Prediction and Modelling of Extreme Sloshing Pressures on Liquid Cargo

Abstract

Sloshing is a well-known phenomenon that has attracted attention of researches over the last few decades. Sloshing in LNG cargo tanks had a new turn with changes in the LNG market at the end of 1990s. As a result, increase in tanks sizes and changes in operational conditions were inevitable which brought some technical concerns regarding sloshing problem. There are a great number of studies in the area of sloshing including analytic, experimental and numerical studies. Since sloshing is a complex liquid motion, the computational effect required for numerical analysis is very high. Therefore, experimental method is widely used in determination of slosh-induced loads. Accurate prediction of maximum pressure in a designated return period is a crucial step in structural design of LNG cargo containment system. In order to determine the maximum pressure, statistical post-processing must be carried out. In this step, it is important that an appropriate statistical distribution is used to describe the peak pressures. Traditionally, Weibull and generalized Pareto models are used in short term prediction

however, there is a need for a wider investigation in this area to find better alternatives for long term prediction. Another issue about sloshing impact pressures is the idealization of peak pressure signals. In the current procedure, peak pressure signals are modelled as triangular shapes for the simplicity of structural analysis. Triangular modelling that passes through rise and decay times at a certain ratio of peak pressure value is used most commonly. Since accurate modelling of peak pressure signals and determination of rise and decay time are significant in terms of structural response, the modelling of peak pressure signals must be studied in more detail. In this thesis, statistical analysis of sloshing impact pressures is carried out. To this end, various statistical models are applied to peak pressure data which were acquired from sloshing models tests of 5hrs duration (in real scale) repeated 20 times in 3 filling levels, and, for further analysis, the best 4 distributions are chosen which are Weibull, generalized Pareto, generalized extreme value and log-logistic distributions. Using different distribution fitting methods, these statistical models are applied to the data sets of peak pressures. The fits are evaluated using probability-of-exceedance curves and goodness-of-fit tests according to different filling levels. Another evaluation is carried out by comparing the squared error between accumulated peak pressure data (100hrs test data) and short duration test data (5hrs test data) fittings in different zones of return period. This evaluation results are also displayed in long term, being plotted to understand the behaviors of distributions in case of long term prediction. In addition, taking 100hrs test data as a reference, another comparison is made for the current short term prediction procedure of the classification societies. In the next part of the thesis, analysis on triangular modelling of impact pressure signals is carried out. The rise and decay times in 9 stations of pressure signals are extracted and utilized for comparing different pressure ratios of triangular signal modelling. The summed absolute difference between the rise and decay times in actual signal and modelled signal are calculated in these 9 stations. The comparison of pressure ratios are displayed in different percentages of highest peak pressures in each filling level. Considering the results, a suggestion is made for pressure ratio of triangular signal modelling.