Centrifugal compressor unstable flow analysis method based on compressed sensing dynamic mode decomposition

Through the dynamic mode decomposition method based on compression perception technology, the problem of identifying the key flow modes of the leafless diffuser in the stability study of centrifugal compressor is solved, which significantly reduces the calculation time and realizes the spectrum of the leafless diffuser and Accurate capture of flow modes.

CN119089819APending Publication Date: 2024-12-06BEIJING INST OF TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411133134.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-19
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

The existing centrifugal compressor stability research methods are difficult to accurately identify the key flow modes of the bladeless diffuser, and the processing of high-dimensional flow field information is long, resulting in limited research on the stall mechanism under complex boundary conditions.

Method used

Using a method based on compression sensing dynamic modal decomposition, a high-dimensional flow field snapshot matrix is ​​established through DMD, singular value decomposition is performed, and the measurement matrix is ​​generated for compression is obtained to obtain a low-dimensional flow field matrix. The dynamic modal decomposition is performed using compression sensing technology. Construct a high-dimensional flow field mode, significantly reducing calculation time.

Benefits of technology

It realizes that the spectrum and key flow modes of the bladeless diffuser are accurately captured under the premise of small guaranteed frequency and mode errors, significantly improving the calculation efficiency of dynamic mode decomposition and shortening the calculation time.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119089819A_ABST
    Figure CN119089819A_ABST
Patent Text Reader

Abstract

The invention discloses a centrifugal compressor unstable flow analysis method based on compressed sensing dynamic modal decomposition, which comprises the following steps of: establishing a high-dimensional flow field snapshot matrix through a DMD (Digital Micromirror Device) method, and calculating characteristic values of high-dimensional flow field snapshots; based on a compressed sensing technology, compressing the high-dimensional flow field snapshot matrix to obtain a low-dimensional flow field snapshot matrix; on the basis of the low-dimensional flow field snapshot matrix, a characteristic value and a DMD mode based on the low-dimensional flow field snapshot are obtained by using a compressed sensing dynamic mode decomposition method; and reconstructing the DMD mode of the high-dimensional flow field snapshot based on the DMD mode of the low-dimensional flow field snapshot. According to the method, the key flow mode of the vaneless diffuser can be accurately obtained, and the calculation time is greatly shortened.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of centrifugal compressor modeling and simulation, and more particularly to a centrifugal compressor unstable flow analysis method based on compressed sensing dynamic modal decomposition. Background Art

[0002] The centrifugal compressor consists of an impeller, a bladeless diffuser and a volute. The study of the key components where flow instability occurs is of great significance for improving the operating stability of the compressor, reducing the negative impact of aerodynamic noise and vibration, and the design of the compressor. When using experimental measurements or numerical simulation methods to conduct detailed research on the flow field of a centrifugal compressor, traditional research methods often rely more on time domain analysis. The Dynamic Mode Decomposition (DMD) method is widely used in the analysis of dynamic characteristics of combustion, jets, turbulence, shock wave boundary layer interference and isolated airfoil flow fields. The use of the DMD method for modal decomposition can take into account the nonlinear effects of the flow to a certain extent, revealing the flow dynamics modes corresponding to specific frequencies, and providing a powerful tool for the study of centrifugal compressor stability.

[0003] In order to solve the centrifugal compressor stability problem, it is necessary to identify the key structures and stall mechanisms that affect the centrifugal compressor stall. Although dynamic modal decomposition can decouple the spatiotemporal coherent structures in complex systems and obtain modes with a single oscillation frequency, it does not have spatial orthogonality and it is difficult to identify the modal subset that has the greatest impact on the flow field. In addition, the current research methods for the stability of centrifugal compressors require a lot of computing time for high-dimensional flow field snapshots containing huge flow field information, which poses a great obstacle to the study of the stall mechanism of bladeless diffuser centrifugal compressors under complex boundary conditions.

[0004] Therefore, there is an urgent need for a research method for the stability of centrifugal compressors, which can change the existing DMD method and accurately obtain the key flow modes of the bladeless diffuser and greatly reduce the calculation time. Summary of the invention

[0005] In view of this, the present invention provides a centrifugal compressor unstable flow analysis method based on compressed sensing dynamic modal decomposition, which can accurately obtain the key flow modes of the bladeless diffuser and greatly reduce the calculation time.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] A centrifugal compressor unstable flow analysis method based on compressed sensing dynamic modal decomposition, comprising:

[0008] The high-dimensional flow field snapshot matrix is ​​established by the DMD method, and the eigenvalues ​​of the high-dimensional flow field snapshot are calculated;

[0009] Based on compressed sensing technology, the high-dimensional flow field snapshot matrix is ​​compressed to obtain a low-dimensional flow field snapshot matrix;

[0010] Based on the low-dimensional flow field snapshot matrix and using the compressed sensing dynamic mode decomposition method, the eigenvalues ​​and DMD modes based on the low-dimensional flow field snapshot are obtained;

[0011] The DMD mode of the high-dimensional flow field snapshot is reconstructed based on the DMD mode of the low-dimensional flow field snapshot.

[0012] Preferably, the high-dimensional flow field snapshot matrix is ​​specifically:

[0013] In the DMD method, the vector matrix form of the high-dimensional flow field snapshot obtained from experiments or numerical simulations is:

[0014]

[0015] Among them, x i , i∈[1,n] represents the flow field snapshot at the i-th moment of each flow parameter. The flow field data is sampled at equal time intervals. The total number of sampling moments is n, and the sampling time interval is Δt. In the above definition, the subscript 1 and superscript n represent the first moment and the last moment in the flow field sequence respectively. It is the matrix of high-dimensional flow field snapshots from the 1st moment to the nth moment.

[0016] Preferably, calculating the eigenvalues ​​of the high-dimensional flow field snapshot includes:

[0017] right Perform singular value decomposition and get:

[0018]

[0019] in, is an m×(n-1)-order matrix, where m represents the dimension of the high-dimensional flow field snapshot, i.e., the number of measurement points in the experiment or the number of grid points in the numerical calculation. X is an m×(n-1)-order matrix, which contains the spatial structure information of the high-dimensional flow field, Σ X is a (n-1)×(n-1)-order matrix, Σ X The singular values ​​in are arranged from large to small, representing the energy distribution in the high-dimensional flow field. X It is a (n-1)×(n-1)-order matrix, which contains the temporal structure information of the high-dimensional flow field;

[0020] Thus we get:

[0021]

[0022] in, The matrix form is as follows:

[0023]

[0024] Solving the Matrix The eigenvalues ​​and eigenvectors are as follows:

[0025]

[0026] Among them, Λ X is a matrix The eigenvalue of W is the eigenvalue of the high-dimensional flow field snapshot; X is a matrix The feature vector of .

[0027] Preferably, based on the compressed sensing technology, the high-dimensional flow field snapshot matrix is ​​compressed to obtain a low-dimensional flow field snapshot matrix, including:

[0028] Generate measurement matrix C∈R q×m , where the number of columns m represents the dimension of the high-dimensional flow field snapshot, and the number of rows q represents the dimension of the compressed low-dimensional flow field snapshot;

[0029] Based on high-dimensional flow field snapshot matrix And by compressing the measurement matrix C, we can get the compressed low-dimensional flow field snapshot matrix As shown below:

[0030]

[0031] in, It represents the matrix of the high-dimensional flow field snapshot from the 1st moment to the n-1th moment. is the matrix of the high-dimensional flow field snapshot from the second moment to the nth moment, Y1 n-1 yes The low-dimensional flow field snapshot matrix obtained after compression of the measurement matrix C is: yes The low-dimensional flow field snapshot matrix is ​​obtained after compression of the measurement matrix C.

[0032] Preferably, obtaining eigenvalues ​​based on the low-dimensional flow field snapshot based on the low-dimensional flow field snapshot matrix and using a compressed sensing dynamic mode decomposition method includes:

[0033] Assume Y1 n-1 and The linear operator A Y , then:

[0034]

[0035] Among them, en-1 ∈R n-1 is the n-1th order unit vector, r is the residual vector, S Y The matrix form is as follows:

[0036]

[0037] For Y1 n-1 Perform singular value decomposition and get:

[0038]

[0039] Among them, Y1 n-1 is an m×(n-1)-order matrix, where m represents the dimension of the low-dimensional flow field snapshot, i.e., the number of measurement points in the experiment or the number of grid points in the numerical calculation. Y is an m×(n-1)-order matrix, which contains the spatial structure information of the low-dimensional flow field, Σ Y is a (n-1)×(n-1)-order matrix, Σ Y The singular values ​​in are arranged from large to small, representing the energy distribution in the low-dimensional flow field, V Y It is a (n-1)×(n-1)-order matrix, which contains the temporal structure information of the low-dimensional flow field;

[0040] Thus we can get:

[0041]

[0042] in, The matrix form is as follows:

[0043]

[0044] Among them, a r is the linear coefficient in linear correlation, r is the order;

[0045] Solution The eigenvalues ​​and eigenvectors of can be obtained:

[0046]

[0047] Λ Y is a matrix The eigenvalue of the low-dimensional flow field snapshot, W Y is a matrix The feature vector of .

[0048] Preferably, the DMD mode of the low-dimensional flow field snapshot is:

[0049] Φ Y =U Y W Y

[0050] or

[0051]

[0052] Among them, Φ Y is the DMD mode of the low-dimensional flow field snapshot.

[0053] Preferably, reconstructing the DMD mode of the high-dimensional flow field snapshot based on the DMD mode of the low-dimensional flow field snapshot includes:

[0054] According to the DMD mode of low-dimensional flow field snapshot and based on the high-dimensional flow field snapshot matrix Directly reconstruct the DMD mode of the high-dimensional flow field snapshot;

[0055] Through the flow reconstruction technology, the low-dimensional flow field snapshots are reconstructed by selecting the modes ranked within the set value in order of energy from large to small.

[0056] Preferably, according to the DMD mode of the low-dimensional flow field snapshot and based on the high-dimensional flow field snapshot matrix Directly reconstruct the DMD mode of the high-dimensional flow field snapshot, specifically:

[0057]

[0058] Among them, Φ X is the DMD mode of the high-dimensional flow field snapshot.

[0059] Preferably, the method further includes calculating the characteristic frequency and modal growth rate of the compressed sensing dynamic modal decomposition mode:

[0060]

[0061] Among them, λ i Represents the eigenvalue of the i-th order low-dimensional flow field snapshot, that is, the matrix Λ Y The i-th element on the diagonal, Δt, represents the physical time interval between two adjacent flow field snapshots, and the real part Represents the growth rate of the compressed sensing dynamic mode decomposition mode, the imaginary part represents the characteristic frequency, μ i That is, the characteristic spectrum of the low-dimensional flow field mode, which includes the growth rate and characteristic frequency of the mode.

[0062] It can be seen from the above technical solution that, compared with the prior art, the present invention discloses a method for analyzing unstable flow of a centrifugal compressor based on compressed sensing dynamic modal decomposition. Aiming at the research on the stall mechanism of the centrifugal compressor with bladeless diffuser under complex boundary conditions, combined with compressed sensing dynamic modal decomposition technology, it can accurately capture the spectrum and key flow modes of the bladeless diffuser while ensuring that the predicted frequency and flow modal errors are small, and significantly improve the calculation time of dynamic modal decomposition. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0064] Figure 1 A flow chart of a centrifugal compressor unstable flow analysis method based on compressed sensing dynamic modal decomposition provided by the present invention.

[0065] Figure 2 A schematic diagram of a centrifugal compressor unstable flow analysis method based on compressed sensing dynamic modal decomposition provided by the present invention.

[0066] Figure 3 The present invention provides a comparison of the total pressure ratio P using the numerical calculation results of the present invention and the experimental results.

[0067] Figure 4 The efficiency η provided by the present invention is compared between the numerical calculation results and the experimental results using the present invention.

[0068] Figure 5 The pressure recovery coefficient C of the diffuser under different working conditions obtained by numerical simulation of the present invention is provided by the present invention. p The pressure rise coefficient C of the impeller f .

[0069] Figure 6 The present invention provides the frequency spectrum distribution and disturbance growth rate of the stall condition of the bladeless diffuser obtained by using the compressed sensing dynamic modal decomposition based on the present invention when the compression ratio is 2%.

[0070] Figure 7 The present invention provides a comparative analysis of the computation time and recognition quality of the compressed sensing dynamic modal decomposition based on the same computing hardware conditions using the present invention. DETAILED DESCRIPTION

[0071] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0072] The embodiment of the present invention discloses a centrifugal compressor unstable flow analysis method based on compressed sensing dynamic mode decomposition, such as Figure 1 and Figure 2 As shown, including:

[0073] The high-dimensional flow field snapshot matrix is ​​established by the DMD method, and the eigenvalues ​​of the high-dimensional flow field snapshot are calculated;

[0074] Based on compressed sensing technology, the high-dimensional flow field snapshot matrix is ​​compressed to obtain a low-dimensional flow field snapshot matrix;

[0075] Based on the low-dimensional flow field snapshot matrix and using the compressed sensing dynamic mode decomposition method, the eigenvalues ​​and DMD modes based on the low-dimensional flow field snapshot are obtained;

[0076] The DMD mode of the high-dimensional flow field snapshot is reconstructed based on the DMD mode of the low-dimensional flow field snapshot.

[0077] Each step of the present invention is described in detail below.

[0078] Step 1: Use the DMD method to establish a high-dimensional flow field snapshot matrix; specifically:

[0079] In the DMD algorithm, the high-dimensional flow field snapshot obtained from experiments or numerical simulations can be written in the following vector matrix form:

[0080]

[0081] Among them, x i , i∈[1,n] represents the high-dimensional flow field snapshot at the i-th moment that stores various flow parameters (such as pressure, velocity, entropy, etc.); the high-dimensional flow field data is sampled at equal time intervals, the total number of sampling moments is n, and the sampling time interval is Δt; in the above definition, the subscript 1 and superscript n represent the first moment and the last moment in the flow field sequence respectively.

[0082] Based on the DMD method, the eigenvalues ​​of the high-dimensional flow field snapshot are calculated; specifically:

[0083] Assuming that there is a certain relationship between any two consecutive flow field snapshots, the flow field at the latter moment can be approximately linearly represented by the flow field at the previous moment through a linear mapping A, that is,

[0084] x i+1 =Ax i (2)

[0085] Assuming that the flow field snapshot comes from a nonlinear process, then this assumed linear relationship is equivalent to the linear Taylor approximation, and the following sequence can be obtained:

[0086]

[0087] According to the assumption, the linear relationship A contains the dynamic characteristics of this set of flow field snapshot sequences, for example: the eigenvalues ​​of A reflect the frequency and time growth rate of the corresponding mode, and identifying these characteristics is the purpose of the present invention.

[0088] when When the number of flow field sequences collected in reaches the point where the main flow structure can be captured, it can be assumed that the vectors in equation (2) are linearly related. In other words, if one more flow field data is added, it will not affect The vector space of , at this time, the flow field data at the nth moment can be expressed by the linear combination of the flow field snapshots at the previous n-1 moments, that is:

[0089] x n =a1x1+a2x2+,…+a n-1 x n-1 +r (4)

[0090] Its matrix form is:

[0091]

[0092] Among them, a T =[a1,a2,…,a n-1 ], r is the residual vector. Then we have:

[0093]

[0094] Its matrix form is:

[0095]

[0096] Among them, V1 n-1 is a matrix The high-dimensional flow field snapshot data matrix at the first n-1 moments, e n-1 ∈R n-1 is an n-1-order unit vector, and the S matrix is ​​in the following form:

[0097]

[0098] First, a snapshot of the flow field Perform singular value decomposition and get

[0099]

[0100] in, is an m×(n-1)-order matrix, where m represents the dimension of the high-dimensional flow field snapshot, that is, the number of measurement points in the experiment or the number of grid points in the numerical calculation. X It is an m×(n-1)-order matrix, which contains the spatial structure information of the high-dimensional flow field. Xis an (n - 1)×(n - 1) matrix, and the singular values in Σ X are arranged from large to small, representing the energy distribution in the high-dimensional flow field. V X is an (n - 1)×(n - 1) matrix, containing the time structure information of the high-dimensional flow field.

[0101] Substituting Equation (8) into Equation (7), we get:

[0102]

[0103] where S Y has the following matrix form:

[0104]

[0105] where a r is the linear coefficient in linear correlation, and r is the order;

[0106] has the following matrix form:

[0107]

[0108] is S X only retaining the first r (r << n - 1) orders and then calculating the eigenvalues and eigenvectors is much faster than directly calculating the eigenvalues and eigenvectors of S X (because it is an r-order approximation).

[0109] Obviously, the matrix and S X have the same eigenvalues. Solving the eigenvalues and eigenvectors of the matrix is as follows:

[0110]

[0111] where Λ X is the eigenvalue of the matrix , that is, the eigenvalue of the high-dimensional flow field snapshot. W X is the eigenvector of the matrix .

[0112] Step 2: Using the compressed sensing technology, compress the original high-dimensional flow field snapshot containing huge flow field information to obtain a low-dimensional flow field snapshot matrix; specifically:

[0113] The compressed sensing dynamic mode decomposition method first needs to generate a measurement matrix C ∈ R q×m, that is, the compression matrix. Among them, the number of columns m of the measurement matrix C represents the dimension of the original high-dimensional flow field snapshot, and the number of rows q represents the dimension of the compressed low-dimensional flow field snapshot. Generally speaking, each row in the measurement matrix C has only one non-zero value, which is equivalent to sampling only the position corresponding to the column where the non-zero value is located. The position (column number) of the non-zero element is randomly generated to ensure the randomness of the sampling.

[0114] The high-dimensional original flow field snapshot matrix obtained in step 1 By compressing the measurement matrix C, we can get the compressed low-dimensional flow field snapshot matrix As shown below:

[0115]

[0116] in, It represents the matrix of the high-dimensional flow field snapshot from the first moment to the n-1th moment, and High-dimensional flow field snapshots at the nth moment, is the matrix of the high-dimensional flow field snapshot from the second moment to the nth moment, and High-dimensional flow field snapshot at the first moment, Y1 n-1 yes The low-dimensional flow field snapshot matrix obtained after compression of the measurement matrix C is: yes The low-dimensional flow field snapshot matrix is ​​obtained after compression of the measurement matrix C.

[0117] Step 3: Based on the low-dimensional flow field snapshot matrix, the eigenvalues ​​and DMD modes based on the low-dimensional flow field snapshot are obtained by using the compressed sensing dynamic mode decomposition method; specifically:

[0118] Assume Y1 n-1 and Y2 n The linear operator A Y , substituting formula (11) into formula (7), we have:

[0119]

[0120] Among them, S Y The matrix form is as follows:

[0121]

[0122] For Y1 n-1 Performing singular value decomposition, we can get:

[0123]

[0124] Among them, Y1 n-1is an m×(n - 1) matrix, where m represents the dimension of the low - dimensional flow field snapshots, i.e., the number of measurement points in the experiment or the number of grid points in the numerical calculation. U Y is an m×(n - 1) matrix, which contains the spatial structure information of the low - dimensional flow field. Σ Y is an (n - 1)×(n - 1) matrix, Σ Y The singular values in it are arranged from large to small, representing the energy distribution in the low - dimensional flow field. V Y is an (n - 1)×(n - 1) matrix, which contains the temporal structure information of the low - dimensional flow field.

[0125] Substituting Equation (13) into Equation (12), we get:

[0126]

[0127] Solving for the eigenvalues and eigenvectors, we get:

[0128]

[0129] The matrix form is as follows:

[0130]

[0131] is S Y Only keep the first r (r << n - 1) orders, and then calculate for the eigenvalues and eigenvectors, which is much faster than directly calculating the eigenvalues and eigenvectors of S Y (because it is an r - order approximation).

[0132] Among them, Λ Y is the eigenvalue of the matrix and is also the eigenvalue of the low - dimensional flow field snapshots. W Y is the eigenvector of the matrix

[0133] After obtaining the eigenvalues of the compressed low - dimensional flow field, the characteristic frequencies and modal growth rates of the compressed sensing dynamic mode decomposition modes can be calculated accordingly. The specific method is as follows:

[0134]

[0135] Among them, λ i represents the eigenvalue of the i - th order low - dimensional flow field snapshot, that is, the i - th element on the diagonal of the matrix Λ Y Δt represents the physical time interval between two adjacent flow field snapshots. The real part represents the growth rate of the compressed sensing dynamic mode decomposition mode, and the imaginary part represents the characteristic frequency, μ​i That is, the characteristic spectrum of the low-dimensional flow field mode, which includes the growth rate and characteristic frequency of the mode.

[0136] The DMD mode of the low-dimensional flow field snapshot can be written as follows:

[0137] Φ Y =U Y W Y (17)

[0138] or

[0139]

[0140] Step 4: Reconstruct the DMD mode of the high-dimensional flow field snapshot based on the DMD mode of the low-dimensional flow field snapshot; specifically:

[0141] 1) According to the DMD mode of low-dimensional flow field snapshot and based on the high-dimensional flow field snapshot matrix Directly reconstruct the DMD mode of the high-dimensional flow field snapshot:

[0142]

[0143] Among them, Φ X is the DMD mode of the high-dimensional flow field snapshot;

[0144] 2) The flow reconstruction technology is used to reconstruct the first few modes with relatively high energy in the low-dimensional flow field snapshots, and the normalized mean square error (MSE) is introduced to quantitatively measure the recognition quality of compressed sensing dynamic modal decomposition to obtain the final centrifugal compressor flow field analysis results; specifically:

[0145] The key flow modes obtained by DMD modal decomposition can be used to reconstruct the flow of the first few modes with relatively high energy through flow reconstruction technology.

[0146] The overall evaluation criteria of compressed sensing dynamic mode decomposition are calculation speed and recognition quality. By introducing the normalized mean square error (MSE), the recognition quality of compressed sensing dynamic mode decomposition is quantitatively measured. Taking flow velocity as an example, the root mean square error is defined as follows

[0147]

[0148] Among them, v rec and v original Represent the reconstruction speed and original speed respectively. Using different compression ratios σ, the computation time and recognition quality of compressed sensing dynamic mode decomposition under the same computing hardware conditions (Intel-Xeon-CPU E5-26862.3GHz, 128GB DDR5 RAM) are compared and analyzed.

[0149] Table 1 Structure and flow parameters of centrifugal compressor

[0150]

[0151] The results of the centrifugal compressor unstable flow field calculation example are described below. This embodiment carries out unsteady numerical simulation and experimental verification of the centrifugal compressor, and uses the dynamic mode decomposition method to perform spectrum and modal analysis on the transient flow data. Then, combined with the compressed sensing technology, the calculation time of DMD is significantly reduced under the premise of ensuring that the predicted frequency and flow modal errors are small. The centrifugal compressor mesh geometry and parameters used in this embodiment are shown in Table 1. Fine-Turbo software was also used to conduct a numerical study of the centrifugal compressor β based on the URANS solver. In the numerical simulation, the centrifugal compressor structure includes the calculation areas such as the inlet pipeline, rotating impeller, bladeless diffuser and volute, in which the impeller adopts the O4H type structure grid and the volute adopts the butterfly topology. In order to ensure that the grid resolution does not affect the accuracy of the numerical calculation, the grid number is verified by the steady numerical calculation of the centrifugal compressor under the condition of 60k rpm speed. In order to take into account both calculation accuracy and efficiency, a centrifugal compressor grid model with a grid number of 7.06 million was selected from the five grid number resolutions, and the y+ value was limited to 5.

[0152] In order to verify the accuracy of the numerical results, the explicit Runge-Kutta fourth-order format is used for time discretization in the steady-state simulation, the implicit double time step method is used to solve the equations in the unsteady-state simulation, and the Spalart-Allmaras model is used as the turbulence model. Figure 3 and Figure 4 The comparison results of the total pressure ratio P and the efficiency η are shown respectively. The results show that the numerical calculation results of the embodiment are very close to the experimental results, and the numerical calculation accuracy is within an acceptable range.

[0153] Figure 5 The figure shows the pressure recovery coefficient of the diffuser and the pressure rise coefficient of the impeller under different working conditions obtained by numerical simulation. When the dimensionless mass flow rate is 0.5, a sharp drop in the pressure recovery coefficient of the bladeless diffuser can be observed. Under the same operating conditions, for the impeller, the pressure rise coefficient continues to decrease with the decrease of the flow rate, but no obvious fluctuation of the coefficient is observed, indicating that the flow instability may come from the bladeless diffuser rather than the impeller. This also shows that the occurrence of flow instability has a significant negative impact on the pressure recovery performance of the bladeless diffuser.

[0154] Figure 6The figure shows the spectrum distribution and disturbance growth rate of the bladeless diffuser stall condition obtained based on the standard DMD and the compressed sensing dynamic modal decomposition when the compression ratio is 2%. For the spectrum related to the impeller rotation and stall mode, the blade passing frequency captured by the compressed sensing dynamic modal decomposition is consistent with the standard DMD. Compared with the stall frequency of 683Hz, the stall frequency obtained by the compressed sensing dynamic modal decomposition is quite close to the original stall frequency, which is 681Hz. However, since the number of calculated eigenvalues ​​has been greatly reduced, the density of the growth rate distribution has changed significantly. Under the same hardware conditions, the main flow modes can still be observed, and the calculation time is shortened from 55.4s to 6.6s.

[0155] Depend on Figure 7 It can be seen that even if the flow field data is compressed to only 0.1% of the original data volume, the recognition quality remains at a relatively high level. When the compression ratio is less than 10%, the calculation time is significantly shortened. When the compression ratio drops to 1% and the MSE is still lower than 0.01, the calculation time of compressed sensing dynamic mode decomposition is 6.4s, which is only 11.6% of the standard DMD, greatly shortening the numerical simulation time required by the existing method, verifying that the present invention greatly shortens the numerical simulation time.

[0156] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0157] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A centrifugal compressor unstable flow analysis method based on compressed sensing dynamic mode decomposition, characterized in that: include: The high-dimensional flow field snapshot matrix is ​​established by the DMD method, and the eigenvalues ​​of the high-dimensional flow field snapshot are calculated; Based on compressed sensing technology, the high-dimensional flow field snapshot matrix is ​​compressed to obtain a low-dimensional flow field snapshot matrix; Based on the low-dimensional flow field snapshot matrix and using the compressed sensing dynamic mode decomposition method, the eigenvalues ​​and DMD modes based on the low-dimensional flow field snapshot are obtained; The DMD mode of the high-dimensional flow field snapshot is reconstructed based on the DMD mode of the low-dimensional flow field snapshot.

2. The centrifugal compressor unstable flow analysis method based on compressed sensing dynamic mode decomposition according to claim 1 is characterized in that: The high-dimensional flow field snapshot matrix is ​​specifically: In the DMD method, the vector matrix form of the high-dimensional flow field snapshot obtained from experiments or numerical simulations is: Among them, x i , i∈[1,n] represents the high-dimensional flow field snapshot at the i-th moment of each flow parameter. The high-dimensional flow field snapshot data is sampled at equal time intervals. The total number of sampling moments is n, and the sampling time interval is Δt. In the above definition, the subscript 1 and superscript n represent the first moment and the last moment in the flow field sequence respectively. It is the matrix of high-dimensional flow field snapshots from the 1st moment to the nth moment.

3. The centrifugal compressor unstable flow analysis method based on compressed sensing dynamic mode decomposition according to claim 2 is characterized in that: Compute the eigenvalues ​​of a high-dimensional flow field snapshot, including: right Perform singular value decomposition and get: in, is an m×(n-1)-order matrix, where m represents the dimension of the high-dimensional flow field snapshot, i.e., the number of measurement points in the experiment or the number of grid points in the numerical calculation. X is an m×(n-1)-order matrix, which contains the spatial structure information of the high-dimensional flow field, Σ X is a (n-1)×(n-1)-order matrix, Σ X The singular values ​​in are arranged from large to small, representing the energy distribution in the high-dimensional flow field. X It is a (n-1)×(n-1)-order matrix, which contains the temporal structure information of the high-dimensional flow field; Thus we get in, The matrix form is as follows: Among them, a r is the linear coefficient in linear correlation, r is the order; Solving the Matrix The eigenvalues ​​and eigenvectors are as follows: Among them, Λ X is a matrix The eigenvalue of W is the eigenvalue of the high-dimensional flow field snapshot; X is a matrix The feature vector of .

4. The centrifugal compressor unstable flow analysis method based on compressed sensing dynamic mode decomposition according to claim 3 is characterized in that: Based on compressed sensing technology, the high-dimensional flow field snapshot matrix is ​​compressed to obtain a low-dimensional flow field snapshot matrix, including: Generate measurement matrix C∈R q×m , where the number of columns m represents the dimension of the high-dimensional flow field snapshot, and the number of rows q represents the dimension of the compressed low-dimensional flow field snapshot; Based on high-dimensional flow field snapshot matrix And by compressing the measurement matrix C, we can get the compressed low-dimensional flow field snapshot matrix As shown below: in, It represents the matrix of the high-dimensional flow field snapshot from the 1st moment to the n-1th moment. is the matrix of the high-dimensional flow field snapshot from the second moment to the nth moment, Y1 n-1 yes The low-dimensional flow field snapshot matrix obtained after compression of the measurement matrix C is: yes The low-dimensional flow field snapshot matrix is ​​obtained after compression of the measurement matrix C.

5. The centrifugal compressor unstable flow analysis method based on compressed sensing dynamic mode decomposition according to claim 4 is characterized in that: Based on the low-dimensional flow field snapshot matrix and using the compressed sensing dynamic mode decomposition method, the eigenvalues ​​based on the low-dimensional flow field snapshot are obtained, including: Assume Y1 n-1 and The linear operator A Y , then: Among them, e n-1 ∈R n-1 is the n-1th order unit vector, r is the residual vector, S Y The matrix form is as follows: For Y1 n-1 Perform singular value decomposition and get: Among them, Y1 n-1 is an m×(n-1)-order matrix, where m represents the dimension of the low-dimensional flow field snapshot, i.e., the number of measurement points in the experiment or the number of grid points in the numerical calculation. Y is an m×(n-1)-order matrix, which contains the spatial structure information of the low-dimensional flow field, Σ Y is a (n-1)×(n-1)-order matrix, Σ Y The singular values ​​in are arranged from large to small, representing the energy distribution in the low-dimensional flow field, V Y It is a (n-1)×(n-1)-order matrix, which contains the temporal structure information of the low-dimensional flow field; Thus we can get: in, The matrix form is as follows: Among them, a r is the linear coefficient in linear correlation, r is the order; Solution The eigenvalues ​​and eigenvectors of can be obtained: Λ Y is a matrix The eigenvalue of the low-dimensional flow field snapshot, W Y is a matrix The feature vector of .

6. The centrifugal compressor unstable flow analysis method based on compressed sensing dynamic mode decomposition according to claim 5 is characterized in that: The DMD mode of the low-dimensional flow field snapshot is: F Y =U Y W Y or Among them, Φ Y is the DMD mode of the low-dimensional flow field snapshot.

7. The centrifugal compressor unstable flow analysis method based on compressed sensing dynamic mode decomposition according to claim 1 is characterized in that: The DMD mode of the high-dimensional flow field snapshot is reconstructed based on the DMD mode of the low-dimensional flow field snapshot, including: According to the DMD mode of low-dimensional flow field snapshot and based on the high-dimensional flow field snapshot matrix Directly reconstruct the DMD mode of the high-dimensional flow field snapshot; Through the flow reconstruction technology, the low-dimensional flow field snapshots are reconstructed by selecting the modes ranked within the set value in order of energy from large to small.

8. The centrifugal compressor unstable flow analysis method based on compressed sensing dynamic mode decomposition according to claim 7 is characterized in that: According to the DMD mode of low-dimensional flow field snapshot and based on the high-dimensional flow field snapshot matrix Directly reconstruct the DMD mode of the high-dimensional flow field snapshot, specifically: Among them, Φ X is the DMD mode of the high-dimensional flow field snapshot.

9. The centrifugal compressor unstable flow analysis method based on compressed sensing dynamic mode decomposition according to claim 5, characterized in that: It also includes the calculation of the eigenfrequency and modal growth rate of the compressed sensing dynamic mode decomposition mode: Among them, λ i Represents the eigenvalue of the i-th order low-dimensional flow field snapshot, that is, the matrix Λ Y The i-th element on the diagonal, Δt, represents the physical time interval between two adjacent flow field snapshots, and the real part Represents the growth rate of the compressed sensing dynamic mode decomposition mode, the imaginary part represents the characteristic frequency, μ i That is, the characteristic spectrum of the low-dimensional flow field mode, which includes the growth rate and characteristic frequency of the mode.