Data driven uncertainty quantification for computational fluid dynamics based ship design

Maritime transport is responsible for an annual emission of around 1000 million tonnes of CO2, which is around 2.5% of the global greenhouse gas emissions. Nowadays, ships are designed using simplified operational profiles representing the expected operational profile during the lifetime of the ship. However, there is a discrepancy between these simplified profiles used for design and the actual full operational profile of a ship during its lifetime. This discrepancy leads to inefficient hydrodynamic ship design resulting in a waste of fuel and an increase of greenhouse gas emissions. The amount of available data on actual operational conditions of ships is rapidly increasing. The Automatic Identification System (AIS) and onboard monitoring systems produce a huge amount of historical data on ship operations. These developments call for efficient data-driven methods that account for this data. Knowledge of operational conditions can be used for Computational Fluids Dynamics (CFD) -based probabilistic uncertainty quantification leading to robust design: A hull shape that is optimal with respect to uncertain operational conditions. Robust design is a promising approach since it makes ships energy efficient for the real usage situation. ThreeUQ-methods are discussed: The perturbation method, the Polynomial Chaos Expansion(PCE) method and the multi-fidelity PCE method. The methods are applied to a simple one-dimensional test case to compute the stochastic moments of the effective power. The multi-fidelity Polynomial Chaos Method is found to be the most efficient UQ method. Moreover, the multi-fidelity PCE can be used as a surrogate for efficient Monte Carlo integration. This makes the method suitable for an Optimisation Under Uncertainty (OUU) algorithm leading to robust design.