Matching Time Step

02-Analytical capabilities
Post Reply
TroyH
Posts: 3
Joined: 01 Nov 2016, 14:32

Matching Time Step

Post by TroyH »

Is it possible to complete the spectral matching against a user input time sequence for the spectrum? I have come across the change in time step under Settings>Response Spectra, but I am still finding that there is a lot of variability in the matched spectrum when I compare it to the spectrum generated in other programs for the same time history record. Ideally I would like to add in the US NRC set where every period differs by 10% from then previous. This appears to provide a more robust match at the lower periods. Is there a way that I can do this?

Thanks
User avatar
seismosoft
Posts: 1184
Joined: 06 Jul 2007, 04:55

Re: Matching Time Step

Post by seismosoft »

Have you tried to run the program with smaller tolerance value? he default value 0.3 is relatively large, so that to achieve better convergence.
Further, note before the analysis the the min. and max. periods are specified, where the matching is applied (outside these limits no maching is expected).
Can you please send us some references for the US NRC set that you specified? Finally, which are the other programs that have smaller variability from SeismoMatch? It might be interesting to check the algorithms that they are using.
Best Regards,
Seismosoft Support
TroyH
Posts: 3
Joined: 01 Nov 2016, 14:32

Re: Matching Time Step

Post by TroyH »

Hi, thanks for the speedy reply.
I have tried the tighter tolerance noted below and still get the same variability. Also, as you note, I have only been comparing this for the range between the max and min periods that I have specified for matching.
The US NRC (United States Nuclear Regulatory Commission) time set that I was referring to is a 10% increase in the period starting from 0.010464697253675 sec and finishing at 10.0 sec (73 data points, 74 including 0). This would give a 'linear' increase on a log10 period scale. This is because the bandwidth of the response is a ratio of the forcing frequency to the natural frequency, hence fewer points are required at the long periods (low frequencies) to obtain a smooth spectral match but significantly more points are required at the high frequencies to obtain a smooth match.

The other program that I have is an excel spreadsheet with a subroutine for generating the response spectra from a time series. This subroutine is called RESPCT from Bjorn Sweinsson that is based on original Caltech code and has been well validated. This subroutine identifies a lot of variability at the higher frequencies.
Regards,
Troy
User avatar
seismosoft
Posts: 1184
Joined: 06 Jul 2007, 04:55

Re: Matching Time Step

Post by seismosoft »

Can it be that for the given record and spectrum, convergence is not achieved within the specified number of itenrations?

Seismosoft Support
TroyH
Posts: 3
Joined: 01 Nov 2016, 14:32

Re: Matching Time Step

Post by TroyH »

I had a bit of a play with that and used a record that was already a close match to the target spectrum so that I could use a tight tolerance. I think I had a tolerance of 0.15 and it would converge in ~10 iterations over a relatively small period range. It just appears that it is not matching very closely between the time steps. This is picked up when I regenerate the matched spectra with a finer time step in my excel sheet.
huffte
Posts: 978
Joined: 22 Jul 2011, 10:19
Location: Cookeville, Tennessee, USA
Contact:

Re: Matching Time Step

Post by huffte »

I wonder if increasing the maximum number of iterations, wavelets, etc. might do the trick? I have had success with this strategy. Also, how many records will you be using? Often, I perform the matching and while any single record might not be a close fit, the mean of all records used ends up being a very close fit. This can happen even when some of the records don't converge using the default settings of SeismoMatch. And remember, most target spectra represent a sort of envelope of multiple earthquake scenarios, so matching any single record so tightly to such a target can produce unrealistic ground motions. Examine not only the acceleration histories of the matched records, but also the integrated velocity and (especially) displacement histories to see if they are realistic. Best of luck.
Tim Huff
Rana
Posts: 6
Joined: 30 Mar 2012, 08:49

Re: Matching Time Step

Post by Rana »

It seems that you are trying to achieve spectrum matching over a big range of period in one pass. This shall destroy the nonstationary characteristics of accelerogram even if you get the match. You need to imperatively increase the range in multiple sweeps. In addition, you need to select a motion which is somewhat proximate to your target based on such objective criteria (may be consistency in terms of M-R-epsilon). Arbitrary selection of motion may often fail to achieve a match.

Regards
Rana
Post Reply

Return to “02-Analytical capabilities”