The presentation focuses on efficient optimization based on uncertainty estimations during simulation runtime. They introduce "Explorative (In-Situ) Data Analytics for CFD" with the SCAI NDV approach. This method aims to extract knowledge from every simulation to gain physical insights and identify global data-driven features. The process supports the simulation by providing stop criteria to save computational efforts, detecting outliers or redundant simulations, and identifying regions of interest for post-processing.

They emphasize the role of data-driven and machine learning methods to reduce data size and derive high-density features for automated analysis. The focus is on interpretability and 'Grey Box' methods to enhance trustworthiness and accuracy. The in-situ analysis allows for visualizing data during runtime, reducing data input/output during and after simulation, and providing earlier availability of results for simulation monitoring.

The presentation covers the challenges and approaches to estimate time-average uncertainty in turbulent flow simulations. They discuss the in-situ estimation of uncertainties, exemplified by flow around an airfoil, comparing different approaches like NOBM and MACF. They also explore explorative analysis in automotive HVAC ducts, focusing on anomaly detection, monitoring solutions, and setting stop criteria.

A significant part of their approach is based on the Laplace-Beltrami operator for simulation monitoring and validation, and they utilize techniques like Proper Orthogonal Decomposition (POD) for data-driven feature extraction. The presentation concludes with a discussion on clustering of results and prediction of flow behavior, highlighting the use of local surrogate models for improved predictions. They stress the potential of their method in predicting characteristic flow behavior depending on cylinder position and other parameters. The presentation ends with thanks and contact information for further inquiries.