guest ::
| Home > Publications database > Automated Generation and Evaluation of FDS Simulations for Optimizing Parameters with Dakota |
| Home > Publications database > Automated Generation and Evaluation of FDS Simulations for Optimizing Parameters with Dakota |
| Contribution to a conference proceedings | FZJ-2015-02305 |
; ;
2015
Abstract: Some questions in Fire Safety Engineering require multiple execution and evaluation of firesimulations with changed parameters or a modified setup: A thorough sensitivity study requiresthe execution of one simulation per sample point. For a design of a fire safety system takinginto account n different scenarios, n simulations are needed. And for an automated parameteroptimization, repetitive simulations are necessary to determine the optimal parameter values.However, one of the most popular tools for fire simulation – the Fire Dynamics Simulator –lacks a functionality for automated batch processing of parameter spaces.In this paper, FDSgeogen is presented, a tool that enables the user to automatically generateFDS input files implementing freely selectable variables, e.g. for material properties orgeometrical measures. The automated input file generation is based on a file containing informationabout the simulation setup, which is formatted in Extensible Markup Language (XML),a file containing the parameter sets for the different simulations, usually formatted as commaseparatedvalues (CSV) and a python parser for the generation of the FDS input files on the basisof the XML and CSV files. Since the XML file contains most of the relevant information aboutthe setup of the simulation, the boundary conditions and the output values, it may seem similarto an FDS input file. However, in combination with the parameter file and the python parser itis not only possible to include freely selectable variables but also to make them dependent oneachother. That is why the presented method provides a lot of flexibility and adaptability forthe automatic input file generation for different applications. The authors strongly encourageothers to make use of the presented method and participate in further development of the code.In order to demonstrate the applicability of the proposed method, an automated parameteroptimization for cone calorimeter experiments is presented. Based on measurements of the heatrelease rate (HRR) of polyurethane foam subjected to two different heat fluxes, the materialproperties emissivity, ", thermal conductivity, k, and specific heat, cp, of polyurethane wereoptimized. For the automated parameter optimization Dakota [1] was utilized. It is an opensource software toolkit that amongst other things includes algorithms for design optimization,parameter estimation, and sensitivity analysis. For determining the loss function value theeuclidean relative differences between the HRR in experiment and simulation were calculated.
|
The record appears in these collections: |
PDF