They say you should trust your instincts. If you’ve worked in the same factory for years, you’ll know the machinery like the back of your hand and have an idea of when it’s likely to go on the fritz. But such guesswork could cost a business huge amounts of money in unscheduled downtime.
A factory I visited recently shocked me when they told me their maintenance procedure. They said they know when changes need to be made because the control room starts to shake.
At the heart of this way of thinking is a reactive approach to maintenance, and indeed data. Making estimates of failure based on such warning signs does no justice to the individual components that make up a manufacturing plant.
Too many businesses make judgements on an entire population of machines. And even when a company does use data, it’s often using techniques that deal in averages across large populations. Using historical data, they predict the life of a component, based on average stresses, inputs and operating environment.
In the real world, it’s unlikely the components to be “average” in every important quality. And these analytical techniques, based on the past, can be a good predictor of the future, but it doesn’t have to be: environments and inputs can change radically within a short time period.
With IoT and machine learning techniques, however, you work with real-time data to understand individual components in the context of their own environment and to predict the focused future of that element, resulting in targeted maintenance schedules.
Can we predict failures accurately, and can we attribute the cause?
A study by Deloitte, published in 2017, found that by using predictive analytics, companies could cut both material and overall maintenance costs by up to 10% [1].
What makes the cost of maintenance so high is that many businesses are doing the wrong kind of maintenance. Plants with the ‘shaking control room’ approach are only one example. There are also those who play it safe by carrying out scheduled maintenance before serious problems occur. Scheduled downtime still costs money and should only be carried out when there are insights that recommend doing so.
Conversely, IoT can tell you when to carry out maintenance and what to repair. A steel factory I visited had a problem they thought was to do with the compression rollers that process the steel. They tightened the relevant nuts and bolts they thought were the source of the problem, but they had to keep doing the same thing every few years.
What they didn’t realise was that the rollers were just a symptom, the problem itself was way upstream. They were producing steel at the wrong tolerances, which was damaging the elements of the factory it was passing through. Had they seen the data that pinpointed the exact source of the problem sooner, they would have saved a lot of time, money, and hassle.
Can we predict failures accurately, and can we attribute the cause?
When I meet with manufacturers my advice is: start small. Find one component in your plant that has costly scheduled maintenance and build a business case to address the problem. What you’re looking for specifically is what the value of reducing unscheduled downtime by a certain percentage could be.
Afterwards, take stock of all the data available from across the business, whether that’s work orders, enterprise resource management or manufacturing execution systems, assess it and find out where it lives.
Then, answer two questions: can we predict failures accurately, and can we attribute the cause? Once you answer those questions, you’re well on your way to improving operations in your organisation.
As you’re starting out with one component, you’re beginning a small pilot project that can scale as needed when you want to solve more problems. This is not designed to be a final production execution system to be deployed across the organisation with an $80m change programme attached. We’re just looking to solve a problem.
Predictive analytics, powered by AI and IoT, allows you to gather a wide range of variables, understand how these variables interact with each other to determine device performance, and monitor system performance in real-time.
This makes it possible to move beyond understanding average population-level device performance and longevity in the past, to building models for future device reliability that are accurate down to the level of individual devices.
Using these insights, you can avoid planned downtime while also predicting and pre-empting the kinds of problems that cause unplanned downtime. This leads to lower costs, higher reliability and increased profits.
Wael Elrifai is VP for Solution Engineering at Hitachi Vantara
Promoted content: Does social media work for engineers – and how can you make it work for you?
So in addition to doing their own job, engineers are expected to do the marketing department´s work for them as well? Sorry, wait a minute, I know the...