Forums

A difficult deconvolution...

Started by tramoman June 30, 2005
Hello,

I am trying to model the behaviour of a reactor vessel when some
particles inside the container strike the inner wall using FINITE
ELEMENT MODELLING software.

I reproduced several random particles hitting the wall at different
times and position across the surface of the inside wall.

Now I would like to get rid of the TRANSFER FUNCTION of the container
from the random particles data using the deconvolution to extract only
the particle informations.

I apply an inpulse inside the vessel (the middle) and I acquire the
output on the outer surface.

THe PROBLEM: How can I deconvolve the spectra in a right way?
In fact the TRANSFER FUNCTION of the vessel changes according with the
position of the impact.

The tranfer function of the vessel will be completely different when
considering the position A or position B.


        | |
 posB>  | |
        | |
 posA>  | |> output
        | |
        | |






Can somebody help me please?
What kind of DSP tool should I use?
it is really important for me to solve this problem.

Thanks a lot
M

Your transfer function is not unique, it is spatially varying.

You would need to either:
1. compute it analytically
2. estimate it numerically, like in a finite but finely sampled grid of
points over the surface of your vessel (using your finite element
software? You would need to run a number of different simulations, in
each of which the impact point on the inside face of the wall would be
different)

tramoman wrote:
> Hello, > > I am trying to model the behaviour of a reactor vessel when some > particles inside the container strike the inner wall using FINITE > ELEMENT MODELLING software. > > I reproduced several random particles hitting the wall at different > times and position across the surface of the inside wall. > > Now I would like to get rid of the TRANSFER FUNCTION of the container > from the random particles data using the deconvolution to extract only > the particle informations. > > I apply an inpulse inside the vessel (the middle) and I acquire the > output on the outer surface. > > THe PROBLEM: How can I deconvolve the spectra in a right way? > In fact the TRANSFER FUNCTION of the vessel changes according with the > position of the impact. > > The tranfer function of the vessel will be completely different when > considering the position A or position B. > > > | | > posB> | | > | | > posA> | |> output > | | > | | > > > > > > > Can somebody help me please? > What kind of DSP tool should I use? > it is really important for me to solve this problem. > > Thanks a lot > M
Thanks for the answers:

Estimate it numerically seems a good option. I can perform different
simulations across the inner surface of the vessel. The level of the
discretization is 500 points in 10 cm. I could run 100 simulations, but
at the end it is not completely clear:
 how can I compute the data of the vessel together?
 In fact I need to extract the acoustic signature of the vessel in
order to use it to correct the acoustic real experiments with a real
vessel.

Can you please be more detailed?
Once I run 100 simulation, I have 100 transfer functions of the vessel,
how can use to obtain a valid frequency tranfer function of the vessel
and so apply the deconvolution of my data?

Thanks a lot
Manuel

"tramoman" <tramoman@yahoo.it> wrote in message 
news:1120473231.052615.136520@g43g2000cwa.googlegroups.com...
> Thanks for the answers: > > Estimate it numerically seems a good option. I can perform different > simulations across the inner surface of the vessel. The level of the > discretization is 500 points in 10 cm. I could run 100 simulations, but > at the end it is not completely clear: > how can I compute the data of the vessel together? > In fact I need to extract the acoustic signature of the vessel in > order to use it to correct the acoustic real experiments with a real > vessel. > > Can you please be more detailed? > Once I run 100 simulation, I have 100 transfer functions of the vessel, > how can use to obtain a valid frequency tranfer function of the vessel > and so apply the deconvolution of my data? > > Thanks a lot > Manuel >
Hi Manuel - could you describe what you are trying to do in a bit more detail? As far as I can see you have : 1) run simulation to see how your vessell will 'ring' when hit with an impulse 2) determined that the way the vessel 'rings' depends on the position of the impacting particle. 3) checked that there are no dead zones at your chosen sensor position so you get a nice clear ring for every particle impact no matter what the position. 4) Ignored background noise for now. 5) Treat everything as elastic so superposition applies. What do you want to do now? Do you : a) want a count of the number of impacts - don't care about position or magnitude of the impact - not worried about the exact time that it happened at? b) need a good estimate of time impact happened? c) desire an estimate of magnitude of impact? d) hope for an estimate of position of impact? I may be completely wrong but when I read your posts I get the impression that by "transfer function" you mean a sort of frequency domain description of how the vessel rings when hit by a particle - if hit in one position you get different modes of vibration set up compared with a strike in another position so the relative strength of different harmonics of several different fundamental modes may give you an indication of particle position provided that impacts only happen infrequently - is this what you are trying to do? I suspect that you have lots of impacts happening all the time and really, eventually, would like to know how accurately you can estimate energy transfer to the vessel per unit time. Can you give an estimate of the number of impulses you expect per cycle of your lowest frequency mode of oscillation, frequency range of the modes that you need to monitor and just how damped they are? If you can clarify what your aims are I'm confident that someone here can help even if its just to say use an HPF to remove all the low frequency ringing and see how closely total power out of the HPF relates to energy into the vessel (assuming some distribution of particle arrival location , direction and energy) but I suspect that you need to give them a bit more information to get their teeth into. Best of Luck - Mike
Hello Mike,

first of all thanks for you time spent of writing me.

I will try to be more detailed:


I want to figure out  the particles concentration of my vessel. To do
this I can simulate random impulses (random time of arrival and random
position of impact across the inside surface of the vessel). Because
the transducer is suppose to be outside in the midde of the vessel,
this will be the monitoring point.

2) determined that the way the vessel 'rings' depends on the position
of the
impacting particle.

Yes, I would like to extract all the possible informations of the
vessel and to apply the deconvolution and extract the particle
concentration only.
Because  the position of the impact it isn't not just in front of the
monitoring point, but it is distant dependent, I would LIKE TO KNOW A
DSP WAY TO EXTRACT THE INFORMATION THAT I NEED. IT IS NOT JUST A SIMPLE
DECONVOLUTION, because frequency impulse response of the vessel varies
with the position of the impact!!


3) checked that there are no dead zones at your chosen sensor position
so
you get a nice clear ring for every particle impact no matter what the
position.

I have already did it. No dead zones!

 4) Ignored background noise for now.
5) Treat everything as elastic so superposition applies.

Ok, thanks.



Answers:

a) want a count of the number of impacts - don't care about position or

magnitude of the impact - not worried about the exact time that it
happened
at?

YES, I want to know roughly how many particles are hitting my vessel
(10000, 20000..etc..)

b) need a good estimate of time impact happened?
I don't mind about the position or the time happened.

c) desire an estimate of magnitude of impact?
In a way, yes because the magnide is directly correlated to the number
of impacts.

d) hope for an estimate of position of impact?
No positions, only concetration information.

Sorry, my mistake, I meant TRANSFER FUNCTION as the frequency impulse
response. So, a sort of frequency domain description of how the vessel
rings when hit by a particle.


AS you understood, I have a lot of impacts all the time and
really, eventually,I  would like to know how accurately you can
estimate
energy transfer to the vessel per unit time.

I can simulate 10000 particles hitting the wall in 800 usec. I must
find a DSP way to estimate that. The spectra obtained considering the
pressure variation of the vessel (tranducer surface) it is between 10
and 150 Khz.

Can somebody suggest me a way to estimate the vessel characteristics
and extract the number of particle information?

Thanks a lot.
I would like just a direction, like : use blind deconvolution or
something else.
Thanks
Manuel

"tramoman" <tramoman@yahoo.it> wrote in message 
news:1120553307.760692.99130@g44g2000cwa.googlegroups.com...
> Hello Mike, > > first of all thanks for you time spent of writing me. > > I will try to be more detailed: > > > I want to figure out the particles concentration of my vessel. To do > this I can simulate random impulses (random time of arrival and random > position of impact across the inside surface of the vessel). Because > the transducer is suppose to be outside in the midde of the vessel, > this will be the monitoring point. > > 2) determined that the way the vessel 'rings' depends on the position > of the > impacting particle. > > Yes, I would like to extract all the possible informations of the > vessel and to apply the deconvolution and extract the particle > concentration only. > Because the position of the impact it isn't not just in front of the > monitoring point, but it is distant dependent, I would LIKE TO KNOW A > DSP WAY TO EXTRACT THE INFORMATION THAT I NEED. IT IS NOT JUST A SIMPLE > DECONVOLUTION, because frequency impulse response of the vessel varies > with the position of the impact!! > > > 3) checked that there are no dead zones at your chosen sensor position > so > you get a nice clear ring for every particle impact no matter what the > position. > > I have already did it. No dead zones! > > 4) Ignored background noise for now. > 5) Treat everything as elastic so superposition applies. > > Ok, thanks. > > > > Answers: > > a) want a count of the number of impacts - don't care about position or > > magnitude of the impact - not worried about the exact time that it > happened > at? > > YES, I want to know roughly how many particles are hitting my vessel > (10000, 20000..etc..) > > b) need a good estimate of time impact happened? > I don't mind about the position or the time happened. > > c) desire an estimate of magnitude of impact? > In a way, yes because the magnide is directly correlated to the number > of impacts. > > d) hope for an estimate of position of impact? > No positions, only concetration information. > > Sorry, my mistake, I meant TRANSFER FUNCTION as the frequency impulse > response. So, a sort of frequency domain description of how the vessel > rings when hit by a particle. > > > AS you understood, I have a lot of impacts all the time and > really, eventually,I would like to know how accurately you can > estimate > energy transfer to the vessel per unit time. > > I can simulate 10000 particles hitting the wall in 800 usec. I must > find a DSP way to estimate that. The spectra obtained considering the > pressure variation of the vessel (tranducer surface) it is between 10 > and 150 Khz. > > Can somebody suggest me a way to estimate the vessel characteristics > and extract the number of particle information? > > Thanks a lot. > I would like just a direction, like : use blind deconvolution or > something else.
O.K. - thanks for the clarification Manuel - I can't see that you need to deconvolve anything at all at the moment then. You have a simulation that should be capable of giving you the total acoustic energy coupled into your transducer for a particle impact, the amount of energy will vary with position but you know the distribution of your particles on average so you know what fraction couples in X energy what fraction couples in Y energy etc. The distribution of impact positions doesn't vary with pressure (I'm assuming) just the total number of impacts per second; but the system is, to a first aproximation, elastic so having a large number of impacts per second doesn't change the spectrum generated by each impact and there's no non-linear coupling mechanism to create new frequencies, you should be able to calibrate your simulation such that total power out of the transducer is proportional to <impact energy>*<number of impacts per second>. If you have a secondary source of information that allows you to determine the distribution of impact energies (like the temperature inside your vessel) then you can get an average value of impacts per second too can't you? If so then the only DSP you need is a moving averager, if not then there's some subtle point that I still don't understand : maybe increasing pressure shifts the resonant frequencies further outside your transducers bandwidth so reducing the average fraction of impact energy coupled into the transducer? Certainly your transducer should be operating in its linear range. Please can you describe why the above average power scheme won't work? I don't know if you've tried cross-posting to alt.sci.physics.acoustics ... it may well be worthwhile. Best of Luck - Mike