Predictive maintenance is a lot like a menu at a Chinese restaurant. Everyone gets the gist of it, but no one knows exactly what is in it. However, a big difference is already apparent here: with predictive maintenance, you don’t need to know what it contains (and as is the case with many restaurants – regardless of the cuisine – I don’t really want to).

Where does the magic happen?

Everything to do with predictive maintenance takes place in a black box. It is in this black box that the magic happens. High-performance computers calculate your ridiculous mega algorithms within a few nanoseconds  – so quickly, that you practically know when and how that transformer is going to spit the dummy before you’ve even pressed the button. This report will be garnished with the information telling you that the best time to service the transformer is in twenty years on a foggy August morning at 5:47. You don’t even have to do anything else. Press the little button, drink some tea, put your feet up and it’s all sorted. This is obviously exaggerated, but where does the crux of the matter lie in reality?

The previous example is just an exaggerated case, but the path towards it can be long and rocky. For good prognosis, you need good data. If you are trying to forecast whether it will be a white Christmas, there is a huge difference between just looking at Christmas Eve from the year before or looking over the last forty years. The validity of a prediction based on forty years of data rises drastically, even though the chance of having a white Christmas does not. The same applies to predictive maintenance. Correlations between input and output data are searched for in a statistical data model. Using the analysis methods of the data model, links will be drawn between several variables from which information relevant for servicing and maintenance will be derived.

What you can conjure up with your data

The validity of predictive maintenance increases with the use of a time frame and a large volume. There is no limit to how far this can go: absolutely all of the data that is associated with a piece of operating equipment can be relevant to predictive maintenance. Absolutely all of it. Take, for example, the luggage conveyor belts in an airport and the results from the football. If someone’s club spontaneously managed to get a spot in an international competition, they might choose to go on a last-minute trip to Milan or Madrid. In other words, the luggage conveyor belts will get used more. Something like this, which at first glance seems absurd, actually makes a lot of sense. Everything can be thought through and the statistics never lie.

The expenses related to data collection should stay in a good cost-benefit ratio – after all, you want to see the wood for the trees. Also, sticking with the analogy, not every forest is the same – there are conifer forests, deciduous forests or mixed woodland. What works with one energy provider won’t necessarily be successful for another.

Predictive maintenance doesn’t suit every piece of operating equipment to the same extent. It does, however, make sense for a lot of people seeing as the benefits of predictive maintenance tend to outweigh the costs.

Statistics – how we learn from our mistakes

Maybe your operating equipment is just too good. If you keep your operating equipment up and running without any malfunctions – how are the statistics supposed to show how to avoid disruptions in the future? The algorithm always learns from mistakes, and that is exactly the idea behind predictive maintenance. Future success can be predicted based on data models and the laws of statistics, using data collected about the faults.

I don’t mean to say that you should let your network go completely haywire so that you can collect malfunction data. You wouldn’t do that at a nuclear power plant (at least I hope not). A lot of operating equipment is so trouble-free, that there is no malfunction data available.

Collecting data is not sorcery

If you are able to gather all relevant data conveniently, predictive maintenance not only takes the cake as the best maintenance concept in theory, but also in practice. This is ultimately due to the fact that all correlations can be identified in a model which has all data present. It is not easy for people to admit that computers are much smarter and more objective than us humans. Does this mean we should throw all existing concepts overboard just like that?

No, predictive maintenance should be seen more as a supplement to existing maintenance concepts. There is nothing wrong with beginning to collect and process data relating to malfunctions and other information. By doing so, you can gain important insights into your maintenance. If we now think back to the Chinese restaurant at the beginning, sometimes it isn’t such a bad thing to know the ingredients.

Predictive maintenance  – better than fortune telling

The cost and benefit of predictive maintenance should not be overlooked. Predictive maintenance will surely be a part of our daily lives. Take a moment to think about where predictive maintenance could already be a useful addition and what groundwork for it (e.g. data collection) is already under way! Since 2011, me and my company here at meliorate are here to answer any questions that infrastructure owners or operators might have.

FURTHER BLOG ENTRIES BY OLIVER D. FÖRSTER

Indiana Jones and the Optimal Maintenance Strategy

Maintenance? – We have always done it like that!