Road Input Generation Using PSD's for Model

[Slowzuki]

Good afternoon, I am modeling a solid axle, multilink suspension and I am attempting to generate an input to the model.

I'm using formula 5.1 in Gillespie's Fundamentals of Vehicle Dynamics along with a random nuber generator as he descibes. The problem is he makes no mention of the range limits on the random number so the formula is generating huge high frequency bumps every so often.

Anyone know of any other methods to generate input? I may move back to constant frequency sine waves with amplitudes matching the PSD distribution.
Ken

 

[GregLocock]

The problem is that the random numbers you generate using a random number generator are NOT suitable for driving a rig. The reason is that they are not gaussian, that is, the cumulative frequency plot of amplitde is not a noraml distribution, it is a straight line.

You have a few choices.

1) Generate pseudo Gaussian white noise by adding a whole bunch of random numbers together. ie RG=1/5*(RND()+RND()+RND()+RND()+RND())

However, this still may give problems.

(2)The best way of generating random numbers that are correctly distributed is to set up a spectrum of the desired magnitude, and with random phase for each frequency. Then do an inverse fourier transform of this. The resulting time series will contain the amplitudes and frequencies you expect, and will have the correct statistical properties as well.

(3) an alternative approach is to filter the time series before you use it t drive your model. I can't think why this would be better or easier than (2)

It is surprising how many people don't know about the dangers of using random numbers to drive test rigs.



Cheers

Greg Locock

 

[IRstuff]

While random numbers are indeed interesting, you'll still need to run some semi-real profiles into the suspension.

For military vehicles, there are prescribed courses that are used to validate performance and lifetime. Munson, Perryman, Belgian block are specific road configurations designed as particularly stressing on vehicles. TTFN

 

[GregLocock]

True, but the road spectrum is a surprisingly good approximation for straight line running with no events. It would however be a very bad choice for a fatigue test.

Cheers

Greg Locock

 

[Slowzuki]

Thanks for the info, the random number generator creates the wavenumber in cycles per feet to feed into the equation. Obviously this number needs to be a Gausian function but I have no idea of the mean or std dist of it.

In regards to item 2, I will take a look at doing that in Matlab tonight. I should note I am not driving a test rig but rather a computer model in Simulink/SimMechanics.

Thank-you for the advise, I am new to the forum. Much more helpful than the Mathworks forum.
Ken

 

[GregLocock]

If your model is any good it will behave like a test rig! Seriously.

In practice where I work the line between physical test rigs and analytical test rigs is getting fuzzier by the month. We are rapidly moving towards hardware in the loop for my area (ADAMS models of suspensions, mostly) and the Test guys already use ADAMS models of their test rigs to sort drive signals out.
Cheers

Greg Locock

 

[Slowzuki]

I hope the model will behave as a test rig since there aren't any for about a 1500 km radius of my house! I was concerned you might think I would attempt to drive one with my not so good signal!

I was trying for Adams but all I could get was UG NX with the Adams module. Not useful for coupled or non-linear elements.

With regards to the method (2) above, I'm looking at that now, are you suggesting setting up several key fixed frequencies across the spectrum I would expect (~0.2-50 hz)?

I have very limiting signal processing experience/knowledge.

 

[GregLocock]

I don't know what format data you are using. Here it is in pseudo code, for 0.1 /ft intervals, for eqn 5.1

dim mag(2048), phase(2048)
G0=1.25*10^5
v0=0.05
for ifreq=1 to 2047
v=ifreq/10.0
mag(ifreq)=G0*(1+(v0/v)^2))/(2*pi*v)^2
phase(ifreq)=RND(1)*2*pi
next ifreq

Then you will probably need to turn that into a real and imaginary array, and then do the IFFT



Cheers

Greg Locock
Cheers

Greg Locock

 

[mloew]

Slowzuki,

I recently attended this technical paper at the SAE 2003 World Congress & Exhibition:

2003-01-0218

http://www.sae.org/servlets/productDetail?PROD_TYP=PAPER&PROD_CD=2003-01-0218

Terrain Roughness Standards for Mobility and Ultra- Reliability Prediction
By David J. Gorsich, David Gunter, Robert Karlsen, Milton Chaika
US Army Tank Automotive and Armaments Command (TACOM)

There was some useful information presented at this presentation.


Best regards,

Matthew Ian Loew

Please see FAQ731-376 for tips on how to make the best use of Eng-Tips Fora.