The error in predictive maintenance

Predictive maintenance has come an awful long way since the early days of visual inspections and rough oil analysis. Today, organizations are working with online condition monitoring, strategic NDT techniques, and using the emerging technologies to reduce costs, reduce risk, and increase revenues.

Fair enough...

But there are a few fundamental issues that sometimes slip through when designing a predictive maintenance program.

1) Not just for predicting failure

I think this is the primary issue facing all of us, and it is one of culture. We are trained from the beginning to use predictive maintenance for detecting the onset of failure. Even RCM encapsulates this thinking.

The problem is that there is still a lot of science that is needed in this area. Online CM systems, and advances in the technology for storing and querying this information have given some companies a helping hand - but most of us are still foraging in the dark.

The key to this, however, is that predictive maintenance is not only useful for predicting failure, but also for predicting the likelihood of failure. For example...

• Overloading a motor will run it to failure quicker, wasting a lot of its early life. 

• Ambient vibration will cause early life bearing failures

• Ambient heat will cause early life failures of many components

• Using rams outside of their design cycle rates or pressures, and so on... you get the pitcure right?

Predictive maintenance is often best used as a predictor of behaviors leading to early life failure, I have personally seen some phenomenal results from focusing on this sort of thinking alone.

2) Masking the original equipment problem

You predict the failure, change the bearing, and voila! Unplanned large failure costs avoided. And then, 6 months later, you pick it up again.

I posted on this several months ago when I used to write for PlantServices.com - the error of predictive maintenance is that it can mask the causes of failure. In this case it could be misalignment (something I really think it underdone in asset maintenance) , or using mallets instead of presses for puching the bearing on, or the habit of putting too much heat into the bearing during installation...whatever the cause, predictive maintenance isn't going to find it.

Getting around this is pretty simple, 1st make sure your RCM analysis actually goes to the right level of detail. E.g. the level where we can do something about it. And second, make sure that you understand the real causes for each failure, not just "the bearing failed".

3) Misapplication

Ever heard someone tell you that "all assets are showing the signs of the onset of failure"?

There are two causes for this. a) they have misunderstood the whole condition monitoring approach and are applying it to a time based failure where the signs of wear are evident from the early stages.

This goes all the way to a misunderstanding and misapplication of NDT technologies as a form of condition monitoring...something that most of the industry hasn't caught onto yet.

Or b) they are using a P-F interval that is too wide, and they are not using this information for any purpose other than indication.

Easy stuff, learn your RCM well and this problem goes away with practice.

I hope this is of some use to you...