MTBF
Hey,
does anyone know much about Mean time between failure (MTBF) or reliability in general? I have to calculate an MTBF for a company I'm working for on placement and I'm struggling to get my teeth into it. I know what MTBF is and the basic ideas of it, so don't just direct me to a google search result for MTBF. I'm just wondering if anybody has any personal dealings with anything like this, and if they have any words of wisdom they could pass on to me.
Thanks,
Meerkat
does anyone know much about Mean time between failure (MTBF) or reliability in general? I have to calculate an MTBF for a company I'm working for on placement and I'm struggling to get my teeth into it. I know what MTBF is and the basic ideas of it, so don't just direct me to a google search result for MTBF. I'm just wondering if anybody has any personal dealings with anything like this, and if they have any words of wisdom they could pass on to me.
Thanks,
Meerkat
Do you mean that you are writing Quality Control Software to measure the performance of a comapany or it's employees? I have some second hand knowledge in this area, what are you stuck on?
The first thing to do is to establish\write down what counts as a failure, then you simply need to count and time stamp them in some kind of chart or graph that makes it look shiney and important.
The first thing to do is to establish\write down what counts as a failure, then you simply need to count and time stamp them in some kind of chart or graph that makes it look shiney and important.
It really depends on what you are trying to find the MTBF of. In the field of reliability, one of the underlying assumptions is that events occur in a probabilistic manner; as such, you can model the likelihood of an event occurring if you can fit a probability distribution to it. There is a wealth of information on the probability distribution functions (pdf) that best fit certain events. For example, the lifetime of capacitors are best modeled with an exponential distribution, which has a pdf of f(t) = lambda*exp[-lambda*t], where t is time and lambda is the failure rate. The MTBF is the expected value of the pdf, which is the integral of t*f(t)dt evaluated over the domain of the pdf. In the case of an exponential distribution, the domain is from 0 to infinity, so you integrate from 0 to infinity. The result is lambda, which means that the MTBF for a failure that follows an exponential distribution is constant (independent of time). The trick in this case is finding lambda. Some manufacturers give the failure rate on their product data sheet. If you don't have this information readily available, see if anyone reputable performed an experiment to predict the lifetime of the unit. If this hasn't been done before, then you need to design an experiment to estimate lambda. Computergeek01 mentioned one way of doing this, but exact failure times require continuous monitoring and can result in more costly experimental set ups. It's easier to rely on read out data (counting the number of failures that occur in an interval). So if you wanted to test capacitor failures under electrical stress, you place a bunch of them in a chamber then check every half hour, hour, day, etc to see how many failed. You can then estimate the pdf of the data by constructing a histogram and fitting a pdf to it or you can try to estimate the pdf by using a Goodness of Fit Test and estimating the confidence interval.
I hope I didn't go way off topic for you, as I don't know how in depth you want to go. In general, your first step it to find a pdf that sufficiently models whatever it is you are trying to estimate. If the pdf isn't given that you can estimate it by designing an experiment and constructing a histogram of the data. Once you know the pdf of what you're trying to model, you need the parameters of the pdf. They may either be found through searching the Internet/related literature or estimated experimentally. It is important to do a Goodness of Fit test and establish confidence intervals so your client can believe your numbers. Once you know the pdf and have the parameters the you should be able to calculate the MTBF. You can find the MTBF for many distributions online.
I hope this helps. Let me know if you have anymore questions.
I hope I didn't go way off topic for you, as I don't know how in depth you want to go. In general, your first step it to find a pdf that sufficiently models whatever it is you are trying to estimate. If the pdf isn't given that you can estimate it by designing an experiment and constructing a histogram of the data. Once you know the pdf of what you're trying to model, you need the parameters of the pdf. They may either be found through searching the Internet/related literature or estimated experimentally. It is important to do a Goodness of Fit test and establish confidence intervals so your client can believe your numbers. Once you know the pdf and have the parameters the you should be able to calculate the MTBF. You can find the MTBF for many distributions online.
I hope this helps. Let me know if you have anymore questions.
Yea sorry I wasn't very clear with my question. It's an MTBF for an electronic system, so it's a lot to do with taking the MTBFs of capacitors and other components and coming to an MTBF of the whole system. What Kabiru has said seems to be really helpful I'm still in the process of compiling a list of all the different components that are to be used in the project so that I can work from there. Once I do that though, I'll have a re-read of your post and I'm sure I'll have questions once I start with the actual calculations.
Thanks.
Thanks.
Topic archived. No new replies allowed.