Fire Science Physics Informed Machine Learning. In Data Science & Engineering Master of Advanced Study (DSE MAS) Capstone Projects

Next generation fire models provide the basis to understand fire physics in detail, leading the way for emulators to model potential fire behavior. This project uses data from hundreds of coupled fire-atmosphere simulations using the QUIC-Fire model to develop a reduced-order emulator. Such an emulator can be used to predict the wildfire spread, which can help the fire agencies to take necessary steps to reduce the damage. Additionally, the predictions can be utilized to mitigate the risk of controlled fires escalating into wildfires.
The project encompasses several objectives: developing a system to evaluate the performance of deep learning models through benchmarking, conducting the benchmarking process, and leveraging the obtained results to enhance the accuracy of the models. Our product utilizes cloud infrastructure to provide a scalable and robust solution for managing the entire lifecycle of machine learning. This infrastructure enabled us to conduct numerous experiments, analyze results, and optimize various deep learning models, including U_Net, TF_Net,PhyDNet, and ConvLSTM.