Data-driven pressure field prediction for ships in regular sea states

Abstract

Merchant shipping is responsible for more than 90% of the global trade and has a significant environmental impact, accounting for over 2% of global greenhouse gas emissions. Therefore, fuel-saving measures are becoming increasingly important in reducing the ecological footprint and increasing the fuel efficiency of maritime transport. Routing optimization systems, which require a rapid prediction of ambient-dependent fuel consumption, represent an essential pillar here, e.g. to reduce added resistances due to seaways and/or wind. The paper aims to predict the added resistance due to seaways. In contrast to conventional methods the goal is achieved by surrogate modeling of the entire pressure fields. To this end, an online/offline-procedure is applied to an exemplarily free-floating container vessel. The online approach to be trained consists of two building blocks, namely a convolutional autoencoder (CAE)-based order reduction step and a neural network-based (NN) regression step that links the reduced space of the autoencoder with three control parameters that describe the sea state (wave height/steepness, encounter angle and wave length). Training data is obtained from time-averaged values for simulating instantaneous ship motion and pressure fields. During the offline phase, the combined CAE/NN is trained to capture the time-averaged pressure fields for a variety of sea-state conditions. During ship operation (online phase), the surrogate model predicts the three-dimensional pressure fields in response to sea state conditions, projects the pressure fields onto the ship hull, and integrates the corresponding resistances to guide the route. The evaluation of the method shows promising results for the different building blocks and the concept could therefore represent an attractive approach for cost-effective surrogate modeling of complex multiphase flow fields.