Korea’s engineers are using GenAI-powered drones to inspect aging tunnels, solving data scarcity issues and improving safety
Korea’s civil engineers are tackling tunnel maintenance challenges with generative AI-powered drones, creating synthetic data to overcome inspection limitations. Aging tunnels require constant monitoring, but a shortage of skilled inspectors makes this difficult. Traditional AI models struggle due to limited real-world damage data, making deep learning impractical. The Korea Institute of Civil Engineering and Building Technology (KICT) has developed an AI model capable of generating realistic synthetic images of tunnel damage, significantly improving AI training efficiency. In just 24 hours, the system can create 10,000 high-fidelity images of concrete cracks, rebar exposure, and delamination. This approach reduces reliance on extensive field data and lowers training costs while enhancing model accuracy.
To further streamline inspections, KICT integrated this AI with autonomous drones. These drones navigate tunnels with a 20cm margin of error using advanced positioning sensors. Unlike traditional methods, where inspectors rely on high-altitude vehicles to assess tunnel ceilings, the AI-powered drones perform real-time damage detection safely and efficiently. The AI continuously learns from collected video footage, improving its ability to detect defects with precision. This innovation significantly reduces the risks and costs associated with manual inspections.
By enabling AI to generate training data rather than requiring large datasets, KICT is breaking new ground in AI-driven construction safety. The successful field tests demonstrate how GenAI can transform infrastructure maintenance, ensuring the longevity and safety of urban tunnels with minimal human intervention.