1: \begin{abstract}

2: We propose a novel cluster-based 

3: reduced-order modelling  (CROM) strategy  for unsteady flows.

4: CROM combines the cluster

5: analysis pioneered in Gunzburger's group \citep{Burkardt2006cmame} 

6: and transition matrix models

7: introduced in fluid dynamics in Eckhardt's group \citep{Schneider2007pre}. 

8: CROM constitutes a potential alternative to POD models

9: and generalises the Ulam-Galerkin method classically used

10: in dynamical systems to determine a finite-rank approximation of the 

11: Perron-Frobenius operator.

12: The proposed strategy processes a time-resolved sequence of flow snapshots in two steps.

13: First,

14: the snapshot data are clustered into a small number

15: of representative states, called centroids, in the state space. 

16: These centroids

17: partition the state space in complementary 

18: non-overlapping regions (centroidal Voronoi cells).

19: Departing from the standard algorithm, 

20: the probabilities of the clusters are determined, 

21: and the states are sorted by analysis of the transition matrix.

22: Secondly, 

23: the transitions between the states are dynamically modelled 

24: using a Markov process.

25: Physical mechanisms are then distilled 

26: by a refined analysis of the Markov process,

27: e.g.\ using finite-time Lyapunov exponent and entropic methods.

28: This CROM framework is applied to

29: the Lorenz attractor (as illustrative example), 

30: to velocity fields of the spatially evolving incompressible mixing layer

31: and the three-dimensional turbulent wake of a bluff body.

32: For these examples,

33: CROM is shown to identify non-trivial quasi-attractors and transition processes

34: in an unsupervised manner.

35: CROM has numerous potential applications

36: for the systematic identification of physical mechanisms of complex dynamics,

37: for comparison of flow evolution models, 

38: for the identification of precursors to desirable and undesirable events,

39: and for flow control applications exploiting nonlinear actuation dynamics.

40: \end{abstract}

41: