Search

CN-122021415-A - Asynchronous flow field data alignment and high-resolution reconstruction method based on multi-physical field feature consistency constraint

CN122021415ACN 122021415 ACN122021415 ACN 122021415ACN-122021415-A

Abstract

An asynchronous flow field data alignment and high-resolution reconstruction method based on multi-physical field feature consistency constraint relates to the field of fluid mechanics, and can automatically correct phase errors of low-precision data to realize high-quality flow field prediction and reconstruction. The method solves the problem that the existing data-driven flow field reconstruction method is difficult to process asynchronous (phase deviation exists) multi-precision CFD data, and by introducing physical consistency constraint containing speed, pressure and vorticity characteristics, a training sample pair with high confidence is automatically constructed, so that accuracy of flow field time sequence prediction and physical reality of high-resolution reconstruction are improved.

Inventors

  • TANG BIN
  • WANG HAODONG
  • LIU CHUANFENG
  • LIAO XIAOFENG
  • QIN SHUO
  • Xu Buyang

Assignees

  • 青岛哈尔滨工程大学创新发展中心
  • 哈尔滨工程大学

Dates

Publication Date
20260512
Application Date
20260115

Claims (9)

  1. 1. An asynchronous flow field data alignment and high resolution reconstruction method based on multi-physical field feature consistency constraint is characterized by comprising the following steps: S1, acquiring a high-precision data set , , Is a time step Is used for the data of high precision of (a), , Is the number of time steps; s2, utilizing a high-precision data set Constructing low precision data sets ; S3, calculating a measurement function for quantifying physical state difference between low-precision data and high-precision data ; S4, utilizing a measurement function Data pairs obtained by sliding window optimization ; S5, data are paired Middle (f) Low precision data Inputting the low-precision data prediction value into a flow field time sequence prediction model, and outputting the low-precision data prediction value; S6, inputting the low-precision data predicted value into a high-resolution reconstruction model, and outputting the reconstructed high-precision data ; S7, high-precision data reconstruction Data pair Time step in (2) High-precision data of (a) Training a flow field time sequence prediction model and a high-resolution reconstruction model.
  2. 2. The method for data alignment and high resolution reconstruction of an asynchronous flow field based on multi-physical field feature consistency constraint of claim 1, wherein the step S1 uses direct numerical simulation data of two-dimensional square column bypass as a high-precision data set Time step High-precision data of (a) The grid resolution of (2) is 128×256.
  3. 3. The asynchronous flow field data alignment and high resolution reconstruction method based on multi-physical field feature consistency constraint of claim 2, wherein step S2 comprises the steps of: S2-1, high precision data Input into an average pooling layer with a pooling core of 16×16 to obtain low-precision data with a resolution of 8×16 All of The low-precision data form a data set , ; S2-2 from the dataset Random decimation in Low-precision data to form low-precision data set 。
  4. 4. The method for asynchronous flow field data alignment and high resolution reconstruction based on multi-physical field feature consistency constraint of claim 3, wherein in step S3, the method is characterized by the following formula Calculating to obtain a measurement function In the following 、 、 Are all the weight coefficients of the two-dimensional space model, In order to be an L2 norm, Is a low-precision data set Middle (f) The velocity field in the low-precision data, , For high-precision data sets Middle time step High-precision data of (a) Is provided with a velocity field in the direction of the velocity field, , In order for the down-sampling operator to be able to perform, Is a low-precision data set Middle (f) The pressure field in the individual low-precision data, High precision data set Middle time step High-precision data of (a) Is used as a pressure field in the air conditioner, Is a rotation operator.
  5. 5. The method for asynchronous flow field data alignment and high resolution reconstruction based on multi-physical field feature consistency constraint of claim 4, wherein step S4 comprises the steps of: S4-1, setting the search interval as , , Is the maximum phase tolerance; s4-2. Traversing time step To the point of All high precision data in A measurement function of low-precision data, selecting the time step corresponding to the measurement function of the minimum value in all measurement function values As the optimal physical matching time; s4-3. Will be Low precision data And time step High-precision data of (a) Composing pairs of data 。
  6. 6. The asynchronous flow field data alignment and high resolution reconstruction method based on multi-physical field feature consistency constraint of claim 5, wherein: the value is 10.
  7. 7. The method for asynchronous flow field data alignment and high resolution reconstruction based on multi-physical field feature consistency constraint of claim 1, wherein the flow field time sequence prediction model is ConvLSTM model.
  8. 8. The method for asynchronous flow field data alignment and high resolution reconstruction based on multi-physical field feature consistency constraint of claim 1, wherein the high resolution reconstruction model is ESRGAN.
  9. 9. The method for asynchronous flow field data alignment and high resolution reconstruction based on multi-physical field feature consistency constraint of claim 1, wherein step S7 comprises the steps of: S7-1. High precision data with reconstruction Data pair Time step in (2) High-precision data of (a) Calculating an RMSE loss function; S7-2, training a flow field time sequence prediction model and a high-resolution reconstruction model by using an Adam optimizer and utilizing an RMSE loss function.

Description

Asynchronous flow field data alignment and high-resolution reconstruction method based on multi-physical field feature consistency constraint Technical Field The invention relates to the field of fluid mechanics, in particular to an asynchronous flow field data alignment and high-resolution reconstruction method based on multi-physical field feature consistency constraint. Background Computational Fluid Dynamics (CFD) is an important tool to study the flow laws of fluids. In practical engineering, to capture turbulence details (e.g., vortex shedding, boundary layer separation), a very high resolution grid (DNS or LES) is typically required, but this incurs significant computational costs. Thus, the industry often uses coarse meshes (RANS or coarsened LES) for fast simulations, but this can lead to loss of flow field details and accumulation of numerical errors. In recent years, a Super-Resolution reconstruction technique (Super-Resolution) of a flow field based on deep learning becomes a research hotspot. However, the prior art has significant drawbacks in processing actual engineering data: 1. The prior art assumes "time-critical synchronization" in that the low resolution input data of most methods is directly downsampled from the high resolution data, both of which correspond exactly in time. 2. "Asynchronous" pain point of real data in real coarse grid CFD simulation, the evolution speed of the flow field tends to be unsynchronized (with phase lag or lead) with the fine grid simulation due to numerical dissipation and truncation errors. For example, the vortex shedding period calculated by the coarse grid may be slower than that of the fine grid. 3. If the physical asynchronous characteristic is ignored, the neural network directly carries out pairing training on the low-precision data and the high-precision data forcefully according to the time stamp, and the neural network learns the wrong mapping relation, so that the reconstruction result is fuzzy, artifacts exist and even the fluid mechanics rule is violated. Disclosure of Invention In order to overcome the defects of the technology, the invention provides a method capable of automatically correcting the phase error of low-precision data and realizing high-quality flow field prediction and reconstruction. The technical scheme adopted for overcoming the technical problems is as follows: An asynchronous flow field data alignment and high resolution reconstruction method based on multi-physical field feature consistency constraint comprises the following steps: S1, acquiring a high-precision data set ,,Is a time stepIs used for the data of high precision of (a),,Is the number of time steps; s2, utilizing a high-precision data set Constructing low precision data sets; S3, calculating a measurement function for quantifying physical state difference between low-precision data and high-precision data; S4, utilizing a measurement functionData pairs obtained by sliding window optimization; S5, data are pairedMiddle (f)Low precision dataInputting the low-precision data prediction value into a flow field time sequence prediction model, and outputting the low-precision data prediction value; S6, inputting the low-precision data predicted value into a high-resolution reconstruction model, and outputting the reconstructed high-precision data ; S7, high-precision data reconstructionData pairTime step in (2)High-precision data of (a)Training a flow field time sequence prediction model and a high-resolution reconstruction model. Further, in step S1, direct numerical simulation data of two-dimensional square column streaming is adopted as a high-precision data setTime stepHigh-precision data of (a)The grid resolution of (2) is 128×256. Further, step S2 includes the steps of: S2-1, high precision data Input into an average pooling layer with a pooling core of 16×16 to obtain low-precision data with a resolution of 8×16All ofThe low-precision data form a data set,; S2-2 from the datasetRandom decimation inLow-precision data to form low-precision data set。 In step S3, the formula is passedCalculating to obtain a measurement functionIn the following、、Are all the weight coefficients of the two-dimensional space model,In order to be an L2 norm,Is a low-precision data setMiddle (f)The velocity field in the low-precision data,,For high-precision data setsMiddle time stepHigh-precision data of (a)Is provided with a velocity field in the direction of the velocity field,,In order for the down-sampling operator to be able to perform,Is a low-precision data setMiddle (f)The pressure field in the individual low-precision data,High precision data setMiddle time stepHigh-precision data of (a)Is used as a pressure field in the air conditioner,Is a rotation operator. Further, step S4 includes the steps of: S4-1, setting the search interval as ,,Is the maximum phase tolerance; s4-2. Traversing time step To the point ofAll high precision data inA measurement function of low-precision d