World's first AI-planned high-speed rail tunnel tested in China's Wufeng

China is trialling an AI system that selects excavation methods for each section of a new high-speed rail tunnel in Wufeng, marking the world's first such project

A remote county in central China is now being used to test a big new idea in building infrastructure. This place was earlier known only for its tough landscape, but it is suddenly getting worldwide attention. The reason: a new breakthrough that uses artificial intelligence to help with one of the hardest jobs in engineering — making tunnels for high-speed trains, South China Morning Post reported.

 

Wufeng Tujia autonomous county in Hubei province is famous for its mountains, thick forests and deep valleys. But this has also left the region isolated for decades. Steep slopes and unstable ground have slowed progress and made building transport networks extremely difficult.

 

That is now changing. Heavy machinery, explosives and advanced software systems are working together to carve out the Yangcun Tunnel — a key link on a high-speed rail line designed for trains reaching 350 kmph.

 

This project is the world’s first high-speed railway tunnel where an AI model made most of the construction-method decisions, the news report said.

 

Why is Wufeng one of the hardest places to build?

 

The mountains of Wufeng have been shaped by hundreds of millions of years of geological activity. The area sits in the Wuling Mountain range, which contains deep fractures, karst formations, fault zones, underground water systems and unpredictable rock layers.

 

More than 120 geological relics show how unstable the ground can be.

 

For tunnel builders, this means unpredictable troubles underground. As Wu Jiaming, senior engineer with the China Railway Siyuan Survey and Design Group, explains, crews “could encounter a cavity, water pocket or loose shale”. Traditional design methods depend on the experience of experts, but even experts cannot accurately predict hundreds of individual tunnel sections.

 

A single error can lead to collapses, flooding or long delays.

 

How did AI take the lead in tunnel design?

 

In the Yangcun Tunnel, AI made the decision on how to excavate each part of a tunnel — through full-face blasting, multi-step cutting or the careful CD (centre diaphragm) method.

 

The team gathered 1,700 construction sections from 251 high-speed rail tunnels. Each entry contained 19 data points, such as rock strength, groundwater conditions, fault lines and the tunnel’s alignment. And this database is only a small sample of the information China has built over decades of construction, the news report said.

 

How exactly did the AI model make its decisions?

 

The model used a “multi-scale convolutional neural network” with attention mechanisms (ACmix) and a specialised loss function called Focal Loss. This helped it analyse both common situations and rare but high-risk conditions.

 

Instead of giving one plan, the AI divided the tunnel into hundreds of segments and recommended a method for each. Some areas required full-face excavation, others needed three-step approaches, and dangerous zones were marked for CD-method use.

 

The model had an accuracy rate of 89.41 per cent, nearly three percentage points better than Random Forest and SVM models.

 

How are humans and AI working together on the project?

 

In the Yangcun project, senior engineers reviewed the AI-generated plan and approved it. The tunnel’s building information modelling (BIM) system now embeds the AI’s recommendations, guiding workers and equipment in real time.

 

Researchers say this signals a new stage where AI can assist with major engineering decisions once considered too risky for automation, the news report said.