Validation of AI model for predicting heat release rate in static and wind driven pool fires

Author(s)

Abstract

This work focuses on validating an Artificial Intelligence (AI) based machine learning model to estimate the heat release rate (HRR) from outdoor fires where the feasibility of performing calorimetry or any other similar technique is challenging or not viable. The only data that can be measured in such cases is the visible flame height and flame tilt (in wind driven scenarios). It is well known that the flame height is strongly dependent on HRR of the fire. Hence, using an AI model to ingest the visible flame images to estimate the HRR is extremely useful particularly for large outdoor fires. The AI model must be first validated for predicting the HRR data from well characterized fire sources to ensure reliability in the model predictions. Large-scale pool fire tests carried out in well-controlled environmental conditions can serve this purpose. In static pool fires, the HRR is correlated with the flame height and diameter of the fire source while in wind driven fires, it also depends on the flame tilt relative to the fire source. Following this, a series of lab-scale experiments have been conducted with static and wind-driven pool fires fueled by a mixture of diesel and canola oil. The mass loss rate of fuel is measured using a load cell and the visible flame height is recorded using cameras. In wind driven tests, the pool fire is placed outside the test section of a wind tunnel and a constant wind is blown upstream of the fire. The wind velocity is varied to change the flame tilt, and hence the HRR of the fire. Further, point measurements of temperature, velocity, and heat flux have also been measured. Preliminary results showed that the AI model is able to predict the HRR from static pool fire tests. The model is now trained to predict the HRR from wind driven tests. The experiments are used for validating the numerical model in Fire Dynamics Simulator (FDS), an open-source CFD code. This model will be used to generate synthetic data (by varying the external conditions) to further train the AI model in the future.

Keywords

Wildland urban interface fire and Fire dynamics and science

Citation

Issues

If you have any questions about this publication or are having problems accessing it, please contact [email protected].