Effizienzsteigerung im Machine Learning im CAE: Schon mit wenigen Variantenrechnungen zu optimalen Resultaten

The presentation highlighted Hexagon's capabilities in sensor, software, and autonomous solutions. The presentation also covered the application of Machine Learning across various types of CAE calculations. It emphasized predicting CAE results, optimizing parameters, utilizing smart superelements, image-based learning, and data mining. Hexagon's Odyssee software, which employs Reduced Order Methods (ROM), was showcased. ROM simplifies models to their most crucial relationships, allowing for efficient data analysis and prediction. The modal method's capability to operate with a smaller Design of Experiments (DOE) was explained. This method can effectively represent data with fewer modes, giving it an advantage over other methods that might require either a high polynomial approach (potentially leading to overfitting) or many points for piecewise linear fitting. A case study with Satven was presented, demonstrating how Odyssee helped reduce the time required for NVH (Noise, Vibration, and Harshness) simulations and optimizations from days to minutes. The presentation further illustrated the prediction of entire animations, interpolating various output formats like d3plot and hdf5. Hexagon also showcased SimManager and graph databases for learning from existing models and finding correlations between model and result changes. The capability of image-based prediction was highlighted, showing how an image can often convey more information than many parameters. This approach predicts sound pressure curves from images of tire tread patterns. The presentation concluded with an example of standalone image prediction. In this case, a gearbox simulation showed contact stress over the tooth surface. Here, result images were interpolated based on result files, with the image exported as a matrix, interpolated, and reassembled as an image.