What is Predictive Maintenance?

What is Predictive Maintenance?

Production depends on machines, and they need to work well.

When a machine breaks, this costs time, money,

…lives.

That is why we preventively maintain them.Better to fix one hundred parts that could break than to have one of them broken at a critical time.

In a perfect world, we could predict which part will break and exactly when.

But in most cases prediction is only stochastic.

Out of 1000 items, we know about 10 will break.However, we don’t know which ones.

Predictive (PM): replace the 10 lightbulbs that will actually break.

But how to find those 10?

Preventive: replace all 1000 lightbulbs.

MAINTENANCE

Traditional predictive maintenance using the most versatilesensor in the world: human eye

…and human ear: vibration analysis

Today: Sensors + Big Data + Analytics

One hope from Big Data:

Massive amounts of information will substitute for poor quality.

Not necessarily!

Data science will find

the 10 lightbulbs.

Can’t get something

out of nothing

This is what data science is not.

Prediction is about having correct information. Place relevant sensors at right places.

find the crack when it is small

Each PM problem is unique.

There is no one-approach-fits-all in PM.

But PM is not only about predicting.

PM is a combination of:

Understanding the economics of the industrial process

Discovering the proper sources of information

Creating a predictive model

Economics:costs increase with PM effort; number of sensors, quality of prediction, etc.

as a resultsavings and profits rise, but not linearly

sweet spotPM costs are low, savings are large

PM can be more expansive than savings

The person to askSo

urc

es o

f in

form

atio

n

Reserve time for experimenting.You will encounter failures before success.

Finally, a data scientist can create a model. Remember: machine learning is a tool, not a solution.

Putting all aspects of PM together is a collaborative work.

in conclusion

Predictive Maintenance is as much research as it is data handling.

It is rather a process and a mindset.

It has no single best practice.

Data Science Team

Prepared by: Danko Nikolic & Davor Andric