A Data Clustering Approach to Identifying Slamming Types in Irregular Waves

A data clustering approach is presented to identifying types of slamming types in irregular waves. The approach is based on the k-means clustering method and is applied to data organized to include peak values of hydrodynamic loads, rigid- and elastic-body responses. Namely, pressure and resistance, acceleration and motions, strain, and finally wave height are investigated, focusing on the wave sequences causing severe slamming events. The test case under investigation is an 8 ft generic prismatic planning hull (GPPH) equipped with bottom grillage panels, towed at Fr = 1.8 in irregular waves, characterized by a significant wave height of 7.4 in and a modal period of 2.4 s. Data are provided by fluid-structure interaction computations with the CFDShip-Iowa code, combining computational fluid dynamic and computational structural dynamic solvers (CFD/CSD FSI). In order to identify the number of data clusters present in the data, and therefore the slamming types, two metrics are used, namely the within cluster sum of squares and the silhouette. In addition, the t-distributed stochastic neighbor embedding (t-SNE) is used to visualize data clusters in a reduced dimensionality space. The paper discusses how the proposed approach allows to investigate what type of wave sequences causes severe acceleration, pressure, strain peaks.