Statistical Models of Environmental Data for Marine Structure Design

Prince-Wright, Robert (1992) Statistical Models of Environmental Data for Marine Structure Design. PhD thesis, University of Glasgow.

Full text available as:
[thumbnail of 13834017.pdf] PDF
Download (12MB)


In this thesis univariate and multivariate statistical inference is examined and then used to model the joint distributions of the environmental winds, waves and currents recorded by the DB1 data buoy. This model is then used to examine the return period responses of a tension leg platform using a linearised frequency domain solution. The thesis is arranged into eight chapters each of which has its own nomenclature, conclusions, tables, and figures. The references use a name and year system and are given at the end of the thesis. Chapter 1 reviews the contents of the thesis and outlines the analysis methodologies used to synthesise a joint probabilities model for wind, wave, and current magnitudes and directions. The use of this model in a level III, time-variant reliability analysis is then discussed to illustrate the two different design philosophies used by the American Petroleum Institute and the United kingdom Department of Energy. Chapter 2 summarises the wind, wave and current data recorded by the United Kingdom Offshore Operators (UKOOA) DB1 data buoy. This data has been assembled into a multivariate dataset and screened to assess if there is any underlying structure in the data. The marginal distributions of the population and monthly componentwise maxima are then examined to assess if the data result from the mixture of more than one population. Chapter 3 reviews both parametric and intrinsic estimators for univariate samples of data. The desirable characteristics of an estimator are examined and then used to select maximum likelihood (ML) as the best estimator for this project. One major advantage of this method is that the sampling covariance matrix for the model parameters can easily be estimated from the sample information matrix. The ML estimators and sample information matrices for the Weibull and Generalised extreme value distributions are then developed and applied to both the DB1 data and a sample of structural response time series. A comparison of population and extreme asymptotic methods is then made to determine which approach is most suited to environmental datasets. The results indicate population modelling is reasonable when the correct model is used and that the asymptotic approach can lead to poor estimators in the small sample case. Chapter 4 examines both intrinsic and parametric estimators for multivariate samples of data. The multivariate kernel density estimator is discussed and then used with the bivariate pairs of DB1 data to confirm the multivariate sample is unstructured in the statistical sense. The transformation of the marginal DB1 data to near-Normal distributed variates is then examined and extended to the multivariate case using the method of maximum likelihood. This method is applied to bivariate and multivariate sets of the DB1 data and the results are then used to select the best set of transformation parameters. The selection criterion for the best parameters is the accuracy of the extreme value predictions from the population model. The results demonstrate the transformed Normal estimator has margins that give accurate predictions for the 50 year return period values. In addition, when the modal value of say zero- up-crossing period conditioned on significant height is checked against the scatter plots it is found the results are in close agreement. The chapter then concludes with a review of the currently available multivariate extreme value models. Chapter 5 deals with the modelling of directional probabilities and in particular uses circular statistical theory with standard directional wave analysis theory to infer the parameters of cosine and von Mises models of directional spreading. The robustness of simply equating the angular moments of the data to the angular moments of a model is examined using simulation. The results indicate the second angular moments are more robust to noise in the buoy response. Consequently they are used with the directional spectra recorded in seastates with significant wave heights greater than 6.0 metres to determine if the spreading is more narrow in extreme seas than predicted by the Hasselmann and Mitsuyasu models. This comparison indicates that the Hasselmann study is applicable to extreme seas. Chapter 6 describes the frequency domain model of a tension leg platform that is used in a subsequent reliability study. The stochastic wind, stochastic first and second order wave, and steady current loading calculations are explained and then a series of parametric sensitivity studies are discussed. This identifies the winds, and waves as the primary causes of the response of the platform. The response calculation considers all six degrees of freedom and allows for the coupling of some modes of motion. Chapter 7 brings together all of the previous chapters into a time-variant reliability analysis of the tension leg platform developed in chapter 6. The effects of spectral shape, wind speed, and directional spreading on the within seastate exceedance probabilities for a variety of thresholds are examined to assess which parameters have a significant influence on the levels of structural reliability.

Item Type: Thesis (PhD)
Qualification Level: Doctoral
Additional Information: Adviser: Douglas Faulkner
Keywords: Ocean engineering, Statistics
Date of Award: 1992
Depositing User: Enlighten Team
Unique ID: glathesis:1992-76285
Copyright: Copyright of this thesis is held by the author.
Date Deposited: 19 Nov 2019 16:10
Last Modified: 19 Nov 2019 16:10

Actions (login required)

View Item View Item


Downloads per month over past year