Keli ZhangSchool of Aeronautic Science and Engineering, Beihang University, Beijing, 100191, ChinaLHD, Institute of Mechanics, Chinese Academy of Sciences, Beijing, 100190, ChinaChangping YuLHD, Institute of Mechanics, Chinese Academy of Sciences, Beijing, 100190, ChinaPeiqing LiuSchool of Aeronautic Science and Engineering, Beihang University, Beijing, 100191, ChinaXinliang Lilixl@imech.ac.cnlpq@buaa.edu.cnLHD, Institute of Mechanics, Chinese Academy of Sciences, Beijing, 100190, China
(December 26, 2023)
Abstract
In this paper, a permeable surface nondimensional FW-H (Ffowcs Williams-Hawkings) acoustics analogy post-processing code with convective effectand AoA (angle of attack) corrections, OpenCFD-FWH, has been developed. OpenCFD-FWH is now used as post processing code of our finite volume CFD solverOpenCFD-EC (Open Computational Fluid Dynamic code for Engineering Computation). However, OpenCFD-FWH can also be used by other CFD solvers with thespecified data interface.The convective effect is taken into account by using Garrick Triangle to switch the wind tunnel cases coordinate system to a moving model with fluidat rest coordinate system, which simplifies the FW-H integration formulation and improves the computational efficiency of the code. The AoA effect isalso taken into account by coordinate transformation.In order to validate the code, three cases have been implemented. The first two cases are a monopole and a dipole in a mean flow with AoA, andthe results of the code and the analytical solution are practically identical. The third case is the well-known 30P30N configuration with aReynolds number of 1.71 and an AoA of . OpenCFD-EC with IDDES (Improved Delayed Detached-eddy simulation) is utilizedto obtain the flow field, and the result shows relative good agreement when compared to JAXA experiments. Moreover, the code is implementedin a hybrid parallel way with MPI and OpenMP to speed up computing processes (up to 538.5 times faster in the 30P30N validation case) andavoid an out-of-memory situation. The code is now freely available on https://github.com/Z-K-L/OpenCFD-FWH.
††preprint: AIP/123-QED
I INTRODUCTION
With the escalating demands for environmental protection, aeroacoustics noise hasreceived considerable attention from both the industrial and academic sectors,especially in the aviation sector. Aircraft noise is restricting the development ofairports. Hence, it is very important to conduct far-field noise evaluation duringaircraft development and design stages. One of the most practical ways to evaluate far-fieldnoise of the aircraft is the hybrid CAA (computational aeroacoustics) approach, since the DNS(direct numerical simulation) for far-field noise in engineering problems is unrealistic.
The hybrid CAA method involves obtaining unsteady flow field through CFD solvers and thenemploying acoustic analogy equations to calculate far-field noise, which is widely adopteddue to its substantial reduction in computational complexity.For example, Molina et al. Molinaetal. investigated tandem cylinder noise throughDDES (Delayed Detached-eddy simulation) and FW-H (Ffowcs Williams-Hawkings) acoustic analogy.Ma et al. Ma, Shi,andSong investigated aeroacoustic characteristics of Swept Constant-ChordHalf model with four different types of high-lift devices through IDDES(Improved Delayed Detached-eddy simulation) and FW-H acoustic analogy. Hu et al. Huetal. utilized implicit wall-resolved LES (Large Eddy Simulation) and FW-H acoustic analogyto explore the noise reduction mechanisms of TE (trailing edge) serrations. Chen et al.Chen, Zang,andAzarpeyvand also used the hybrid method of LES and the FW-H acoustic analogy, tostudy the noise of flow across a cylinder with varying spanwise lengths.Souza et al. Souzaetal. (a) carried out LBM (Lattice Boltzmann Method) simulation on the30P30N high-lift configuration and applied FW-H acoustic analogy to compute the associated acoustic field.Teruna et al. Terunaetal. analyzed the noise reduction effect of a fully resolved3-D printed porous TE utilizing LBM and FW-H acoustic analogy as well. DNS and FW-H acoustic analogywere conducted by Turner and Kim TurnerandKim to assess the importance of quadrupolenoise in aerofoil flow separation or stall conditions. They acquired the quadrupole noise by calculating therelative difference between the FW-H results of the solid and permeable surface.
Currently, there are only a few open source codes available for FW-H acoustic analogy, such as libAcoustics developedby Epikhin et al. Epikhinetal. for OpenFOAM written in C++, SU2PY_FWH developedby Zhou et al. Zhou, Albring,andGauger for SU2 written in python, and a Farassat 1A solver developed byShen et al. ShenandMiller for HiFiLES written in C++. However, they all have some problems. Firstthey do not support MPI (Message Passing Interface) parallel to accelerate the computing processes and reducememory usage by distributing the computing tasks across multiple nodes/computers, which is very important when facinglarge datasets. Second, they only support FW-H integration solutions for solid surface, which do not account forthe quadrupole noise and unable to address cases with porous materials. Third, libAcoustics and SU2PY_FWHrequire the installation of OpenFOAM and SU2 software, respectively. Fourth, libAcoustics and SU2PY_FWHdo not consider inflow with an AoA (angle of attack). Finally, they lack comprehensive tutorials, making it difficult forothers to use their codes with other CFD solvers.
Hence, the OpenCFD-FWH code has been developed for our compressible finite volume CFD solver OpenCFD-ECLi, Fu,andMa (2010) (Open Computational Fluid Dynamic code for Engineering Computation). More importantly, it can be utilizedby any other solvers with the right data structures. Alternatively, one can modify the data reading module of the code accordingly.The code is based on a permeable surface FW-H integration solution with the Garrick Triangle GarrickandWatkins (1953) appliedto simplify the equations. The inflow with an AoA is also taken into account. Additionally, the code is implemented in a hybridparallel way to accelerate the computing processes and reduce the memory requirement for a single node/computer.The deployment of the code demands only an MPI library and a Fortran 90 compilation environment.Furthermore, Matlab programs for monopole and dipole validation are provided to generate the required input data for tutorial purposes.
The rest of the paper is organized as follows. Sec.II derived the permeable surface FW-H acoustic analogy methods withconvective effect and AoA correction. Sec.III depicted the code structure and parallel implementation. Sec.IVpresent the results of the code for three different validation cases. Finally, conclusions are given in Sec.V.
II FW-H Acoustic Analogy
II.1 Permeable Surface FW-H equation
The FW-H equation FfowcsWilliamsandHawkings (1969) for permeable surface has the form of:
(1)
where the is the D’Alembert operator, c is the sound speed, is thedensity, is the density of the undisturbed medium, and are the Dirac delta and Heaviside function, respectively.The moving surface is described by such that is theunit outward normal of the surface Farassat (2007). The , and Lighthill tensor stress are defined as:
(2)
(3)
(4)
where is the component of the velocity of the moving surface, is the component of the velocity of the fluid, isthe Kronecker delta, and is the stress tensor:
(5)
where is the ambient pressure, is the viscous stress tensor. Usually is a negligible source of soundand is neglected by almost any other FW-H implementations Epikhinetal. ; Zhou, Albring,andGauger ; ShenandMiller ; Farassat (2007). Hence, is used in this paper.
II.2 Integration solution for general cases
Neglecting the quadrupole term in Eqs.(1), and following the derivation procedure of Farassat 1A formulation Farassat (2007),the integral solution of the FW-H equation for permeable surface can be derived as:
(6)
(7)
(8)
where is the observer coordinate vector, t is the observer time, r is the distance between observer and source, is the sound speed of the undisturbed medium, the superscript "·" means derivative over the source time , and the subscripts T and L represent the thickness and loading components, respectively. M is the Mach number vector of the moving surfacewith component . , , and are defined as:
(9)
(10)
(11)
where is the component of the unit radiation vector.
The subscript in Eqs.(6), and (7) means thequantities inside the square brackets are evaluated at retarded time:
(12)
Despite the quadrupole term in Eqs.(1) is omitted, the quadrupole source inside the permeable FW-H surface is still beaccounted for by Eqs.(8) according to Brentner and Farassat BrentnerandFarassat (1998).
II.3 Integration solution for wind tunnel cases
Eqs.(8) is derived in a coordinate system that, the source is moving in a stationary medium with observers at restin the far-field. In a wind tunnel case, where both the source and observer are stationary within a uniform flow with an AoA,which is the common scenario in the majority of aircraft CFD cases, the Garrick Triangle GarrickandWatkins (1953) can beapplied to transform the coordinate system. In this new coordinate system, the source is now moving in a stationary medium,while observers remain stationary relative to the source. This will lead to a large simplification of the formulation andincrease the computational efficiency of the code.
First, let us assume that the mean flow has a velocity of along the positive axis direction. The retarded time ofEqs.(12) will be changed to:
(13)
where R is the effective acoustic distance between the source and the observer Brès, Pérot,andFreed (2010):
(14)
(15)
(16)
(17)
where is the component of distance between the observer and the source.
The component of the unit radiation vector is now altered to:
Next, consider a mean flow with an AoA in the x-y plane, and its velocity magnitude remains equal to . By using the 2D planecoordinate transformation
(21)
(22)
and bring them into the and of the Eqs.(14), and Eqs.(15) yields:
(23)
(24)
(25)
(26)
The component of the unit radiation vector is also changed:
(27)
(28)
(29)
(30)
(31)
Then, replacing all the in Eqs.(6) Eqs.(10) by Eqs.(23).In addition, both the moving surface and fluid velocity need to subtract the mean flow velocity, sincethe coordinate system has changed. Now, the distance is a constant and can be calculated in advancefor each observer, rather than in every sampling frame. Moreover, the source time derivative of ,and will be zero because the surface is in uniform rectilinear motion. Therefore, the simplifiedversion of Eqs.(6), and Eqs.(7) for wind tunnel cases take the following form:
(32)
(33)
with
(34)
(35)
(36)
(37)
Notice that the quantities inside the square brackets in Eqs.(32), and (33) are now evaluatedat the retarded time calculated by Eqs.(13). And the necessary inputs for far-field noise calculationfrom the CFD solver include the coordinate, unit outward normal, and area of the FW-H surface, along with the density,velocity, and pressure pulsation at each sampling frame.
II.4 Nondimensionalization
Since OpenCFD-EC utilizes dimensionless Navier-Stokes equations, OpenCFD-FWH is based on a nondimensional version ofEqs.(32), and (33) to avoid data conversion errors and computational expenditures.
The reference quantity used for the dimensionless transformation is the mean flow quantity, with the exception thatthe pressure is nondimensionalized by , and the coordinate is nondimensionalized by the units of the mesh,which yields:
(38)
(39)
(40)
Then the dimensionless FW-H integration solution for wind tunnel cases can be obtained by replacing the variables inEqs.(32), and (33) to their corresponding nondimensional counterparts:
(41)
(42)
with
(43)
(44)
(45)
Note that , and remain unchanged whether the formulations arenondimensional or dimensional. Thus, the superscript "*" will not be necessary.
III Implementation of OpenCFD-FWH
OpenCFD-FWH can be divided into 4 main parts: initialization, pressure signals calculation, data output and finalization,as illustrated in Fig.1.
III.1 Initialization
The first step of the code is to initialize all the MPI processors. This involves control file reading,surface geometric data acquiring, assigning surface to the corresponding MPI processor, allocating memory,and reading the location of the observers as well as the FW-H dataset.
The essential parameters for the code are specified in the control.fwh file, such as Mach number, AoA, time step,number of observers, number of sampling frames, and number of OpenMP (Open Multi-Processing) threads.An example of the control.fwh file is given in AppendixA.1.
The coordinate , unit outward normal , and area of each subface in the FW-H surface are included inthe FWH_Surface_Geo.dat file. These quantities are specified at the center of the subfacesand split in different Faces, due to OpenCFD-EC is a cell center solver for multiblock structure mesh.A detailed description of the file can be found in AppendixA.2.
With the Faces information acquired, a partition method deployed by OpenCFD-EC for block splitting is applied forload balancing as shown in Fig.2. The method will first rank the Faces by their cellnumbers, and then assign the Faces to the processor with the least number of cells in order. As a result, the upper limitfor the utilization of MPI processors by the code corresponds to the number of the Faces. To achieve a faster MPIacceleration result, one can divide FW-H surface as much and as equally as possible during the mesh generation stage,or segment the output FW-H dataset of the CFD solver. Besides, when using a new big.little architectureCPU of Intel, or parallelizing an old system with a new one, one can adopt a partition method that considers theperformance of the processors to accomplish optimal load balancing.
Following the MPI partition, each processor will allocate memory for FW-H data at all the sampling frames according tothe assigned Faces. Then the root processor reads the observers.dat file, in which the coordinates of each observer occupya single row. Then, they are broadcasted to every other processor.
Subsequently, the FW-H dataset is read by the root processor and distributed to the corresponding processor.An illustrative description of the FW-H dataset can be found in AppendixA.3. Finally,memory is allocated for interpolated observer time across all processors, along with final pressure signal resultat the root processor.
III.2 Pressure signals calculation
The second step of the code is to calculate the pressure signals at one observer, rather than computing at all observersat once for the sake of conserving memory usage. Additionally, OpenMP parallel is deployed on all processors to expeditethe calculation procedure. Further details regarding the MPI and OpenMP mixing parallel implementation will be expoundedin Section III.4.
The "compute R" subroutine in Fig.1 is responsible for calculating the effective acousticdistance based on Eqs.(23) Eqs.(31).
The "compute Noise" subroutine in Fig.1 is responsible for calculating the pressure signalsat each subface during respective source time, based on Eqs.(41) Eqs.(45). It isnoteworthy that and remain unchanged in different observers, but they are both not stored to save memory.In addition, and are computed using second-order schemes, employing a one-sided scheme for thefirst and last frame, while employing a central scheme for the other frames.
The "compute t" subroutine in Fig.1 is responsible for calculating the observer time based on:
(46)
(47)
where and are computed by making used of the MPI_ALLREDUCE function. Consequently,the observer time period, during which all subfaces collectively contribute to the observer pressure signal is, as shown in Fig.3.
The "interp pressure signal" subroutine in Fig.1 is responsible for interpolating thepressure signal of each subface depending on the source time, into pressure signal depending on the observer time, withthe help of cubic spline interpolation. After that, both the observer time and source time pressure signal stored atevery subface will be deallocated to conserve memory as well.
III.3 Data output and Finalization
The third step of the code is to conduct surface integration across all the subfaces to obtain the pressure signal at aobserver and output it in the p_observer-xxx.log (xxx stands for the observer No.) file located within the/FWH_result-mpi folder.
The final step of the code is to verify whether all observers have completed their calculations, if the answer is no,the code will loop back to the second step for the subsequent observers. If the answer is yes, all the processors willcall MPI_Finalize to end the code.
III.4 Parallelization
OpenMP is a widely used API (application programming interface) that supports shared-memory parallelization in multi-coreand multi-processor systems. It is developed to facilitate parallel programming in C, C++, and Fortran, and can be easilydeployed without extensive modifications to the existing serial codes.
OpenCFD-FWH is implemented in a hybrid parallel way that the FW-H data surface is spread to many MPI processors,and OpenMP is deployed to split the loop in the computing stage among each MPI processor. This will result in an enormousreduction of the computation time, as shown in Table1. With the use of 31 nodes, each with 32 CPUcores for OpenMP parallelization, a remarkable acceleration of up to 538.5 times is achieved in comparison to the serial condition.When the number of MPI processors (only MPI parallel) and OpenMP threads (only OpenMP parallel) is almost equal, their accelerationeffect is nearly the same.
MPI processors
OMP threads
init time/s
computing time/s
total time/s
computing acceleration ratio
total acceleration ratio
1
753.7
18120.3
18873.9
31
887.2
811.3
1698.5
22.3
11.1
1
32
823.9
774.3
1598.1
23.4
11.8
31
32
717.5
33.7
751.2
538.5
25.1
Additionally, Table1 illustrates that the predominant portion of the execution time is spent oninitialization in the hybrid parallel condition. This is attributed to the fact that I/O operations can not be accelerated,as the data reading process requests sequential operations.
Furthermore, the 30P30N validation case costs a maximum of 62.6 GB of RAM (Random Access Memory). Without MPI parallelization,the computational demands for larger FW-H datasets can pose significant challenges for nodes and computers with limited memorycapacity. Hence, the MPI parallelization ensures successful execution on memory-constrained systems, or for even larger datasetsthat can easily consume hundreds of RAM.
IV Validation
Stationary monopole and dipole in a uniform flow with analytic solutions, along with a 30P30N case computed byOpenCFD-EC solver are used to validate OpenCFD-FWH.
IV.1 Stationary monopole in a uniform flow with AoA
The complex velocity potential for a stationary monopole in a uniform flow is given by Najafi et al.Najafi-Yazdi, Brès,andMongeau (2011) as:
(48)
where A is the amplitude, is the angular frequency of the monopole, and is the imaginary unit. In contrast to Najafi et al.Najafi-Yazdi, Brès,andMongeau (2011), Eqs.(23) and (24) are used here to calculate and , respectively,taking into account the AoA effect of the uniform flow.
The velocity, pressure, and density pulsations induced by the monopole are:
(49)
(50)
(51)
(52)
The parameters used for the monopole are given in Table2. To avoid any errors introduced by the CFD solver,the FW-H dataset for the code is generated by Eqs.(48) (51).
340
0.6
1
45
1
5
The permeable FW-H data surface is a sphere with a radius of 2 meters. Its center is located on the monopole. The sphere isdivided into 18 segments at the polar angle direction and 36 segments at the azimuth angle direction, resulting in a total of648 subfaces, as shown in Fig.4.
The observer locations are evenly distributed along a circle with a radius of 340 meters in the x-y plane, and its center is alsolocated on the monopole. A Matlab code is written to generate 1000 sampling frames, covering a time period of 2.5 seconds, in just26 seconds on a laptop equipped with an Intel i7-13620H CPU. Subsequently, the OpenCFD-FWH code processes the dataset in less than2.5 seconds with 12 OpenMP threads on only one MPI processor for 20 observers. Since the sphere FW-H surface is generated as asingle Face.
The comparison between the exact monopole solution and the result obtained from the OpenCFD-FWH code of far-field RMS(root mean square) pressure directivity and pressure signal of the right below observer are shown in Fig.5 and Fig.6, respectively. Very good agreements are observed betweenthe exact solution and the code. It is worth noting that results with even smaller errors can be achieved by using finerFW-H surface mesh and higher time sampling frequencies, but for the sake of simplicity, these results are not presented here.
Moreover, Fig.5 demonstrates that the directivity pattern of the monopole is diverted towardsthe inflow due to the convective effect.
IV.2 Stationary dipole in a uniform flow with AoA
The complex velocity potential for a stationary dipole, with the polar axis coinciding with the x2-axis in a uniform flowis given by:
(53)
The velocity, pressure, and density pulsations induced by the dipole are acquired by Eqs.(49) (51) as well. And the parameters used for the dipole are given in Table3.
340
0.5
1
10
1.5
7.5
The FW-H data surface remains consistent with the monopole case, while the observer locations have been relocatedto a radius of 50 meters in the x-y plane. Another Matlab code has been developed to generate 1000 sampling frames,covering a time period of 2 seconds. The time required for generating the FW-H dataset and post-processing it isbasically the same compared with the monopole case.
Fig.7 and Fig.8 present the result of the dipole far-field RMS pressuredirectivity and pressure signal of the right below observer, respectively. Excellent agreements are also achieved betweenthe exact solution and the code. By applying finer FW-H surface mesh and higher time sampling frequencies, results withessentially no error can be achieved. Again, for the sake of simplicity, these results are not presented here.
The mean flow leads to a reorientation of the maximum RMS pressure, resulting in a larger RMS pressure in the inflowdirection as shown in Fig.7.
All the Matlab programs used for the monopole and dipole validation cases are provided in the Tutorials folder of theOpenCFD-FWH project on GitHub. One can change the parameters in these programs to validate our code and get a betterunderstanding of OpenCFD-FWH.
IV.3 30P30N far-field noise prediction
The 30P30N configuration was developed by McDonnell Douglas (now Boeing) in the early 1990s. It has been extensively used in thestudy regarding the aeroacoustic characteristics of high-lift devices, especially for slat noise Murayamaetal. (a); Terracoletal. ; Souzaetal. (b); ChoudhariandLockard ; Terracol, Manoha,andLemoine ; Murayamaetal. (b); Souzaetal. (2019); Himenoetal. .
The JAXA modified version 30P30NMurayamaetal. (a, b) is utilized here to validate the code.The airfoil profile of the 30P30N configuration is shown in Fig.9, with a stowed chord length of .Both the deflection angles of the slat and flap are , with the chord lengths of the slat and flap being and, respectively.
IDDES based on SA turbulence model is carried out on the OpenCFD-EC solver. The inflow Mach number is , with an AoA of .The Reynolds number based on the stowed chord length is .
Fig.10 depicts a sketch of the computational domain. It extends in theforward and vertical directions and in the rear direction. Its length in the spanwise direction equals to, following the recommendation in the BANC-III workshopChoudhariandLockard (the 3rd AIAA Workshop on Benchmark Problems for Airframe Noise Computations). A periodic boundary condition is appliedin the spanwise direction. The permeable FW-H data surface is indicated by the blue line in Fig.10,which is one stowed chord length away from the 30P30N airfoil and stretches in the wake flow direction. No endcap isused to avoid the spurious (numerical) noise created by wake flows crossing the permeable FW-H data surfaceRibeiroetal. (2023). The spanwise length of the FW-H surface is identical to the computational domain. Besides, a spongelayer is deployed at the boundaries of the computational domain, where the viscosity is adjusted to 100 times the value used inthe physical domain to mitigate reflections at the domain boundaries, in accordance with the approach taken by Himeno et al.Himenoetal. .
A multiblock structure mesh with C-type topological is created, yielding a total cell count exceeding 43 million. Eachplane of the 2.5D mesh comprises approximately 0.25 million cells, and the entire mesh is composed of 175 planes withequal spacing in the spanwise direction. Close-up views of the mesh around FW-H surface and slat cove area are shown inFig.11. Additionally, the average value of the dimensionless wall distance of the mesh is below unity.
The well-known Roe scheme is employed to decompose the inviscid flux, with the third-order MUSCL scheme for variablereconstruction. The implicit dual-time LU-SGS method is applied for time advancement, with a time step of .Five inner subiterations are used, with local time-stepping approach to accelerate the convergency process. And the FW-H datasampling interval is . An RANS simulation with SA turbulence model is carried out to initialize the flow field.Subsequently, approximately of physical time is calculated by IDDES-SA, with available for data processingafter removing the initial transient.
A time average of is obtained, with a difference less than compared to the average outcomes of the BANC-IIIChoudhariandLockard . Fig.12 presents the time average distribution obtained over the last seconds. While there is a slight underprediction of negative pressure on the suction side, a reasonably good agreement canbe seen, especially in the slat cove region, when compared with the JAXA Kevlar experimentMurayamaetal. (b) under AoA.
The scaling method used by Avallone et al.Avallone, Vander Velden,andRagni (2017) is utilized to take into account the difference betweenthe spanwise acoustic integration size of the numerical simulation and experiment:
(54)
where and are equal to and , respectively.
The pressure signal obtained by the OpenCFD-FWH code is segmented into blocks with a overlap, anda Hanning window is employed. A total of 6535 sampling frames are input to the code, and the running time with differentparallel strategies can be found in Table1. As shown in Fig.13, the PSD(Power Spectral Density) result of the code is in good consistency with the JAXA hard-wall experiment at frequenciesbelow . Both the results exhibit a slightly higher noise level in the low-frequency range compared to the JAXAkevlar-wall experiment. Furthermore, the humping noise originating from the high-frequency vortex shedding from the slatTE of the reduced-scale wind tunnel model is absent in the FW-H result. This is attributed to the relatively coarse meshat the slat TE, which is unable to capture the high-frequency vortex shedding. Overall, the result validates that thefar-field noise can be accurately evaluated by the OpenCFD-FWH code.
V CONCLUSIONS
This paper presents the methodology, parallel implementation and validation of a post-processing code: OpenCFD-FWH,designed specifically for predicting far-field noise in wind tunnel cases, encompassing nearly all scenariosencountered in aircraft CFD cases. It is developed to use the flow field results of our OpenCFD-EC solver as input.However, it can be readily deployed for use with other solvers by modifying the data reading part of the code orconverting the FW-H dataset to the required format. Moreover, the deployment of the code only required an MPI libraryand a Fortran 90 compilation environment, without the need to install OpenCFD-EC or other affiliated libraries.
The code is based on the integration formulation of a nondimensional FW-H equation for permeable surface with convectiveand AoA effects corrected by Garrick Triangle, and 2D plane coordinate transformation, respectively. This formulation willincrease the computational efficiency compared to the original one. Additionally, the nondimensionalization of the FW-Hequation is the same as the nondimensionalization of the Navier-Stokes equations in the OpenCFD-EC solver.
MPI-OpenMP mixing parallelization is implemented to accelerate the post-processing process and reduce memory usage on a singlenode/computer when deploying the code on distributed computing systems. When dealing with very large datasets, as is common inaeroacoustic noise research related to landing gear or high-lift devices with LES, it can avoid an out-of-memory situation.On the CAS SunRising platform, by utilizing 31 nodes, each with 32 OpenMP threads, the computing time of the code is 538.5times faster compared to the serial implementation. This demonstrated the high operational efficiency of OpenCFD-FWH.
Three validation cases are considered in this paper. The monopole and dipole cases are compared with exact analytical solutions,and excellent agreements are achieved. The 30P30N configuration is used in the third case, with the flow field variable producedby IDDES-SA simulation via the OpenCFD-EC solver as input. For frequencies below , the far-field PSD result demonstratesrelatively good agreement with JAXA experiments, particularly with the JAXA hard-wall experiment. However, the result of thecode does not present the high-frequency hump observed in the experiments. This is due to the inability of the coarse mesh tocapture the high-frequency vortex shedding at the slat TE. Overall, the code is deemed validated.
The code is openly accessible on GitHub, along with the Matlab codes for the monopole and dipole validation cases tofacilitate its utilization by readers.
Acknowledgements.
This work was supported by the National Natural Science Foundation of China (Grant No. 12272024),National Key Research and Development Program of China (Grant Nos. 2020YFA0711800, 2019YFA0405302, 2019YFA0405300)and NSFC Projects (Grant Nos. 12072349, 12232018, 12202457), National Numerical Windtunnel Project, Science Challenge Project(Grant No. TZ2016001), and Strategic Priority Research Program of Chinese Academy of Sciences (Grant Nos. XDCO1000000 and XDB0500301).The authors thank CNIC (Computer Network Information Center), CAS (Chinese Academy of Sciences) for providing computer time.
The data that support the findings of this study are available from the corresponding author upon reasonable request.
Appendix A File structures for OpenCFD-FWH
A.1 An example of the control.fwh file
OpenCFD-FWH utilized namelist method to read in control.fwh file. An example of the file is presented in Fig.A1, with the values therein representing the default settings in OpenCFD-FWH. Kstep_start and Kstep_end determine the start and final steps for FW-H post-processing, respectively.With the step interval: delta_step, OpenCFD-FWH can calculate the number of sampling frames. FWH_data _Format decided whether the FW-H dataset is in binary or ASCII format(0 for binary and 1 for ASCII).
A.2 Structure of the FWH_Surface_Geo.dat file
FWH_Surface_Geo.dat file is in ASCII format in convenient for data checking. It followsa structure similar to the Generic boundary description .inp file. It starts with a first line of text: variables=x,y,z,n1,n2,n3,dS,as illustrated in Fig.A2. The second line contains a single number indicating the total number ofFaces. Subsequently, is the nx, ny, nz for one Face, along with the x, y, z, n1, n2, n3, dS values for its subfaces, until the last Face.Note that each Face is described in a block way, consequently one of the nx, ny, nz will be 1.
A.3 Structure of the FW-H dataset
The FW-H dataset used for OpenCFD-FWH comprised multiple FWH-xxxxxxxx.dat files, where xxxxxxxx denotes the iteration steps.The file can be in either binary or ASCII format, with binary format being recommended for its significantly smaller data size.A schematic of the file structure is provided in Fig.A3.
REFERENCES
References
(1)E.Molina, B.Y.Zhou, J.J.Alonso, M.Righi, andR.G.Silva,“Flow and noise predictions around tandem cylinders using DDES approach with SU2,”inAIAA Scitech 2019 Forum(American Institute of Aeronautics and Astronautics).
(2)X.Ma, Y.Shi, andW.Song,“Aerodynamic and aeroacoustic analysis of SCCH models of four high-lift configurations near stall angle of attack,”inAIAA AVIATION 2020 FORUM(American Institute of Aeronautics and Astronautics).
(3)Y.-S.Hu, Z.-H.Wan, C.-C.Ye, D.-J.Sun, andX.-Y.Lu,“Noise reduction mechanisms for insert-type serrations of the NACA-0012 airfoil,”941,A57.
(4)G.Chen, B.Zang, andM.Azarpeyvand,“Numerical investigation on aerodynamic noise of flow past a cylinder with different spanwise lengths,”35,035128.
Souzaetal. (a)D.S.Souza, D.Rodríguez, F.H.T.Himeno, andM.A.F.Medeiros,“Dynamics of the large-scale structures and associated noise emission in airfoil slats,”875,1004–1034 (a).
(6)C.Teruna, F.Avallone, D.Ragni, A.Rubio-Carpio, andD.Casalino,“Numerical analysis of a 3-d printed porous trailing edge for broadband noise reduction,”926,A17.
(7)J.M.TurnerandJ.W.Kim,“Quadrupole noise generated from a low-speed aerofoil in near- and full-stall conditions,”936,A34.
(8)A.Epikhin, I.Evdokimov, M.Kraposhin, M.Kalugin, andS.Strijhak,“Development of a dynamic library for computational aeroacoustics applications using the OpenFOAM open source package,”66,150–157.
(9)B.Y.Zhou, T.Albring, andN.R.Gauger,“Aeroacoustic prediction and optimization capabilities in SU2,”.
(10)W.ShenandS.A.E.Miller,“Validation of a high-order large eddy simulation solver for acoustic prediction of supersonic jet flow,”28,1950023.
Li, Fu,andMa (2010)X.Li, D.Fu, andY.Ma,“Direct numerical simulation of hypersonic boundary layer transition over a blunt cone with a small angle of attack,”Physics of Fluids22 (2010).
GarrickandWatkins (1953)I.GarrickandC.E.Watkins,“A theoretical study of the effect of forward speed on the free-space sound-pressure field around propellers,”Tech. Rep.(1953).
FfowcsWilliamsandHawkings (1969)J.E.FfowcsWilliamsandD.L.Hawkings,“Sound generation by turbulence and surfaces in arbitrary motion,”Philosophical Transactions of the Royal Society of London. Series A, Mathematical and Physical Sciences264,321–342 (1969).
Farassat (2007)F.Farassat,“Derivation of formulations 1 and 1a of farassat,”Tech. Rep.(2007).
BrentnerandFarassat (1998)K.S.BrentnerandF.Farassat,“Analytical comparison of the acoustic analogy and kirchhoff formulation for moving surfaces,”AIAA journal36,1379–1386 (1998).
Brès, Pérot,andFreed (2010)G.Brès, F.Pérot, andD.Freed,“A ffowcs williams-hawkings solver for lattice-boltzmann based computational aeroacoustics,”in16th AIAA/CEAS aeroacoustics conference(2010)p.3711.
Najafi-Yazdi, Brès,andMongeau (2011)A.Najafi-Yazdi, G.A.Brès, andL.Mongeau,“An acoustic analogy formulation for moving sources in uniformly moving media,”Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences467,144–165 (2011).
Murayamaetal. (a)M.Murayama, K.Nakakita, K.Yamamoto, H.Ura, Y.Ito, andM.M.Choudhari,“Experimental study on slat noise from 30p30n three-element high-lift airfoil at JAXA hard-wall lowspeed wind tunnel,”in20th AIAA/CEAS Aeroacoustics Conference(American Institute of Aeronautics and Astronautics).
(19)M.Terracol, E.Manoha, M.Murayama, andK.Yamamoto,“Aeroacoustic calculations of the 30p30n high-lift airfoil using hybrid RANS/LES methods: Modeling and grid resolution effects,”in21st AIAA/CEAS Aeroacoustics Conference(American Institute of Aeronautics and Astronautics).
Souzaetal. (b)D.Souza, D.Rodríguez, L.Simões, andM.Medeiros,“Effect of an excrescence in the slat cove: Flow-field, acoustic radiation and coherent structures,”44,108–115 (b).
(21)M.M.ChoudhariandD.P.Lockard,“Assessment of slat noise predictions for 30p30n high-lift configuration from BANC-III workshop,”in21st AIAA/CEAS Aeroacoustics Conference(American Institute of Aeronautics and Astronautics).
(22)M.Terracol, E.Manoha, andB.Lemoine,“Investigation of the unsteady flow and noise generation in a slat cove,”54,469–489.
Murayamaetal. (b)M.Murayama, Y.Yokokawa, H.Ura, K.Nakakita, K.Yamamoto, Y.Ito, T.Takaishi, R.Sakai, K.Shimoda, T.Kato, andT.Homma,“Experimental study of slat noise from 30p30n three-element high-lift airfoil in JAXA kevlar-wall low-speed wind tunnel,”in2018 AIAA/CEAS Aeroacoustics Conference(American Institute of Aeronautics and Astronautics).
Souzaetal. (2019)D.S.Souza, D.Rodríguez, F.H.Himeno, andM.A.Medeiros,“Dynamics of the large-scale structures and associated noise emission in airfoil slats,”Journal of Fluid Mechanics875,1004–1034 (2019).
(25)F.H.Himeno, D.S.Souza, F.R.Amaral, D.Rodríguez, andM.A.Medeiros,“SPOD analysis of noise-generating rossiter modes in a slat with and without a bulb seal,”915,A67.
Ribeiroetal. (2023)A.F.Ribeiro, M.R.Khorrami, R.Ferris, B.König, andP.A.Ravetta,“Lessons learned on the use of data surfaces for ffowcs williams-hawkings calculations: Airframe noise applications,”Aerospace Science and Technology135,108202 (2023).
Avallone, Vander Velden,andRagni (2017)F.Avallone, W.Vander Velden, andD.Ragni,“Benefits of curved serrations on broadband trailing-edge noise reduction,”Journal of Sound and Vibration400,167–177 (2017).
Introduction: My name is Errol Quitzon, I am a fair, cute, fancy, clean, attractive, sparkling, kind person who loves writing and wants to share my knowledge and understanding with you.
We notice you're using an ad blocker
Without advertising income, we can't keep making this site awesome for you.