Unsupervised clustering and performance prediction of vortex wakes from bio-inspired propulsors

Abstract

An unsupervised machine learning strategy is developed to automatically cluster the vortex wakes of bio-inspired propulsors into groups of similar propulsive thrust and efficiency metrics. A pitching and heaving foil is simulated via computational fluid dynamics with 121 unique kinematics by varying the frequency, heaving amplitude, and pitching amplitude. A Reynolds averaged Navier-Stokes model is employed to simulate the flow over the oscillating foils at Re = 10 6 , computing the propulsive efficiency, thrust coefficient and the unsteady vorticity wake signature. Using a pairwise Pearson correlation it is found that the Strouhal number most strongly influences the thrust coefficient, whereas the relative angle of attack, defined by both the mid-stroke and maximum have the most significant impact on propulsive efficiency. Next, the various kinematics are automatically clustered into distinct groups exclusively using the vorticity footprint in the wake. A convolutional autoencoder is developed to reduce vortex wake images to their most significant features, and a k-means++ algorithm performs the clustering. The results are assessed by comparing clusters to a thrust versus propulsive efficiency map, which confirms that wakes of similar performance metrics are successfully clustered together. This automated clustering has the potential to identify complex vorticity patterns in the wake and modes of propulsion not easily discerned from traditional classification methods.

A Hybrid Computer Vision and Machine Learning Approach for Robust Vortex Core Detection in Fluid Mechanics Applications

“…In data science, clustering is a method for őnding clusters in a data set that are characterised by the greatest similarity within a cluster and by the greatest dissimilarity between clusters [37].…”

A Direct-Forcing Immersed Boundary Method for Incompressible Flows Based on Physics-Informed Neural Network

“…The effects of the DNN architecture on the convergence of training and velocity error are tested by using DNNs with nine different configurations. To be more specific, three numbers of hidden layers (5,6,7) and three numbers of neurons per layer (90, 100, 110) are considered. The results obtained by the nine are summarized in Table 2.…”

“…Over the past few years, machine learning (ML) has permeated into various research areas of fluid mechanics [1], e.g., reduced-order modeling [2,3], wake-type clustering and classification [4][5][6][7], development of turbulence closure model [8][9][10], flow optimization and active control [11][12][13][14][15][16], to name a few. The successful application of ML to these areas relies on the availability of large-scale data, which are obtained from CFD simulations or experimental observations.…”

PIV Snapshot Clustering Reveals the Dual Deterministic and Chaotic Nature of Propeller Wakes at Macro- and Micro-Scales

“…Examples of data-clustering analyses applied to propeller performance and, in general, vortex dynamics are also limited. Calvet et al [20] developed an unsupervised machine learning strategy to automatically cluster vortex wakes of bio-inspired propulsors into groups of similar propulsive thrust and efficiency metrics. Doijode et al [21] introduced a clustering approach for optimizing propellers by directing the search for the optimal design towards design clusters with good performance, i.e., high propulsive efficiency and low cavitation.…”