Intelligent Data-Driven Maintenance Planning for Marine Renewable Energy Systems

Abstract

This thesis addresses marine renewable energy (MRE) maintenance literature gaps, thus lowering the levelized cost of energy (LCOE) for MRE and promoting its adoption. Theme 1 reviews MRE maintenance literature to identify trends and gaps in academic work. Key findings include condition-based maintenance (CBM) being valuable and mathematical optimization methods being scant. Theme 2 then uses machine learning in CBM for rotational systems. Minimalist Bi-LSTM architecture is tuned for bearing diagnostics, outperforming state-of-the-art (SOTA) models. A hybrid deep learning model then estimates system reliability for a chance-constrained selective maintenance problem (SMP). Performance matches SOTA models while yielding survival-focused maintenance plans. Lastly, for Theme 3, an SMP tailored for MRE minimizes LCOE and introduces work location probabilities. A deterministic SMP is developed, along with an uncertain variant solved through robust optimization. Results highlight the LCOE’s balancing of competing objectives and the robust model’s mitigation of onshore work impacts.