
Re: Increasing number of decimal places on Frequency Reponse file
Jared Conway Oct 8, 2014 7:36 PM (in response to Graeme Hyson)i think this came up recently and it was that excel was reducing the number of digits by default rather than the output missing the info
did you open the file directly?
how many digits are you seeing?

Re: Increasing number of decimal places on Frequency Reponse file
Graeme Hyson Oct 9, 2014 3:15 AM (in response to Graeme Hyson)I can see five digits, the number of decimal places depends on how large the value is, it displays 35.568 or 746.77 or 1001.5 for example.
To save the data I am probing the results plot at the relevant point and then clicking the graphclock icon to get the frequency response plot. I then do 'saveas' on the response graph and saving as a .csv. When I view it in Excel I simply open the .csv file and values are already in columns, that in itself is slightly odd because I expected a .csv file to just have the values with commas  as the name suggests!
Just for clarity I need the extra DP on the frequency values, i.e. the Xaxis numbers in the frequency response plot.

Re: Increasing number of decimal places on Frequency Reponse file
Jared Conway Oct 10, 2014 6:14 PM (in response to Graeme Hyson)what happens when you open the csv in notepad? maybe post here


Re: Increasing number of decimal places on Frequency Reponse file
Graeme Hyson Oct 13, 2014 3:20 AM (in response to Graeme Hyson)Opening it in notepad gives me the same result  I get 5 digits. Unfortunately the data and model are commercially sensitive.
I've managed to find a workaround of sorts. The SW help file tells you how it calculates the spread of points between the natural frequencies at which it analyses, therefore I can put the same calculation into Excel and 'fill in the blanks' so to speak. Fortunately SW tells you the natural frequencies (via results>list resonant frequencies) to a large number of DP, so you can use those as a starting point. However this is obviously somewhat more laborious than just getting more DP from SolidWorks!

Re: Increasing number of decimal places on Frequency Reponse file
Mike Pogue Oct 13, 2014 11:57 AM (in response to Graeme Hyson)Graeme,
These modal calculations are just not that accurate. SolidWorks is giving you 1/10000 (5 significant figures), which is probably 1000 times more accurate than the actual result. If your conclusions are sensitive to the fifth decimal place of this input, you should not trust them.


Re: Increasing number of decimal places on Frequency Reponse file
Graeme Hyson Oct 13, 2014 12:18 PM (in response to Graeme Hyson)Mike, perhaps if I explain what I'm trying to do it'll make more sense...perhaps there's even a way to achieve my desired end result without this DP issue cropping up.
I'm performing random vibration analyses of components. The input is specified as a PSD, as are the outputs. We're interested in estimating the worst case displacement response vs. frequency. This is not something you can read directly from the PSD output by SolidWorks, it's a random output after all.
In order to arrive at a worst case displacement we take the PSD data and use them to calculate RMS displacement, then multiply the RMS by a factor (normally 3) to work out the biggest absolute displacement you're likely to see within the random response. PSD is related to RMS via bandwidth, so I need to know bandwidth. To work out bandwidth I'm looking at the interval between each frequency analysed by SW. Say for example it performs one analysis at 10Hz, one at 15Hz and one at 20 Hz, I assume SW used a bandwidth of 5Hz around the 15Hz result, i,e, 12.5Hz to 17.5Hz, when calculating the PSD value, thus I can work backwards from the output PSD value to get the associated RMS value.
The trouble I have is that SW groups the analysis frequencies more tightly around the resonances. That means when I'm trying to work out bandwidth I'm often getting a width of zero or otherwise losing way too much accuracy because of the loss of DPs. I know SW is capable of using more DP because it must be using higher DPs in its own calculations in order to get the nice smooth frequency vs. response plots it generates....and I'm simply trying to 'reverse' that calculation.
I've tried messing with the number of analysis frequencies and the biasing parameters to change the number of analysis points and/or their spcings, but this results in either prohibitively long solve times (due to the vast number of points) or 'missing' the peaks in the response because the points are too spread out.
If there is a way to output RMS displacement vs. Frequency directly then I can skip this whole process!

Re: Increasing number of decimal places on Frequency Reponse file
Mike Pogue Oct 15, 2014 9:53 AM (in response to Graeme Hyson)I've used exactly the technique I think you are using. I'm not exactly sure about the bandwidth you are looking for, so let me outline what I did to see where we are different. I don't have this version of simulation at my new job, so I'm going from memory/google search on result options. SW should plot PSD stress, but it doesn't. (yes it does)
SO,
If you plot the displacement PSD at the point of interest, then integrate over the entire frequency domain, you'll get RMS displacement. Multiply that by 3 and you'll get 3 sigma displacement, which is typically where it's chopped. Pedants like to point out that in "reality" this should technically go to infinity, but that is utter nonsense. Its the statistical model that goes to infinity, nothing in physics ever does. From the 3 sigma displacement, you have to move to handcalcs to convert it to stress, because, as I mentioned, I don't think SW is going to do that conversion for you. This requires engineering judgment to pick the points to plot, and to convert displacement samples to stress.
I think this is exactly what you are looking to do. I think the problem is the output options you've chosen, unless I'm misunderstanding. You should get the PSD displacement, not the displacement at some frequency, which you've pointed out is not meaningful. I think you need to get the whole PSD and integrate over the whole domain. Sorry if I'm adding confusion.

Re: Increasing number of decimal places on Frequency Reponse file
Jared Conway Oct 14, 2014 1:40 AM (in response to Graeme Hyson)I Don't have the software with me right now but the help says rms is an option. 2012 SOLIDWORKS Help  Random Vibration Analysis
the help also explains how the points are chosen.

Re: Increasing number of decimal places on Frequency Reponse file
Mike Pogue Oct 14, 2014 9:28 AM (in response to Jared Conway)That will save some timeconsuming postprocessing. On the other hand, if you do a study with broadband white noise and output the displacement PSD, you can apply multiple input PSDs to one solution. You couldn't do that with RMS outputs. That would save some timeconsuming calculations.
It would still be better if Simulation calculated RMS stress or PSD stress (unnamed other packages do). Once Simulation has the displacement, it should be straight forward to back out stress. It's a pain to do it by hand, and the results are cheezy. I'm kind of keeping this thread alive in the hope that somebody has a better way than I've suggested.

Re: Increasing number of decimal places on Frequency Reponse file
Jared Conway Oct 14, 2014 8:15 PM (in response to Mike Pogue)am i missing something? i haven't looked in the software but the help says those options both exist. maybe a version/service pack thing? 2014 SOLIDWORKS Help  Random Vibration Analysis
After running the study, you can plot rootmeansquare (RMS) values, or psd results of stresses, displacements, velocities, etc. at a specific frequency or graph results at specific locations versus frequency values.
Modal, Rayleigh, and Composite modal damping options are available for this type of analysis. See Damping Effects.

Re: Increasing number of decimal places on Frequency Reponse file
Graeme Hyson Oct 15, 2014 7:14 AM (in response to Jared Conway)Jared....I read the same thing in the help file, but I can't work out how to plot RMS against Frequency. I just cant' find the setting/button in SW to produce it!
This http://www.vibrationdata.com/tutorials2/psd.pdf link pretty much explains what I'm trying to do.
Table 2 essentially summaries the process I'm adopting. SW analyses response around discrete frequency points, these points are the "band center frequencies" in Table 2. The Bandwidths I'm referring to above relate to the info in the first and fifth columns of Table 2. I know the PSD values (from SW) and a know the band centre frequencies (also from SW) thus I should be able to work out RMS in any given band.
SW outputs a CSV file which lists discrete frequencies and a PSD at each. You can't produce a PSD at a single frequency, is doesn't make any mathematical sense!, so I have to assume SW is actually telling my PSD in a group of bands, and that those bands are centred around the discrete frequencies listed by SW. Where I ultimately end up is with a bar chart which says betweem 1020Hz RMS is XXX, between 2030Hz RMS is YYY etc. Thus I effectively have a frequency vs. RMS response plot, albeit one wich is broken down into finite bands rather than a continuous line. That bit I'm OK with, this is FEA after all.
Where I'm tripping up, as per my previous posts, is working out the bandwidths. In some cases SW analyses between say 1034.4778Hz and 1034.5403Hz, thus with the rounding error I get an PSD and RMS value between 1034.5Hz and 1034.5Hz...i.e. a bandwidth of "zero" due to the rounding error. I can force SW to use more widely spaced frequencies, however because I'm dealing with a lightly damped and quite springy (it's a long thin piece of carbon fibre) structure it has quite 'tight' natural frequencies, thus if I spread the analysis points out I miss the response peaks.
I'm doing this all in terms of acceleration input and displacement output by the way. I can't talk in terms of stress output because I'm dealing with an anisotropic composite part, which is another layer of complexity!

Re: Increasing number of decimal places on Frequency Reponse file
Mike Pogue Oct 15, 2014 9:43 AM (in response to Graeme Hyson)Graeme,
Solid works is not giving you discrete bands, it's giving you a curve representing PSD vs frequency. It's only computing enough points on the curve to represent it faithfully.
The most accurate way to integrate these is splines, which you can look up. If you are math phobic, you can plot the points in AutoCAD and use splines, regions and massprops to get the area.
The second most accurate is probably centered rectangles, in which case each rectangle would be centered on a point and have edges halfway between points.
The third most accurate is triangles (I think), in which case you connect all of the points with straight lines and integrate under the resulting triangles.
The worst way to integrate is forward or backward rectangles, in which case you'd have rectangles with edges on the points.

Re: Increasing number of decimal places on Frequency Reponse file
Mike Pogue Oct 15, 2014 9:48 AM (in response to Graeme Hyson)Why can't you get stress from an anisotropic part?

Re: Increasing number of decimal places on Frequency Reponse file
Jared Conway Oct 15, 2014 6:08 PM (in response to Graeme Hyson)can you post some screenshots of the outputs you are getting and also what version/service pack that you are using and a sample that shows an example of your setup?


Re: Increasing number of decimal places on Frequency Reponse file
Mike Pogue Oct 15, 2014 9:47 AM (in response to Jared Conway)I see it now. I actually had to do CtrlF because I still couldn't see it.



