How to solve the inverse problem in modal analysis?

aenea · May 13, 2017, 08:59

I'm modeling ultrasonic horn (picture attached) and need to get a resonance at the frequency 25 kHz. I've imported CAD geometry to ansys with dimensions, finded after calculations in Matlab. And the mode of longitudinal oscillations is a little bit less than 25 kHz. I've been сhanging the length of part x (on the picture) to reach a resonance at the frequency 25 kHz. But may be I can find required length automatically in ansys? In summary: I know, what frequency I need, and I have to find optimal length for it (solve inverse problem).

LuckyTran · May 13, 2017, 13:40

Can't you just do it manually? I really don't think it's necessary to start a Manhattan project for this purpose... Do you really need the automation? You're falling into the classical trap of going straight to the 100 in the 1-10-100.

But anyways, there is an adjoint solver that does these types of things. You could also approach it using response surfaces. You're only tuning the length it seems so all these tools are overkill. You could solve the inverse part externally using basic least squares in excel, matlab, etc...

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Problem, analysis (ansys 13 and 14)	castaway	ANSYS	0	June 13, 2012 06:54
[ANSYS Meshing] help me for modal analysis of steel structure	Eng medo	ANSYS Meshing & Geometry	0	December 5, 2011 07:43
Can I solve this problem by Fluent?	Kai_kc	FLUENT	1	October 27, 2010 06:29
CFD software capable of modal analysis?	Jim-gineer	Main CFD Forum	4	January 14, 2010 10:29
extremely simple problem... can you solve it properly?	Mikhail	Main CFD Forum	40	September 9, 1999 10:11

May 13, 2017, 13:40		#2
LuckyTran Senior Member Lucky Join Date: Apr 2011 Location: Orlando, FL USA Posts: 5,761 Rep Power: 66	Can't you just do it manually? I really don't think it's necessary to start a Manhattan project for this purpose... Do you really need the automation? You're falling into the classical trap of going straight to the 100 in the 1-10-100. But anyways, there is an adjoint solver that does these types of things. You could also approach it using response surfaces. You're only tuning the length it seems so all these tools are overkill. You could solve the inverse part externally using basic least squares in excel, matlab, etc...