Your browser is no longer supported

For the best possible experience using our website we recommend you upgrade to a newer version or another browser.

Your browser appears to have cookies disabled. For the best experience of this website, please enable cookies in your browser

We'll assume we have your consent to use cookies, for example so you won't need to log in each time you visit our site.
Learn more

Digital inspection prejudice is hindering game-changing potential

Img 4607 crop

Digital images can improve asset inspections, so why is the industry adoption so slow?

Increasing the use of digital imagery in asset condition surveys has game changing potential.

Technology which can collect data about the condition of, for example a tunnel, using arrays of digital cameras fixed to moving vehicles, is already available.

Processed digital images

The digital images can be processed to produce visualisations of the tunnel’s condition. Machine learning tools can be trained to automatically identify defects, particularly through object recognition technology.

It initially sounds like a simple route to huge savings in time, increased health and safety and increased assessment accuracy. But as with many new technologies, the headline benefits belie the complexities of getting industry adoption.

A new study undertaken by Arup has been examining the barriers to greater uptake of computer vision  technology.

With funding from industry innovation group i3P, the consultant was commissioned to investigate the potential of integrating computer vision techniques with current and future asset management processes.

Arup associate director Mike Devriendt led the study. “If you go back 10 or 20 years it was all very manual – an engineer with a clipboard would stare at things,” he says. “We found that when you then go back in successive years it is very difficult, for example, to look at a tunnel after construction works have been carried out close to that asset, and see whether the condition of that asset changed with time.

Record keeping is notoriously not great in terms of being able to tell how things have changed

 “The record keeping is notoriously not great in terms of being able to tell how things have changed.”

Of course, digital imagery is already in widespread use in the industry through mediums such as laser scans, mobile sensors and thermal scans. However, the study wanted to examine the potential for use of digital images combined with machine learning, getting to a point where a physical inspection would only be used in the event of an anomaly.

In the UK 40% of construction expenditure goes on repair and maintenance, with infrastructure brining in a £8bn bill. Inspections are key to making decisions about how this money is spent.

The Arup team decided to examine the barriers to adoption with a particular focus on the tunnels sector, undertaking interviews with clients such as London Underground, High Speed 1 and Network Rail.

The advantages of this technology are clear. From a health and safety point of view, using this technology means fewer engineers have to set foot into the potentially dangerous environment of tunnels. In addition, there is a level of automation which, if managed, can be a basis for planning maintenance.

Tunnel inspection

Tunnel inspection

The goal is to automate data collection in previously dangerous environments

“Rather than taking the odd photo of different defects, you capture entire images over however many kilometres of tunnel you’re looking at. So, you always have a source of truth of what the condition of the asset is at any given stage.

“It could be video, it could be a sequence of images which you then stitch together, which you can then the present in an immersive Google Streetview type view, or in CAD based view – there are various different softwares that can do that. This helps if you’re trying to develop augmented reality-based models or want to use it as part of a building information model,” says Devriendt.

Although using the technology initially sounds like a no-brainer, to Arup’s surprise, the study found many major clients are still undertaking asset inspections in a very physical way. And the barriers to adopting  inspection technology are complex.

The first issue to overcome was the perception of “black box” technology.

“As soon as I mentioned machine learning, peoples’ eyes glazed over,” says Devriendt. “But what asset owners want is 100% surety. Some of the interviewees had had suppliers turn up saying ‘we can run these algorithms through’, but when the client has compared the outcome to manual inspection, it has not been favourable.”

Rather than taking the odd photo of different defects, you capture entire images over however many kilometres of tunnel you’re looking at

Devriendt is concerned that if clients get a poor first experience of using this technology it will damage take up of computer vision technology and ultimately mean its benefits are not realised.

Some of the interviewee replies might suggest the technology is not quite ready if the clients said physical inspections were better, but this also might suggest there is a need for an industry bench test of the technology, according to Devriendt.

A benchmark test could involve an asset owner or supplier providing thousands of images, which are all fed into an algorithm.

The asset owner already knows what the asset’s defects are (based on previous detailed manual inspection), so the technology supplier’s claims and trained machine learning algorithms can be tested against this knowledge. If the technology comes up with the same answer, the asset owner can be sure of the effectiveness of the technology.

Benchmark tests

Arup is now keen to continue working with i3P to get more funding to take the project on to develop benchmark tests for the industry.

 “We are trying to work to set industry standards. Otherwise there is a risk of undermining the potential in the industry with asset owners putting their toe in and having bad experiences. I fear unless these things are set up properly, you’ll get a loss of confidence,” says Devriendt.

The work to set up a benchmark test also raises the issue of intellectual property (IP). How willing would clients be to share their images and data with the industry to bring about industry-wide benchmark tests?

If a contract says that all background and arising IP is owned by the client, that may limit the supply chain’s willingness to offer up machine learning algorithms they have developed. This is particularly so considering the algorithms will naturally develop arising IP from the training process.

The study found that interviewees thought getting agreement about IP would be challenging, particularly if rigid and non-collaborative contractual terms are used. Using terms that allow arising IP to be commercialised by all parties or use of preferential licence agreement was mooted.

Adopting new technology does not automatically happen when the software is developed. As Arup’s study found out, if the industry is to innovate, there must be discussion and collaboration to ensure that take-up works for everyone.  

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

Have your say

You must sign in to make a comment

Please remember that the submission of any material is governed by our Terms and Conditions and by submitting material you confirm your agreement to these Terms and Conditions. Please note comments made online may also be published in the print edition of New Civil Engineer. Links may be included in your comments but HTML is not permitted.