You’ve heard of the Kelvin scale, we’re pretty sure. But what you might not know is that its inventor, mathematical physicist William Thompson – later to become Lord Kelvin – was from Belfast. Point one in his favour, as far as we’re concerned. Point two? He’s famed for saying, “if you can’t measure it, you can’t improve it."
And that’s something we firmly believe in.
Although the phrase was coined more than a century ago, it’s still relevant now – especially when it comes to compiling data for a predictive maintenance regime.
We’ve talked before about how predictive maintenance is revolutionising diverse industries such as healthcare and the automotive business, but it makes sense fiscally too. Recent studies show that unplanned downtime costs industrial manufacturers an estimated $50 billion each year.
But before you get to the savings, you need to see that you have the right data to implement a successful predictive maintenance regime. In general, you’ll need to collect three types of datasets to get the most out of a predictive maintenance regime.
This is the baseline data that shows the normal behaviour of your machinery. This reference data informs the 'learning phase' of a predictive maintenance regime whereby historical or reference data is used in order to generate the predictive model.
It includes design and test data, maintenance data and old operating data. The more in-depth your reference data is, the more of a guideline it can give you for predicting variances.
Likewise, the more information you have on a piece of machinery and how it works, the easier it is to tell when something is malfunctioning or showing signs of distress.
Many of the critical factors that will eventually help you predict failure or a replacement requirement may be buried in structured data like the component details of your machinery. This includes the equipment year, make, mode and warranty details.
But this reference data also includes many unstructured data types too – including log entries, sensor data, error messages, temperature records and repair and maintenance reports.
Operating data is the granular data your machinery produces every second of every day. In order to produce the predictions, this operating data is integrated into the prediction model during the 'learning phase' of the process.
The operating data is also integral to the 'learning' aspect of your maintenance programme as it gives the model capabilities to update, embodying new information in the algorithm(s).
By its nature, the sheer amount of data can be staggering. For example, on the latest Airbus A350 airplane, 6,000 sensors across the plane generated 2.5 terabytes of data every day. To put that into perspective, if you were to print out all that information, you’d need to fell 125,000 trees to produce enough paper. Airbus' next model of the A350 will produce three times as much data – so the ability to efficiently read and store it will be paramount.
Operating data is crucial to a predictive maintenance regime because the software uses historical operational signatures for each piece of machinery. It can detect even the subtlest changes in the behaviour of the machine. The main benefit of predictive maintenance is that, once properly implemented, it can identify and flag changes in system behaviour well before those variables reach a point where they could cause real operational damage. Instead, there is time for analysis and corrective action.
Domain data is a looser categorisation, though it still matters. The information gathered as part of this 'adoption phase' informs and refines the predictions that are made.
Many organisations make the mistake of discarding legacy data in favour of optimising for output for the new regime – but that’s a mistake.
Domain data includes maintenance manuals that enable engineers to act on failures and promptly fix issues. It also involves codifying your staff’s in-built knowledge. For example, your team might know that a particular machine always breaks down when it snows. They might not know why – but that's what predictive maintenance can determine with a structured data analysis.
In some instances, these three types of data may not be readily available to your organisation – but you can start working towards them with the intention of implementing a predictive maintenance regime. If your current system doesn't produce enough operating data, sensors can be added to your machines. Missing reference data can be found from similar machines or by asking the manufacturer, as well as taking detailed logs.
The key is to assemble your data so that it is clean and organised. From there, you can start to put together your predictive maintenance regime – and then the big savings kick in.
If you want to harness the power of artificial intelligence in your business, Statwolf’s data science service can help, with advanced online data visualisation and analysis simply running in your web browser.
We offer a range of custom services to suit your needs: advanced data analysis and modelling, custom algorithm creation and implementation with a particular focus on Predictive Maintenance.
We’ve helped a host of clients get the most out of their data. Are you going to be next? We don’t just deal in data analytics; we offer data solutions.
Want to make sense of your data? Download our comprehensive guide: The Pedictive Maintenance Cookbook.