Quantifying Epistemic and Aleatoric Uncertainty in the Ampair 600 Wind Turbine

Abstract

Determining the uncertainty in a mechanical joint is very important and very difficult. This paper presents two methods of determining the uncertainty in the joint: maximum entropy approach and sampling methods. Maximum entropy is an approach that can quantify the aleatoric and epistemic uncertainty independently. This approach is applied on a rigid connection of the Ampair 600 Wind Turbine and shows that the epistemic uncertainty of the system is very high. Sampling methods are used on an simplified representation of the wind turbine as a lumped mass approximation. The sampling methods are able to treat the joint in a nonlinear sense by using a discrete four-parameter Iwan model as the joint model. This is able to predict accurately the data within the uncertainty bounds when considering epistemic uncertainty. The Iwan joint model is then implemented on the high fidelity model and preliminary results are presented. Keywords IntroductionDetermining the uncertainty in a mechanical joint is vital to model correctly the global system. The dissipation of energy through a joint can lead to higher damped systems than what is designed. While characterizing the uncertainty, there are two main types of uncertainty that are important to consider: aleatoric (parameter-based) and epistemic (model-based) uncertainty. Aleatoric uncertainty accounts for part to part variability and epistemic uncertainty corresponds to the uncertainty due to the unknown physics in the system such as model form error. In order to characterize this uncertainty, two different methods are used in this research: maximum entropy and sampling methods.Maximum entropy is an approach to quantify uncertainty that can determine the distribution of both the input and output variables [1]. This is combined with random matrix theory to create a set of model variables that maintain a positive definite matrix [2] and used for linear systems. Nonlinear systems can also use this method with some slight changes [3]. This approach is able to treat the aleatoric and epistemic uncertainty independently by determining the distribution of the parameters using experimental data to select an optimal dispersion parameter, which can be thought of similar to the coefficient of variation [4]. New advances in the maximum entropy approach are ongoing [5].