Steel corrodes. The combination of time, weather, and degradable materials makes that inevitable. When it comes to the skeletons of our infrastructure, such as bridges, towers, and guardrails, corrosion can be slow and subtle, or dangerously quick. And it's up to maintenance teams to spot the rate of deterioration before it becomes a hazard.

But knowing when to check is as important as knowing where. Working out the best timing for checking steel corrosion has been both difficult and expensive for a long time.

Researchers at NTT think they've found a better, smarter way to predict it.

Predicting Corrosion

NTT's new technology, set for commercial release in fiscal 2025, is able to forecast how corrosion will spread years into the future by analyzing nothing more than photographs and local weather data. It's the first system in the world that can generate predictive images able to show how corrosion will likely develop at the level of individual structures. This opens up the possibility of allowing maintenance teams to move away from the current uniform inspection cycles, which check everything at fixed intervals, regardless of actual condition.

It's Tricky! Steel Does Not Corrode Predictably

One reason that existing systems don't work well is that corrosion does not behave predictably. It's not a simple case of doubling every "x" number of years. In a recent field study undertaken by NTT Access Network Service Systems Laboratories, researchers looked at dozens of examples of roadside infrastructure in Ibaraki Prefecture, Japan, measuring the growth in corroded area over time. They found that after three years, some surfaces had corroded by around 60%, while others had corroded by nearly 150% after four years. After six years, some had grown by only 80%. That kind of inconsistency makes it hard to plan an efficient and timely maintenance program.

Other studies have tried to estimate corrosion progression using macro-level averages, things like temperature, rainfall, or sunlight hours across an entire city or region. But they were too general and lacked the granularity needed to understand how a specific bridge or pole might be degrading.

NTT Has Found a Way

NTT's new system is different. It uses three inputs:

From those metrics, it then generates a realistic image of how the corroded surface is expected to look after a certain number of years.

Generative Adversarial Network

How does it do this? The core of the system is a type of deep learning model known as a GAN, or Generative Adversarial Network. In simple terms, it's a model that learns by trying to fool itself: one part generates synthetic images, while another tries to spot the difference between synthetic and real ones. Over time, both improve. Through a series of training using inspection images and weather records collected over several decades, the model has now learned how rust typically forms. How it spreads, changes color, and eats away at edges and corners.

Over 90% Accuracy

Validation tests on 20 examples of roadside infrastructure showed a strong correlation between predicted and actual corrosion levels after an average of 4.4 years, with a margin of error under 10%. Over ninety percent accuracy means that we can now think about changing how infrastructure maintenance is scheduled. Slow-corroding structures might be checked less often, saving money and time, while fast-corroding ones could be prioritized.

Better Maintenance Planning, For Less Money

Along with much greater safety, accurately forecasting the pace of corrosion will allow repair teams to plan workloads and budgets over multiple years. Instead of scrambling to fix things as they fail, infrastructure managers will be able to plan ahead, line up projects more evenly, avoid cost spikes, and reduce strain on limited personnel.

NTT plans to apply the technology initially to road bridges managed by its own group companies, but hopes to expand it over time to other structures like towers, and to other types of damage, such as cracking or material fatigue.

By using images and weather reports to make predictions, the system brings long-range thinking to a job that has traditionally relied on routine, experience and, frankly, guesswork. As infrastructure ages and inspection costs rise, tools like this could help shift the balance from reactive to proactive maintenance, keeping our built environment safer, for longer, and at lower cost.

Innovating a Sustainable Future for People and Planet

For further information, please see this link:
https://group.ntt/en/newsrelease/2025/04/30/250430a.html

If you have any questions on the content of this article, please contact:
Public Relations
NTT Information Network Laboratory Group
https://tools.group.ntt/en/rd/contact/index.php

Daniel O'Connor

Daniel O'Connor joined the NTT Group in 1999 when he began work as the Public Relations Manager of NTT Europe. While in London, he liaised with the local press, created the company's intranet site, wrote technical copy for industry magazines and managed exhibition stands from initial design to finished displays.

Later seconded to the headquarters of NTT Communications in Tokyo, he contributed to the company's first-ever winning of global telecoms awards and the digitalisation of internal company information exchange.

Since 2015 Daniel has created content for the Group's Global Leadership Institute, the One NTT Network and is currently working with NTT R&D teams to grow public understanding of the cutting-edge research undertaken by the NTT Group.