Artificial Intelligence Enabled Smart Firefighting

Summary

The dynamic conditions of life-threatening fires require rapid decision-making by firefighters and civilians. This project aims to develop novel AI-driven technologies to provide real-time actionable information. Using artificial intelligence (AI) and machine learning (ML), fire services and the public can respond to fire emergencies more efficiently, reduce risks to firefighters and civilians, and optimize their operational effectiveness and decision-making, ultimately saving lives and reducing property losses.

Description

Firefighters douse flames bursting from a building as a flashover occurs during an experiment.

Credit: NIST

Objective

This project uses AI and ML to develop data-driven solutions that enable real-time forecasting and provide actionable information to enhance safety and situational awareness.

Research Plan

The research plan for the project includes four activities: 1) collection of high-fidelity data, 2) data preprocessing and enhancement, 3) model development and validation, and 4) model deployment and testing. These activities drive the advancement of methodologies to obtain, prepare, and enhance the data for developing models that can be used in real-world emergency response applications. Using state-of-the-art AI/ML paradigms and cross-validation techniques, the developed models provide predictions and/or optimized recommendations. The project team engages with key stakeholders to deploy and test the model in controlled settings. Feedback and additional data are collected to enhance the model’s performance.   

Current Thrust Areas:

1. Flashover predictions: Firefighters rely solely on their experience to recognize flashovers, and traditional numerical approaches cannot be used in real-life fire scenarios. This research thrust is developing ML-driven tools to provide real-time forecasts of flashover in a residential home with up to 14 compartments. These new tools hope to eliminate the need for simplifications (i.e., unrealistic temperature sensing limits and prior knowledge about the fire and building interior conditions).
2. Dynamic path optimization for evacuation planning:  Evacuation plans prioritize the shortest route to exits, often neglecting the impact of fire growth and smoke spread. In case of a fire emergency in a large commercial building, escape paths may become hazardous, leading to confusion, delays, and even casualties (click for image of a representative building layout). This research thrust will create a ‘smart’ recommendation system that uses reinforcement learning to help people escape fires safely and quickly (click to launch animation). The system will analyze changing fire conditions and find the best evacuation routes.
3. Physiological monitoring for firefighter’s heart health: Sudden cardiac death has been a leading killer for U.S. firefighters. However, current National Fire Protection Association (NFPA) standards are limited to offline guidance and available tools from AI and medical research communities are only for the general population. This research thrust has collaborated with U.S. academic/medical institutions to collect firefighters’ electrocardiograms during active training. It is developing an ML-based model to provide real-time, continuous, and reliable heart health monitoring for firefighters in firefighting and emergency response training. Additional research is underway to advance the model to be deployed.

Major Accomplishments

• The National Institute for Occupational Safety and Health (NIOSH) Fire Fighter Fatality Investigation and Prevention Program cites the NIST paper for future training to protect firefighter health.
• Two fire data generators, 1) CData and 2) FDGen, have been developed to facilitate the collection of synthetic data for machine learning model development.