Generative Artificial Intelligence (AI) has found various applications in domains like computer vision and natural language processing. However, limited research exists in the engineering domain, where prevailing challenges involve mixed tabular data, data scarcity, and imbalances. This paper focuses on generating synthetic offshore jacket designs to improve the data quality of a scarce and imbalanced existing dataset. Data quality is quantified by evaluating the machine-learning efficiency of the synthetic data on a domain-specific downstream task. An integrated method is proposed for generating jacket designs, combining modern data-driven techniques with traditional multi-objective-driven approaches. The method addresses challenges related to mixed attributes, data scarcity, and class imbalances. Experimental results demonstrate improved predictive performance on the downstream task when models are trained on synthetic data compared to using only real data. These findings contribute to the advancement of generative AI in offshore engineering and related fields, offering valuable insights and potential applications.
