Code for our paper titled
"Generative AI based Augmentation for Offshore Jacket Design: An Integrated Approach for Mixed Tabular Data Generation under Data Scarcity and Imbalance"
The dataset of the 100 real jacket substructure designs described in our paper can be found under real_structures.json
To compute all objectives for a given population, run evaluate_designs.ipynb, placing the .json file of the population under Genetic_Algorithm/util/data/to_evaluate.
All results and figures in the paper can be generated by this codebase.
Figure 2: Dataset distributions for all features in our dataset. The y-axis represents the
percentage of data.
| (a) | (b) | (c) |
|---|---|---|
 |
 |
 |
Figure 6: Graphical representation of the plausibility objective and threshold calculation for a one-class SVM. (a) depicts the learned hyperplane of the SVM. (b) shows the distribution of the signed distances of the real data points to the hyperplane, and the chosen threshold. (c) represents an example Pareto-optimal set of GA-generated solutions constrained by a threshold.
Figure 7: Histogram distributions for continuous and discrete features of our database and of synthetic data generated by all data-driven methods.
Figure 8: Improvement of MLE metrics (CatBoost model) due to increased number of
synthetic samples.
Figure 9: Histogram distributions of the brace type feature (C) for real and NSGA-II generated data. Plausibility-constrained solutions (yellow) align more closely with the real data, while non-constrained solutions (red) explore alternatives more extensively.
Figure 10: Joint scatter plot of our evaluation metrics for different runs of NSGA-II, by gradually increasing (left → right) the plausibility threshold.