Is there any detailed information available concerning the RESP command?


Yes, please see the discussion below for analyses conducted in ANSYS 11.0 SP1 and also refer to Solution 335519 "Question on generating response spectrum in ANSYS 8.1 from array parameter."

A customer sent in a model and noted that our response spectrum calculations did not match the generally expected values predicted by some other spectrum software package. I created my own model (resp.inp) of a single COMBIN14 element and a single MASS21 element. I ran a full transient with an equivalent damping ratio via BETAD (see input file for details). I used his input data (testing just the UX values), and I got much lower accelerations than what our RESP command predicted.

I passed the attached plots of our calculated spectrum response along with the corresponding acceleration plots for several different frequencies (as labeled) to our Development staff. They dug through all of this material and provided the information below. The referenced plot for the first five seconds of the transient is labeled "embedded_plot.png".

Essentially, the ANSYS RESP command is designed to work with data coming from ANSYS, which is already smooth and not raw data like was provided for this exercise. With all of the noise in the data, the spectrum results are not as expected. By smoothing the data, as is done in those other commercial codes, the spectrum results will be better. I filed Defect 67167 "POST26 RESP command should use average acceleration and not Newmark value directly" to improve the results we get out of ANSYS, but you still need to smooth the data before issuing the RESP command to get the best results possible.

Development noted that most codes use the "Ramp Invariant (Smallwood) Step Response Matching Algorithm", like is used in DADiSP. More information about DADiSP/SRS can be found at: <a target=_blank href="http://www.dadisp.com/srs.htm">http://www.dadisp.com/srs.htm</a>http://www.dadisp.com/srs.htm

Development also recommends reviewing the Digital Data Acquisition paper <a target=_blank href="http://www.ttiedu.com/articles/srs-standardization.pdf">http://www.ttiedu.com/articles/srs-standardization.pdf</a>http://www.ttiedu.com/articles/srs-standardization.pdf from the May/June, 1996 IES Journal paper entitled: "A Proposed Method to Standardize Shock Response Spectrum (SRS) Analysis" by Bill Hollowell and Strether Smith. they also reviewed the original work by David O. Smallwood at the Sandia National Laboratories entitled "An Improved Recursive Formula for Calculating Shock Response Spectra".

In summary, the RESP command will be improved in 12.0 to use average acceleration data and not the Newmark values directly, but the raw data still needs to be smoothed beforehand for the most accurate results.


Discussion from ANSYS Development Staff:
==============================

1. The /POST26 RESP command should be using the average acceleration and not the Newmark value directly. The Newmark value is known to have a lot of noise in it and for noisy input will produce noisy output. In the solver we report the average acceleration (OUTRES,A and in the inertial load computation M*A; see eqn 17-9 of the TM). Please file a defect on this and I'll make this change.

2. The input data is very noisy with high frequency content. Here, for example, is the data from 0-5 seconds:
<< OLE Object: Picture (Enhanced Metafile) >> (embedded_plot.png)

This induces the high accelerations at the high-frequency range (even when using the average acceleration values).

3. Also at high frequencies, the output equals the input. The relative displacement is on the order of 1e-4, but only 5 decimal points are given for the input!

4. These other codes, e.g. DADiSP (based on the Smallwood method), filter out this noise. I gather that this data came from a lab measurement and that is why it is so noisy. These other codes know this and know that filtering the raw data is a necessity. The ANSYS RESP command, however, was not meant for this kind of data but to instead take the output of an ANSYS transient (which will be by nature smooth) and calculate the corresponding response spectrum. You could try using EDNDSTN to filter the input data and use that then in the RESP command.

I've attached 2 Excel spreadsheets that looks closely at the data for f=1000Hz. One is for the entire time range and the other for the first 5 seconds. They take the data directly from POST26 (*with* the average acceleration change!) and plot the input displacement, the POST26-RESP-calculated velocity, the POST26-RESP-calculated acceleration and the POST26-DERIV vel and DERIV accel. The velocity matches pretty well and while the accelerations follow the same trends, the Newmark RESP values are much higher due to the induced noise that gets amplified as the transient progresses.

<< File: trans to time 5.xls >> << File: trans.xls >>





Show Form
No comments yet. Be the first to add a comment!