03 April 2017 - ABN AMRO Innovation

Bart Vredebregt | Aiir Innovations | Innovation Friday Centaurs | Innovation Centre ABN AMRO

Innovation Friday – the monthly 'open house' of ABN AMRO's Innovation Centre – will take you on a journey towards the future of the financial sector.


no history lesson this time so let's just get started everybody here has flown recently probably last year and this is what you want to end up place just sitting at gate no problems but it can happen that you end up with this this is an engine that has a blowout and it can happen when the turbine blades are inside the engine and are not inspected properly there can be cracks in their dens and other defects and during maintenance they actually so your maintenance to take out the engine they have all the time in world to do so but inside they are a little bit and they can see them right now so they have two options take all engine apart the text three days to for us births cup this or kind of camera on mistake to inspect the engine that looks like this so this guy is looking at that screen together with colleague for about 24 hours to inspect the engine for cracks dents and other stuff and during the time it's really hard to stay focused all the time it's easy to miss something so let's say my neighbor walks by and I say hi and then I could have missed the thing that caused the blowout we have seen earlier so that's one thing a human is really

bad at staying focused a long time and don't get distracted and yeah paying attention to the small details but actually a computer is really good at that it stays focused the whole time and never gets tired it never gets distracted so that's the perfect deal so then you say okay let's do it completely automatic that's actually what we're aiming for in the end to do so then you see this inside of the engine and the system just says like okay I found some anomaly on the top of light and defects at the bottom perfect yeah nice a theory but for stuff like this for deep learning actually need a lot of data so how do we get here in a safe way because we can just say yeah we did this little experiment we sent the blend into the air and now it crashed happens good luck next time so basically what they're doing right now is they just make video human leukocyte video and writes little reporting birds like everybody else does it so that takes up to 24 hours and right now what we're doing is okay let's first help human do it's a little bit easier okay skip the part where you need to enter stuff in words just optimize

the workflow just make your life easier and what mechanic is doing at the same time is actually helping us train the system because mechanic is saying oh I see something here and system is like thanks thanks that its use that for later on it doesn't say anything yet it's just there in the background learning like little toddler that looks over your shoulder in fact what you're doing and at some point when it grows up we can move to the next stage where we actually do the following thing we use the software to look over and shoulder or and XE say oh we see something right here I start really working together that's the perfect team because the human is really good at judging in the cracks the defects and other stuff it finds but the computer is really good at finding everything so it will never blink it will never get tired and all the other stuff making and in fact that you actually only one person looking at that screen together with software and then you will find best every damage and then in the future it might be the case that you move on to the third phase but that actually up for discussion right now doesn't make sense to completely

remove the human form loops so Chad you get to a point where we best make video the software does all the work send a PDF through your desk the next morning you're like okay cool that's it then we actually get to the discussion of the human scent hour or is it some sometimes it's actually better in my opinion for a human and computer to work together in harmony then to have the computer do all the work and then create a lot of extra work in the end because computer will say like oh dear you have ten dollars in damages and there is maybe then them it is actually in there so then it doesn't work out so it might be the case that this is the way to go and then it takes eight hours instead of four hours which began business it could be the guys that we will end up doing this for the smaller inspections like you know there might be something here let's do reconnaissance fears and doing this for the real deal now but where are we doing this we're actually working together with Calum right now to develop this stuff they are really helping us to make this work because as a start-up in the aviation industry it's kind of hard to fit an engine in it inside small

office of started and actually get your ends on it but that is really cool that they are helping us with this and especially if you consider that we just started a year ago with five students and a professor that were like ah this computer stations that it's really cool but we need to apply it somewhere actually our professor moved to the day you delve but he still evolved income needs really cool you have improved from the most artificial intelligence and I'm doing risky stuff by showing you guys numbers about sectors but you guys know that it's basically the fact that the industry we are working it in at one big issue the number of experts in maintenance is staying about the same but the amount of airplanes in the world is doubling in becoming 20 years so you have two options or a lot of people how to inspect act engines or you use opportune our software or other solutions to actually solve this issue for once am pro pro and that's good thing for us because then we can actually grow a business and try to get maybe sensitive that yeah for 47 billion yeah it would be nice but you never know so we're taking off and thank you for

your attention you

XML Transcript: