The present paper introduces outlier analysis of sloshing-induced random impact pressure. Analysis of outliers is one of the essential steps in the study of conducting experiments and data analyses. However, it is difficult to find systematic study of outlier analysis in sloshing load problem due to the statistical characteristics of probabilistic distribution of sloshing impact pressure. In this paper, outlier analysis, which is a method to improve convergence through statistical analysis on experimental sloshing pressures lacking convergence, was developed. In order to validate the outlier analysis method, a series of 1:50 scale sloshing model test was carried out for 138k LNG cargo hold. From this study, it can be found that existing statistical analysis method is likely to lead to misinterpretation results due to the outliers due to lack of convergence of model experiments. Also, when the outlier analysis is performed, it is possible to reduce the distortion of the analysis result caused by the outliers and to predict the sloshing load more reliably based on the same model test data.
Sloshing is one of the key issues in the design of cargo holds for Liquefied Natural Gas (LNG) Carriers and other LNG-related offshore structures such as LNG-FPSO and FSRU. The interpretation of the sloshing phenomena within an LNG cargo is generally the use of experimental methods rather than analytic or numerical methods due to the complexities of physical phenomena.
Earlier, Mathiesen (1976) and Gran (1981) made pioneering achievements in statistical analysis of the sloshing problem. Mathiesen (1976) conducted a model test of irregular swaying motion and applied statistical analysis to the Weibull distribution function. Gran (1981) performed statistical analysis by applying the model test results to the Frechet distribution function. Graczyk et al. (2006, 2008) presented a systematic method of analyzing sloshing problems covering ship motion analysis, model test, structural response, and statistical analysis. Based on previous studies, many international classifications have published their guides or procedures for estimating sloshing design loads. The procedure of the International Classification Society is similar in that it estimates the sloshing load through the model test. However, each Classification is building its own procedure in the method of performing model test, analyzing experimental data, and estimating sloshing design load. The International Classification Society states that long-term model tests should be carried out to obtain sufficiently converged results for sloshing load prediction, but there is no clear standard (LR 2009). It is known that the minimum recommended time of the proposed model experiment is significantly lower than the convergence time (Kuo et al., 2009; Fillon et al., 2011; Kim et al., 2014).