An explainable deep learning method for the prediction of the vertical wave bending moment

Deep Neural Networks (DNNs) can be essential in determining shiploads and forces. Data-driven powered systems can be trained using extensive data sets of ship design parameters, environmental conditions and historical performance data to develop predictive models that can accurately estimate the loads and forces acting on a ship. The scope of this work is to predict the vertical wave bending moment (VWBM) for a ship (chemical oil tanker) whose operational and design parameters, obtained via numerical methods, are unknown to the trained algorithm. Such an approach may make it possible to bypass the costly computational time required for such numerical predictions. This research paper examines the role of Explainable AI in the context of deep learning for VWBM prediction. SHAP (Shapley Additive Explanations), based on the game-theoretically optimal Shapley values, was applied. The relevant inputs (e.g., pitch, heave) were analysed and processed in this framework to understand how the DNN achieved a particular output based on the limited features available.