Validation science for large scale pool fire simulation

The process of verification and validation, also known as validation science, plays an important role in software development. This dissertation demonstrates the process of verification by using the parallel linear solver as an example and validating different levels of simulations to quantify sources of error in the Arches fire simulation tool. Arches is a software tool for computing pool fire performance using massively parallel computers. Each component from each level of the Arches code has to undergo the same verification and validation process. The demonstrations in this dissertation are intended as a template for future validation efforts. During the verification process, numerical and algorithmic errors are quantified. In addition, the performance and the scalability of the software is tested. The process of constructing the validation hierarchy is different from the traditional visual comparison. A well structured validation hierarchy can help determine the sources of errors.; The implementation of the parallel linear solvers is part of the verification demonstration. In order to limit the time spent in debugging and coding verification by writing the parallel linear solvers of Arches, the author has implemented existing solver toolkits. The advantage of using existing software packages is that most of these toolkits were developed at national laboratories and have been tested on different platforms. Therefore, more time has been spent on studying the performance and making large runs. The structure of Arches conforms to the evolving Common Components Architecture (CCA) standard, meaning that all the routines are modularized and it is easy to implement with other packages that also conform to CCA. With this architecture, the extensibility and flexibility of the solvers in Arches is, in principle, unlimited. In addition, the modeling errors in the physics behind the simulation and the results that it produces have been studied.; Numerical simulation, experiment, and validation are closely knit. Experiments can provide data to validate simulation codes, and simulation can produce results at resolutions that experiments cannot obtain. While the efficiency of parallel implementation is progressing, the accuracy of the overall simulation will also be investigated. Various mixing, reaction, turbulence, radiation and numerical differencing schemes are used to study the accuracy of the simulations. Different levels of the validation schemes have been used. The different levels of validation are based on the complexity and the physics coupling. Validation at the lower levels can give a quantitative validation metric for the higher levels. Higher level validation can produce qualitative feedback and determine sources of errors for the lower levels. The main objective is to prove that Arches can give reasonable qualitative and quantitative results for large-scale pool fires.