Session: 02-05-04 Extreme Loads and Responses 4

Paper Number: 131595

131595 - Extreme Structural Response of Wind Turbines to Environmental Loading 

Reliable estimation of extreme structural response statistics is crucial for the design and operation of marine structures. For this, both the long-term variability of the metocean environment and the short-term variability of structural responses must be considered. Traditionally, this process involves fitting known probability distributions to environmental data and the corresponding physical dynamic responses. A limitation of this approach is its sensitivity to the choice of distribution, which introduces uncertainty and potential nonconservative inaccuracies in the estimated extreme structural response.
In principle, a full long-term analysis is preferred, where all possible environmental conditions are considered. However, this is often not feasible due to high computational cost, and approximate methods need to be used, where response calculations are only performed for selected load cases. The environmental contour method is one approach that is often used in ocean engineering applications, where a limited number of environmental conditions is selected for response calculations based on their exceedance probability. More recently, an approach based on sequential sampling and a Gaussian processes regression has been proposed for extreme structural response analysis.

In this paper, we investigate the application of scenario theory both as a distribution-free alternative and to obtain conservative extreme value distributions. We propose two novel methods: first, a risk-based and chance-constrained optimization approach to learn functions that bound the joint environmental and structural data with varying degrees of tightness. For instance, we might seek a function that bounds most of the responses (so the prediction is more informative but riskier after eliminating a few outliers), or might seek a function that bounds all available responses (so the prediction is more conservative but less risky).

Then, we propose a second method focused on establishing extreme value distributions that are conservative with respect to tail probabilities. In this setting we consider moment-matching distributions that maximize the probability of the response exceeding a fixed limit. In addition, we use scenario theory to rigorously bound the probability of unseen data exceeding the predicted upper bounds.

We explore how these strategies can be applied in the case where real measurements of the physical response can be obtained, and for the case where the physical response is computed using physics-based simulation. In the latter case, multiple realizations of the stochastic response can be simulated for selected environmental conditions. For numerical experiments we use a simulator of aeroelastic response of a floating offshore wind turbine.
Our findings suggest that the proposed methods can provide a resilient framework for dealing with the variability and uncertainties inherent in the assessment of extreme structural response.

 

 

Presenting Author: Luis Crespo NASA Langley Research Center

Presenting Author Biography: Dr. Crespo is a senior research scientist at the Dynamic systems and Control branch of NASA Langley Research Center. His research interests include: system identification, uncertainty quantification, stochastic systems, adaptive and data-based control, stochastic optimization and learning theory. He is the author of 100+ peer reviewed conferences and journals in these areas.

Authors:

Luis Crespo NASA Langley Research Center
Christian Agrell DNV
Erik Vanem DNV