Machine Learning And Digital Custodianship Could Help Network Rail With Asset Inspections

Machine learning and digital custodianship could help Network Rail with the “systemic issues” that exist within its structures inspections regime, according to an artificial intelligence (AI) tech company.

Mind Foundry is “a machine learning company which spun out of Oxford University’s machine learning research group in 2015”, explained its director of civil infrastructure Tom Bartley. He added that one of Mind Foundry’s “key areas of research is around aging infrastructure”.

This is has been a growing problem for Network Rail. Recently, the regulator Office of Rail and Road (ORR) penned a scathing letter calling out “systemic issues” within Network Rail’s compliance with asset examination requirements last month. ORR believes Network Rail has been “noncompliant” in a number of areas where the safe management of its assets is concerned. These areas cover the regularity and efficiency of its structural assessments, knowledge of the capacities of some its structures and the many missing risk assessments.

This is worrying not only for Network Rail, but also users of the UK’s rail network.

Bartley believes machine learning could help Network Rail with its asset inspections issues – and in fact could also aid all asset owners.

“While Network Rail has problems with its aging assets, so do owners across the UK and developed world,” he said.

“Most of our modern infrastructure was built during a 20 year period, during the during the 50s through to the 70s, and they’re all reaching an age now where deterioration is starting to really accelerate.”

Bartley stated how the management of ageing assets is not an easy task; especially as decline in structural health happens over a long period of time and needs to be monitored regularly.

“We used to talk about custodians and how there was a real strong sense of custodianship, where an asset manager will spend their career managing a small number of assets and really knowing the history of those assets,” he said.

“What’s changed over the past 20 years or so is that these asset managers have started retiring, and owners are starting to rely more and more on consultants to do a lot of the work. This means they’re losing out on a lot of that longevity and custodianship.”

In response to this, Mind Foundry has been developing the idea of digital custodianship which will utilise AI to help fill in some of the gaps in knowledge.

“One of the things we’re thinking about is how we can create digital custodianship using all the expertise the consultants bring by also using AI to get that continuity of knowledge back,” Bartley said.

“How do we look at historic records and how do we project that into the future?”

Aside from accessing that knowledge, Bartley believes using machine learning and AI to aid with inspections and examinations will take out a lot of the dirty work. Mind Foundry’s inspection application aims to do just that.

“What our work is really about is around the examination process,” he said.

“Currently, a lot of the challenges that owners have is that their examinations are very manual; it’s about people visiting a site on a periodic basis.

“They are doing this visually and they’re experts; they know bridges. But what they end up doing is they take photos, and a bridge is quite complex structure. They come back and the engineer looks and goes, ‘where on the structure is this?’

Despite the shortcomings of the approach, the photos are used to give the structure a severity rating on a scale to describe its state.

“These scales do not really give you a sense of change over time,” Bartley admitted. “What engineers want to know is how the structure is changing.”

Another shortcoming of the photos is the time consummation for the limited data provided.

“There’s so much time spent taking those photos, putting them into reports and then someone trying to interpret these reports,” Bartley said.

“After all that, the data is actually not very useful because often people will not know where the defect on the asset is as well as a whole host of other important information.”

He said that Mind Foundry is able to use AI “to help locate the defect on the structure and automate the production of the report to save time but ultimately to introduce really snazzy features around change over time”.

These features include using computer vision to quantify literal measurements from the photo, which could work out how long a crack/defect is. Mind Foundry’s technology can then overlay historic photos to see how the defect has changed over time giving engineers a real sense of what is happening with the asset.

Bartley explained to NCE how Mind Foundry is hoping to be able to scale up small changes to allow for a better system in the digital custodianship space.

“Our long-term vision is about predicting deterioration and being able to help owners prioritise work interventions they need to do,” he said.

“We’re starting with this inspection app because that’s currently the bottleneck in the process.

“Ultimately, the outcome is high quality data but the kind of the reason they would acquire it now is that it saves them time. It saves them effort in the short term.”

All this sounds like the step in the right direction and a really positive movement within an industry that has a persistent problem but inertia towards incorporating this new technology still exists.

“When we’re talking about digital custodianship, it is a way of replacing 20th century ways of working with 21st century ways in terms of how deterioration, evaluations and conditions are currently modelled,” Bartley said.

“There is some inertia, I think, in terms of changing the standards that underpin some of the ways things get measured and reported.”

Bartley admitted this isn’t necessarily the industry’s fault though.

“I think we’re at a point where technology has just changed rapidly in the past five years and things that weren’t feasible now are,” he said.

“We’re now at the point where your mobile phone has got good enough cameras to take those photos at two o’clock in the morning with the right resolution.

“We’re at a point where the machine learning actually is viable, where the big data systems are too.”

While the technology is there, its adoption has been slow. Bartley has pondered what this could mean if structure inspection regimes aren’t enhanced and the defects themselves aren’t discovered and monitored.

“We don’t really have a choice when it comes to examining ageing assets because they’re all going to fail at the same time,” he said.

“We don’t have the resources and it is far too disruptive to not try and make these assets live longer than their design lives.”

To avoid this from happening, Bartley thinks it’s paramount asset managers embrace the future of the space.

“The second part is then around the funding and how we unlock resources,” he said.

“At the moment, because the data is so crude, it’s really hard for asset managers to come to convince the politicians that the works are actually needed.

“The machine learning has this predictive approach that takes all the historical information for the structure and brings it into its current and then predicts what’s going to happen.

“That alone is going to help us prolong the life of assets because we will know when the right time to make an intervention is.”

Like what you've read? To receive New Civil Engineer's daily and weekly newsletters click here.