Sample Size to get Statistically Valid Data

[farasm5]

We are looking for stabilization time during the leak test to check the amount of decay. (Pressure is the same for all stabilization times.)

According to some unofficial testing, 4 stabilization times (10second, 15s, 20s & 30s) were selected and 5 times were tested per stabilization time. It was confirmed that the amount of decay is smaller when the stabilization times were increased and the variations of each tests (5 times per stabilization time) were similar.

Stabilization time 15s looks good to us. But only 5 times were tested for this time (15s).
My question is how many times we have to test with 15s to get statistically valid data for the amount of decay?

I look forward to hearing from you.

Thanks.

Peter Kim

 

[Steve Prevette]

It depends. In your original scenario, did you run the data through ANOVA and find out the the decay rates were significantly different between the stabilization times? This would give you some confidence you are on the correct track.

When you say you want to get "statistically valid data for the amount of decay" what does that mean to you? That indeed 15 seconds is the best stabilization time? That you want to know what the decay rate is within a certain error band? That you want to declare this process to be "in control" (that is, stable and predictable)? You will get different answers for sample size depending upon what your need is.

 

[dgriffith]

We are looking for stabilization time during the leak test to check the amount of decay. (Pressure is the same for all stabilization times.)

According to some unofficial testing, 4 stabilization times (10second, 15s, 20s & 30s) were selected and 5 times were tested per stabilization time. It was confirmed that the amount of decay is smaller when the stabilization times were increased and the variations of each tests (5 times per stabilization time) were similar.

Stabilization time 15s looks good to us. But only 5 times were tested for this time (15s).
My question is how many times we have to test with 15s to get statistically valid data for the amount of decay?

I look forward to hearing from you.

Thanks.

Peter Kim

*Forward: At what point will you decide there is no more stabilization needed and how many iterations will you need to find the pressure value that best represents the remaining stable pressure in the system? I recommend a visual approach first, rather than jumping in with statistics and then wash rinse repeat. Data visualization will tell you more in one iteration. Done properly, you can record the results and perform analysis on them while plotting. Monitor the pressure with a good transducer and plot the results in real time--about one data point per minute to start--and you will be able to see how much stabilization time you need.*

Perhaps you already know that abiabatic's are the culprit here. The apparent gain or loss of pressure due to temperature change within the system (heat of compression) without external heat added or removed. So, yes, the longer the stabilization time, the less impact it will have on any apparent leak.

The volume of the system may determine how long the stabilization need be. So will the delta pressure and delta time. So will the precision of the measurement--do you care about ten-thousandths or just tenths.

For a small system of less than a ft.^3, our wait time is 5 minutes, then leak check timing of either 1 or 5 minutes. This is based on two things--manufacturers leak check procedure, and visual observation--you can actually see the displayed pressure change its rate during the wait time. I would have to know more about your system to have any opinion other than 15 seconds is too short.

But we have never tried to statistically characterize ours, perhaps because we have the luxury of waiting until we're satisfied, with no time constraint. I would be interested in your findings.
I would add a reproducibility component to your testing. Perhaps repeating a few tests after dis-assembly and re-assembly of your plumbing connections.

 

[farasm5]

Hi, Steve & Dgriffith

I appreciate for your response. Let me explain in details.
We did not run through ANOVA. The product we are testing has volume, 8? (L) x 6?(W) X ?3?(H). It is not big.

According to the collected data ( 5 times per stabilization time), when the stabilization time was 15s, the amount of decay was max. 0.023psi, min. 0.018 psi, average 0.02psi and standard deviation 0.0019.

For example, I will set up a 0.035 psi as a trigger point. After the stabilization time (15 s), if the leak tester detects the amount of decay is greater than 0.035psi within certain time, the tested product will be rejected.

My concern is that alarm is falsely triggered not because of a leak but because the leak tester is not stabilized within 15 seconds yet.
I want to be confident that the amount of decay doesn?t exceed 0.035 psi after 15 stabilization time unless there is a leak. In order for me to be confident for this, how many times I have to test the amount of decay with 15s stabilization time?

Thanks.
Peter

 

[Steve Prevette]

For example, I will set up a 0.035 psi as a trigger point. After the stabilization time (15 s), if the leak tester detects the amount of decay is greater than 0.035psi within certain time, the tested product will be rejected.

My concern is that alarm is falsely triggered not because of a leak but because the leak tester is not stabilized within 15 seconds yet.
I want to be confident that the amount of decay doesn’t exceed 0.035 psi after 15 stabilization time unless there is a leak. In order for me to be confident for this, how many times I have to test the amount of decay with 15s stabilization time?

Thanks.
Peter

We are still faced with what is your definition of "stabilized". If you can express that as less than a certain rate of pressure change per time interval, and if you indeed know your standard deviation then I'd probably define it as measuring a rate of change less than that value, taking into account the standard deviation of the measurement error. This will relate to sample size in that the standard deviation of the average leakage rate can be better estimated the more samples you have - the standard deviation of the average is the standard deviation of the individuals divided by the square root of the sample size.

 

[dgriffith]

Steve, my post explained a little of it as far as 'stabilized' goes. Pressure change in a closed system, where one would perform a leak test (like the air conditioning system on your car-it's not allowed to leak into the environment), the measureable pressure is influenced by adiabatic effects. As the temperature of the system returns to its quiescent point, the pressure also changes, and that can be seen as a system leak, which it is not. Stabilization is the point at which adiabatic effects are no longer present, and the pressure only changes (ideally) if the system leaks. He's trying to find that point and prove it statistically, I believe.

 

[Steve Prevette]

However, without a numerical explanation of what "stabilized" means, there is no way to numerically calculate or justify a specific sample size ahead of time.

I think what you are going to be faced with is collecting data in small increments, analyze the cumulative data to that point and subjectively decide "is that good enough". If the answer is no, collect another set of data. Basically you will have to keep going until you can definitively say - the test passes or the test fails.