Cluster-based reduced-order modelling of the flow in the wake of a high speed train

Abstract

The flow field in the wake of a high-speed train is studied by cluster analysis and a cluster-based reduced-order model (CROM) is derived. The CROM strategy is a generalization of the Ulam–Galerkin method for the approximation of the finite-rank Perron–Frobenius operator and constitutes a data-driven approach to extract physical mechanisms in an unsupervised manner. Time-resolved data is first clustered into groups by using the k-means clustering algorithm to yield a small number of representative flow states, the cluster centroids. Then, the cluster transitions are modelled as a Markov process. A further analysis of the derived dynamic model provides information on the interaction of the dominant structures in the flow. The flow field around a generic high-speed train model, here the Aerodynamic Train Model, is obtained from a large-eddy simulation. This train model is designed to reproduce the geometrical features of the ICE2 train. The extracted flow structures can be associated with longitudinal vortices and vortex shedding. Furthermore, these structures are found to be associated with either states of low or high drag of the train.