Data-driven prediction of ship heel in turns

AbstractInternational Code on Intact Stability (ISC) sets regulatory standards to ensure ship stability and prevent capsizing. One key criterion introduces an equation to estimate steady heel in turning passenger ships, enforcing a safety threshold. In practice, two issues arise: existing literature indicates that the equation produces results that deviate considerably from observations, and maximum heel is an important safety measure that is unaddressed by the ISC. In this study, we re-examined the ISC steady-heel formulation and developed a practical method to predict maximum heel. We first confirmed that the ISC steady-heel equation output can deviate substantially from observations. We then analysed its assumptions, recalibrated and extended its parameters, and investigated machine learning-based predictors. For the available dataset, none of these yielded sufficiently accurate steady-heel predictions, indicating that steady-heel is difficult to predict from early-stage design information. In contrast, maximum heel proved more suitable for modelling with the same inputs. We trained neural network predictors targeted at maximum heel and introduced an easily implementable adaptation of the ISC equation. Both approaches showed good agreement with measurements. Overall, our study clarifies the limitations of the current steady-heel criterion and provides a practical tool for maximum heel predictions during the early design stage.