Fatigue reliability using a multiple surface approach

Abstract

Reliability analysis for offshore wind turbine structural fatigue is an effort demanding task. The new trends in the design of these systems, such as, the usage of alternative computational fluid dynamics or finite element methods, are expected to further increase the effort required to assess fatigue in the design phase. There is a growing demand for techniques that enable practical fatigue design procedures. The present paper researches on how to use fatigue damage surfaces in order to assess stress-cycle (SN) fatigue reliability. A Gaussian process predictor model is applied as a surrogate of the fatigue damage, allowing the interpolation of multiple Gaussian distributed surfaces. Probabilistic SN curves are considered in the implementation, creating a double surface model where the Gaussian process model is built on top of the SN curve. Evaluation is performed on a 5MW turbine on a monopile foundation. Results of the implementation show that there is a significant advantage in using a surrogate of fatigue damage. These only require a limited number of time domain simulations to be defined. Moreover, the predictor surrogates accurately the design procedure within different material probabilistic characteristics, and accounting for loading uncertainty. Fatigue reliability assessment with Gaussian process models may be performed with approximately 10% to 40% of the computational effort in relation to the fatigue assessment using binned environmental conditions. The approach presented can be applied to any component and system, with the only requirement being the definition of a representative fatigue indicator to surrogate.