Part A: Innovative Data Augmentation Approach to Enhance Machine Learning Efficiency—Case Study for Hydrodynamic Purposes

These days, AI and machine learning (ML) have become pervasive in numerous fields. However, the maritime industry has faced challenges due to the dynamic and unstructured nature of environmental inputs. Hydrodynamic models, vital for predicting ship responses and estimating sea states, rely on diverse data sources of varying fidelities. The effectiveness of ML models in real-world applications hinges on the diversity, range, and quality of the data. Linear simulation techniques, chosen for their simplicity and cost-effectiveness, produce unrealistic and overly optimistic results. Conversely, high-fidelity experiments are prohibitively expensive. To address this, the study introduces an innovative feature engineering that incorporates uncertainty into features of linear models derived from higher fidelity modeling. This enhances productive data entropy, positively enhancing feature classification and improving the accuracy and feasibility of ML models in hydrodynamic responses of floating vessels. Tested with data from a known geometrical shape exposed to regular and irregular waves, the technique employs Ansys Aqwa for linear models. The results demonstrate the efficiency of the proposed technique, expanding the applicability of ML models in realistic scenarios. The application of the proposed approach extends beyond and can be further applied to any stochastic process, which expands the ML application for realistic use cases.