Artificial Intelligence Helping Us Work Smarter
Words by Nikki Stefanoff
Photos by Björn Rust
This content is produced in partnership with Monash Materials Science and Engineering
Word on the street is that the robots are coming, so look busy. But are they really going to leave us as bewildered and jobless as we think? According to Monash University’s Will Nash, that might not be such a bad thing.
They say that the best way to find a gap in a market is to experience the void and be sufficiently annoyed by it that you’re motivated to do something about it. Will Nash from Monash University’s Department of Materials Science and Engineering felt both – tedious condition assessments for structures such as large buildings and bridges are the driving force behind his work. He’s already deep in the planning stages for this technological takeover, hoping that he can hand over this arduous, but also dangerous and expensive, work to our future ’bot colleagues.
“I finished my undergraduate degree in materials engineering back in 2006 and then spent 10 years working in the industry as a consultant engineer, something I still do on a casual basis,” he explains. “I worked for AECOM in Canada before coming back to Australia and working for hydro-electrical SMEC, and then onto a very small consultancy, called MEnD, where I still work. My career involved doing condition assessments for a lot of different structures like large buildings and bridges and I very quickly found out that it’s part of the job that can be very long and uncomfortable. It’s so tedious to have to go and map all the cracks on, say, a bridge and while you’re there you’re always saying to yourself: surely this could be automated?”
Turns out Nash wasn’t the only one thinking it and after bumping into Nick Birbilis, head of the Materials Science and Engineering Department, and chatting with him over a beer, Birbilis told Nash that he’d been looking for someone to work on a PhD in this very area.
Nash began work on his PhD in 2016. “What I’m doing at the moment is teaching the computer to spot corrosion using AI methodology,” Nash says. “In layman’s terms, what I’m trying to replicate is the technology behind how Facebook, for example, can recognise faces in all photos. Apple has the same ability built into its photo application. I want to use that same technology so we can get to the point where we can fly a drone over a bridge, scan it from the air and have it give feedback on where the corrosion is.
“The problem lies in the fact that a face has distinct features, which are arranged in a fairly distinct pattern that for the most part don’t change. If you were asked to draw a face you would have eyes, nose and a mouth because, well, a face is a face. Corrosion, however, has this problem where it doesn’t share that many features and under different conditions, different light for example, it will look different. It has a distinctive colour, but even the texture can change depending on how long it’s been corroding or its underlying sub state. If I asked you to draw me corrosion, you wouldn’t be able to. Even corrosion engineers would find it hard to find a representation that a computer would find useful.”
Nash isn’t the only one researching the concept – to get this right could be a real industry game changer. And while there’s no-one out there doing quite what Nash is, the closest parallel to his work would be what’s happening in the medical research field surrounding imaging for brain scans. “There is a lot of work being done with medical image segmentation,” he says. “The similarity between what they are doing and my research lies in the fact that no brain tumour is the same. They don’t tend to have a set shape and finding them isn’t that easy. The medical industry will need to teach the computer to work in the same way I am trying to. The hard part comes in how you are teaching the computer to recognise things, whether a tumour or corrosion, because if you teach it the wrong way, it’s never going to understand what it’s looking for. It’s a problem it just won’t be able to solve.”
Nash’s intention is to solve this problem by the end of 2018, making for an incredibly fast-moving PhD. But once he’s worked out how to detect corrosion, there’ll be no stopping him. “I’m keen to look at other materials we can put AI to and think more about how we can use reinforcement learning to develop more efficient structures,” he says. “The future looks limitless and it could be that we end up training robots to build structures from scratch. If we could use a computer to design a whole building then it could revolutionise both the architectural and building industries, although, it could take the fun out of them! I think that the problem at the moment is that designers still design the same way they always have, which is down to habit, but it then results in a limited mindset. Unless, of course, you’re Frank Gehry who has pushed all the boundaries on building forms that don’t follow the orthogonal form.”
Nash puts it all down to working out the right reward for the computer if it does what it’s supposed to do as well as a punishment if it doesn’t, which is a similar motivation system to humans – and somewhat terrifying. But unlike humans, the main benefit is that AI doesn’t get tired or bored doing tedious work and could play a huge role to keep people safe. “I gave a presentation at a corrosion conference last year and my first slide was a selection of selfies I had taken in the weirdest places I had been sent to do corrosion assessments,” laughs Nash. “If we can get automated corrosion detecting right, it could have a real impact on those in the industry. It would mean that you’re not putting someone in a precarious position when you send them out to do an assessment, which automatically reduces the risk to the inspector as well as cutting costs. This would mean that more money could be spent on keeping the building as safe as possible and that could make a huge difference to the parts of the world where safety checks are kept to a minimum because of affordability.”
“I’m keen to look at other materials we can put AI to and think more about how we can use reinforcement learning to develop more efficient structures.”