Data analysis and modelling for onboard support of marine operations

Abstract

The ever-increasing exploration of ocean resources has led to more frequent and intensive marine operations. However, marine operations are vulnerable to accidents due to unpredictable environmental factors and human decisions. Therefore, providing onboard support for operators' decision-making is crucial to ensuring safe and sustainable marine operations. To achieve this, advanced sensors have been deployed on various offshore platforms to collect real-time data. As time goes on, this marine data accumulates and forms marine big data, characterized by five high Vs - volume, velocity, variety, veracity, and value. Marine big data speeds up the transition towards digitalization and automation, enabling onboard support for marine operations, such as ship motion prediction, path planning, and structural health monitoring.

The utilization of massive marine data to drive digitalization has become a significant topic in both research and industry. Data analysis and modeling offer a promising solution to address this issue. Data analysis is mainly divided into four categories: descriptive analytics, diagnostic analytics, predictive analytics, and prescriptive analytics. From a macro perspective, predictive analytics or modelling is a component of data analysis. Descriptive analytics answers the question: ‘What happened' while diagnostic analytics addresses the problem of ’why it happened'. Both these categories focus on discovering historical information in marine data. In contrast, predictive analytics and prescriptive analytics tend to analyze the future behaviours and events of a system by answering the questions: what will happen and how to make it happen. These four categories of data analysis involve a large variety of approaches, theories, and tools. Hence, they have been widely used to mine valuable knowledge and critical insights for onboard support of demanding marine operations.

Complex data types and different applications pose challenges to marine data analysis for onboard decision support. How to combine different analysis approaches to figure out these issues is the main concern in this dissertation. To show the importance of data analysis for onboard support of marine operations, three case studies are highlighted: ship dynamic positioning (DP) capability analysis, structural health monitoring, and Automatic Identification System (AIS) data analysis and modelling.

Ship DP capability is subject to the impact of environmental factors and the thruster's failure. Understanding the interaction between thrusters and environmental factors can provide support for DP capability improvement to prevent the occurrence of a loss of position. The objective of this study is to analyze the thrusters' significance under the influence of the thruster's failure and environmental disturbance via descriptive analytics, predictive analytics, and prescriptive analytics such as statistical analysis, Machine Learning (ML), and sensitivity analysis (SA). The experiment results show the feasibility of the proposed method.

Structural health monitoring aims to identify the modal parameters of offshore structures imposed by drifting ice. The covariance-driven stochastic subspace algorithm (SSI-cov) was proposed to identify physical modes (natural frequency, damping ratio, and mode shape). Many uncertain parameters of SSI-cov bring uncertainties to the identified modal parameters. To address this issue, diagnostic analytics, such as clustering, and prescriptive analytics such as uncertainty analysis (UA) and SA, are combined to quantify the uncertainty of the identified modal parameters. The results present the proposed method can achieve an efficient and accurate uncertainty quantification of the identified modal parameters. Additionally, it outperforms the traditional slack values-based SSI-cov.

AIS data contains rich information about ship status, which has been widely used for ship behaviour analysis. To take full advantage of information-rich AIS data, this study gives three applications: COVID-19 impact analysis, probabilistic ship route prediction, and short-term ship trajectory prediction. First, descriptive analytics and diagnostic analytics are used to extract important features and analyze the statistics of these features in the case of the Coronavirus disease 2019 (COVID-19) raging. This study mainly analyses the interaction between ship behaviours and COVID-19 impacts for the support of marine traffic management. Based on this work, the extracted features are further applied to the next application such as ship route prediction. Diagnostic analytics, including clustering and dynamic time warping (DTW), is chosen to make probabilistic ship route prediction. It is carried out in two steps. The first step is to cluster ship trajectories using clustering to render routes and the next step is to classify ship trajectories into different routes based on trajectory similarity estimated by DTW. Finally, the obtained route information is then used as prior knowledge for the ship trajectory prediction. A hybrid model is constructed based on historical trajectory information and online predicted ship positions obtained by Gaussian Process. The results demonstrate the proposed model outperforms the data-driven model and can obtain more accurate ship trajectory prediction.