Title: Ensemble Visualization and Analysis of Machine Tools
Name: Mathias Hummel
Phone: +49 (0)631 – 205-4194
The design process of machine tools can be supported by virtual prototyping. The performance of a specific design with a given set of parameters is assessed using simulation methods such as computational fluid dynamics (CFD) simulation. A straightforward workflow is to run a simulation for a specific design and parameter setting, analyze the result, and then make changes that would likely improve the performance. This process is repeated multiple times. However, running the simulation and analyzing the results can be very time consuming.
With the improved availability of high performance computers, so-called ensemble methods are becoming increasingly relevant in scientific computing: the same simulation is performed in parallel for a large number of different, pre-defined parameter settings.
The goal of this project is to develop new methods and techniques for ensemble visualization and analysis to improve and accelerate the virtual prototyping workflow of machine tools. By considering a large space of design parameters, design iteration will be accelerated and the expanded perspective to designers will yield results of increased robustness.
In our preliminary work (Comparative Visual Analysis of Lagrangian Transport in CFD Ensembles), we introduced a method that can be used to interactively explore similarities and differences in the transport behaviour of flow simulation ensembles, and to identify outliers and trends.
We will expand on this work with methods that leverage high performance computing capabilities for the visualization and analysis of ensembles to allow the identification of promising candidate prototypes and anomalies, and to facilitate fast, interactive visualization and analysis of individual simulation runs.
To mitigate the bandwidth limitations that arise in the context of high-performance computing with large data sets, we will develop new storage mechanisms for simulation output using Lagrangian interpretations of flow simulation data as well as compression techniques. Further, we are developing new methods for the analysis of large vector and tensor fields and ensembles that leverage the advantages of Lagrangian representations to provide new insights with both increased performance and accuracy.
Finally, we will apply our methods and build on simulations developed within the IRTG to investigate ensemble prototyping workflows for machine tools.
As a result of this project, we expect to obtain new methods and tools that leverage high-performance computing capabilities for visualization and analysis of simulation ensembles and enable the fast analysis of simulation results.
Illustration 1: In this ensemble visualization, 20 simulation results for a stirring apparatus are analyzed using a classification space that is spanned by two variances (left column). Areas of similar and dissimilar fluid transport behavior are exposed by this classification (center column). Further, the method can be used to detect areas with diverging trends and outliers (top right). Details can be found in Hummel, Obermaier, Garth, Joy: Comparative Visual Analysis of Lagrangian Transport in CFD Ensembles