A Machine Learning Framework for Predicting Structural Failures in Ship Recycling: Overcoming Data Gaps Across Recycling Methods

This paper introduces a machine learning framework to predict structural failures during ship dismantling, focusing on four common recycling methods: Dry Docking, Pier Breaking/Alongside, Landing/Slipway, and Beaching. The framework leverages advanced techniques, including generative data imputation (WFGAIN-GP), GAN-based augmentation, and a multi-model approach combining Random Forests and Graph Neural Networks, to address challenges like incomplete data and method-specific risks. Results indicate that beaching poses the highest failure risks due to uncontrolled stresses, while Landing/Slipway methods demonstrated the lowest failure probabilities in this study, attributed to ship-specific factors such as age and corrosion rates. The model also highlights the influence of recycling methods on structural risks and showcases the potential for integrating real-time monitoring to enhance safety and operational efficiency. Future work will focus on validating the model with real-world data and incorporating advanced degradation models to improve predictive accuracy. This framework provides a robust foundation for improving safety and risk management in the ship recycling industry.