Machine learning-based techniques for marine structures: A state-of-the-art review

The ocean, as one of the vast territories on the Earth, holds abundant resources to be explored, while the high-quality construction and life-cycle maintenance of marine structures are fundamental for advancing this progress. Compared to inland engineering, more severe challenges are posed for marine structures, such as structural durability, variable loading, and harsh environments, making it challenging for traditional analysis and simulation techniques to meet the growing demands of this field. Artificial intelligence, representative by machine learning (ML) and deep learning (DL) algorithms, exhibits robust capability to handle highly nonlinear and complex problems, providing a promising alternative for addressing such challenges in the marine structures. The trend of implementing ML techniques in marine structures is rising, driven by updated algorithms, enhanced computing powers, and high-quality databases. This paper provides a systematic review of the application of ML in marine structures, paying attention to the typical ML and DL algorithms, the popular development platforms, the modeling process, and the application status in terms of design, construction, and maintenance stages. Given that ML algorithms heavily rely on data patterns while generally ignoring the mechanical principles underlying prediction problems, a novel structural mechanisms-based modeling process is proposed to achieve reliable and reasonable predictions. Based on the findings from the review, further research directions and challenges are highlighted and discussed. This paper aims to provide valuable resources for structural engineers and researchers to understand and engage in this promising domain.