Integrating artificial intelligence into structural maintenance and management

You're listening to The Engineers Collective, the podcast by New Civil Engineer. New Civil Engineer is the leading publication covering the Civil Engineering sector, a trusted and authoritative voice for over 50 years. The Engineers Collective comes to you with monthly news analysis from our editorial team and in-depth interviews with industry experts and thought leaders. If you're a private sector organisation looking to reach tens of thousands of listeners, you can find out more about sponsoring an episode of the podcast by visiting newcivilengineer.com/podcast. Hello and welcome to The Engineers Collective from New Civil Engineer, I'm your host and Ian and I'm joined by our senior reporter Tom Pashby, how's it going, Pash? Good, thank you. How are you? Yeah, not too bad. Friday again, thankfully. I don't want to be Friday, but it's time you listen to this. Well, it might be next Friday. I mean, that's Friday anyway. I'm not going to be a good afternoon because it won't be an afternoon, well, it could be morning. It could be the middle of the night. Maybe people listen to this too. Yeah, to get to sleep. Anyway, coming up, we have my interview with Mind Foundry, Director of Civil Infrastructure Tom Bartley. Mind Foundry is kind of a start-up working in the AI space that has a focus on infrastructure and is doing lots of interesting things and bending the use of AI into infrastructure because actually when it comes to maintenance, so Tom spoke very well and very interestingly about that and for the people who are skeptical about AI, especially when it comes to infrastructure, I think it would be a good one to listen to to maybe get a good picture of what exactly it offers. So that's coming up. But before all that, we are going to talk about size well C, which has been in the news a lot lately. I mean, we, I think we mentioned when the government committed 14.2 billion to one was that a month or so ago. That was spending. Yeah, so that was mid-June and we've been waiting for the final investment decision or the Fed to determine who else is going to chip in some money to pay for this thing and we finally got it this week. The headline that has come out of it is the government kind of admitting that it's going to cost a lot more than expected. They put around 38 billion whereas the last cost estimate was 20 billion, they said in 2020. And the 38 billion is quite a lot of breakdown of how this cost is going to be covered which you wrote about. So I'm going to hand over to you to explain it all. Yeah. Yeah. It's extremely complicated partly because there's so many different government support packages to fund size well C. There were loads that were announced way before the Fed. The Fed should have been achieved, I think, before some recess last year, but then obviously we had the election. So very delayed, final investment decision, EDF's proportion of ownership has been squeezed a lot since then because every time the government has put in more money under their various subsidy schemes, it has meant that EDF's proportion of ownership has gone down. So, I mean, as of four days ago, the government owned something like 90% of size well C. And the FT kept on reporting different investors showing interest in putting in various amounts of money, various proportions of equity stakes. There was rumors about various different asset management firms putting in money from, I guess, private equity. Central Kerr eventually did commit 1.1 billion of equity investment, which gives them a 15% ownership stake and a few other investors have also involved. But the massive headline figure from my take as not a financial journalist was the 36.6 billion of debt financing that the National World Fund, which is a UK government arms length or independent body that is funded by the Treasury, put in size well C to get it over the line. And the whole point of the National World Fund is to put public money into projects that would otherwise not be successful because the private sector used them as too risky. That's the kind of fundamental USP of National World Fund's role, so that therefore implies that it was too difficult to get private investors to commit to investing in size well C without massive government support. And normally, I mean, I don't really understand the difference between equity investing and debt financing, but normally when the National World Fund gets involved with the project, they want to fund it like 1-3 as in the National World Fund putting in, let's say, 10 million of investment and then the private sector putting in 30 million of investment, but that's not the case with this. So that's kind of a small crash course, I would say. I don't know how long I've taken there. What was that? Two minutes? Three minutes? Yeah. So I assume even though you might not know the answer to this, but I'm a bit boring, the National World Fund is putting up 36.6 billion debt financing on a 38 billion around 38 billion projects, but there's also these private companies involved. So is the National World Fund going to recruit some of the debt financing? Or is it just, is it only going to go into that debt financing if the project over runs? I mean, there's maybe two or three questions there. I guess the short answer is I don't know, but I'm pretty sure that the assumption is that government, all the parties involved in investing in size will see hope to get return on investment. So they hope to get the money back and a bit more. The normal rate of return, I think, the National World Fund aims for something like 3%, I might have just pulled their figure out of the net. I did speak to them recently and I'm going to be writing up an interview soon about their kind of, their ways of operating and that was actually the day before the FID. So we didn't talk about size, we'll see. And of course, all the private sector investors are hoping for a healthy return as well because they wouldn't be putting the money in without their expectation. I don't, as I said before, I don't understand how debt versus equity works. I think that the whole thing is quite exotic as in it's not something, it's not like a normal type of investment because there's so many different inputs and outputs and ways of making inputs happen. The government and size will see said that the cost to consumers will be £1 per month on bills. So the cost to bill payers is that. That's the regulated asset-based model through which they'll get the money back that they've invested. Yeah. So there's the regulated asset-based. There's about four or five different names for different budgets that are involved in size will see. And the regulated asset-based is one of the big ones. But then down the line, size will see, and the government have said that if size will see as completed, it will save the energy system £2 billion a year. I don't understand how they've come to that figure. They have released some documents like the value for money assessment, but I'm not sure if they've actually explained at what point consumers will be better off as a result of size will see. When it will change from being a cost on bills to bills becoming cheaper, yeah, I don't know if they know that. And I'm pretty confident they haven't told the public about that because they might not tell us no. We should just say the other shareholders are LaCasse who's taken 20% centrica who owns British gas, taken 15% amber infrastructure with an initial 7.6% and EDF with 12.5% and the government has 44.9% state, which is a minority, but it is the biggest single shareholder. Yeah, so actually about the proportion of public ownership, I think that there was an expectation that the government would retain its majority stake, and I know that campaigners wanted the government to maintain its majority ownership to, I guess, in their eyes provide more democratic oversight of the project. And the last thing that I could find in terms of government statements about public ownership was a minister of state from the Department for Energy Security and Net Zero in the House of Lords, and I think it was either March or May this year saying that the position was for the government to have a majority stake. So the government, as far as I can tell, hasn't really explained why it's allowed that to happen because, you know, in the scheme of things, a few percent isn't that much compared to how much they've already put in, especially as the amount of public exposure that there is to all the risks. So yeah, I don't really understand that. It really feels like they were juggling deals right up to the deadline until they announced it, but that's just my layman's perspective. Before we go even further out of our adapted finances, let's claw it back towards something we understand. And you've also done another piece on size while looking at, although the final investment decision has only just been reached. There is actually a lot of construction going on on site and you did kind of an interesting piece that hit satellite imagery from three years ago this year and seeing how much has changed. Yeah. It's quite dramatic. If anyone goes on Google Maps or preferably Google Earth and looks at the size of our site, which is near a town called Lacedon, you can really obviously see all the activity that's happening. It's kind of like when you look at any part of the root of HS2, you can see all the kind of yellow areas where top soil has been removed. So I kind of combined these before and after images of the site with some graphics that were produced by size will see itself, the graphics from size will see show that the main areas are the main construction area, which is the area that's just north of size well A and size well B. And they need to bridge a gap over an SSI where there's two waterways that will then connect with an area called the temporary construction area. And that's the really big yellow area that's visible. If you zoom in on maps or earth, you can see girders and vehicles and roadways and water storage areas that have all been made. And then as you zoom out, you can see to the southwest of the temporary construction area, there's a kind of spur that comes off, which is, I believe, highlighting the future root of the rail line extension, which is going to be built, which tends to be of interest to lots of civil engineers. And then in the south, you can see the and silire a construction area, which is where the size will see administrative office is. So this is going to keep changing. Of course, I do hope that Google flies over again and takes a bunch more images because it will be interesting to do a further comparison before construction is finished to see how everything progresses. I also hope that size will see itself might release some more like drone footage, because I'm pretty sure they have that. So it's really interesting to see how all of the bits kind of come together. Yeah. Rather than go on Google Earth and Google map yourself, I would suggest going on patches story on all that. Of course. All that, of course. But the slide is because it does actually works really well to see how much has changed. Yeah. Yeah. Hopefully Google Maps will go over again before another three years, although this project is going to be decades probably in the construction. So we'll get many more years down the line. And yeah, hopefully now that they've got the feed, they've got the green light and everything they can start releasing more about what's actually happening on site. The other big thing that's happened, even though, as we said, there's already lots of construction happening. They hadn't actually appointed a main contractor for constructing the power plant until recently. And they've named Belfabeti, where we can travel publics and Lengor Rock as the civil works alliance will work together to build it. All three of those contractors are already working on Henkley Point C, which is basically exactly the same design as size will see. So it makes sense to just bring them all together again and put them on the next one. Hopefully, a lot of the same workforce will go over. And hopefully things will go a lot more smoothly than they have at Henkley, you know, if life has a wave, like the best lay plans of mice and men, all that nonsense. But it should work, isn't it, Erie? That's what they're saying. Yeah. And we have to trust them. That's an extent. Anyway, yeah, and there's been lots of other size well things going on and there will be many more going on. But thanks, Pash, for that quick rundown on the recent peak developments there. You're very welcome. And listen, if you are a magazine subscriber or a website subscriber, our new issue is out. It's a road issue. I wrote about the A417 missing link. Even that's our road issue, Pash, you did a big deep dive into SMRs and what they might take to actually build them. Yeah. The link was that I was hopefully going to be talking a bit about roads, the roads that lead up to SMRs and the kind of perimeter roads that you need for the, you know, civil nuclear canstabbery's Jeep to drive around on. But I can, I was struggling a little bit to pin people down and telling me exactly what a civil's work is needed because people just don't know really. Yeah, it's all a myth here. It's all a moment. And it's not like you spoke to like chances, you spoke to them, we're doing a little Balfour BT, like the big name, so the people who are involved in these projects don't even really know yet what it's going to take, which is kind of interesting. No, the thing, another thing I was trying to do was to pick out the differences between how it will work on a nuclear brownfield site like Wilfer or Oldbury versus a non-nuclear greenfield site, because if you look at the renderings produced by Rolls-Royce SMR, it's very lush, like a green, like tele-toppy land rolling hills. Yeah, and the new, this is getting into it a bit, but the new National Policy Statement, looking at nuclear sighting, opens up more types of site. So there's a wide range of possibilities about what civil's will be needed, and I was attempting to work out what that is, but it's quite difficult because no one really knows. No, maybe in a year or so, they might be a clearer picture. Anyway, it's still a really interesting piece, so you can see that in the magazine online now, as well as some other good stuff. Alright, thanks again, Pash, and to the list coming up in a second, you'll hear my interview with Tom Bartley, all about AI and infrastructure. The Engineers Collective comes to you with monthly news analysis from our editorial team, an in-depth interviews with industry experts and thought leaders. If you're a private sector organisation looking to reach tens of thousands of listeners, you can find out more about sponsoring an episode of the podcast by visiting newcivilengineer.com forward slash podcast. Alright, welcome back to the Engineers Collective, and now joined by Mind Foundry's Director of Civil Infrastructure Tom Bartley. Hello Tom. Hey Rob, great to be here. Thanks for joining me, so Mind Foundry, let's get the spiel, who are Mind Foundry, who are you? Let's give us the lowdown. Okay, great, so let's start with Mind Foundry. Mind Foundry, a machine learning company, spun out of the machine learning research group at Oxford University about 10 years ago, our founders still are active professors at the university, specialising in applied applications for kind of all the emergent tech and making sure it works for the real world. The founding use case was all about predictive maintenance of mining equipment, and 10 years ago that lots of data and sensors coming off when it's the right time to repair, replace, et cetera, parts. And we've been on a journey since, there's about a hundred of us now. We organise it into kind of two verticals, primarily, so we have a defense and national security vertical, and then I head up our infrastructure and build environment vertical. So yeah, we're about, as said, about a hundred of us, sixty of us have our kind of engineers or data scientists and really about the applied application of machine learning into kind of the real world, real world situations. And then your job is directly sort of civil infrastructure, and what is the overlap between what Mind Foundry does and infrastructure? Sure, so yes, I'm Tom, I'm a civil engineer, I've kind of come through a kind of a very digital career path, where I kind of started my career working with an engineering consultancy, doing a doctorate in how to apply BIM to major projects, and that's kind of spent. It's been a convoluted route through various startups and kind of in and out of consultancies, and now I'm with Mind Foundry where I joined about fifteen months ago to establish a new business unit for the infrastructure vertical. The key thing that we are like working on, our mission is about the world's most important problems, and I'm sure you'll listen and recognize that one of the most important problems we have now is the age of infrastructure and how we care for our aging assets. So we are building products and solutions to help asset owners understand, first understand the condition of their assets, getting to much more quantified, less subjective, more consistent definitions of what condition are assets are in and then using that to develop predictive models that will help identify what we call the divents and maintenance renewals are required and when the right time to make those is. And so yeah, we've kind of at the startish of our journey, we launched our first product about a month ago, it's called Windwood Inspect and there's an inspection application for bridge engineers to collect much richer information about the quality of the, so the condition of our structures, and crucially couple that with the kind of short term benefits in terms of workflow optimization and improvements. Nice. So we'll definitely dive into that, but let's start with the root of the problem which is current maintenance regimes, how do you see, what do you see as the main issues to how we currently maintain our infrastructure? So I think we had a perfect example come out over the weekend around the bridges on the M6 in Cumbria and the major renewals that are going to be required six years of disruption for what could have been a 10 to 50,000 pounds deck joint renewal. It's really hard for asset owners to understand what's happening with their structures that are out there in the real world, and across a large portfolio, identify the work, prioritize the work, get the funding for the work and actually implement the work, and we're at a point now, as I'm sure your listeners are aware, it's been on the pages of NCE a lot recently, where we're kind of approaching what we call a condition crisis and we need to really start addressing that and be proactive and doing things the same way is not going to be the way to do it, we can create, and a kind of at-mind foundry our proposition is using artificial intelligence to assist with, as I say, understand the condition and then identify plan, prioritize the maintenance in a really proactive sense to save money of the life cycle assets, but crush the key part on networks operational. Yeah, is it an issue if we don't have enough manpower to do as much regular inspections we need? There is a key challenge around, I think there's been two key trends that have happened over the past 10, 20 years, one is that we've seen a shift in the way that asset owners manage their assets and a shift away from in-house, what we call custodians looking after their assets and having a long-term perspective, the job of an inspection and asset management has become, have been outsourced a lot, we've engaged consultancies to both do the inspections but also do the assessments and the appraisals around that, and that has the benefit of accessing expertise and knowledge transfer, getting the latest innovations into the asset organization but the challenges that we've lost that long-term perspective of what is happening with the structures and particularly individuals who say, "Oh, I know that bridge when it was constructed and I've kind of followed it all the way through and I remember when we did those maintenance work 15 years ago and all that one's been a bit funny." So that's been a big change away from kind of a personal responsibility. The second key change is just the aging workforce and again, kind of a major, kind of a bigger trend across the industry but we are seeing our inspection workforce is reaching retirement age and therefore the reason that there's a real kind of constraint on the availability of inspectors to actually go out and do that, I don't think, but it's kind of against the fundamental part of your question, Rob, I don't think it is a personal issue. I think whilst they are kind of things that have happened, we're really kind of talking about finances and the government and kind of budget holders not prioritising the requirement to collect this information as a thing that they need to do and so in a world of unlimited resources we just be able to hire the inspectors and train them and kind of have a continuation. So there's something much bigger in terms of how we compel funding to recognise that this is a problem and do something about it. Yeah, so it is quite a critical issue. So AI potentially has the solution or a solution to help with these constraints. So how is Mind Foundry kind of approaching it? So yeah, we think it's part of the solution that certainly human AI collaboration is one of the kind of the key words that we use and so it's not replacing engineers or inspectors, it's about augmenting their role to make them more efficient and more effective in what they do and there are two parts that one is the two parts. So one is the capacity to kind of undertake these works and that's two for one of the people but two do we have enough, if you're in the railway, for instance, actually getting the possessions to be in there and actually look at the bridge is quite a narrow talk about time. So actually you've just got kind of a hard constraint around that and the second is about the quality and the consistency of the data and so we kind of seeing opportunities for artificial intelligence to assist with both of those, both of those pieces and I guess we, you know, innovating in this sector is really hard and particularly making kind of, we've got to balance short-term benefits with long-term outcomes and so we kind of see the workflow automation, efficiencies as a key part of it and then the outcome being the data and so how we can assist with data collection, get much richer data from the site, what we, those have been involved in inspecting and maintaining assets, it's fairly consistent across, I'll talk about bridges or not but it's I think it's a fairly consistent approach across different asset types that someone will go and look at a bridge, their job is then to try and find what might be wrong with that bridge and then document that down and ultimately what comes down is whatever, whatever notes and photos were taken with a very kind of simple scale of one to five of severity, how bad is it and one to five, how much of the element is affected by that and that kind of if you think about the complexity of the real world being boiled down to these very very simple measures we kind of lose all that detail in the process and then we have the challenge that what is a three to one person is a two to another and a four to another so we kind of have then this inconsistency and subjectivity that comes into it and so we think artificial intelligence plays a big role in containing a lot of the context to the condition, understanding what's actually happening and not having that huge information loss to the inspection report what goes into the asset management system but also then assisting to say actually when each person describes something as a three A or a three C they're doing that consistently across across the piece, now ultimately that condition information feeds two processes one is about actually identifying what the maintenance works are, what should we actually be doing and we see there's kind of a very quick big funnel between what gets recommended at inspection stage to what actually gets implemented and that forgets actions and then it goes into another process which says on a long-term forecasting perspective what's our budget mean going to be in three five three five ten fifteen years time and using that information for kind of portfolio level deterioration. One of the key so what one of the key challenges is so we have this kind of what gets recommended what gets implemented this funnel but if the data that comes out to inform the long-term budgeting there's no connectivity really in that in that deterioration modeling and forecasting back to the individual assets and individual pieces of work so whilst with the law of big numbers and averages the deterioration modeling at a portfolio level averages out we don't then say we've allocated the budget and we know exactly what works when we to fund it then comes back to a human to then say okay here's my budget one we're going to spend it on this financial year and so the kind of the bigger vision for us is really about that downstream piece about understanding what interventions to make and how to make sure we prioritize the right preventive stiff works against against the other kind of needs of the network. So it all sounds quite complex but it actually is kind of simple who we're going to talk about it on a very basic level it's kind of like here's a picture of the defect three years ago here it is one year ago here it is now the AI can say this is what's happened this is what needs to happen basically that's I mean that's what it starts with it starts with photos 90% of asset condition information can be gleaned visually there are hidden elements and there are kind of other bits where we've got kind of advanced forms of scanning and kind of penitentious of tests but for the most part actually we can see what's happening and so we're starting with computer vision as you say we're up to understand change over time and just visually say has this got worse since the last time we looked at it the chances out of someone else that looked at it two years ago six years ago and comparing two photos side by side is really hard but with our image matching and our object recognition and change modeling. We've got this nice really simple thing that says, you know, our sprawling has grown by 11% since the last time we looked at it, it is actually getting worse. Yeah, so as you say, it kind of boils down into quite a simple use of functionality at the shotgun. And is this what you call digital custodianship? Yeah, so digital custodianship is our vision for bringing AI into the loop as kind of what we see. We see a digital custodian as being part of the team with with real engineers. We've got we've got an AI engineer the cost of the digital custodian really having that holding that long term perspective where we can start to understand what happened in the past and bring that into models that forecast the future. And we we've kind of we've worked with WSP on this vision for digital custodianship and you know, Steve Denton has noted that the kind of the changes that have happened across his career and as we get into this kind of crisis of around aging infrastructure, the need to kind of do things radically different and the opportunity and the technologies that we're talking about are at the cutting edge even three or four years ago, we didn't have, we think about our inspection application, cameras work good enough on mobile phones to be the primary source of data capture on site. Now that they're there, we can we can start to really clever things with how we locate the effects and assets, etc. And so yeah, digital custodian digital custodian ship is a is a vision paper that we published Corridor and with with Steve and others at WSP and it's available publicly to to kind of engage the industry on this this journey. It doesn't it's not a proprietary mind foundry or WSP thing. It's really about kind of do we need to think about this differently and how can we be engaging AI in a pragmatic way. Interesting. And people can find that by just searching. Yeah, yeah, I think if you google digital custodian ship we will be one of the top hits. So it's yeah and we really you know we're really open to a conversation I think we thinking ahead to tech fest later this year we'll be putting together kind of the next iteration of our thinking around digital custodian ship and leading up to that we'll be doing a lot more engagement with various asset owners and others in the industry to kind of ground that in kind of a real roadmap for the industry and kind of understand the real constraints. And so by the time we get to tech fest in November we'll be launching the next kind of the next publication that sits alongside that to kind of give some practical guidance to to the industry. Nice. So yeah you mentioned there about how phone cameras are now good enough to do asset inspectors or to take pictures and you mentioned that you recently launched the first phone app so tell me more about that and how it works. Right, thank you. So yeah, like capturing data from size and bringing it back to the office is both and as I've talked about that information loss challenge but it's also it's also very inefficient and often not very joined up and so but we're not the first people to invent an app for for going out and taking capturing data about the real world and bringing that back but we have got a different perspective I think in terms of being a camera first app rather than the format that allows you to attach photos and so yeah we launched last month and WSP again have been our early adopters and they've been working on rail assets and some highways with local authorities to to really help us kind of fine tune and now we're we're exploring opportunities with a range of local authorities consultancies et cetera across the industry. The it's a mobile phone app it's it's called Windward Inspect and the key thing is when you go out and take a photo it's all about understanding the context in which we're capturing the the defect information and the condition information so we should start by saying here are some context photos here's the here's a full photo of that span or here's the here's that abutment or whatever it might be and as you go through you would tag on that element this is where that defect is this is what the defect is here are the supporting photographs we we use kind of voice typing and and kind of language processing to make it really efficient to capture those notes and what comes out of the process is report automation so what would usually take days to take those site notes and refactor them restructure them from scribbles and apps and point-of-shoot cameras et cetera put them into a report in a really clean structured way pds aren't going away for archive purposes but we also then get this really rich environment that allows you to say look I can navigate the structure in a different way I can see all the elements I can see whether defects are I can see how they interact with each other I can upload a photo of the previous inspection that we did five years ago and I can see has it got worse since then and so yeah we're we're at the start of that journey now the key thing with with women's inspect is it's a workflow solution that makes inspections more efficient today and it gives owners and the engineers that didn't go visit the bridge much much much more context and richness about about the structure but then that all becomes training data for those machine learning models that I've been talking about for for for predicting deterioration and planning and prioritizing different works so that the kind of the whole system works as a it's kind of a self-supporting thing where we can then use the use of the tool then becomes training data that can then make the process even more efficient and unlock future capabilities for for the industry very interesting and people can go and find that now for free on that store it's not free sorry no but if you if you contact us we we definitely pace for itself every time you use it so be very keen to speak to anyone that's involved with inspections and feeling the pain of the data coming back that that is neither actionable or and it's just very slow to work with we are kind of I think is that it does pay for itself every time you use it and that that's kind of a key part of our pricing nice so this is kind of just like the latest step of AI's journey with infrastructure that Mind Foundry has been working on quite a lot from recent years I think you've done some some work with HS2 is that right well we we were successful with the what's called the HS2 Business Accelerator Program which is a a program designed to engage startups and de-risk startups coming in and working in what is very complex to the re-environment last year they changed their focus in the in the innovation space to be thinking about right that these things are now being built how do we actually maintain them how do we they're going to be stuck at the assets are going to start to be hand over so actually we responded with windward inspects and if you think about the life cycle of a structure or kind of any infrastructure there's there's kind of an early TV period and then there's probably 20 or 30 years where very little actually happens across and then we get to the point where we are now with with our existing aging infrastructure problem and so the key thing HS2 are going to have to adopt these adopt these assets and and understand the teething problems etc and understand all the hand the quirks in early life and so windward inspect was so that's just one of the the innovations that they wanted to explore with with that and we're we're still in conversation about about how we can support them as they start their asset management journey when delivery and delivery is finished so you know incredibly complex program so yeah we've been really fortunate to work very closely with the innovation teams and and across that asset management teams in HS2 to at this stage understand the opportunity and really kind of shape it for them interesting and and the what other areas of UX explored with with mind foundry and AI this is something with a insurance companies are interested in it as well yeah I think there's a huge crossover here and this is kind of visionary so so one of our strategic investors in mind foundry is Ioannis and Dawa insurance there are Japanese headquarters but global insurance company and and we have a joint venture in Oxford called the I the IOER and D lab where we kind of explore ideas and so there is a huge crossover with insurance both in terms of how we describe and classify risk but also visionary end of the spectrum like how we currently manage and kind of asset owners own risk is it's all really self-insured you know we kind of the taxpayer ensures our bridges and and and actually there's this kind of an interesting thing to say well it's currently an uninsurable asset class because we don't just we just don't have the data to understand what conditional structures are in or what risk is associated with it what we want to start using tools like windward to to understand condition do we under do we unlock new categories of insurance for our for our clients and so we do have some work in Japan where we are using the same inspection application to engage with local authorities and to start thinking about other new models. I in this idea where leaders in the use of telematics and cars to adjust your premium as you will be driving and so that the question is can we create a similar black box for infrastructure that would really reduce the risk that asset owners hold and manage that. So you can imagine that's a very long-term piece but we're fortunate to work with a visionary organisation to be kind of exploring the only steps. It all comes down to data at the start and so we kind of it turns out we still need to work out how to do bridge inspections and or asset inspections to get the data that support that. Yeah but it has some potentials to be very beneficial for insurers and asset owners alike. Absolutely I mean I think you just you just see so many asset owners are sitting on a risk that could be catastrophic for the network and hopefully not in day-to-lives but because we tend to catch failures before they before they act particularly in terms of bridge structures we kind of see the capacity is reduced before we get to collapse. We're fortunate not to see many collapses particularly in the UK but if we could change it so that actually there is a financial instrument that means that the owners had certainty about how much they'd be paying and you know that some of that would come with some predicted you know it has to include prevent stiff works you can't just allow your structure to get to the point where it falls down and the insurer will replace it. The kind of the thinking around in an actuarial sense could be transformative for owners and particularly taking that risk off their balance sheet or off their conscience and making sure that you know are you sleeping well at night well we know that we were ensured and the product means that we are going to keep on networks operating safely. Pretty interesting and in some of the other applications you've been working on a belief you did some work with Sir Robert Macalpine on the A19 is that right? Yeah we did so we again this was a project co-funded with the Io Ennis the the R&D lab and with the National Quantum Computing Centre and so there's a big challenge around maintenance about how do you actually how do you actually fit it in in your working windows how do you make it work and so this project was with the A19 Sir Robert Macalpine DBFO and they have got amazing records spanning pretty much the whole length of the DBFO and they were kind of very early to digitize their records and so we've been able to use those records retrospectively go could in this case quantum computing but could AI based planning assist with this complex basically it's a packing problem in terms of how do you do all this maintenance within the windows that we have and against all the different maintenance activities that we have hugely multidimensional like literally on a map but there is this kind of a 2D map that you have to fit it on to but then all these different criteria and so yeah we did a project earlier this year with with SRR I'm using their data to demonstrate how different you know kind of emerging computing and this is kind of bringing together machine learning based planning with emergent computing capability into the form of quantum computers quantum processors could could really kind of transform how many works we could fit into the same window for the same amount of money but can we bundle works together so that if we do take possession or we have a road closure these different these different works can happen at the same time and and we can fit them together in a more cohesive way and I think it's really part of the answer in terms of we have to be more proactive with our maintenance and we have to find ways to fit more work into the same amount of capacity that we have and we're Michal fine interested or are they impressed by the results? Yes certainly I mean it's the end of the dbfl contract so it's kind of we're looking retrospective to see how we would do things differently next time it's certainly proven the concepts and we're still in conversations about how we take it forward to the next steps there's this kind of a real nice alignment for these term maintenance contracts with or term you know the dbfl and that kind of the alignment of of interest and so we haven't yet found the actual one we can deploy it with but we're certainly having conversations about how it takes the next steps? Interesting so considering all these varied and potential applications for AI and infrastructure there also arises questions around safeguarding its use which is something also that MindToundry is involved in working with the advanced research and invention is it invention agency? Invention agency yes this is Arya for short so what can you tell me about the work you're doing there? Sure so if we're going to start relying on AI we're going to start bringing it into the loop you know everyone will be kind of aware of examples of hallucinations or you're using a predictive kind of a probabilistic approach to making decisions around maintenance and you know kind of the whole the whole kind of range there's a risk there's a risk in a probabilistic approach that it just has a blind spot and it just makes a mistake classic computing has techniques called verified verification systems like really strong testing and robust frameworks and like making sure it can't fail with AI it's much more complex because we're not we're not putting in the input rules by which you can pass off any other test we're providing this data and the same predict these kind of outcomes and so Arya the advanced research and invention agency have they are they're a government agency that they're fairly young and basically where innovate UK is really good at innovating at near-term near-to-market solutions Arya is much more about bigger bets that require first principles research and safeguarding AI is one of those projects that they have and that's kind of not just about infrastructure but that is just if we are going to be putting AI to safety critical applications maybe that's kind of the operation of kind of nuclear plants or taking decisions around medicine or whatever it might be how do we do that how do we do that in a verifiably safe way and they have this safeguarded AI program and where there's kind of two sides to it one is building the safeguards for the process and then the other is having real world applications for AI where safeguarding is appropriate and so we've been successful in working with Arya on on a project for safeguarding the digital custodian and it we we've just finished water two or 15 month program to to understand where we would need to implement safeguards to to the digital custodian and particularly we've been it's been fantastic we've had we've got a steering group that's got some really ambient people on it so we've got Julian Staden from Network Rail who is in the technical authority and we have Rob Engle from Leeds City Council who is the bridge manager there and we've been working across a number of other consortancies and local authorities to and authorities to kind of get their data and really kind of start modeling how can we predict deterioration structures how can we say you know you don't need to do an inspection the next five years because nothing is nothing is off note if you're going to take those kind of decisions based on AI you want to make sure it's safe and so yeah it's been it's been fantastic it's a it's a it's a it's what they call the feasibility project at the stage in terms of the opportunity to do some really deep R&D with our machine learning scientists collaborating WSP's engineers getting really deep into that the different types of modeling and when you think about the big Silicon Valley AI companies doing deep research and inventing new forms of science that's what we're doing here for for aging infrastructure hopefully we'll be successful for phase two which is actually to implement implement the the ideas that we've been working on but in the meantime we are you know really getting into the data really kind of getting deep into here's 20 years of bridge information from from a lesser owner and what can we gleam from that in order to predict what will happen next using you know the most advanced data science methods nice it is quite encouraging to hear that this is going on because the government at least here in the UK is really going all in on AI talking about mainlining it into our the way the country works so it's good to hear that there are safe guiding procedures underway as it's quite important I believe. It's hugely important particularly what the you know safety critical applications where there is a risk to life or risk populations that we we can't blindly start trusting AI systems and and as the world gets more complex and the kind of the number of criteria that we have to deal with increases are kind of human brains ability to cope with those is becomes more and more stretched and so yeah like the the academics that are on the other side of the program which are inventing these ways of safeguarding, you know, a whole range of artificial intelligence technologies. It's really fascinating and it's a really strong commitment from the government that the UK is going to be leaders. As we have been with engineering standards and, you know, we kind of have a role in the world to be in the responsible, you know, bringing kind of responsibility and kind of growing approaches to things. And so, yeah, it really compliments our strategy, I think, as a nation to be leading on, not just, could we get a computer to do this? It's, could we get a computer to do this in a reliably safe way? And they're kind of an extra order of magnitude in terms of the complexity of the question. But from a civil engineering context, can you trust the computer or not? If you can't trust the computer, you're not going to adopt it fundamentally. You might kind of pay for license fee, but we'll just end up having engineers taking the decisions anyway. And we need that collaboration to work if we're going to address the scale of the challenges that we have. Has it been quite simple or complex, determining where the safeguards need to be on when it comes to the built environment? Well, so the advantage of this project is they call it a curriculum of problems that we have to develop. And so, we've done some really kind of extensive value stream mapping across all the different decisions that are taken from inspections and kind of capturing that asset information right through to the decision making of which interventions do we make? And how do we prioritize against all the other interventions? So it's, it's a, what we call a, is a, is a curriculum of problems? We have the kind of the approach is kind of limited to where is that data today? Like, if we're going to do like demonstrators? And so we, part of that curriculum, we do need to create models that, that kind of prove the ideas that we're working with. And so, there's been a bit of kind of finding the synergies between what's available today and, and, and what the curriculum says the future requires. And then the other parts of the mind find your own map are about making sure we can enable those, those kind of longer term, those longer term decisions. So it's been, it's been, it has been challenging actually, because I think we, that, that, when you constrain to the existing data, you, you then kind of, there's a risk that you constrain yourself to existing ways of working. And so that's been, that's been the balancing act. But we certainly have found, we, we're doing some predicted modeling around state, state transition in bridge conditions. So at what point will this go from A to A to A to B or whatever, whatever the numbers are? And that's been network rail data. It's been really fantastic to, to be working with them. And as we go forward, the, the kind of, the richness of the data and how we combine data sets from highways and local authorities and bring that together into kind of a master data set that is, is big enough that actually we can start to train reliable models. One of the key challenges in infrastructure is no individual owner has enough, has enough bridge stock, frankly, or enough asset stock to, of, here's the material type, here's the construction type, here's the, the duration, here's the age, et cetera, to train reliable models. And so this kind of a scale thing that needs to happen here in terms of industry pulling together and creating a, a much larger pool of data that we can train, we can train these models on so that everyone gets to benefit. Interesting. So it's still a long road to go. So the long road, slow and long road. Yeah, I think. But we're, as we've said, the conditions are, the conditions in 2025 are right, we've got computing power and urgency in terms of the problem space are kind of, are, have aligned. If you know, if we were at this age infrastructure challenge 10 years ago, there's no way, no quantum compute didn't exist in the, in a way that we could just tap into it and other techniques as well. So, and, and the fact that asset management systems have been deployed for about 20 years now. So there is 20 years worth of data that we can work with, for instance. There's an alliance of stars that mean we can, we can certainly accelerate where we want to be able to, yeah, even two or three years ago. So, clearly, there is a lot of AI use in the built environment happening already and then it's a constant constantly evolving and there's probably going to be a lot more going forwards. But some people might be a bit resistant to the idea, especially those who've been in the industry for a long time or those who have doubts about AI, you know, what, what it's good for. What would you say to the people who are very resistant about this development? Got, first, like I sympathize, it's such a fast moving field. I think there are what we see as very, very capable advanced off the shelf solutions that have generated a lot of hype with a very, very shallow basis of application. You know, there's, the, the, the scaled AI approaches are, you know, based on, based on things that are globally available, you know, video or kind of text processing, et cetera. And so, there's a real challenge here with the engineering depth and engineering knowledge that is required to actually apply it to engineering applications. So, I really sympathize, but I think there are a whole host of companies, both startups and companies, vendors that are already at scale and, you know, there's big data science capabilities in a lot of organizations now that are proving that there is a role for advanced computing in the form of artificial intelligence machine learning, et cetera. We're seeing some really strong, compelling use cases at the, at the functional level. Rob, I kind of said I started my career doing a lot of change management around BIM. When building information modeling was a new thing for infrastructure in 2012 was, was when I entered, was when I entered the industry. And so I spent the first six or seven years of my, my career going around and talking about, hey, there's this thing called BIM, and we should think about doing it. And here's the, here's the, the dimensions, there was the same level of cynicism in those times. I guess what happened, and it was, again, it was at a point where actually the technology was very nice and it didn't, it didn't really work in some settings. It did work in other settings, which I'm transforming, transferring a, a building's tool into an infrastructure context is, is a challenge, right? But we, we can see that like there has been a load of, a load of progress in that and that model models are the default approach now. But I think the thing is we haven't fundamentally changed, we haven't really done transformation, we haven't fundamentally changed the way that delivery happens through, through BIM qualities higher, et cetera. But AI is going to fund them into the change from things. And so I think it's really important that people at least become conversant with the range of technologies and the range of approaches. And what does it mean to move from a deterministic type of programming to a probabilistic type of programming? And, and how do you do that safely and responsibly? And, and what are the advantages? And, you know, everything is just going to go some, it's going to continue to get faster all the time. And so it's really important that people do try and keep up to speed with this emerging field or emerged field that has kind of hit the mainstream. Yeah, it's, I mean, for younger people, like, it is going to be your career for older people, you, like, older people that have, that have reached senior grades, you're going to be managing this, this stuff, and competitive advantage is going to mean that you're going to have to keep pace in some way. And so yeah, we've, we published a white paper actually at TechFest last year. That was, that's what we went to TechFest with. And it was AI for civil engineers, what you need to know to build the future. And what I did in the same way that I used to talk about BIM 15 years ago, 12 years ago, you know, just breaking it down is just very simple components and saying, look, in engineering, in civil engineer friendly language, here is, here is what AI is. Here's one it means. This is what we'd mean by a model and a prediction. And, you know, there's, there's kind of AI and then there's machine learning and there's computer vision and there's all these different techniques that you might hear about. And it's very quickly, it's very easy to quickly get lost and bamboozled by all the different languages people talking about random forests. And some, someone talked about a hairy ball probably the other day. I don't know what one of those is. But like, you know, it very quickly gets very mathematical and very, very scientific and what we need to be able to do as the buyers of this technology is basically develop that spider sense for saying, do I trust this or not? And how do I, how do I leverage it? And so that white paper, again, is available on the Mind Foundry website. And I'm sure we can signpost it in other ways as well. But, but really it's, you know, 20 or 20 odd pages of kind of simple here examples across the asset life cycle of where different AI technologies can be applied. And how to, how to think responsibly and openly about those about the technologies. Nice. Yeah. I will definitely put the link to that paper in the in the show notes because I think it's a good place for people to continue from this conversation into discovering more about AI and infrastructure if they want to. Great, thank you. Great, so thanks for joining me today, Tom, has been a really interesting chat, we covered quite a lot, and I'm sure there's a lot more to be discovered, but hopefully this is a good primer for people who are interested in this subject area. Really, thank you Rob, yeah, we're just on the start of the journey. At Mind Foundry, and I think as an industry, but you can really see there's been, in the past couple of years, a real engagement with this topic, and a real urgency around age and infrastructure, so yeah, hopefully this is only the start. Yeah, and if you are going to go to NC's Tech Fest this year, you can come down and speak to Tom in person and learn that. That's right, yeah, we'll be on the asset management stage, and we'll have a store there at Mind Foundry. We're kids. Alright, that's that for this month's episode, thank you for joining us on the Engineers Collective, we'll see you again next time. For Listening to the Engineers Collective, the podcast by New Civil Engineer. The Engineers Collective comes to you with monthly news analysis from our editorial team, and in-depth interviews with industry experts and thought leaders. If you're a private sector organization looking to reach tens of thousands of listeners, you can find out more about sponsoring an episode of the podcast by visiting newcivilengineer.com forward slash podcast.