A web-based approach for post-processing large ensembles

Abstract

Recent improvements in the field of Computational Fluid Dynamics (CFD) allow for quick creation of large ensembles of CFD results, making data analysis the bottleneck of the design process. Accelerating analysis of large CFD simulation ensembles is key to accelerating future design campaigns. The goal of CFD ensemble analysis is to support decision making. Decision making rests on previous knowledge, new data, and analysis and synthesis of both. Human input is required in ensemble analysis to create new knowledge, and make informed decisions. Data visualisations are a powerful interface between CFD data and humans that can be improved to accelerate CFD ensemble analysis in support of decision making. This thesis presents several visualisation-based techniques to improve this interface, and thus accelerate the analysis. The backbone of this thesis is the ‘divide-and-explore’ data-analysis strategy, which accelerates the analysis by reducing the number of comparisons made between individual ensemble members. During analysis, engineers are interested in understanding the flow behavior (via flow visualisations), and how it relates to the inputs (e.g. geometry parameters) and the outputs (e.g. performance parameters); the relations between these three data segments form avenues along which cause-and-effect relationships may be found. To find cause-and-effect relationships the user must first observe some correlated changes by comparing individual CFD ensemble members. To understand the entire dataset many comparisons need to be made. This thesis shows how existing ensemble analysis approaches can be combined into the ‘divide-and-explore’ strategy, which accelerates the analysis by reducing the number of comparisons required to generate understanding of the ensemble. The software developed identifies and addresses the gaps preventing existing software from supporting this strategy in practice. Unsteady simulations are increasingly being used. Unsteady data is typically visualised as a series of static images, perhaps combined in a video file, which restricts the interactivity with the underlying data, and is relatively cumbersome to share. The second contribution of the thesis is the development of a web-based unsteady data visualisation which supports interactivity, and allows easy sharing of unsteady data visualisations, thus accelerating unsteady data analysis. x During design campaign efforts, experimental data also needs to be compared with CFD data. Approaches to merge experimental flow visualisations and corresponding geometry in a 3D environment exist, but require bespoke experimental apparatuses. To allow experimental flow visualisations made with widespread user electronics, such as smartphones, to be merged with 3D CFD data, the existing methods need to be extended to allow interactive merging. The third contribution of the thesis is a 3D software environment, in which experimental flow visualisations can be interactively merged with CFD data to form a 3D aerodynamics data ‘log-book’ that supports easier comparison of experimental and CFD data, and analysis of the differences. Existing knowledge is an important input to the decision making within a design iteration, and new knowledge created, as well as the observations it is based on, will be important in the next design iteration. Newly created knowledge is typically captured in written reports, with static images illustrating the explanations presented in the text. However, static images may not allow knowledge re-users to answer questions not expected when the report was written. To allow this, the connection between knowledge contained in text, and the underlying data supporting it needs to be preserved. The fourth contribution of the thesis is a system that allows knowledge to be captured on-the-go alongside the data, thus allowing better re-use of the data and the knowledge. The system allows collaborative work to further accelerate the analysis.