Modelling vortex-induced loads using machine learning

Dynamic fluid loading is of major concern to the off shore industry. Impacted structures may exhibit accel- erated fatigue and even failure. Load monitoring has therefore become an integral part of structural health monitoring. Obtaining direct load measurements can be cumbersome, e.g. obtaining strain gauge or pres- sure taps measurements. In this work, an alternative indirect method is proposed. The method revolves around building a mapping between more easily accessible wake velocity measurements and the induced loads. Classical machine learning techniques are adopted and relatively compact models, able to be run in real-time, are obtained. The method is illustrated on vortex-induced loads on a submerged cylinder.