in the present invention, wherein: in the step 3, the physical model of the micro-mixer is converted into a digital matrix model, and the on and off of the 80 branch channels of the micro-mixer are represented by the numbers 0 and 1. When randomly on and off, the disconnected channel is set to 0, and the reserved channel is set to 1. The order and direction of 0 and 1 are the same as the branch channel number. In some positions of the 10×10 digital matrix where there is no branch channel corresponding to it, it is represented by Complement them to form a complete 10×10 matrix structure.

[0014] As a preferred solution of the micro-mixer automated design and performance prediction method based on big data and machine learning in the present invention, when training the convolutional neural network model, Replace with 0 to form a 10×10 pure digital matrix model.

[0015] As a preferred solution of the micro-mixer automated design and performance prediction method based on big data and machine learning in the present invention, wherein: in the simulation database, the first 100 columns of data of each item represent the digital matrix model of the micro-mixer, and the 101st column of data represents the mixing efficiency value of the corresponding micro-mixer.

[0016] As a preferred solution of the method for automated design and performance prediction of a micro-mixer based on big data and machine learning in the present invention, data with mixing efficiency between 50% and 100% in the simulation database are selected as a subset of the simulation library, 80% of the data are randomly selected as a training set, and the remaining 20% ​​of the data are used as a test set.

[0017] Compared with the prior art, the present invention has the following technical effects: the present invention selects a method for randomly switching on and off the branch channels of the micro-mixer, and establishes a micro-mixer database by randomly switching on and off the branch channels, so that the database contains as many designs of micro-mixers as possible, and users can select from the database according to their needs; when establishing the database, the plug-in COMSOL Multiphysics with MALAB's joint simulation tool automatically realizes the random on and off of the micromixer, and imports the randomly on and off micromixers into COMSOL for microfluidic simulation. Finally, the channel information of each micromixer and the mixing efficiency parameters of the outlet are exported. Then, the machine learning (convolutional neural network CNN model) method is introduced, and the channel information of the micromixer is used as the input of the CNN model, and the mixing efficiency value of the outlet is used as the output of the CNN model. In this way, the CNN model is trained so that the CNN model can learn what kind of micromixer channel corresponds to what outlet mixing efficiency value. Once the CNN model has this learning ability, the channel information of the new micromixer can be input into the CNN model, and the mixing efficiency value of each corresponding micromixer can be predicted within a few seconds, so as to achieve the purpose of quickly and accurately recording the design parameters of the micromixer, thereby establishing a micromixer database and improving the design efficiency of the micromixer. In the established micromixer database, there are more micromixers with a mixing efficiency of not less than 95% to ensure the design effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them: Figure 1 The diagram is a transformation diagram of the determination of the grid size of the micro-mixer of the present invention and the input information of the CNN model; (A) is a design of 5 grid micro-mixers of different sizes, including the total number of chips and channel numbers generated by the random on-off of the corresponding channels; (B) is a design generated after the grid micro-mixer d in (A) is randomly on-off; (C) A complete grid micro-mixer model with a grid size of 4×4; (D) A micro-mixer design after the complete 4×4 grid micro-mixer is randomly on-off; (E) is a 10×10 matrix structure corresponding to (D), the positions of 0 and 1 correspond to the branch channel number positions one by one, and the positions of the non-branch channels are represented by To supplement; (F) After replacing with 0, the micro-mixer solid model is converted into a 10×10 pure digital matrix model.

[0019] Figure 2The microfluid simulation performance diagram of the micromixer under different simulation parameters in the present invention, (A) 6 cross-section cut positions of the grid random micromixer; (B) Fluid mixing performance at 6 cross-section positions under different Reynolds numbers; (C) The mixing efficiency change trend at the cross section f×f when Re increases from 0.05 to 10; (D) The mixing efficiency change trend at the cross section f×f when Re is between 0 and 1.

[0020] Figure 3 Schematic diagram of the CNN model structure in the present invention (a 10×10 digital matrix composed of 0 and 1 is used as the input of the CNN model. After two convolution operations, two maximum pooling operations and a fully connected layer, the outlet mixing efficiency value Cout of the micro-mixer is finally used as the output of the CNN model).

[0021] Figure 4 The CNN model iteration process of the full simulation library and a subset of the simulation library in the present invention and the expansion diagram of the database; (A) 1000 iteration training results of the full simulation library; (B) 1000 iteration training results of the subset of the simulation library; (C) 9691 micro-mixer channel number and outlet mixing efficiency parameter diagram in the simulation database; (D) 47030 micro-mixer channel number and outlet mixing efficiency parameter diagram after prediction by the CNN model, the 47303 data items include 9691 simulation library data and 37339 CNN model prediction data.

[0022] Figure 5 The microfluidic simulation performance of the sampled micromixers in the present invention is used to verify the predicted performance of the CNN model; (A) A concentration variation diagram of 12 micromixer designs and 6 micromixer designs with mixing efficiency greater than 90% randomly selected from the CNN model prediction database for microfluidic simulation according to the mixing efficiency gradient; (B) Comparison of the simulated values ​​and predicted values ​​of 18 micromixers. DETAILED DESCRIPTION

[0023] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0024] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0025] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0026] Example 1 Reference Figure 1 to Figure 3 This embodiment provides a micro-mixer automated design and performance prediction method based on big data and machine learning, which can quickly expand the number of micro-mixer design parameter libraries and quickly obtain more micro-mixer designs with a mixing efficiency of not less than 95%.

[0027] A method for automatic design and performance prediction of a micro-mixer based on big data and machine learning, comprising the following steps: 1. Select the random on-off micro mixer; 2. Joint simulation builds a grid random on-off micro-mixer design parameter library to obtain the original database; 3. Build a convolutional neural network model, adjust the digital structure of the original database, obtain a simulation database, and use the simulation database to train the convolutional neural network model to predict the outlet mixing efficiency value Cout of the micromixer.

[0028] In step 1, the branch channels constituting the mixing channel are randomly switched on and off. Each branch channel has a 50% probability of being retained in the design of the micromixer, but also a 50% probability of not being retained. If a micromixer is composed of i branch channels, then 2 i A random micromixer design.

[0029] In this application, a grid random micromixer design parameter library is established based on a 4×4 grid micromixer. The cross-sectional dimensions of the 4×4 grid random micromixer channel are , the length of the internal branch channel is 3 mm, the internal quadrilateral is a rhombus with an angle of 90°, the external boundary triangle is an isosceles right triangle, and the length of the external boundary branch channel is mm, which has 80 branch channels, thus generating 2 80 A random micromixer design.

[0030] In step 2, COMSOL Multiphysics and MALAB are combined through the LiveLink for MATLAB interface to jointly simulate a part of the characteristic design of the micromixer population. In the joint simulation tool, the MALAB program controls the random on and off of the mesh micromixer, continuously generates new micromixer designs, and records channel information. The COMSOL Multiphysics software is responsible for microfluidic simulation of each newly generated micromixer, quantifies the mixing efficiency value at the outlet of each micromixer, and establishes a two-dimensional laminar steady-state microfluidic simulation model in COMSOL Multiphysics to determine the simulation parameters and realize automated joint simulation. The mesh of a randomly selected micromixer is densified, and the mesh size is adjusted to the critical value of the maximum Newton iteration number of the steady-state solver by adjusting the parameters of the free triangular mesh. If the mesh is further refined, the returned solution does not converge, and when the mesh size parameter is adjusted near the mesh size parameter, it is found that the simulation result has not changed much and can be ignored, so as to determine the mesh independence parameter of the mesh mixer.

[0031] In step 3, the physical model of the micro-mixer is converted into a digital matrix model. The on and off of the 80 branch channels of the micro-mixer are represented by the numbers 0 and 1. When randomly on and off, the disconnected channels are set to 0 and the remaining channels are set to 1. The order and direction of 0 and 1 are the same as the branch channel numbers. In some positions of the 10×10 digital matrix where there are no branch channels corresponding to them, The micro-mixer is randomly switched on and off, and its spatial position corresponds to the matrix position. When used for training CNN model input data, it can more accurately reflect the mixing performance in actual situations and more accurately predict the micro-mixer information parameters.

[0032] When training a convolutional neural network model, Replace with 0 to form a 10×10 pure digital matrix model.

[0033] In order to meet the input and output requirements of the CNN model, the digital structure of the original database is adjusted. The adjusted database is defined as a simulation database, in which the first 100 columns of data of each item represent the digital matrix model of the micromixer, and the 101st column of data represents the mixing efficiency value of the corresponding micromixer.

[0034] The present invention selects a method for randomly switching on and off the branch channels of the micro-mixer, and establishes a micro-mixer design parameter library through the random switching on and off of the branch channels, so that the database contains as many designs of micro-mixers as possible, and users can select them in the database according to their needs; when establishing the database, the plug-in COMSOL Multiphysics with MALAB's joint simulation tool automatically realizes the random on and off of the micromixer, and imports the randomly on and off micromixers into COMSOL for microfluidic simulation. Finally, the channel information of each micromixer and the mixing efficiency parameters of the outlet are exported. Then, the machine learning (convolutional neural network CNN model) method is introduced, and the channel information of the micromixer is used as the input of the CNN model, and the mixing efficiency value of the outlet is used as the output of the CNN model. In this way, the CNN model is trained so that the CNN model can learn what kind of micromixer channel corresponds to what outlet mixing efficiency value. Once the CNN model has this learning ability, the channel information of the new micromixer can be input into the CNN model, and the mixing efficiency value of each corresponding micromixer can be predicted within a few seconds, so as to achieve the purpose of quickly and accurately recording the design parameters of the micromixer, thereby establishing a micromixer database and improving the design efficiency of the micromixer. In the established micromixer database, there are more micromixers with a mixing efficiency of not less than 95% to ensure the design effect.

[0035] Example 2 Reference Figure 2 This embodiment provides an automated design and performance prediction method for a micromixer based on big data and machine learning. The difference between this embodiment and Embodiment 1 is that it can further improve the fluid mixing effect of the micromixer.

[0036] In the specific joint simulation, the Reynolds number Re is 4.

[0037] When determining the microfluidic simulation parameters of the micromixer, COMSOL Multiphysics 6.2 is used to simulate the mixing situation in the micromixer, with the Reynolds number (Re) as a variable. The calculation formula of Re is as follows: ; In the formula, dynamic viscosity ,density , v is the fluid velocity, m / s, D h is the characteristic length of the mixing channel, m.

[0038] In order to compare the fluid mixing effect under different Reynolds numbers, the outlet section is selected for evaluation and divided into an infinite number of unit surfaces when dividing the grid. The mixing effect is evaluated by calculating the standard deviation of the volume fraction of the component on each unit surface. The fluid mixing performance evaluation equation (2) is: ; In the formula, is the volume fraction when the fluid is completely mixed, n is the number of points on the cross section, is the volume fraction at point i. When M<80%, the fluid mixing effect is poor, when 80%≤M<95%, the fluid mixing effect is poor, and when M≥95%, the fluid mixing effect is good.

[0039] In order to improve the computational efficiency, save computational resources, and quickly obtain the outlet outflow parameters of 9691 micromixers, a two-dimensional laminar steady-state microfluidic simulation model was established in COMSOL Multiphysics 6.2 to determine the simulation parameters and realize automated joint simulation. In order to study the mixing performance of the micromixer and simulate the fluid mixing under different Re, this embodiment randomly selected a micromixer from the randomly generated 9691 micromixers to explore parameters such as mesh size and normal inflow velocity.

[0040] When performing microfluid simulation, the grid size parameter directly affects the accuracy and computational efficiency of the simulation. Generally speaking, the denser the grid, the more accurate the simulation result, but the higher the computational cost. Since 9691 randomly generated micromixers are different, it is impossible to perform a detailed analysis of the grid density for each micromixer when performing microfluid simulation. In order to make the simulation result as accurate as possible, the present embodiment performs grid densification processing on a randomly selected micromixer. By adjusting the parameters of the free triangular grid, the grid size is made to reach the critical value of the maximum Newton iteration number of the steady-state solver. If the grid is further refined, the returned solution does not converge, and when adjusting near the grid size parameter, it is found that the simulation result has not changed much and can be ignored. To determine the grid independence parameters of the grid mixer, detailed microfluid simulation parameters are given in Table 1.

[0041] To determine the normal inflow velocity at the inlet, Figure 2 (A) Mixing conditions of six 0.4 mm × 0.4 mm rectangular sections cut from the mixing channel of the micromixer. Figure 2 It can be seen from (B) that with the increase of Re, the mixing conditions of cross sections a×a and b×b did not change significantly, while the mixing efficiency of cross sections c×c, d×d, e×e, and f×f was significantly improved. When Re=4, the mixing efficiency of cross section f×f reached 98.13%, and the mixing effect was already excellent at this time.

[0042] exist Figure 2(C) and (D) show the changes in the mixing efficiency at the cross section f×f when Re increases from 0.05 to 10. It can be seen from the figure that the mixing efficiency at the cross section f×f shows a trend of decreasing first and then increasing. When Re<1, the mixing of microfluids mainly depends on the diffusion effect between molecules. Since the length of the mixing channel of the randomly switched micromixer is long or short and not fixed, for the micromixer with a long mixing channel, a smaller Re may require a longer mixing time, which limits the application of microfluidic chips in instant detection equipment. Fluids under too high Re are not only difficult to control, but also easily damage the microfluidic chip channels made of PDMS. In addition, the mixing channel length of some micromixers is short, and a larger Re will make the mixing time of the two fluids too fast, and the ideal mixing efficiency cannot be achieved in actual operation.

[0043] In this embodiment, Re=4 is selected as the final flow parameter, and the normal inflow velocity at the corresponding inlet is 10 mm / s.

[0044] Example 3 Reference Figure 4 and Figure 5 This embodiment provides an automated design and performance prediction method for a micro-mixer based on big data and machine learning. The difference between this embodiment and embodiments 1 and 2 is that it can further improve the efficiency of mixing efficiency prediction.

[0045] Specifically, the data with mixing efficiency between 50% and 100% in the simulation database are selected as the simulation library subset, 80% of the data are randomly selected as the training set, and the remaining 20% ​​of the data are used as the test set.

[0046] Figure 2 As shown in Figure 1, CNN mainly consists of input layer, convolution layer, pooling layer, fully connected layer and output layer. In order to better train the CNN model and predict the mixing efficiency parameters, a 10×10 digital matrix ( Figure 1 , F) as input, the 1×1 outlet mixing efficiency value C OUT As output. By continuously optimizing the CNN model parameters to better predict the fluid behavior of the grid random micromixer, Table 2 gives the specific parameter configuration of the CNN structure.

[0047] Table 2 CNN parameter configuration In the process of training CNN models, mean square error (MSE), mean absolute error (MAE) and mixed accuracy (QUOTE ) as the evaluation parameter of the CNN model, the formula includes, ; ; ; In the formula, n represents the total number of items in the training set or test set, i represents the index of each item in the training set or test set, y i represents the simulation value of the mixing efficiency of each micromixer, Represents the CNN model prediction value of the corresponding micromixer mixing efficiency.

[0048] During the training process, the learning rate of the CNN model was set to 0.0005, and MSE was used as the loss function of the CNN model. The full set of simulation database and the subset of simulation database were iteratively trained respectively, and the loss of each iteration was recorded.

[0049] Figure 4 (A) shows the results after 1000 iterations of training of the entire simulation library, where The loss rate is 87.3%, and the loss rate is 7.92×10 -3 , MAE is 7.12×10 -2 , the test set The loss rate is 79.2%, and the loss rate is 2.19×10 -2 , MAE is 1.20×10 -1 . Figure 4 (B) shows the results of the simulation database subset after 1000 iterations of training. The loss rate is 91.2%, and the loss rate is 6.07×10 -3 , MAE is 5.93×10 -2 , the test set The loss rate is 89.2%, and the loss rate is 8.31×10 -3 , MAE is 7.62×10 -2 .

[0050] From the above, we can see that the training effect of the simulation library subset is significantly better than that of the simulation library full set. It can be seen that the CNN model trained with the subset It is higher, indicating that the fitting of the micromixer mixing efficiency value on the subset is better and the prediction effect is better, which can be used for subsequent prediction tasks. Although the training effect of the entire simulation library is poor, it can also provide a screening basis for subsequent prediction tasks.

[0051] In order to quickly obtain more micro-mixer design parameters without using joint simulation, the trained CNN model was used to predict the mixing efficiency values ​​of 37,339 regenerated micro-mixer outlets. Using the CNN model prediction and screening method, the micro-mixer simulation database can be expanded by nearly 4 times in less than 30 minutes, which is nearly 4,800 times faster than joint simulation. Figure 4 (C) and (D) show the trend change diagram of the grid random micromixer design parameter library from 9691 simulation database data to 47030 CNN model prediction database data. It can be seen from the figure that with the expansion of the micromixer database, the distribution points of the mixing efficiency data are more dense. The CNN model method significantly improves the data coverage of the loose space in the simulation library, providing more possibilities for the diversified selection of micromixers.

[0052] The design of the micromixer in the CNN model was spot-checked to further prove that the micromixer designed using the present application meets the error requirements and the accuracy of the predicted mixing efficiency.

[0053] 18 micromixer designs were randomly selected from 37,339 micromixer designs for traditional microfluidic simulation (i.e., each micromixer was imported into COMSOL Multiphysics for microfluidic simulation), and the mixing efficiency at the micromixer outlet was quantified. The concentration changes of the 18 micromixer simulations are shown in Figure 2. Figure 5 (A) As shown. Figure 5 In (B), the simulation values ​​of the corresponding micromixers are compared with the predicted values. It is found that among these 18 micromixers, the relative deviation between the simulation value and the predicted value is as small as 0.245% and as large as 7.162%. It is 89.2%, so the relative deviation is within 10.8% and meets the requirements, which also proves that the present application can be used to predict microfluidic mechanical properties and build a grid random micromixer design parameter library.

[0054] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for automated design and performance prediction of micromixers based on big data and machine learning, characterized in that: It includes the following steps, 1. Select the random on-off micro mixer; 2. Joint simulation builds a grid random on-off micro-mixer design parameter library to obtain the original database; 3. Build a convolutional neural network model, adjust the digital structure of the original database, obtain a simulation database, and use the simulation database to train the convolutional neural network model to predict the outlet mixing efficiency value Cout of the micromixer.

2. The method for automated design and performance prediction of a micro-mixer based on big data and machine learning as claimed in claim 1, characterized in that: In the step 1, a grid random micromixer design parameter library is established based on a 4×4 grid micromixer.

3. The method for automatic design and performance prediction of a micro-mixer based on big data and machine learning as claimed in claim 2, characterized in that: The cross-sectional dimensions of the 4×4 grid random micromixer channel are , the length of the internal branch channel is 3 mm, the internal quadrilateral is a rhombus with an angle of 90°, the external boundary triangle is an isosceles right triangle, and the length of the external boundary branch channel is mm.

4. The method for automatic design and performance prediction of a micro-mixer based on big data and machine learning as claimed in claim 1, characterized in that: In the step 2, COMSOL Multiphysics and MALAB are combined through the LiveLink for MATLAB interface to perform a joint simulation on a part of the characteristic design of the micromixer population. In the joint simulation tool, the MALAB program controls the random on and off of the grid micromixer, continuously generates new micromixer designs, and records channel information. The COMSOL Multiphysics software is responsible for performing microfluidic simulation on each newly generated micromixer and quantifying the mixing efficiency value at the outlet of each micromixer.

5. The method for automatic design and performance prediction of a micro-mixer based on big data and machine learning as claimed in claim 4, characterized in that: A two-dimensional laminar steady-state microfluidics simulation model is established in COMSOL Multiphysics to determine the simulation parameters and realize automated joint simulation. The mesh of a randomly selected micromixer is densified, and the mesh size is adjusted to reach the critical value of the maximum number of Newton iterations of the steady-state solver by adjusting the parameters of the free triangular mesh.

6. The method for automatic design and performance prediction of a micro-mixer based on big data and machine learning as claimed in claim 5, characterized in that: During the joint simulation, the Reynolds number Re is 4.

7. The method for automatic design and performance prediction of a micro-mixer based on big data and machine learning as claimed in claim 1, characterized in that: In step 3, the physical model of the micro-mixer is converted into a digital matrix model. The on and off of the 80 branch channels of the micro-mixer are represented by the numbers 0 and 1. When randomly on and off, the disconnected channels are set to 0, and the remaining channels are set to 1. The order and direction of 0 and 1 are the same as the branch channel numbers. In some positions of the 10×10 digital matrix where there are no branch channels corresponding to them, Complement them to form a complete 10×10 matrix structure.

8. The method for automatic design and performance prediction of a micro-mixer based on big data and machine learning according to claim 7, characterized in that: When training a convolutional neural network model, Replace with 0 to form a 10×10 pure digital matrix model.

9. The method for automatic design and performance prediction of a micro-mixer based on big data and machine learning according to claim 8, characterized in that: In the simulation database, the first 100 columns of data for each item represent the digital matrix model of the micromixer, and the 101st column of data represents the mixing efficiency value of the corresponding micromixer.

10. The method for automatic design and performance prediction of a micro-mixer based on big data and machine learning according to claim 9, characterized in that: The data with mixing efficiency between 50% and 100% in the simulation database are selected as the simulation library subset, 80% of the data are randomly selected as the training set, and the remaining 20% ​​of the data are used as the test set.

Images