Evaluation of impact vibration with level vs time integration time

[Hefaistos]

While measuring sound impact noise about 30dB(A) I want to evaluate the produced impact acceleration. The question is how do this the simplest way.

I've used a integration time of 125ms but 65ms does show a 40% higher level. Since the comparison is just between object measured at the same time under about the same conditions it's not of very high importance.

The impact (measurable above the background) is about 10ms but analysing with integration at 35ms (impulse) gives "funny" results. 10ms actually gives a nice graph but very far from the 125ms result.

Does anyone have an idear on how to best analyse impact sound and vibration to make it easy to interpret and to visualize? And also how to think while evaluating sound and vibration levels at the same time. Different integration time? as short as possible?

In the hope that someone has come up with a good method before me.

//Jonas Löfhede

 

[electricpete]

I assume the "level" plot is acceleration integrated over a sliding window.

I'd say you must have a dc bias on your input acceleration signal.

When you use a longer integration signal, you pick up more of the dc bias in the input.

I don't know if this is expected for your accel... I recall reading on the forum that some respond to gravity.

Whether it is expected or not, you can remove it by filtering the acceleration signal prior to integrating.

You might try to implement the following HP filter: H(s) = s / (s-Tau)

1 - It looks like 1 at high frequencies.
2 - It will not chop your peak as long as Tau is much larger than the "rise time" of your impact.
3 - It has an added benefit that the peak will in fact decay away to zero after the impact, with time constant Tau. That's somewhat similar to your current behavior except endpoint will be zero instead of negative number and decrease will be exponential instead of what I think you have is linear (?).

=====================================
(2B)+(2B)' ?

 

[electricpete]

Correction in [highlight]highlight[/highlight]
H(s) = s / (s [highlight]-[/highlight] Tau)
should've been
H(s) = s / (s [highlight]+[/highlight] Tau)



=====================================
(2B)+(2B)' ?

 

[electricpete]

Also regarding #3, the filtered signal would decrease exponentially to zero after the peak.

The integrated signal would rise higher than the true peak of the input signal. That's probably not a good choice of filter.

I gather you are really interested in the area under the initial impact acceleration. There may be a better filter to do that without distorting your result, but I can't think of one off-hand. Maybe you can combine peak detector with zero crossing detector. Look for peaks above a certain threshhold, record their time-of-occurence and magnitude. Then estimate area as 1/2 of the peak times the time between the zero crossing before and the zero crossing after the peak. Or for that matter, buffer the signal and integrate between those two zeros. Just thinking out loud. There may be other ways.



=====================================
(2B)+(2B)' ?

 

[GregLocock]

The first step in any NVH test is to produce a repeatable test. The second step is to find a measurement that correlates with your subjective assessment. That's pretty much where you are now. In this case I'd be looking at some sort of RMS level in 20 ms, or even working in the frequency domain. Failing that yes use the impulse repsonse, that's what it is there for.





Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?