Authors: Rainer Splechtna, Kresimir Matkovic, Denis Gracanin, Mario Jelovic, Helwig Hauser
Abstract: Multi-level simulation models, i.e., models where different components are simulated using sub-models of varying levels of complexity, belong to the current state-of-the-art in simulation. The existing analysis practice for multi-level simulation results is to manually compare results from different levels of complexity, amounting to a very tedious and error-prone, trial-and-error exploration process. In this paper, we introduce hierarchical visual steering, a new approach to the exploration and design of complex systems. Hierarchical visual steering makes it possible to explore and analyze hierarchical simulation ensembles at different levels of complexity. At each level, we deal with a dynamic simulation ensemble --- the ensemble grows during the exploration process. There is at least one such ensemble per simulation level, resulting in a collection of dynamic ensembles, analyzed simultaneously. The key challenge is to map the multi-dimensional parameter space of one ensemble to the multi-dimensional parameter space of another ensemble (from another level). In order to support the interactive visual analysis of such complex data we propose a novel approach to interactive and semi-automatic parameter space segmentation and comparison. The approach combines a novel interaction technique and automatic, computational methods - clustering, concave hull computation, and concave polygon overlapping - to support the analysts in the cross-ensemble parameter space mapping. In addition to the novel parameter space segmentation we also deploy coordinated multiple views with standard plots. We describe the abstract analysis tasks, identified during a case study, i.e., the design of a variable valve actuation system of a car engine. The study is conducted in cooperation with experts from the automotive industry. Very positive feedback indicates the usefulness and efficiency of the newly proposed approach.