You are hereSecond-order load and response models for floating structures: Probabilistic analysis and system identification

Second-order load and response models for floating structures: Probabilistic analysis and system identification

Todd C. Ude
Principal Advisor: 
C. Allin Cornell
Year Published: 
Sun, 01/01/1995 (All day)

Wave forces on large-volume structures are nonlinearly related to the incident wave elevation. The nonlinear transformation gives rise to forces which oscillate at sums and differences of the incident wave frequencies, driving resonant response in moored, floating structures. Second-order models based on diffraction analysis have been used increasingly in recent years to model these forces.

Our first objective is the development of efficient methodologies to perform probabilistic analyses for quantities relevant to design, namely fatigue damage accumulation and extreme motions. We focus on the propagation of a stationary, random wave input through a deterministic second-order system. Methodologies for integrating these stationary-input results over long-term environmental variation are presented. These allow us to develop a global view of the importance of the nonlinearity and the implied non-Gaussian nature of the response. We demonstrate the application of these methodologies using a realistic model of an operating structure (the Snorre TLP).

Our second objective focuses on the use of data in verifying and identifying second-order models. Two approaches are considered: in the first, we begin with our best theoretical model as a framework, allowing us to use all of the observed data to robustly estimate a few unknown parameters to best explain the observations. Using model test data, we find that the diffraction-based model can predict the observed behavior satisfactorily, when appropriate damping values are identified from the data. The second approach proposes to fit by linear regression the optimal second-order model to explain the observed data. This non-parametric approach provides greater flexibility, but requires significantly more data to robustly fit the increased number of unknowns. The application of this approach to model test data is demonstrated and discussed.