Implementation of detailed chemistry and load balancing in the fire dynamics simulator

Published

March 16, 2025

Author(s)

Chandan Paul, Marcos Vanella, Randall McDermott, Jason Floyd

Abstract

The Fire Dynamics Simulator (FDS) is a widely used software package for engineering and research applications, designed to simulate diverse fire scenarios. With advancements in high-performance computing, the focus is shifting towards making fire models more predictive rather than prescriptive. Currently, FDS employs an infinitely fast one- or two-step reaction mechanism to simulate fire. Ignition and extinction are modeled heuristically, and critical species such as soot and carbon monoxide (CO) are predicted using prescribed yields (e.g., kilograms of soot or CO per kilogram of fuel consumed). To enhance its predictive capability, detailed chemistry has been implemented in FDS using the Sundials CVODE ordinary differential equation (ODE) solver. This approach provides accurate source terms for the heat release rate (HRR), eliminating the need for separate ignition and extinction models. Moreover, CO concentrations can be directly obtained from the reaction mechanism, while precursor species like acetylene (C\textsubscript2}H\textsubscript2}), predicted by the mechanism, can serve as inputs for soot modeling. To accelerate these calculations, an analytical Jacobian formulation is employed. The chemistry implementation is validated against Cantera software package for constant volume ignition delay problems. To address the computational cost of detailed chemistry, two load-balancing strategies are implemented and compared. The first, All-to-All Communication, where all Message Passing Interface (MPI) processes communicate with every other MPI process ensuring an even distribution of chemically active computational cells and chemistry stiffness across all MPI processes. The second strategy, Minimal Communication, reduces communication costs by limiting MPI interactions. While this approach evenly distributes the number of chemically active cells, it results in an uneven distribution of chemistry stiffness among MPI processes. These strategies are evaluated using two slot burner experiments, one 50\,cm by 5\,cm and the other 25\,mm by 8\,mm, involving 84 and 3600 MPI processes, respectively. Results demonstrate that the All-to-All Communication strategy outperforms Minimal Communication, as the cost of communication is relatively insignificant compared to the computational cost of detailed chemistry calculations.

Conference Dates

March 16-19, 2025

Conference Location

Boston, MA, US

Conference Title

14th U.S. National Combustion Meeting (USNCM)

Pub Type

Conferences

Download Paper

Local Download

Keywords

Fire modeling, FDS, Detailed chemistry, Chemistry load balancing

Citation

Paul, C. , Vanella, M. , McDermott, R. and Floyd, J. (2025), Implementation of detailed chemistry and load balancing in the fire dynamics simulator, 14th U.S. National Combustion Meeting (USNCM), Boston, MA, US, [online], https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=959383 (Accessed October 8, 2025)