We recently sat down with Dr. Moray Kidd and picked his brains on what he believes is the future of AI in maintenance engineering.
Dr. Kidd has over 20 years of experience developing artificial intelligence engineering applications and previously held engineering roles in the power distribution, aerospace and automotive industries. Dr Kidd is currently a senior lecturer at a leading UK University.
In the course of the conversation we discussed the commercial growth of AI in manufacturing environments, how next generation predictive maintenance could save companies millions and why civil aero engines and electric vehicle platforms OEMs may hold the solution.
It’s obviously an area of interest for me. The idea of using AI to solve engineering and manufacturing problems isn’t that new though – it’s been applied in maintenance research for decades. What’s new is the growing commercial focus, that has been building over the last ten years.
The term ‘Industry 4.0’, which has accompanied this, has been great. It’s done a lot to alter the perception of maintenance engineering.
The ability to intelligently predict when machinery is going to fail – even when it appears to be within specification following inspection – can be hugely significant from a commercial perspective. For example, if a critical component on an oil platform fails this could lead to an outage that could shut down operations for a number of days – that could equate to around millions of pounds in lost oil production. But, the cost of the part that failed is normally insignificant in comparison. If we knew that was going to fail ahead of time and replaced it, in a managed process, it would provide significant saving.
"The ability to intelligently predict when machinery is going to fail – even when it appears to be within specification following inspection – can be hugely significant from a commercial perspective."
- Dr Moray Kidd
Unexplained part failures can lead to major commercial consequences. This might include a vehicle recall by an automobile manufacturer or fleets of aeroplanes being grounded. Next generation proactive maintenance can reduce this risk of this happening.
There have been a number of initial attempts to roll this out but many have failed for two main reasons. The first is the impractical approach adopted by those attempting to implement the technology. For example, if you have a group of computer scientists asking operators to shut down production lines, in order to seed failure modes to train algorithms, it’s never going to happen. The second is an over-reliance on experts. If that’s the case it’s never going to be scalable. What we are now moving towards is scalable solutions that are non-intrusive and easy to roll out.
If you have a machine that is already rich in data, you can gather the information and train an algorithm to identify what normal looks like. You can then monitor the machine to identify when you have an anomaly. When it detects a problem it can then alert the maintenance engineers, so they can resolve the problem before the failure occurs – thus, maximising availability. This gets exciting when you can apply it to standard pieces of equipment, like conveyor belts used in modern warehousing.
The big one is cultural. When you have machine learning being applied to engineering maintenance, you are bringing the computer science and engineering communities together. They are culturally very different – and up until about 10 years ago the idea of them working together was unusual.
It will take time to build trust and for engineers to believe what the data in the black box is telling them. For example, the algorithm might identify an anomaly – a vibration caused by off specification bearing – but following inspection it may be incorrectly determined that the bearing is within specification. An engineer might then ask, why do we need to replace the part? But, if the bearing fails the repair cost of the consequential failure could be more than 100 times the cost of the part.
What the holy grail will be for OEMs and operators / manufacturers – and I know companies like SamsonVT are working towards this – is an accurate prediction for ‘remaining life’. This is the estimated time when a part will fail. There will always be a big challenge with that, as those predictions will depend on the future load and environmental conditions. In some cases due seasonal demands, assets can be loaded higher than average and this can negatively impact the useful remaining life.
"What the holy grail will be for OEMs and operators/manufacturers – and I know companies like SamsonVT are working towards this – is an accurate prediction for ‘remaining life’. This is the estimated time when a part will fail."
- Dr Moray Kidd
I can see OEMs really embracing this technology. But, where we currently have black boxes being integrated post production, in the future they will be working with companies, like SamsonVT, to embed this technology during production. In the past, manufacturers have resisted this approach as it adds costs. But, as more companies embrace a service-led approach they will want next generation predictive maintenance to be integrated during the design phase.
You can already see this happening in automotive and aerospace. Where historically automotive manufacturers would build a car then allow independent dealerships to handle sales and maintenance, you now have electric vehicle companies providing the full service – from production to maintenance, and even power. Many manufacturers are moving more towards the ‘power by the hour’ approach. The big aero engine manufacturers don’t sell assets anymore, they sell a service provision and engineering maintenance is integral in this business model. Perhaps the future of assessment management will shift much more to an OEM cross sector complete service provision.