Author(s): Boyang Jiao; Fang Li; Li Zhou; Liangliang Lu; Shifeng Ding; Pentti Kujala
Linked Author(s):
Keywords: Ice force peaks; Mixture distribution; Polar ship; Probabilistic modelling
Abstract: Accurate modelling of ice force peaks is essential for the structural design and safety assessment of polar ships. Aimed at the uncertainty of short-term ice force peaks, this paper presents a probabilistic analysis of full-scale ice force measurements on the bow of the polar vessel S.A. Agulhas II during Antarctic operations. Four commonly used single distributions, exponential, Weibull, gamma, and lognormal, together with their corresponding same-family mixture distributions, are systematically compared. A total of 200 candidate fits are examined for 10 measurement cases, and model performance is evaluated in terms of overall fit, upper-tail fit, mean relative tail-quantile error, and numerical stability. The results show that the best model in all 10 cases is a mixture distribution, indicating a consistent improvement over single distributions in both overall and upper-tail fitting. In the representative case, the best model, a six-component Weibull mixture, improves the overall coefficient of determination from 0.9478 to 0.9979 and the upper-tail coefficient of determination from 0.8738 to 0.9946, while reducing the mean relative tail-quantile error from 0.1282 to 0.0340. However, high-order Weibull mixtures show limited numerical robustness, whereas exponential mixtures exhibit the most stable performance.
Year: 2026