For the last 10 years, the hot topic in the world of technology has been big data and advanced analytics.
When we talk about these technologies in the manufacturing and utilities sectors, the conversation tends to focus on how they will speed up production and make operations more efficient.
We don’t always talk about what it means for maintenance.
But advanced analytics and Big Data can have a massive impact on the maintenance process. They can help you to learn more about your assets, so you can predict when they are going to break down, and get there before they do.
This has untold benefits for manufacturers – as I’m going to explain in this blog post.
Predictions not recommendations
Normally when you buy an asset it comes with recommendations from the manufacturer You carry out maintenance based on these technical recommendations; if it’s recommended that you should complete an overhaul every six months, that’s the time scale you will use.
The problem is that machines behave differently depending on their environment and context.
Having a wind turbine in Spain is very different to having one in Denmark, or even another place in Spain. And in these instances, the recommendations for the asset’s maintenance won’t be exactly right or as accurate as they could be in terms of operational efficiency
You’ll either end up doing maintenance too early, when it’s unnecessary, or too late, when the asset has broken and cost a lot of money in downtime.
This is where big data and advanced analytics can work really well. You can use machine learning to analyse the asset’s data, and work out when it is likely to fail ahead of time.
Being able to predict an asset’s maintenance needs will not only reduce the risk of the asset breaking, but it will also make the maintenance work cheaper.
If an asset is in a controlled environment like a factory it’s not so hard to take care of it. But when it’s out in the field (which is literally the case for a wind turbine) it’s difficult to monitor.
Sending engineers out for potentially unnecessary checks is expensive – but not as expensive as having to do an emergency repair in a remote location, which might take a long time if you have to wait to order parts.
And all the while the turbine is broken, it isn’t producing anything.
All of this can be prevented completely with predictive system, which can tell you when an asset needs a repair – so you can schedule in a repair and order the parts in advance, eliminating the need for downtime as well as having unnecessary stock
This is part of an approach we call intelligent engineering. It’s a shift from the traditional ‘break-fix’ model, towards something far more proactive and therefore smart.
Keeping wind turbines up and running
Predictive maintenance can deliver a return on investment of over 200%, depending on the number of assets that need to be maintained.
It can also lower the cost of maintenance 10 and 20%, alongside reducing planned maintenance time by 40%.
This is what we’ve found by applying predictive maintenance to wind turbines in the EMEIA region.
Our aim was to use advanced analytics and big data to improve the maintenance operation around three main components within a wind turbine: the gear box, rotor and generator
And we’ve achieved it; the predictive approach has enabled us to spot an upcoming failure a month ahead of time.
This means we are increasing the machines’ lifetimes by 1-2 years. And since they last longer, we improve the financial costs associated to the investment.
How do we do it?
Using deep learning, we’ve developed an anomaly detection algorithm capable of finding the tiny mistake in a really big data set.
This is called operational data management analytics (ODMA). It already works for predicting failures on certain models of wind turbines, but it can easily be adapted to other assets using the same algorithm and methodology.
Ultimately, we’re not interested in creating a generic piece of software. We want to learn how to create good algorithms that solve big maintenance problems – no matter what type of problem.
Still, the thing that always amazes me is the fact that there is so much behind the technology.
Normally, when I start a predictive maintenance project with a customer, we start by talking about thinks like data availability and maintenance costs and operations.
And then it always happens: you start digging into the problem, and you end up talking about stock management and operational issues around insurance conditions and cultural change.
So, predictive maintenance actually involves many other issues. This topic – the importance of change at the wider cultural level – is something we explored in our recent white paper ‘Transforming Manufacturing: Co-creating the digital factory’.
It just shows you need a holistic approach to make predictive maintenance work – both in terms of the technology and everything else that surrounds it.
Maintenance powered by machine learning
Machine learning and data analytics will have a transformative impact on the maintenance process.
Where previously maintenance was expensive, time consuming and difficult, it will become quicker, simpler to schedule and more cost effective.
This will make a serious difference to manufacturing and utilities, since maintenance is so integral to these asset-heavy industries.
The power of prediction makes all the difference. If you can spot a problem forming, you can prevent it.
A problem-free environment is where we are heading in Industry 4.0 – and artificial intelligence is taking us there.