Predicting nonlinear motion instabilities in wave energy converters using data-driven methods

• Four data-driven methods are explored to analyze and predict motion instabilities of a wave energy converter. • A local predictability index can be used to predict instability magnitude in the near future. • Statistical features based on heave data are identified to predict sway motion instabilities. Floating structures can be subject to dynamics arising from nonlinear coupling between multiple degrees of freedom. This may result in unexpected behavior and give rise to instabilities. The phenomenon has been observed for traditional offshore structures and for emerging offshore renewable energy systems such as floating wind platforms and wave energy converters. The dynamics may alter the expected performance of the devices, and lead to reduced life time as the fatigue loads in critical components are affected. Understanding, predicting, and to some extent controlling these large nonlinear motions is important to guarantee the performance and survivability of these systems. Theoretical frameworks have been developed and are important to understand the underlying physics and mathematical description of the behavior. However, physics-based models are often subject to assumptions limiting their applicability. Moreover, experimental noise or prototype inaccuracies and imperfections may result in deviations from the predicted behaviour based on idealised theoretical models. Data-driven modelling can be a useful tool to complement the physics-based models, and are used increasingly in both industry and science. Here, we develop and explore four different data-based models to analyze and predict nonlinear dynamical instabilities for a point-absorbing wave energy converter. The models are compared to theoretical predictions as well as experimental data, and their capabilities differ significantly. Our findings show that a local predictability index can be used to predict instability magnitude in the near future, and that the time-varying stiffness coefficient driving the instability can be used to classify and predict motion instabilities.