# What can you do if everyone across you industry is using MTBF?

- First, stop using MTBF yourself.
- Second, show others the information that is found in using Reliability directly rather than using MTBF.
- Third, translate your work back to MTBF and be very clear about duration and other assumptions.

## Stop using MTBF yourself

You do not have to announce the change or ask permission. Just take the data you already have and instead of calculating the MTBF, calculate appropriate reliability function. Fit to Weibull or Lognormal or whatever is appropriate.

See if you can see the difference.

Calculate the reliability at points in time that are important to your product or system. Then calculate the same value use the Exponential distribution reliability function based on the MTBF you would have used.

Different? Why is that?

Let’s say you are working on a motor and gear box assembly and have times to failure data for a few machines. You and your maintenance team knows that the system will wear out over time and as it ages, the chance for failure increases.

This data will most likely follow a Weibull distribution with a beta (shape parameter) that is greater than one. Let’s assume the MTBF is 5 years of operating time. Note: You could also use a mean cumulative function to understand the effect of repairs over time, if the equipment is repairable. MCF is preferred for repairable systems and Weibull for non-repairable when I’m first looking at life data.

And, let’s say we’re interested in the chance of successfully operating for a year. The MTBF (exponential distribution based) will underestimate the probability of the system to operate for a year, whereas the Weibull will correctly reflect the relatively low probability of failure in the first year with an increase is chance to failure as it wears in later years.

Does your data show a difference? If so would it alter any decisions you would make based on the data? This first step is really just to convince you to use the appropriate distribution or statistical technique to convert your data into information.

## Show others your results

After convincing yourself and becoming comfortable with the calculations, show those around you.

Especially show those making decisions based on the data. Show your customers, vendors, suppliers, and engineering teams. Show your marketing and finance teams, too. Talk to sales and anyone else that will listen.

Show them that using the data to make decisions is a part of how we operate. Show them using an accurate reflection of the data permits better decisions. It will save you time, money, resources, and frustration.

The first time I did this, I simply provided the MTBF calculation along with a short calculation and plot of the same data using a Weibull fit. It provided a quite comparison and illustrated the difference in the view of the same data the two techniques provided.

I was amazed at how quickly others understood and correctly used the cumulative distribution function (CDF) or plot showing time vs probability of failure. (For the marketing group I used the Reliability function plot as it is a bit more positive being the probability of success. They liked ‘success’ over ‘failure’.

Even managers got it.

## Translate back to MTBF (if forced to do so)

Only if forced to do so.

When facing customers and an entire industry of inertia and common use, just switching away from MTBF may not be possible.

In this case, continue to use your data to extract the best information by using the appropriate techniques. Make good decisions and recommendations.

When required to provide an MTBF value – as a minimum provide:

- The MTBF value with a duration over which it is appropriate
- The impact of using the assumption of constant failure rate when it isn’t true
- The probability of success (or failure) figure over the same period based on Exponential and Weibull, for example.
- And, add a link to the nomtbf.com site (optional)

In short, focus on the value of making good decisions and the cost of making poor decisions. We are using data, estimate, requirements, etc to make decisions. Using the best available information allows us to make better decisions.

Overall, I agree. It occurs to me that there’s an interesting juxtaposition. When I buy tires, I don’t expect to replace them for some time. Yet a tire manufacturer will be perfectly happy with tracking a constant failure rate. I should think that where you are in the “food chain” will have some effect on the models that are useful.

In the same way, there are many processes that we can model using a constant rate (and I believe many of them are Poisson processes) so long as certain conditions are met. Over a year’s time, I will likely buy kleenex at a more or less constant rate. But if I have a cold, then for a time I’ll have a higher usage rate. If the failure is running out of tissue, then the decision is when to buy more. These kinds of decisions are fairly common. The model I choose needs to take into account the decision I will make.

I suspect a smart tire vendor will notice your rate of tire wear and send you a reminder to check thread wear about when the tires are due for replacement. They could even track and predict the perfectly timed advert if they perform regular car maintenance and tire rotations for your vehicle.

Thanks for the comment and examples.

Cheers,

Fred