Whats is the ideal number os samples for a test? 1

[Henry26]

Hello everyone,

I'm performing a FEA calculation in a cast iron component, in order to validate this calculation I've looking for the ideal number os samples to perform the test. I saw some standard for destructive testing but they are for standardized test bodies, someone know how to find this amount?

Thank you!
Henry

 

[dik]

Strength of castings are so dependent on the shape... you might consider casting one of your objects and do whatever tests are required.

Dik

 

[racookpe1978]

There is NO "ideal" number of samples! NEVER can be an ideal number actually.

Do you understand that sampling is a statistical method? IF the testing you decide to do to analyze future failures is actually "statistical" in nature it can only - at best! - provide a statistically correct way to estimate the probability of some future "identically statistically identical" object of failing in real-world use the same way as it was tested.

Now, the test of even only ONE object may show absolutely that the design will always fail (too weak, too flexible, the wrong shape, etc.) Or it may show that the object is absolutely strong enough and big enough, but absolutely too heavy to fly, or tall to lift, or too wide to fit in a doorway or in the shipping container.

But, if you test 5000, the 5001 might break early in service life anyway because the customer didn't "behave like the test" and "didn't install it per procedure" in the "expected design conditions" ...

The nbr of test pieces requires an analysis of the likely failure mechanism, the amount of variance between the fabricated parts (castings are notorious for random unpredictable failures internally due to voids, cracks, stress risers, holes, vents, cavities, and failed molds!) So, what is your standard deviation of each part? What is the expected failure mechanism? How likely is your test going to discover the expected flaw in the product? (If a 0.2 mm pinhole causes a fatal leak, will your test actually find that size pinhole?) How expensive is the test to run? How long must you run the test? How expensive is the replacement part - if it is tested to destruction? What is the cost of failing the test? What is the cost of NOT failing the test (getting a false "positive" result, and then building the bad product based on bad test results? How many parts will you build, and will any test actually find the potential flaws?

A rocket engine and rocket cost 100,000,000.00. The test costs an additional 50,000,000.00 How many times do you test the irreplaceable rocket engine before launching a 750,000,000.00 dollar satellite into orbit? A submarine and its nuclear reactor costs 2.8 billion and has 120 people on board, takes 6 years to make. How many tests do you do before letting it submerge the first, 2nd, and 3rd time?

 

[racookpe1978]

Your login_id says you are "automotive" - If so, your question is easier because your industry (your buyer) purchases a very large number of identical parts and has explicit ISO/ASTM rules for the reliability they demand. The parts too are a bit simpler than a 300 MegWatt gas turbine rotor and generator system. Based on those rules for the number of parts you are fabricating (designing), the QA manuals have tables giving the nbr of tests required to establish a reliability.

But since you have given us no clue about the reliability demanded (the expected accuracy of the prediction of failed parts missed by the test and the number of total parts built), we cannot tell you the number of parts needing to be tested.

 

[CELinOttawa]

Bridge codes typically approach this as a "with this sample size, here are your safety factors" type of approach. That might be the best interpretation available. Effectively you would reduce the stress in your FEA you would allow based upon the number of real world tests you were going to employ. The statistical methods are simply that: Raw statistics, which you could apply to any material.

The NZ manual is available as a free download. See section 7.3.6 and Table 7.2 for what I mean...

https://www.nzta.govt.nz/assets/resources/bridge-manual/docs/Bridge-manual-complete-v3.2.pdf

There is ALWAYS going to be Engineering judgement to apply a la Racookpe1978 (above), but it is nice to be able to apply something specific to get into a reasonable ballpark.

 

[JeanLucPicard]

Shouldn't it be specified by the norm?

For most fabricated structural elements (eg masonry bricks) - min sample size is 3 samples. This is due to the fact that the element is made within factories in controlled environment.



Live long and prosper!

 

[IRstuff]

There are a whole slew of military standards and handbooks on sampling plans; much depends on how much confidence you want, and how much time and money you can spend. See:

MIL-HDBK-108 SAMPLING PROCEDURES AND TABLES FOR LIFE AND RELIABILITY TESTING
MIL-HDBK-53-lA MILITARY HANDBOOK GUIDE FOR ATTRIBUTE LOT SAMPLING INSPECTION AND MIL-STD-105
MIL-STD-105E Sampling Procedures and Tables for Inspection by Attributes
MIL-STD-414 Sampling Procedures And Tables For Inspection Buy Variables For Percent Defective
MIL-STD-690C FAILURE RATE SAMPLING PLANS AND PROCEDURES

Obviously, there's Wikipedia, etc.:

https://en.wikipedia.org/wiki/Lot_quality_assurance_sampling

https://en.wikipedia.org/wiki/Estimation_statistics

https://en.wikipedia.org/wiki/Statistical_hypothesis_testing

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert!

https://www.youtube.com/watch?v=BKorP55Aqvg

faq731-376 forum1529 Entire Forum list

http://www.eng-tips.com/forumlist.cfm

 

[Henry26]

Hello guys,

Thank you ever so much for all your answers. My test will not reach the failure, I just wanna validate the Neuber's rule, in other words, the test must give me the same value of the FEA, I don't need to kknow the probability of failure because I will work in linear region of the material. After some researches I found the "Paired t Test", until now this is the better method that I've found.

Thank you,
Henry.

 

[rb1957]

I think it's usual to test each batch, as process control. I would be surprised if your test results will give you the same value as your FEA ... near enough is the most you can expect.

another day in paradise, or is paradise one day closer ?

 

[Henry26]

#rb1957
I agree with you, because this I'm thinking to use "Paired t Test" that work with the average of differences.