Overview

As the electric vehicle market grows rapidly, a new set of safety challenges is emerging. According to the National Fire Agency, there were 187 EV fires nationwide between 2018 and June 2024, including 16 cases in Seoul alone. According to the EV Charging and Parking Area Fire Prevention and Response Manual published by the Seoul Metropolitan Fire and Disaster Headquarters in December 2024, EV fires involve battery thermal runaway, reach extremely high temperatures, and can cause greater damage than internal combustion vehicle fires. The risk becomes even more severe when fires occur in underground spaces. Because early response is critical in EV fire incidents, organizations need systems that can detect smoke quickly and enable immediate action.

To address the unique risks of EV fires, a leading fire safety research institute in Korea partnered with Superb AI to develop an AI-powered fire detection system.

Challenge: The Limits of Existing Fire Detection Systems

The first challenge in developing an EV fire detection system was the unique nature of battery fires. Thermal runaway in lithium-ion batteries follows a different pattern from conventional fires, making early detection difficult with traditional fire detection systems.

In enclosed spaces such as underground parking lots, rapid evacuation and fire suppression are essential. However, existing systems often take too long to detect fires after ignition. False alarms also create major operational challenges, leading to unnecessary evacuations and response costs.

The key technical requirements included the ability to distinguish between different types of smoke and gas, real-time monitoring and early warning for thermal runaway events, and integration with existing safety infrastructure.

Solution: Building an AI-Powered Real-Time EV Fire Detection System

“We developed an AI model with Superb AI to build an AI-powered real-time fire monitoring system. Using a custom model trained on a broad fire incident dataset, we developed specialized detection algorithms for early-stage heat and battery gas release. For fire-related data that was difficult to secure, we used the generative AI capabilities of Superb Platform to synthesize additional data. As a result, the false alarm rate was reduced by 80%, and average detection time improved by 65%.”

— Representative from a fire safety research institute

Superb AI provided a tailored solution designed around the unique characteristics of EV fires. The key components included the following:

1. Development of a Specialized AI Model

Superb AI developed an AI model trained on the characteristics of EV battery fires and parking lot environments. The team created multiple models using diverse datasets, then selected the model that delivered the best balance of accuracy and speed within the requirements for real-time processing. This enabled early detection of fire risks.

(AI model test screen for fire and smoke recognition)

2. Efficient Data Processing System

The team processed 100,000 images covering fire, smoke, and vehicle objects, and built a training dataset by combining real fire incident data with test data captured in controlled environments. Automated data labeling improved processing efficiency, while rigorous quality control helped ensure model accuracy. For rare cases, the team also used generative AI to create synthetic data, securing enough training data for model development.

(Analyzing patterns and relationships in fire and smoke datasets using the Scatter View in Superb Platform)

3. Real-Time Monitoring Platform

Using edge computing technology, Superb AI built a system capable of real-time analysis on-site. Jetson-based edge devices enabled fast, real-time processing at the deployment location.

The system was also integrated with existing CCTV infrastructure, emergency response systems, and building management systems. A custom dashboard was provided to support monitoring across multiple locations.

Benefit: Improved Safety and Lower Operational Costs

The project delivered the following key outcomes:

1. Enhanced Safety

Detection time was reduced by 65% compared with the existing system, while the false alarm rate decreased by 80%. The system showed particularly strong performance in identifying early signs of battery thermal runaway.

2. Improved Operational Efficiency

Real-time monitoring and automated alerts significantly shortened response time. Seamless integration with existing infrastructure also made operations and management easier.

3. Greater Cost Efficiency

By reducing false alarms, the system helped lower unnecessary response costs. Improved early detection also made it possible to minimize potential damage before incidents escalated.

Conclusion: Opening a New Era of Fire Safety with AI

This project demonstrated how AI technology can help solve new safety challenges in the EV era. Superb AI will continue contributing to a safer society through innovative AI technology.