Sea Spray Icing Prediction: Integrating Experiments, Machine Learning, and Computational Fluid Dynamics

Abstract

<p>Ice accretion is challenging for maritime and offshore operations in the Polar regions. Activities related to tourism, oil and gas exploration, fishing, and offshore wind energy are increasing in the arctic. Ice accretion on vessels and offshore structures pose a threat to structural integrity, vessel stability, and personnel safety in outdoor working environments. Freezing sea spray is the largest contributor to marine and offshore icing, attributing to 80-90% of offshore icing incidents. <p>Sea spray icing is a niche field with comparatively limited research. Theory related to this field is difficult to find from a single source and is spread throughout literature. Having no single source to refer to for theory and standards makes it challenging for new researchers in the field. <p>Models for prediction of sea spray icing are essential for safer maritime operations in the Arctic. Existing models have varying approaches and provide rather varying predictions, making it difficult to say which one is more accurate. Additionally, existing models are heavily dependent on existing empirical formulations developed from limited observations for important variables like spray flux, something pointed out to be the weakest link in any prediction model. Using these formulations, typically based on medium sized fishing vessels, limits the predictions to the type and size of vessel the formulations are based on. ISO35106 points this out with a comment that none of the current models can predict sea spray icing on a wide range of vessels. <p>Full-scale testing of sea spray icing poses significant challenges with respect to personnel safety in extreme weather conditions, as well as the costs associated with such testing. This has resulted in limited full-scale or laboratory data. This in turn makes it difficult for validating prediction models as well as for the development of new and better models. <p>Apart from different approaches, prediction models could have different purposes. Operational prediction models, or forecasting models, predict an icing rate for a given set of metocean conditions without any consideration of the type of vessel or structure. These could be best suited for taking precautionary actions in case heavy icing is forecasted. Comprehensive prediction models, on the other hand, can estimate the spatial distribution of icing over the vessel or structure surface for a set of metocean conditions. In other words, these can estimate the icing rates as a function of location on the surface. These are more suitable at the design stage to take anti-icing measures or design optimization of the vessel/structure itself to minimize icing or that of the heating systems. <p>The main objective of this PhD project was to develop a ‘general’ model for prediction of sea spray icing. After collecting extensive data from laboratory experiments, a forecasting type of model, Spice, was developed ‘bottom-up’ from the experimental data using Machine Learning. An upgrade to this model is later made using Computational Fluid Dynamics to estimate some variables essential for prediction. This model, dubbed Spice2, is a comprehensive prediction model and independent of any existing formulations for estimation of spray. Spice2 was validated with icing measurements from a full-scale test. <p>The study presents a shift from the traditional modelling methods owing to the techniques used. The presented models show promising results with considerable scope for improvements in future research. Main contributions: <ol> <li>Consolidation of theory, standards, and existing prediction models in a single source.</li> <li>Valuable sea spray icing data from the largest set of laboratory experiments to date.</li> <li>SPICE (Sea sPray ICE): A machine learning model for prediction of sea spray icing.</li> <li>SPICE2: An upgrade to SPICE using CFD to estimate ship specific parameters for prediction of icing rates and distribution of icing.</li> </ol>