Learning hydrodynamic coefficient databases for vortex induced vibration prediction of marine risers using sparse sensor measurements

Abstract

Semi-empirical models are currently the state-of-the-art technology for flexible cylinder vortex induced vibrations (VIV) predictive modelling. Accurate prediction of the structural response relies heavily on the accuracy of the acquired hydrodynamic coefficient database. Due to the large number of inputs required, the construction of systematic hydrodynamic coefficient databases from rigid cylinder forced vibration experiments can be time-consuming or even intractable. An alternative approach has been implemented in this work to improve the flexible cylinder VIV prediction by machine-learning optimal parametric hydrodynamic databases using physical measurements along the structure. The methodology is applied to a straight riser in uniform flow and extended to non-straight riser configurations and non-uniform incoming flow profiles. Moreover, database inference is extended to using direct sparse sensor measurements along the structure. Specifically, a 19-dimensional parametric hydrodynamic coefficient database is obtained for: (i) straight riser in uniform flow (using either displacement or strain data) (ii) straight riser in stepped uniform flow (iii) straight riser in sheared flow (iv) catenary riser in uniform flow of various incidence directions between the catenary plane and the incoming flow stream (v) stepped (2-diameter) riser in uniform flow. The predicted amplitude and frequency responses, using the extracted databases, are compared with observed experimental results.