CN112084727A - Transition prediction method based on neural network - Google Patents
Transition prediction method based on neural network Download PDFInfo
- Publication number
- CN112084727A CN112084727A CN202011154867.4A CN202011154867A CN112084727A CN 112084727 A CN112084727 A CN 112084727A CN 202011154867 A CN202011154867 A CN 202011154867A CN 112084727 A CN112084727 A CN 112084727A
- Authority
- CN
- China
- Prior art keywords
- flow field
- transition
- intermittent
- factor
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000007704 transition Effects 0.000 title claims abstract description 79
- 238000000034 method Methods 0.000 title claims abstract description 50
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 33
- 238000004364 calculation method Methods 0.000 claims abstract description 49
- 239000012530 fluid Substances 0.000 claims abstract description 36
- 238000012549 training Methods 0.000 claims abstract description 23
- 238000013507 mapping Methods 0.000 claims abstract description 21
- 230000008569 process Effects 0.000 claims description 15
- 230000004913 activation Effects 0.000 claims description 7
- 210000002569 neuron Anatomy 0.000 claims description 5
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 4
- 238000012937 correction Methods 0.000 claims description 4
- 230000008014 freezing Effects 0.000 claims description 3
- 238000007710 freezing Methods 0.000 claims description 3
- 238000012821 model calculation Methods 0.000 claims description 3
- 230000006870 function Effects 0.000 description 17
- 238000004088 simulation Methods 0.000 description 8
- 238000009826 distribution Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 238000013461 design Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000003062 neural network model Methods 0.000 description 3
- 230000000739 chaotic effect Effects 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 238000010168 coupling process Methods 0.000 description 2
- 238000005859 coupling reaction Methods 0.000 description 2
- 230000006378 damage Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 230000010349 pulsation Effects 0.000 description 2
- 238000000926 separation method Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 210000004027 cell Anatomy 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000004904 shortening Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 230000003746 surface roughness Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000011144 upstream manufacturing Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/28—Design optimisation, verification or simulation using fluid dynamics, e.g. using Navier-Stokes equations or computational fluid dynamics [CFD]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2113/00—Details relating to the application field
- G06F2113/08—Fluids
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/14—Force analysis or force optimisation, e.g. static or dynamic forces
Abstract
The invention discloses a transition prediction method based on a neural network, which comprises the following steps: step A, obtaining a plurality of known intermittent factors and local average characteristic quantity as a training set; step B, training a neural network and obtaining an intermittent factor mapping model by taking the local average characteristic quantity in the training set and the corresponding intermittent factor as an input value and an output value respectively; step C, a computational fluid mechanics solver carries out flow field iterative computation until the flow field computation result is iteratively converged, and a transition flow field prediction result is output; in each iteration step, the computational fluid dynamics solver provides the local flow field average for the pause factor mapping model, and the pause factor mapping model provides the pause factor for the computational fluid dynamics solver. According to the method, the number of partial differential equations needing to be calculated for transition prediction is reduced, the calculation time is greatly shortened, the contradiction that the precision and the calculation efficiency cannot coexist is solved, the calculation efficiency is high, and the calculation precision is high; the invention has good universality because of not depending on empirical formulas.
Description
Technical Field
The invention belongs to the technical field of computational fluid mechanics, and particularly relates to a transition prediction method based on a neural network.
Background
The transition refers to the transition process of the fluid from the layered stable flow to the chaotic turbulence. The boundary layer transition is commonly present in the interior of the fluid machine or on the surface thereof, and is a challenge to be solved in the classical physics. In the operation process of the aircraft, the fluid is influenced by various factors including incoming flow turbulence, object surface roughness, Mach number and the like, and is easy to be converted from layered stable flow to chaotic turbulence. This transition is accompanied by a sharp change in both wall friction and thermal conductivity. Therefore, the rapid and accurate Computational Fluid Dynamics (CFD) simulation of transition flow is of great significance to the improvement of the performance and design efficiency of the aircraft.
The transition prediction can adopt a Direct Numerical Simulation (DNS) method for solving full-scale turbulence pulsation or a Large Eddy Simulation (LES) method for only solving Large-scale pulsation, but the calculation amount increases exponentially with the reynolds number Re, and the development of the current computer technology still cannot meet the calculation requirement of the engineering application. The Reynolds Average Numerical Simulation (RANS) method combined with the transition model still plays a very important role in engineering practice by virtue of the usability and the high efficiency.
The relevance intermittent transition model can relatively accurately predict the transition problem by introducing an 'intermittent factor' to quantitatively describe the generation of turbulence, and is the most popular method in the current engineering transition prediction. In transition flow, the flow field intermittently shows a laminar flow or a turbulent flow at the same spatial position, which is called an intermittent phenomenon. If a function is used to describe this phenomenon and the function value is defined as 0 for laminar flow and 1 for turbulent flow, the pause factor γ is the time average of the function.
The conventional practice of the model is to obtain the required pause factor by explicit equations or by solving additional transport equations. In the early days, Dhawan et al used empirical formulas to describe the flow direction distribution of the pause factor, but the obtained pause factor only depends on the incoming flow and the object plane conditions, does not consider the flow field structure, and is only suitable for simple flow. Libby first introduced a transport equation with a pause factor to calculate turbulence, and then much research work focused on improving the predictive effect of the transport equation on various flows, such as free shear flow, boundary layer transition. However, these models are often started by an improper variable determination transition, that is, when calculating the pause factor of a local grid point, the flow physical quantities in the upstream and downstream regions of the flow field need to be used, and this calculation method is difficult to implement parallel calculation, and is also difficult to be applied in an unstructured grid because the grids are not arranged in sequence. Menter et al uses a Reynolds number with a base strain rate instead of the momentum thickness Reynolds number ReθA four-way SST-gamma-Re based entirely on local variables was developedθA transition model. The model is obtained by introducing two extra Reynolds numbers Re with the intermittent factor gamma and the momentum thickness respectivelyθThe pause factor is calculated for the partial differential equation of the variables and then the SST turbulence model is coupled to simulate the transition process. Many studies at home and abroad show that the model has more accurate prediction capability on transition flow, but the additionally introduced differential equation reduces the solving efficiency of the method. Bas et al propose an algebraic transition model (or BC model) based on the local average quantity, which is more efficient than the algebraic transition model, but cannot accurately simulate the situation of turbulence in the (0.5,2) interval. In addition, the conventional transition model is bound by an empirical formula, is sensitive to a computing platform, needs to calibrate model parameters again for different platforms, and often does not have good universality.
Disclosure of Invention
The invention aims to provide a transition prediction method based on a neural network, aiming at the defects of low calculation efficiency, low calculation precision, dependence on an empirical formula and poor universality of the conventional transition prediction method, so that the calculation time is greatly shortened, the calculation efficiency is high, the calculation precision is high, the empirical formula is not depended on, and the universality is good.
In order to solve the technical problems, the technical scheme adopted by the invention is as follows:
a transition prediction method based on a neural network is characterized by comprising the following steps:
step A, acquiring a plurality of known intermittent factors and corresponding local average characteristic quantities as a training set;
step B, taking the local average characteristic quantity in the training set as an input value, taking an intermittent factor corresponding to the local average characteristic quantity taken as the input value in the training set as an output value, training a neural network and obtaining an intermittent factor mapping model;
step C, a computational fluid mechanics solver carries out flow field iterative computation until the flow field computation result is iteratively converged, and a transition flow field prediction result is output; wherein, the flow field iterative computation process comprises:
step C1, calculating the local average characteristic quantity output by the fluid mechanics solver;
a step C2 of inputting the local average feature quantity obtained in the step C1 to the intermittent factor mapping model;
step C3, solving the intermittent factor mapping model to obtain intermittent factors, and inputting the intermittent factors into a computational fluid mechanics solver;
step C4, judging whether the flow field calculation result of the fluid mechanics solver converges or not based on the intermittent factor obtained in the step C3, and if so, terminating the iteration process; if not, the computational fluid dynamics solver updates the local average feature quantity according to the pause factor value obtained in step C3 and jumps to step C1.
By means of the method, the pure data driven black box mapping model between the local average characteristic quantity and the corresponding intermittent factors is reconstructed by the neural network method based on the known training set data, and the intermittent factor mapping model is obtained. The intermittent factor is solved by a neural network (algebra) method, and on the premise of ensuring the precision, compared with the traditional SST-gamma-Re introducing an additional differential equationθThe model has higher computational efficiency. The unification of transition prediction accuracy and high efficiency is really realized. Book (I)The invention does not need to solve additional differential equations, thereby greatly shortening the calculation time and having high calculation efficiency; meanwhile, the method does not depend on empirical formulas, so that the universality is good.
Preferably, in the step a, the pause factor is derived from known experimental or high-precision transition model calculation data; the local average characteristic quantity corresponding to the intermittent factor is derived from the calculation data of the SA turbulence model which is combined with the existing intermittent factor and subjected to data correction; in the step C, the computational fluid mechanics solver is a computational fluid mechanics solver based on the SA turbulence model. The method combines known data, utilizes the SA turbulence model to control turbulence generation of each space position of a flow field, and designs a data-driven algebraic SA-NN transition model.
In a preferable mode, in the step a, the local average characteristic amount in the local average characteristic amounts includes one or more of a flow field density, a minimum wall surface distance, a free turbulent flow degree, a motion viscosity coefficient, a Q criterion, a normalized strain rate, an SA model flow field variable, a vortex reynolds number, a dimensionless term like a turbulent viscosity, and a pressure gradient along a streamline. The invention utilizes a series of local average characteristic quantities closely related to the intermittent factors and models the mapping relation between the local average characteristic quantities and the intermittent factors by introducing a neural network technology, thereby providing an efficient and accurate method for numerical simulation of transition prediction problems.
Preferably, the neural network comprises six fully-connected layers, two residual blocks containing eight hidden layers and eight fully-connected layers which are arranged from shallow to deep, wherein each hidden layer has 24 neuron nodes, and the activation function of each hidden layer is a linear rectification function (RELU function). Through repeated experimental balance between calculation cost and loss function, the neural network in the form has the optimal effect.
As a preferable mode, in the step C, the flow field calculation result is a flow field density, a flow field velocity, or a flow field pressure.
As a preferable mode, in the step C4, the method for determining whether the flow field calculation result converges includes:if the difference value between the current flow field calculation result and the last flow field calculation result is less than 10-5If not, the convergence is judged.
As a preferred mode, the process of obtaining the local average feature quantity corresponding to the pause factor includes: step A1, substituting the pause factor obtained from the existing data into an SA turbulence model and freezing; step A2, performing iterative computation by a computational fluid dynamics solver until a flow field computation result is converged; in step a3, the computational fluid dynamics solver outputs the local average feature quantity.
Compared with the prior art, the method has the same or even higher prediction accuracy, greatly shortens the calculation time by reducing the number of partial differential equations to be calculated for transition prediction, solves the problem that the accuracy and the calculation efficiency cannot coexist, and has high calculation efficiency and high calculation accuracy; the invention has good universality because of not depending on empirical formulas.
Drawings
Fig. 1 is a schematic diagram of the present invention.
Fig. 2 is a schematic diagram of a neural network structure.
FIG. 3 is a T3A-transition tablet computing grid diagram.
FIG. 4 is a comparison of the T3A transition plate wall friction curves.
FIG. 5 is a grid map of the S809 airfoil profile calculation.
FIG. 6 is a comparison graph of the results of the calculated values and experimental values of the aerodynamic characteristics of the airfoil of S809. Wherein, FIG. 6(a) is a comparison graph of lift coefficient versus angle of attack; FIG. 6(b) is a graph comparing drag coefficient with angle of attack.
FIG. 7 is a comparison graph of the distribution of the friction coefficient of the airfoil surface at different angles of attack of the S809 airfoil. Wherein the angle of attack in fig. 7(a) is 1 °; the angle of attack in FIG. 7(b) is-5 °; the angle of attack in fig. 7(c) is 9 °.
Fig. 8 is a residual convergence curve under the attack angle of 0 ° of the T3A-transition flat plate and the S809 airfoil.
Detailed Description
The invention specifically comprises three phases: firstly, in a data preparation stage, collecting accurate calculation data of basic flow problems obtained by various experimental or model methods to form a training set, wherein the training set comprises but is not limited to plate transition flow, large separation cylinder flow, square column flow, periodic mountain flow, background step flow, square tube flow, airfoil flow and free shear flow; then, in a learning stage, training a neural network model by utilizing a training set in a supervised learning mode to form an intermittent factor mapping model taking the local average characteristic quantity of the flow field as input and an intermittent factor as output; and finally, in a coupling solving stage, in each iteration step of flow field calculation, the intermittent factors are given by an intermittent factor mapping model, an additional differential equation does not need to be solved, and the calculation efficiency of the transition prediction problem is greatly improved.
The invention designs a data-driven SA-NN transition prediction method by combining a neural network technology and an SA full turbulence model. The construction process of the model can be divided into two parts: off-line learning and coupled solution. The off-line learning part mainly comprises training data selection, a neural network model framework and parameter optimization. In the coupling solving part, the trained intermittent factor mapping model is embedded into a Computational Fluid Dynamics (CFD) solver, and intermittent factors pass through the average characteristic quantity q (q) of the local flow field1,q2,…,qn) And calculating and transmitting to a CFD solver, wherein the specific process is shown in FIG. 1.
Specifically, the transition prediction method based on the neural network of the present invention includes the following steps:
step A, acquiring a plurality of known intermittent factors and corresponding local average characteristic quantities as a training set;
step B, taking the local average characteristic quantity in the training set as an input value, taking an intermittent factor corresponding to the local average characteristic quantity taken as the input value in the training set as an output value, training a neural network and obtaining an intermittent factor mapping model;
step C, a computational fluid mechanics solver carries out flow field iterative computation until the flow field computation result is iteratively converged, and a transition flow field prediction result is output; wherein, the flow field iterative computation process comprises:
step C1, calculating the local average characteristic quantity output by the fluid mechanics solver;
a step C2 of inputting the local average feature quantity obtained in the step C1 to the intermittent factor mapping model;
step C3, solving the intermittent factor mapping model to obtain intermittent factors, and inputting the intermittent factors into a computational fluid mechanics solver;
step C4, judging whether the flow field calculation result of the fluid mechanics solver is converged or not based on the intermittent factor gamma obtained in the step C3, and if yes, terminating the iteration process; if not, the computational fluid dynamics solver updates the local average feature quantity according to the pause factor value obtained in step C3 and jumps to step C1.
In the step C, the flow field calculation result is the flow field density, the flow field speed or the flow field pressure. In the comparative test of the embodiment, the density of the flow field is selected to determine whether the flow field converges.
In step C4, the method for determining whether the flow field calculation result converges includes: if the difference (residual) between the current flow field calculation result and the last flow field calculation result is less than 10-4If not, the convergence is judged.
In the step A, the intermittent factor is derived from known experiment or transition model calculation data; the local average characteristic quantity corresponding to the intermittent factor is derived from the calculation data of the SA turbulence model which is combined with the existing intermittent factor and subjected to data correction; in the step C, the computational fluid mechanics solver is a computational fluid mechanics solver based on the SA turbulence model.
An equation SA model control equation is as follows:
wherein v is the viscosity coefficient of molecular motion,
representing a modified vortex-viscosity coefficient, P, related to the vortex-viscosity coefficient of the turbulenceνTo generate an item, DνTo destroy the term, Cb2And σ is a constant.
Referring to the idea of the relevance intermittent transition model, the intermittent factor is applied to the turbulence model to inhibit turbulence generation before transition and in the transition process, so that the original full turbulence result is corrected to the transition flow field. And (3) considering the objectivity definition of the intermittence factor for describing the intermittence state at a certain point of the flow field, describing the potential functional relationship through a neural network, and combining the simulation model with an SA turbulence model to have rationality. For the transport equation of the SA turbulence model, the pause factor is used to suppress the generation and destruction terms of vortex-stick:
wherein, PνAnd DνRespectively, a generating term and a destroying term, gamma, of the original SA modelpreThe method is an intermittent factor predicted by a neural network model, and beta is a damage term correction coefficient.
The acquisition process of the local average characteristic quantity corresponding to the pause factor comprises the following steps: step A1, substituting the pause factor obtained from the existing data into an SA turbulence model according to the form of formula (2) and freezing, namely the value of the frozen pause factor is not changed in the subsequent calculation; step A2, performing iterative computation by a computational fluid dynamics solver until a flow field computation result is converged; in step a3, the computational fluid dynamics solver outputs the local average feature quantity.
In the step a, the local average characteristic quantities in the local average characteristic quantities include a flow field density, a minimum wall surface distance, a free turbulence flow degree, a motion viscosity coefficient, a Q criterion, a normalized strain rate, an SA model flow field variable, a vortex reynolds number, a dimensionless term similar to a turbulence viscosity, and a pressure gradient along a streamline, and the local average characteristic quantities are shown in table 1.
TABLE 1 local average feature quantities as inputs to neural networks
The invention utilizes a series of local average characteristic quantities closely related to the intermittent factors and models the mapping relation between the local average characteristic quantities and the intermittent factors by introducing a neural network technology, thereby providing an efficient and accurate method for numerical simulation of transition prediction problems.
The invention adopts a neural network to construct an intermittent factor mapping model.
A general neural network structure may consist of an input layer, several hidden layers, an output layer and weighted connections, as in fig. 2. The network layers surrounded by the cross-layer connections are called residual blocks. The network layers adopt a full connection form. The input layer is composed of a group of flow field local average quantities q (q) representing different attributes of the flow field1,q2,…,qn) And (4) forming. These input quantities go through weighted connections to the next hidden layer, where the new input values are compared to thresholds at the nodes and then transformed by a non-linear activation function. The feature information is transformed to a new feature space each time it passes through a hidden layer until the result is output. For a new input xiOutput component of the first hidden layer
Can be expressed as:
wherein, the superscript (m) represents the mth hidden layer, phi is a nonlinear activation function, and the RELU function is adopted in the application; w is aijThe connection right between the upper layer and the present layer; c. CjIs a threshold at a neuron node; the subscript i represents the present layer hidden layer and the subscript j represents the upper layer hidden layer. Similarly, the output components of the second and third hidden layers can be written as:
wherein m is1,m2The number of neuron nodes of the first and second hidden layers is respectively represented. It can be seen that the output of the first hidden layer is added to the inactive output of the third hidden layer by a cross-layer connection, the combined data stream passes through the activation function and then passes downward, the cross-layer connection is actually adding an identity map between the layers, and finally the output component of the fourth hidden layer can be expressed as:
thanks to the multi-level, non-linear activation transformation, the network has the ability to describe deep non-linear mappings between input features and output quantities.
The loss function is an important index directly reflecting the model precision, and the loss function adopted by the invention is set as follows:
wherein the content of the first and second substances,
representing the corresponding true label, gamma, in the training setk(qi) Represents the output of the neural network, and N is the total number of data points used for training.
Finally, weighting w of network nodes by a gradient descent methodijAnd a threshold value cjOptimizing:
wherein eta is the optimization step length.
Through repeated experimental balance between calculation cost and loss function, the neural network adopted by the invention comprises six full-connection layers, two residual blocks containing eight hidden layers and eight full-connection layers which are sequentially arranged from shallow to deep, wherein each hidden layer is provided with 24 neuron nodes, and the activation function of each hidden layer is a linear rectification function (RELU function).
The relevance intermittent transition model actually controls the generation of turbulence in a flow field through the intermittent factor which is a parameter changing along with the space, so as to achieve the purpose of simulating transition. For example, for a certain point in space, if the flow state is laminar flow, the pause factor should be 0, and then the generation term of the turbulent flow is multiplied by 0 in the control equation, that is, no turbulent flow is generated at this point. The problem translates into how the pause factor should be calculated, and conventional methods use empirical equations or introduce additional differential equations to calculate the pause factor, which may introduce calculation errors or result in a reduction in efficiency. The method utilizes the neural network to construct the model, calculates the intermittent factors through the local average characteristic quantities of 10 flow fields, does not adopt an empirical formula, reduces the uncertainty of artificial introduction so as to improve the prediction precision, and ensures the transition prediction efficiency by taking the model as an algebraic model.
Compared with a common correlation intermittent transition model, the data driving transition model provided by the invention has the same or even higher prediction precision, greatly shortens the calculation time, does not depend on an empirical formula, and has stronger universality.
The advantages of the process of the invention are demonstrated by the following examples.
1, T3A-transition flat plate
The T3A-transition flat panel experiment will be used to test the predictive capability of the data-driven intermittent factor model transition described above. The computational grid used is shown in fig. 3, the grid is 324 × 108 (flow direction × normal), and 291 grid cells are distributed on the flat panel. The incoming flow Mach number is 0.0577, and the Reynolds number is 1.4 multiplied by 106The turbulence level was set to 0.843.
The turbulent boundary layer has larger friction, so that the transition condition of the boundary layer can be judged through a friction curve. FIG. 4 shows the results of comparison of wall friction, and SST-. gamma. -Re can be foundθThe model and the SA-NN model provided by the invention can better predict the natural transition, but the SA model cannot predict the transition, and the wall friction calculated by the full turbulence is greatly different from the experimental value. The BC model predicts a transition but prematurely predicts the transition position of the T3A tablet. As can be seen from the T3A-transition plate (with the degree of turbulence of 0.843) wall surface friction curve shown in fig. 4, the transition position and the experimental value of the BC model simulation have a large deviation, and the present invention is better in compliance. The reason is that the existing model depends on an empirical formula for artificial calibration to a great extent, which increases the uncertainty of the model in a certain range and influences the applicability. The SA-NN model which is the same as the equation avoids the problems while ensuring the solving efficiency.
S809 airfoil profile
The S809 airfoil is a laminar flow airfoil with the thickness of 21% c, is specially designed for a transverse-axis wind turbine, and is a typical example of a verification transition model. The computational grid is as shown in fig. 5, a C-type topological structure is adopted, about 6.6 ten thousand grid units are totally divided, and the first-layer grid distance of the wall surface reaches 1 multiplied by 10-6c, the far field boundary is 120c, the Mach number of an inlet is 0.1, and the Reynolds number is 2.0 multiplied by 106The airfoil leading edge turbulence was set at 0.2%.
Fig. 6 compares the results of the calculated values and the experimental values of the aerodynamic characteristics of the airfoil S809. As can be seen from the figure, the SA-NN model is related to SST-gamma-ReθThe model is more consistent with the experimental value in the lift-drag characteristic. The lift coefficient predicted by the original SA model without considering transition is small, and the resistance is always predicted too much. The SA-NN model then largely corrects this.
FIG. 7 shows the friction coefficient distribution of the airfoil surface at different attack angles, and it can be seen that the SA model can not predict transition at attack angles of 1 °, -5 °, and 9 °. And SA-NN model and SST-gamma-ReθThe results of the transition model are very close. At an attack angle of 1 degree, the upper and lower wing surfaces of the airfoil are in separation transition around 0.550 and 0.526 respectively. On the upper partThe transition position of the airfoil surface moves forward along with the increase of the attack angle, the transition position reaches about 0.01 at the attack angle of 9 degrees, the wall surface friction is lowered continuously along the flow direction, and the flow is fluidized again. The transition position of the lower airfoil surface continuously moves towards the rear edge along with the attack angle.
The above results verify transition prediction capabilities of the SA-NN model. On this basis, another advantage of the model is the solution efficiency. Compared with the square-range SST-gamma-ReθAnd the transition model and the SA-NN model give intermittent factor distribution through an algebraic black box model, so that an additional differential equation is not required to be introduced, and the time consumption of calculation is reduced. The efficiency improvement is more obvious on the calculation of large grid quantity. Fig. 8 lists the residual convergence curves for the T3A transition plate and the S809 airfoil at 0 ° angle of attack. The SA-NN model compares to SST-gamma-Re when converging to the same precisionθThe calculation time of the transition model is reduced by more than 35%. After the flow expands to three dimensions, the efficiency advantage brought by the algebraic transition model is more considerable.
While the present invention has been described with reference to the embodiments shown in the drawings, the present invention is not limited to the embodiments, which are illustrative and not restrictive, and it will be apparent to those skilled in the art that various changes and modifications can be made therein without departing from the spirit and scope of the invention as defined in the appended claims.
Claims (7)
1. A transition prediction framework based on a neural network is characterized by comprising the following steps:
step A, acquiring a plurality of known intermittent factors and corresponding local average characteristic quantities as a training set;
step B, taking the local average characteristic quantity in the training set as an input value, taking an intermittent factor corresponding to the local average characteristic quantity taken as the input value in the training set as an output value, training a neural network and obtaining an intermittent factor mapping model;
step C, a computational fluid mechanics solver carries out flow field iterative computation until the flow field computation result is iteratively converged, and a transition flow field prediction result is output; wherein, the flow field iterative computation process comprises:
step C1, calculating the local average characteristic quantity output by the fluid mechanics solver;
a step C2 of inputting the local average feature quantity obtained in the step C1 to the intermittent factor mapping model;
step C3, solving the intermittent factor mapping model to obtain intermittent factors, and inputting the intermittent factors into a computational fluid mechanics solver;
step C4, judging whether the flow field calculation result of the fluid mechanics solver converges or not based on the intermittent factor obtained in the step C3, and if so, terminating the iteration process; if not, the computational fluid dynamics solver updates the local average feature quantity according to the pause factor value obtained in step C3 and jumps to step C1.
2. The method for predicting transition based on neural network of claim 1, wherein in step a, the pause factor is derived from known experimental or high-precision transition model calculation data; the local average characteristic quantity corresponding to the intermittent factor is derived from the calculation data of the SA turbulence model which is combined with the existing intermittent factor and subjected to data correction; in the step C, the computational fluid mechanics solver is a computational fluid mechanics solver based on the SA turbulence model.
3. The neural network-based transition prediction method of claim 2, wherein in the step a, the local average feature quantities in the local average feature quantities include one or more of flow field density, minimum wall distance, free turbulence flow degree, kinematic viscosity coefficient, Q criterion, normalized strain rate, SA model flow field variable, vortex reynolds number, dimensionless terms like turbulence viscosity, and pressure gradient along a flow line.
4. The method according to any one of claims 1 to 3, wherein the neural network comprises six fully-connected layers, two residual blocks containing eight hidden layers, and eight fully-connected layers sequentially arranged from shallow to deep, wherein each hidden layer has 24 neuron nodes, and an activation function of each hidden layer is a linear rectification function.
5. The transition prediction method based on the neural network as claimed in any one of claims 1 to 3, wherein in the step C, the flow field calculation result is a flow field density, a flow field velocity or a flow field pressure.
6. The transition prediction method based on a neural network as claimed in any one of claims 1 to 3, wherein in the step C4, the method for determining whether the flow field calculation result converges includes: if the difference value between the current flow field calculation result and the last flow field calculation result is less than 10-4If not, the convergence is judged.
7. The method for predicting transition based on a neural network of claim 2, wherein the obtaining of the local average feature quantity corresponding to the pause factor comprises: step A1, substituting the pause factor obtained from the existing data into an SA turbulence model and freezing; step A2, performing iterative computation by a computational fluid dynamics solver until a flow field computation result is converged; in step a3, the computational fluid dynamics solver outputs the local average feature quantity.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011154867.4A CN112084727A (en) | 2020-10-26 | 2020-10-26 | Transition prediction method based on neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011154867.4A CN112084727A (en) | 2020-10-26 | 2020-10-26 | Transition prediction method based on neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112084727A true CN112084727A (en) | 2020-12-15 |
Family
ID=73731067
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011154867.4A Pending CN112084727A (en) | 2020-10-26 | 2020-10-26 | Transition prediction method based on neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112084727A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112784508A (en) * | 2021-02-12 | 2021-05-11 | 西北工业大学 | Deep learning-based airfoil flow field rapid prediction method |
| CN113051835A (en) * | 2021-04-14 | 2021-06-29 | 中国科学院工程热物理研究所 | Efficient prediction method for unsteady flow field of impeller machinery based on machine learning |
| CN113361173A (en) * | 2021-06-17 | 2021-09-07 | 中国空气动力研究与发展中心超高速空气动力研究所 | Compressible modification method of transition model completely based on local flow field parameters |
| CN113420507A (en) * | 2021-07-08 | 2021-09-21 | 刘维烁 | Fluid flow simulation method, device and medium |
| CN114036869A (en) * | 2021-11-18 | 2022-02-11 | 中国空气动力研究与发展中心超高速空气动力研究所 | Hypersonic speed boundary layer transition prediction method considering head bluntness effect |
| CN114818549A (en) * | 2022-06-29 | 2022-07-29 | 苏州浪潮智能科技有限公司 | Method, system, equipment and medium for calculating fluid mechanics parameter of object |
| CN115859760A (en) * | 2022-04-29 | 2023-03-28 | 中广核风电有限公司 | Method and device for simulating flow field turbulence |
| CN116050242A (en) * | 2022-11-15 | 2023-05-02 | 中国空气动力研究与发展中心计算空气动力研究所 | Transition prediction method, device, equipment and medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070087756A1 (en) * | 2005-10-04 | 2007-04-19 | Hoffberg Steven M | Multifactorial optimization system and method |
| US20120245910A1 (en) * | 2011-03-22 | 2012-09-27 | Desktop Aeronautics, Inc. | Predicting transition from laminar to turbulent flow over a surface |
| CN108733914A (en) * | 2018-05-17 | 2018-11-02 | 复旦大学 | Transonic airfoil Natural Laminar Flow delay based on artificial neural network turns to twist design method |
| CN109033525A (en) * | 2018-06-27 | 2018-12-18 | 浙江大学 | A kind of hypersonic transition prediction method based on simplified three equation transition models |
| CN110702356A (en) * | 2019-10-12 | 2020-01-17 | 空气动力学国家重点实验室 | Hypersonic velocity transition prediction method considering surface roughness effect |
-
2020
- 2020-10-26 CN CN202011154867.4A patent/CN112084727A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070087756A1 (en) * | 2005-10-04 | 2007-04-19 | Hoffberg Steven M | Multifactorial optimization system and method |
| US20120245910A1 (en) * | 2011-03-22 | 2012-09-27 | Desktop Aeronautics, Inc. | Predicting transition from laminar to turbulent flow over a surface |
| CN108733914A (en) * | 2018-05-17 | 2018-11-02 | 复旦大学 | Transonic airfoil Natural Laminar Flow delay based on artificial neural network turns to twist design method |
| CN109033525A (en) * | 2018-06-27 | 2018-12-18 | 浙江大学 | A kind of hypersonic transition prediction method based on simplified three equation transition models |
| CN110702356A (en) * | 2019-10-12 | 2020-01-17 | 空气动力学国家重点实验室 | Hypersonic velocity transition prediction method considering surface roughness effect |
Non-Patent Citations (2)
| Title |
|---|
| 张玉伦;王光学;孟德虹;王运涛;: "γ-Re_θ转捩模型的标定研究", 空气动力学学报, no. 03 * |
| 郑天韵等: ""基于深度残差网络的高精度自然转捩模拟方法"", 《中国知网》 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112784508A (en) * | 2021-02-12 | 2021-05-11 | 西北工业大学 | Deep learning-based airfoil flow field rapid prediction method |
| CN113051835A (en) * | 2021-04-14 | 2021-06-29 | 中国科学院工程热物理研究所 | Efficient prediction method for unsteady flow field of impeller machinery based on machine learning |
| CN113361173A (en) * | 2021-06-17 | 2021-09-07 | 中国空气动力研究与发展中心超高速空气动力研究所 | Compressible modification method of transition model completely based on local flow field parameters |
| CN113420507A (en) * | 2021-07-08 | 2021-09-21 | 刘维烁 | Fluid flow simulation method, device and medium |
| CN113420507B (en) * | 2021-07-08 | 2023-11-21 | 刘维烁 | Fluid flow simulation method, device and medium |
| CN114036869A (en) * | 2021-11-18 | 2022-02-11 | 中国空气动力研究与发展中心超高速空气动力研究所 | Hypersonic speed boundary layer transition prediction method considering head bluntness effect |
| CN114036869B (en) * | 2021-11-18 | 2022-08-12 | 中国空气动力研究与发展中心超高速空气动力研究所 | Hypersonic speed boundary layer transition prediction method considering head bluntness effect |
| CN115859760A (en) * | 2022-04-29 | 2023-03-28 | 中广核风电有限公司 | Method and device for simulating flow field turbulence |
| CN114818549A (en) * | 2022-06-29 | 2022-07-29 | 苏州浪潮智能科技有限公司 | Method, system, equipment and medium for calculating fluid mechanics parameter of object |
| CN114818549B (en) * | 2022-06-29 | 2022-11-29 | 苏州浪潮智能科技有限公司 | Method, system, equipment and medium for calculating fluid mechanics parameter of object |
| WO2024001096A1 (en) * | 2022-06-29 | 2024-01-04 | 苏州元脑智能科技有限公司 | Method and system for calculating fluid mechanics parameters of object, and device and medium |
| CN116050242A (en) * | 2022-11-15 | 2023-05-02 | 中国空气动力研究与发展中心计算空气动力研究所 | Transition prediction method, device, equipment and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112084727A (en) | Transition prediction method based on neural network | |
| CN112016167B (en) | Aircraft aerodynamic shape design method and system based on simulation and optimization coupling | |
| CN110705029B (en) | Flow field prediction method of oscillating flapping wing energy acquisition system based on transfer learning | |
| CN104049000B (en) | The gas-liquid phase content migrating complex network based on mode is measured and verification method | |
| CN110309571B (en) | Wing body fusion underwater glider external shape optimization method based on radial basis function model | |
| CN110222828A (en) | A kind of Unsteady Flow method for quick predicting based on interacting depth neural network | |
| CN105718634A (en) | Airfoil robust optimization design method based on non-probability interval analysis model | |
| CN110516318B (en) | Airfoil design method based on radial basis function neural network proxy model | |
| Shahrokhi et al. | A surrogate assisted evolutionary optimization method with application to the transonic airfoil design | |
| CN114065662B (en) | Method suitable for rapidly predicting airfoil flow field with variable grid topology | |
| Peng et al. | Multi-fidelity nonlinear unsteady aerodynamic modeling and uncertainty estimation based on Hierarchical Kriging | |
| CN110276131B (en) | Wing body fusion underwater glider appearance optimization method based on polynomial response surface model | |
| Zhu et al. | Physics-informed machine learning for surrogate modeling of wind pressure and optimization of pressure sensor placement | |
| Pan et al. | Estimation of power performances and flow characteristics for a Savonius rotor by vortex particle method | |
| Glumac et al. | A multi-fidelity wind surface pressure assessment via machine learning: A high-rise building case | |
| Mukesh et al. | Influence of search algorithms on aerodynamic design optimisation of aircraft wings | |
| Kim et al. | Geometric modification for the enhancement of an airfoil performance using deep CNN | |
| CN105138766A (en) | Adding method based on fuzzy clustering for hypersonic velocity aerodynamic heat reduced-order model | |
| Hoyos et al. | Airfoil shape optimization: Comparative study of meta-heuristic algorithms, airfoil parameterization methods and reynolds number impact | |
| Massegur Sampietro et al. | Recurrent Multi-Mesh Convolutional Autoencoder Framework for Spatio-Temporal Aerodynamic Modelling. | |
| CN111625901B (en) | Intelligent pressure coefficient curve generation method for wing profile | |
| Guo et al. | An aero-structure-acoustics evaluation framework of wind turbine blade cross-section based on Gradient Boosting regression tree | |
| Moshtaghzadeh et al. | Prediction of Wind Turbine Airfoil Performance Using Artificial Neural Network and CFD Approaches. | |
| Liu et al. | Unteady aerodynamics modeling method based on dendrite-based gated recurrent neural network model | |
| Shahrokhi et al. | An efficient aerodynamic optimization method using a genetic algorithm and a surrogate model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |