Rainflow anaysis subsampling

[SupacatJim]

Hi,
I am running a rainflow analysis on some data sampled on a suspension component to help predict the life of the structure. The results from the rainflow analysis is scaled to give component stress (linear relationship assumed) and then using Miner's rule the life of the component is determined. I am however getting some strange results.
The data recorded was sampled along with other channels at a frequency of 5kHz. As this is a much higher frequency than required, I sub-sampled the data to 500Hz (taking every 10th point). I have run both these data through the rainflow analysis and Miner's rule, and get results showing that the sub-sampled data gives a component life ten times worse than the full data set.

Has anyone any possible explainations for this? I could understand the other way as the high frequency noise could have a large effect, but this has got me baffled. Obviously it has large implications for the structure!

Many Thanks

Jim

 

[rb1957]

how much of the intermediate steps can you see ?

do you feed the data in at one end of the balck box, and out pops the life ?

or can you see the rainflow spectrum ? if you can, you can see the difference between the two spectra.

on the face of it, the higher frequency sampled spectrum includes more turning points, producing smaller cycles. that would indicate that your signal (input) hasa higher frequency than you expect. could you have data issues too ? (like alaising?)

 

[SupacatJim]

I understand what you say regarding the sample rate vs event time etc. We work with a standard of 11ms (I've no idea why 11 and not 10) minimum duration of a load for damage to occur. This would suggest a minimum sample rate of 100hz and hence why the 5kHz was reduced to 500Hz. I admit a filtration process would have been better, but I have not carried that out yet and am just trying to understand the results I have at present!
RB - I cannot see any of the intermediate steps from the sampled input data to the rainflow output (i.e. the range and number of cycles for all found cycles), but can see all the other steps. I have a program to sort the cycles into 100 range bins, and then an excel spreadsheet to carry out the Miner's rule calculation.
Having looked into the results more now, the main difference between the raw and sub-sampled data once sorted in bins is an increase in the maximum cycle (approx 50% higher) and a huge increase (10 times) in the quantity of cycles in the first bin.

Thinking and talking about it whilst writing this, I think I can see how missing data can cause larger cycles, but did not think that this would outweigh the overall decrease in total number of cycles (although I can see that with fatigue it is a log relationship)

Overall I guess the things that I want to find out are: a) Is the characteristic I have of the sub-sampling making the fatigue life worse a rule with rainflow analysis, or just something that needs to be checked for; b) Is a higher sampling rate always better and c)which data should I use for my life predictions.

Many thanks so far.

Cheers

Jim