Hi all,

Can anyone give me difference between MTBWSF andMTBF..... anfd how to find MTBWSF from MTBF of a system....

Thank you in advance

Hi deepthi,

Can you explain defination of MTBWSF?

hi

MTBWSF is mean time between wrong side failure

I am not sure but in my opinion it means to say that the system fails leading to a hazard/accident.



It can be better explained using the following example:

Say a system has two states/modes A and B and a & b be their corresponding outputs.

Two cases when the system fails:

1. In state A when the system gives an output b
Say the system fails but will not lead to a dangerous situation (a hazard) but is safe.

2. In state B when the system gives an output a
The system fails and there is a potential for an hazard or an accident. The system fails on the wrong side.


A simpler explanation would be a component in a ckt getting short or open. If the component fails such that it is short circuited and has no effect on the systems functionality, then it is a failure. In the other case if it opens the continuity of the ckt is broken and the component has failed on the wrong side.

I hope I am clear, and wish to know if there is any contradiction.

Thank you.

Hullo Deepthi,
There are several sources of error in computing MTBF. The requirement of independence and identical operating conditions is often not met, so the MTBF calculated is questionable. We are also assuming an exponential distribution when working with MTBF values. This is not always correct, especially when working at the elemental level eg., while doing a design FMEA.To this you want to add a 'refinement'

quote:

MTBWSF is mean time between wrong side failure

I suggest that the quality of source data in most field conditions is such that the answers will be hard to believe or use in a practical sense.
If you have a situation where good quality failure data is available, you will be better off defining the failure distribution, by e.g. using Weibull analysis.
Wrong side or right side is often a matter of chance (there but for the grace of God etc.). You are more likely to manage failures using survival probabilities than MTBWSF.

Caluclating the MTBF arithmetically, i.e number of failures over a given time will result in a figure which will be incorrect. The only time you can do this is if the failure are idenpendantly identical. That is to say you have assumed exponential failure rate.

The best option is to determine the failure characteristic, (Beta) by using the Weibull fucntion.

dvaidr,

I am not sure I understand your post, perhaps you can clarify. My observations on your post are as follows:
1. Computing MTTF inherently implies we are assuming an exponential distribution, or a Webull shape factor of 1.
2. I think you mean Weibull Scale factor, or 'eta', not Shape factor or 'beta'; MTTF (or MTBF for repairable items) has the same value as 'eta' for an exponential distribution, as you are no doubt aware.
3. The requirements of 'independent' and 'identical' applies to all of these calculations, whether MTTF or Weibull parameters.

Perhaps mu English isn't the best. Assuming exponential failure is the same as assuming a constant failure rate. Is the failure rate constant? Is, so fine. If not, them one must determine the failure characteristic, Beta. Now I know that some Engineers use Beta and Eta differently. My Brother who is a Mathematician use eta for shape characteristic whereas I would use Beta.

What I was thrying to say that a lot of people calulate the arimetic mean for MTTF/MTBF - fine is it is constant failure rate. In the rel world though, I have never come across constant failure rate.

dvaidr,
Using MTBF and MTTF means we are assuming a constant failure rate, or an exponential distribution.

At the elemental level of failure modes, i.e at spring broken or clutch plate worn level, you are quite right about the distribution not being exponential.

As soon as we have complex assemblies with many such elemental failure modes, e.g., a ball bearing or centrifugal pump, the distribution tends to be exponential. In the Nowlan and Heap Aircraft Industry study in the 1960s, they found that about 90% of failure modes were exponential (if we ignore early ife failures).

So you're saying that you tend to use constant failure rate always!

From experience, the systems I've worked on have included lognormal, to exponential characteristics. And I stress the 'system' as opposed to components therein.

If one really gets to the nitty-gritty, then for repairable systems one must use the general renewal process to ascertain the condition of the asset when it is returned to service. (Good as new, bad as old...)

You really must be working on different systems to me. Rarely have I found a system to yield exponential failure characteristics.

dvaidr

quote:

So you're saying that you tend to use constant failure rate always!

No I am not.

I have done a study of ca. 35000 failures in the O&G industry; we were able to get over 1000 sets of Weibull parameters. About 70% had shape parameters between 0.9 and 1.1, close enough to 1 to consider them exponentially distributed.
In the Airline study of 1960's, 68% were exponential, while a further 21% had a constant hazard rate if we ignored early life performance. So the total with this simplification was nearly 90%.
There have been similar studies in the Marine and some other industries, with similar results.

So I conclude that at sub-system level, a large proportion of failures tend be exponential.

Back to your question: the small proportion of age-related failures also matter, and must be identified. I do not recommend we treat ALL failures as exponentially distributed. The main point is that we must understand the actual distribution when deciding maintenance strategies.

You said, "Using MTBF and MTTF means we are assuming a constant failure rate, or an exponential distribution".

I still don't grasp this. If you are assuming a beta of 1 then MTBF/MTTF is the arithmetic mean, then you go on to mention Weibull. Weibull is used to determine the shape parameter or failure characteristic, beta (amongst other parameters), when failure rate is perhaps not constant (< or > 1). For repairable assets the general renewal process needs to be used to determine the effects of the repair.

dvaidr

quote:

You said, "Using MTBF and MTTF means we are assuming a constant failure rate, or an exponential distribution".

I still don't grasp this

Please read section 7.3, pg. 34 of the book " The Reliability of Mechanical Systems", IMechE publication, John Davidson, ISBN 0 85298 881 8 or,
Section 2.8.1, Page 39, of Reliability. Maintenance and Logistic Support, Kluwer Academic Press, Kumar et al, ISBN 0-412-84240-8, or
Section on Mean Availability, pg 34-36 of my book Effective Maintenance, details in footer.