To call the first failure of in a daisy-chain of 10 components N10, is quite a stretch. This could be an outlier and thus totally off a proper distribution; you do not have a real value for beta; the width of the confidence band is very wide—I also really would not call it N50 either for the same reasons
Werner
-----Original Message-----
From: Nieznanski, John A - SSD <[log in to unmask]>
To: 'Werner engelmaier' <[log in to unmask]>; [log in to unmask] <[log in to unmask]>
Sent: Thu, 18 Dec 2008 6:54 pm
Subject: RE: [TN] to apply partition correction on individual daisy chains or not?
Hi Werner,
Thanks again for the explanation.
However from my last example, I was only questioning why you switched over to using N50 instead of N10 in my calculation. Was this simply
a typo, or something else?
In my simple example, once I have my first failure in a single daisy chain of 10 components, I think I can claim to calculate Nf (10%,
component) as:
Nf(10%,component) = Nf(10%,m daisy-chained components) * [m]^(1/BETA)
= 800 * (10)^1/4
= 1422
Do you agree? Your calculation replaced my 10% value with a 50% value with no explanation, which puzzled me.
I understand N50.
I agree that tak
ing that first failure in a chain of 10 parts to be N50 is conservative. I also agree that 10 parts is a small sample size and compromises statistical robustness. However, is it so small
that we can’t apply this kind of statistical analysis to a chain of this size?
John
From: Werner engelmaier [mailto:[log in to unmask]]
Sent: Thursday, December 18, 2008 3:07 PM
To: Nieznanski, John A - SSD; [log in to unmask]
Subject: Re: [TN] to apply partition correction on individual daisy chains or not?
Hi John,
No, N50 is the mean cyclic life for a large number of identical components. For practical reasons, the 'large' number is reduced to a minimum of 32 to maintain some semblance of statistical robustness. Anything less than that at the confidence band gets much
too wide.
Sometimes you see N63.2, which simply is an artifact of the Weibull distribution being based on 'e'. It does not have any physical meaning—so N50 is preferred.
Of course, when you do a DfR-procedure (see IPC-D-279), you do not design for N50, but something like N1 (consumer products), N0.1 (industrial equ.), N0.01 (aircraft) depending on the severity of the possible consequences of failure.
N50 has nothing to do with sample size.
However, in what you are doing there may be the danger that your accelerated testing produces tighter failure distributions with larger values of beta—thus you woul
d be too optimistic.
On the other hand, taking the first failure as your N50 might be quite conservative.
Werner
-----Original Message-----
From: Nieznanski, John A - SSD <[log in to unmask]>
To: 'Werner engelmaier' <[log in to unmask]>; [log in to unmask] <[log in to unmask]>
Sent: Thu, 18 Dec 2008 2:33 pm
Subject: RE: [TN] to apply partition correction on individual daisy chains or not?
Hi Werner,
Thanks for confirming this calculation.
Is it correct to conclude that the reason you are calling this N50 (and not some other percentage) is because I have limited samples
of each component type and thus the first failure on the single chain of “m” components essentially defines the characteristic lifetime (or N50) for all “m” components on the chain?
I have some chains with as few as 8 components and other chains which as many as 40 components. These also need to be analyzed appropriately,
so I want to confirm the analysis approach for all the chains in my test.
If this isn’t the reason, please let me know how we justify the claim of calling this N50.
Thanks again.
John
From: Werner engelmaier [mailto:engelmaier@
aol.com]
Sent: Thursday, December 18, 2008 1:38 PM
To: [log in to unmask]; Nieznanski, John A - SSD
Subject: Re: [TN] to apply partition correction on individual daisy chains or not?
Hi John,
The un derlying assumption in Eq. 16 is an adequate sample size (32).
What you are doing is not partitioning a component, however, but daisy-chaining components together.
In this case, what you did is correct, because you record the first failed component of 10, and so your eq. should read like this:
Nf(50%,component) = Nf(50%,m daisy-chained components) * [m]^(1/BETA)
= 800 * (10)^1/4
= 1422
Plotting things like this on a Weibull graph typically is very helpful.
Werner
-----Original Message-----
From: Nieznanski, John A - SSD <[log in to unmask]>
To: [log in to unmask]
Sent: Thu, 18 Dec 2008 12:43 pm
Subject: [TN] to apply partition correction on individual daisy chains or not?
Hi TechNet List Gurus,
If I have several different serial daisy chains undergoing accelerated reliability testing per IPC-SM-785, can I apply partition correction (IPC-SM-785 Eq. 16) to calculate an "effective" Nf @test (X%) for each daisy chain? Each chain only has a single component type, but there are d
ifferent numbers of components in each chain.
For example, if the first chain has 10 two-terminal leadless SMT components (assume BETA=4), and the first IPC-SM-785 failure occurs at 800 cycles, then per Eq. 16:
Nf (component) = Nf (m partitions) * [1/m] ^ (1/BETA)
= 800 * (10) ^ 1/4
= 1422
So with 10 components in the first partition or daisy chain, then Nf @test (component) = Nf @test (10%) = 1422.
A more conservative analysis neglecting the partition correction would be to say Nf (10%) = 800.
Which way makes the most sense?
From Nf @test (10%) and BETA, I can calculate Nf @test (50%), then apply the acceleration factors to get Nf @use (50%), then extrapolate to get Nf @use (1%) and Nf @use (0.1%).
Best regards,
John Nieznanski
This e-mail and any files transmitted with it may be proprietary and are intended solely for the use of the individual or entity to whom they are addressed. If you have received this e-mail in error please notify the
sender.
Please note that any views or opinions presented in this e-mail are solely those of the author and do not necessarily represent those of ITT Corporation. The recipient
should check this e-mail and any attachments for the presence of viruses. ITT accepts no
liability for any damage caused by any virus transmitted by this e-mail.
---------------------------------------------------
Technet Mail List provided as a service by IPC using LISTSERV 15.0
To unsubscribe, send a message to [log in to unmask] with following text in
the BODY (NOT the subject field): SIGNOFF Technet
To temporarily halt or (re-start) delivery of Technet send e-mail to [log in to unmask]: SET Technet NOMAIL or (MAIL)
To receive ONE mailing per day of all the posts: send e-mail to [log in to unmask]: SET Technet Digest
Search the archives of previous posts at: http://listserv.ipc.org/archives
Please visit IPC web site http://www.ipc.org/contentpage.asp?Pageid=4.3.16 for additional information, or contact Keach Sasamori at [log in to unmask] or 847-615-7100 ext.2815
-----------------------------------------------------
|
