Real-Time Prediction of Slamming Loads on High-Speed Planning Craft Using Extended LSTM Networks

Slamming loads, characterized by severe impacts during vessel interactions with the water surface, pose significant challenges to the structural integrity and safety of high-speed marine vessels. This study leverages extended Long Short-Term Memory (xLSTM) networks, which incorporate enhanced memory capabilities and attention mechanisms, to develop real-time predictive models for slamming loads. These models consider critical factors such as hull shape, vessel state, and incoming wave field dynamics.

Utilizing high-frequency time-series data from prior physical tow-tank experiments of the General Prismatic Planning Hull (GPPH) model conducted at the United States Naval Academy (USNA), we train xLSTM networks to accurately predict the occurrence and magnitude of slamming loads. The xLSTM’s ability to capture both immediate and long-term dependencies allows it to focus on critical indicators leading to slamming events, providing a robust framework for real-time prediction.

Our goal is to explore the potential for real-time prediction and to demonstrate the potential improvements in prediction accuracy and computational efficiency compared to traditional methods through this approach. This study will contribute to the ongoing efforts to enhance the design and operational safety of surface vessels by mitigating the risks associated with slamming loads.