Dear @DonVK and @fluid ,
I’am trying to design a diffraction horn for use with a B&C DCX50 compression driver. I’ve read yuichi arai’s book about diffraction horn design and build a CAD model. Now im interested if I can further optimize the design. I’ve tried to simulate the horn in Akabak but I came by the problems you describe in your blog post:
„Especially the individual fins or separate channels and their integration as BEM model at throat and at the end of the fins is tricky. You could either simulate them as individual channels but in this case several interfaces are necessary and you could end up with interface issues (one to many and many to one subdomain). Or the BEM model is one single mesh including the fins but then at fin start and end you encounter very close vertices when the fins are very sharp which is a problem with BEM.“ -> https://sphericalhorns.net/2023/04/09/the-mk3b2-radial-fin-horn/
With the individual channels as separate domains I not only struggle with interface issues, importing the model into akabak is incredible tedious. With all in one domain the fin ends are getting displayed in purple, indicating that the mesh is not sufficient dense.
Which of the modeling options yielded you the best results? What were your mesh settings? Did you use Gmesh for meshing?
For non-fin-horns i found this mesh settings:
"For examples a 400Hz horn simulated up to 10Khz, needs the driver membrane meshed at 5mm (150 elements), the inner horn at 10mm (2800 elements), outer horn at 30mm (1000 elements) mouth interface at 10mm (1000 elements), and edge at 20mm (100 elements). This 5050 element model, with 64 frequencies on an 8 thread 3.5Ghz processor takes approx. 3-4 hrs to solve." ->https://sphericalhorns.net/2019/10/20/bem-simulation-for-a-freestanding-horn/
How did you model the exterior of the horn? As the horn mouth is not straight I placed the horn inside an infinite baffle and modeled the outside of the quarter horn.
I’am trying to design a diffraction horn for use with a B&C DCX50 compression driver. I’ve read yuichi arai’s book about diffraction horn design and build a CAD model. Now im interested if I can further optimize the design. I’ve tried to simulate the horn in Akabak but I came by the problems you describe in your blog post:
„Especially the individual fins or separate channels and their integration as BEM model at throat and at the end of the fins is tricky. You could either simulate them as individual channels but in this case several interfaces are necessary and you could end up with interface issues (one to many and many to one subdomain). Or the BEM model is one single mesh including the fins but then at fin start and end you encounter very close vertices when the fins are very sharp which is a problem with BEM.“ -> https://sphericalhorns.net/2023/04/09/the-mk3b2-radial-fin-horn/
With the individual channels as separate domains I not only struggle with interface issues, importing the model into akabak is incredible tedious. With all in one domain the fin ends are getting displayed in purple, indicating that the mesh is not sufficient dense.
Which of the modeling options yielded you the best results? What were your mesh settings? Did you use Gmesh for meshing?
For non-fin-horns i found this mesh settings:
"For examples a 400Hz horn simulated up to 10Khz, needs the driver membrane meshed at 5mm (150 elements), the inner horn at 10mm (2800 elements), outer horn at 30mm (1000 elements) mouth interface at 10mm (1000 elements), and edge at 20mm (100 elements). This 5050 element model, with 64 frequencies on an 8 thread 3.5Ghz processor takes approx. 3-4 hrs to solve." ->https://sphericalhorns.net/2019/10/20/bem-simulation-for-a-freestanding-horn/
How did you model the exterior of the horn? As the horn mouth is not straight I placed the horn inside an infinite baffle and modeled the outside of the quarter horn.
We (@fluid, @docali) did a lot of model variations with subdomain partitions when looking at fin channel phase coherence. Close edge warning in AKABAK display as purple/pink faces. Some models were total free of warnings. However, It did not make much difference. The development models used a single interior subdomain with a single mouth interface (mesh 8-10mm) using GMSH and this will be sufficient to 10Khz. The exterior shell can be a simple shape (sweep mouth to throat) and at lower resolution, but round overs will always help.
Symmetry is your friend. This horn can be modelled in 1/4 symmetry. AKABAK assigns each frequency to a CPU thread, so the total #frequencies should be a multiple of the number of threads. Yes, it will take alot of CPU time and the time will get exponentially (cubic) longer as the model #elements increases. Lots of memory and CPU cooling helps.
Symmetry is your friend. This horn can be modelled in 1/4 symmetry. AKABAK assigns each frequency to a CPU thread, so the total #frequencies should be a multiple of the number of threads. Yes, it will take alot of CPU time and the time will get exponentially (cubic) longer as the model #elements increases. Lots of memory and CPU cooling helps.
Some models had separate mesh resolutions but not programmed dynamic meshing. Finer meshes when closer to the throat + fins and medium mesh for the inner horn and coarse mesh for the outer horn.