Back to the Future: Predictive Maintenance
Predictive Maintenance is like the menu at a Chinese restaurant. Everyone understands the concept, but no one really knows what’s inside. But here’s the big difference: with Predictive Maintenance, you don’t need to know exactly what’s inside (and at some restaurants—regardless of culture—I’d rather not know anyway).
Where Does the Magic Happen?
In Predictive Maintenance, everything takes place in a “black box,” and this black box is where “the magic happens.” High-powered computers calculate their complex, futuristic algorithms so quickly that it’s as if you know when and how the transformer will fail almost before you press the button. This dramatic prediction comes with a maintenance recommendation to service the transformer at 5:47 a.m. on a foggy Monday morning in August—twenty years from now. And that’s all you have to do. Press a button, relax, have some tea, and put your feet up. Of course, this is an exaggeration, but what’s the real core of Predictive Maintenance?
It’s similar to the exaggerated example above, but the road to getting there can be long and rough. Good predictions require good data. It’s one thing to predict a white Christmas based on the previous Christmas Eve, but another to consider data from the past forty years. With forty years of data, your prediction is more reliable, though you still may not get snow. It’s the same with Predictive Maintenance. In a statistical data model, correlations between input and output data are identified. Through these data model analysis methods, multiple variables are linked, resulting in actionable insights for maintenance.
What You Can Reveal With Your Data
With more time and data volume, Predictive Maintenance’s accuracy increases. And it doesn’t stop there: any data even remotely related to an asset can be relevant. Really, all of it. Think about airport baggage conveyors and sports results. A fan whose team unexpectedly qualified for an international match might spontaneously book a trip to Milan or Madrid, leading to increased demand on baggage systems. What sounds absurd at first glance makes sense on second. Everything is possible, and statistics are always right.
The effort required for data collection should be proportional to the benefit—you don’t want to miss the forest for the trees. Not every “forest” is the same, either; there are pine forests, deciduous forests, and mixed forests. What works for a power grid operator might not work for a railway network. Predictive Maintenance isn’t suitable for every asset to the same extent, but for some, the benefits will outweigh the costs.
Statistics – Learning From Mistakes
Your assets may even be too well-maintained. If they’re so well-kept that no failures occur, how is the algorithm supposed to learn how to prevent future issues? After all, it can only learn from mistakes, and that’s the whole point of Predictive Maintenance. With failure data, you can use data models and statistical principles to forecast future events.
But don’t take that to mean you should blow up your network just to gather failure data. You wouldn’t do that at a nuclear power plant (or so I hope). Some assets are so low-maintenance that there’s little or no failure data for them.
Collecting Data Isn’t Magic
If you can gather all relevant data cost-effectively, Predictive Maintenance is not only theoretically the best maintenance concept, but practically as well. This is because, in a model where all data is available, all correlations can be identified. It’s not easy for everyone to admit, but computers are sometimes smarter and more objective than we are. So, should we throw all existing concepts overboard?
No, because Predictive Maintenance is best seen as a complement to existing maintenance approaches. There’s no harm in starting to collect and prepare failure data along with other relevant information. This will yield valuable insights for your maintenance. And if you think back to the Chinese restaurant from the beginning, it’s sometimes not so bad to know what ingredients are in your meal.
Predictive Maintenance – More Than Fortune-Telling
The costs and benefits of Predictive Maintenance shouldn’t be underestimated. Predictive Maintenance will likely become part of our daily lives. Start thinking now about where Predictive Maintenance could be a sensible addition and which preparatory steps (such as data collection) could be taken today! Since 2011, my company Meliorate has been supporting infrastructure owners and operators in tackling these questions.
Author: Oliver Förster