How Many Test samples would be acceptable for Company Allowable Data used on STC projects

[bigbertha912]

I am currently working as the only stress engineer in a large organization and this is the first time in many years that I am writing test plans for latches/hinges/inserts into the typical honeycomb panels we will be installing them into. This will become part of the analysis handbook that is currently being created and will be used on many STC projects going forward, so the tests need to be good and adequate for FAA approval.

My question is, what is the minimum limit required to get an adequate test that the FAA would be okay with. I've done test with a sampling of 20 specimens per installation before, but that was a very long time ago. Is that still adequate? Obviously more is better, but i am also constrained by budget and time.

Advice is appreciated.

 

[Sparweb]

One way to pose the question you are trying to answer with your specimens is "how much variability can we accept?" There is an actual strength value that you may think these specimens can achieve, but there is also a value that you can accept for design purposes. If the two values are really close, then you will have to test many samples to prove that 95% of the time your insert will have the required strength. OTOH, if you need a design value just 33% of the possible strength of the insert, then you may only need to test a few specimens to have that same confidence. Testing more samples doesn't make the specimens stronger - they just look stronger in the end because the higher confidence in the strength value increases what you're allowed to use for design. Statistics - it can be a game of words.

You will also have to test in different load directions: tension, shear, combined tension+shear and so on.

If you have combinations of numerous sizes of insert and/or panel to test, you may be able to ask for an alleviation in the following way: Say you have 3 inserts of sizes A, B, and C. You also have 2 panel types X and Z, which you use in 2 thicknesses, and expect 8 specimens for each test configuration. Do you have to do all 96 combinations? Maybe not. You may be able to interpolate the properties of B if it's in between sizes A and C. You may be able to select only panel type X for tests if its core is much lighter than panel type Z, maybe doing just a few tests in panel Z to show that it makes stronger insert joints. Same thinking with panel thicknesses, and so on. You may also consider testing a larger number of insert A and then showing that inserts B and C will have the same confidence even if you don't test as many of them...

All fodder for discussion with your FAA engineering authority BEFORE you plan these tests.

Have you looked at Chapter 9 of AR-MMPDS?


STF

 

[bigbertha912]

SparWeb - Thank you for your response and the explanation. I have looked at Chapter 9 of the MMPDS a little, which is why i was wondering about sample size. some of the charts express speciman numbers of 10 to 10,000. I also read somewhere that in order to make 95% or 99% accuracy that you would need a minimum of 200 samples.
But this company definitely falls under the category of 33% of the possible strength being adequate for what we do, that you mention.

I did not think about interpolation between specimans. I will look into this as well.

Thank you for taking the time to respond.

 

[SWComposites]

Look at CMH-17 Volume 1.

For a point design number (one configuration, temperature, etc) typically 3 material batches * 6 replucates = 18 samples is enough.

 

[stressguyMike]

Assume you are looking at allowables for the inserts and panels.

I attached brief FAA guidance on design values.

Need to account for materials from different lots (panels), as well as installation by different mechanics (inserts).

For inserts, the typical batch is 15, with samples made from 3 different lots and the inserts installed by 3 different mechanics.

Also, CMH-17 has a spreadsheet that quickly and easily performs the necessary statistical analysis.

If you don't have that spreadsheet, it can be purchased for $500. Also, copies are floating around and you might be able to get a hold of one.

Good luck.