go back

Structural Behavior Clustering Methods for Topologically-Optimized Designs

Yasuyuki Shimizu, "Structural Behavior Clustering Methods for Topologically-Optimized Designs", Technische Universität München, 2021.

Abstract

This thesis shows how to distinguish the dynamic structural behavior under the crash loading using several distance metrics to compare each time series data and clustering algorithms. Dynamic Time Warping is used to compare time series, which have different time scales. For the clustering method, OPTICS and k-medoids methods are used and compared. The first part deals with some simple examples, such as the movement of a spring system and beam bending, to reveal the method’s validity for calculating the metrics and clustering method. The combination of Dynamic Time Warping and OPTICS is precise to capture different modes, behavior, and periods (for periodic behavior). In the second part, topologically optimized structures are used to represent more realistic cases. This reveals that the suggested method is quite useful to classify several structures into the cluster with similar dynamic behavior. This work can support the designers in selecting representative structures at the beginning stages of design.



Download Bibtex file Per Mail Request

Search