The AI crack spotter helping engineers work smarter

On the whole, cracks never get smaller. Take a glass, for example. Once you spot a crack in it, then you really ought not to use it anymore because it’s only a matter of time before it shatters. Thankfully, this isn’t the case with every slightly fractured material and, for the most part, an early spot can prevent a small crack from becoming a big problem.

Concrete is a really good example of this. It’s incredibly strong and durable, which is why we use it for buildings, roads and bridges. But after around 50 years it begins to show the first signs of deterioration and then civil engineers are quickly deployed to inspect structures, check for cracks and make important recommendations as to the course of action. This is essential work, especially when you think of places such as complex motorway systems. Used by hundreds of thousands, if not millions, of motorists every week, these roads often feature huge numbers of bridges, tunnels and other concrete structures too.

In Japan, one such network is operated by the Nippon Expressway Company (NEXCO) and numbers just under 10,000 kilometres of road, 16,700 concrete bridges and 1,680 tunnels. Around half of all these are over 30 years old, with many dating back to the 1960s. That’s a whole lot of concrete. Especially when you consider that every centimetre needs to be safe and fit for purpose.

Traditionally, a small army of intrepid engineers and experts would have undertaken the inspection work in its entirety, getting up close and personal with cracks, discolouration and concrete spalling (where concrete has broken off the structure). This isn’t just walking around with a magnifying glass and tapping the concrete to check the pitch – think, scaling bridges on scaffold, using elevated platforms and even vehicles with extendable arms, helping them to reach tricky spots. It’s a necessary, but costly business, even before any further analysis or repair work takes place.

With ever expanding road networks and anticipated difficulties in recruiting engineers in the future, NEXCO’s Research Institute needed to explore new and efficient ways to tackle this endless challenge. In doing so, they discovered Canon’s Inspection EYE for Infrastructure, an image-based service which uses deep learning to detect defects.

The inspection images are first captured using a high-resolution Canon EOS DSLR camera, which sits on a motorised panoramic mount, automatically panning and tilting to record a spot from all angles. These images are then stitched together to create a single high-definition version for Artificial Intelligence to analyse. This AI model is trained using data from Tosetsu Civil Engineering Consultant Inc., who are pioneers in image-based infrastructure inspection, and it has learned to detect cracks at a level comparable to that of highly skilled inspection engineers, even identifying wall surface cracks without erroneously detecting dirt or joints.

In a comparison exercise, a civil engineer who was given the same images as the Inspection EYE for Infrastructure took 720 minutes to identify about 500 cracks, then prepare the corresponding inspection data. With the help of our AI tool, the same engineer took only 90 minutes to complete the same task. It was also noted that the system offered not only accuracy, but consistency in results – something which has the potential to be a little variable in during human analysis. However, qualified engineers always assess the final outcomes and, if required, any edits are fed back into the AI model, so it can continue to learn and improve.

When working on Nippon Expressways, “Automated defect detection using AI is only a part of the inspection process,” explains Masakazu Honda, Division Chief of the Maintenance Management and Promotion Division, Infrastructure Development Department at NEXCO RI. “An engineer will have to make the final evaluation,” However, our tests have found that Inspection EYE for Infrastructure demonstrates an extremely high level of detection precision, with accuracy at 99.5%. “While some AI systems detect individual cracks in a piecemeal fashion, Canon’s AI system offers a level of precision that is close to visual checks by inspection engineers, such as detecting successive cracks as one single crack,” Mr Honda adds.

This new technology not only reduces inspection time and cost, but Mr Honda sees huge benefits for engineer training, as it frees up their time and allows them to undertake more site visits – important in developing those vital skills of observation and understanding of different kinds of structural damage. “To mature as an engineer, they need to see the site with their own eyes,” he says. “Even with [AI providing] greater convenience, it remains vital to develop the ability to evaluate via on-site observations.”

In this way, the extra ‘eyes’ provided by Canon are not only an important tool in keeping us safe, but they are supporting the development of careers in an extremely specialist field – one that is essential to the safety of travellers, but also to the economies which rely on well-maintained road systems for commuters and goods transportation alike.

Find out more about how we work with businesses to solve their problems.