Prof. Dr. Jochen Garcke, Dr. Daniela Steffes-lai, Dr. Rodrigo Iza-Teran, Mandar Pathare, Anahita Pakiman, Sara Hahner, and Christian Gscheidle presented their exploratory machine learning (ML) approach for computer-aided engineering (CAE) in vehicle safety optimization at Fraunhofer SCAI. Their research focused on combining ML with domain expertise to provide a data-centric view, assisting engineers in analyzing complex simulation data during virtual product development. The team developed tools for comparative and explorative analysis of data from numerical simulations, addressing automotive crashworthiness with Finite Element Method (FEM), forming processes like cup drawing, and computational fluid dynamics. These tools aid in visualizing deformation behavior modes over time, including for wind turbines under turbulent load, and in analyzing pairwise simulations. Key features of their work include the detection of extended contours, closed holes, thickness and material changes, duplicate parts, and changes in various components of FEM models. Their tools enable automatic documentation of design changes, including PDF reporting and JSON export, and offer a comparison of FE-SurfaceMesh versus CAD-HullMesh. They highlighted the concept of informed machine learning, where geometry-aware data representation simplifies the analysis pipeline and organizes numerous simulation results. Their patented method for low-dimensional representations of simulations aids in structurally organizing several simulation results for further processing. Their approach also includes post-processing of multiple simulation results, identifying distinct behaviors through clustering algorithms, and using anomaly detection workflows to analyze each simulation. This method helps in determining whether a new simulation fits within existing behavior modes. The team's work extends to building a knowledge layer that provides a new representation of data, combining structural and unstructured data, and facilitating quicker decision-making and improved design guidelines. This involves the use of open graph databases and the analysis of energy fingerprints for different development stages. In their use-case setup, they conducted 3D Detached Eddy Simulations with OpenFOAM to study the flow around two cylinders, aiming to find clusters for prediction, set up stop criteria for simulations, and identify anomalies and outliers. Their ML-assistance tools simplify handling data from many simulations, identifying behavior modes or outliers, and investigating correlations between design changes and results. Their research focuses on aligning simulation behavior with design changes, exploiting large language models and foundation models in CAE, and making design suggestions based on identified relevant design changes. Their work demonstrates the potential of ML in enhancing the CAE workflow, offering intuitive overviews of simulation behaviors, and providing predictive insights for vehicle safety optimization.