Inverse modeling of wind turbine drivetrain from numerical data using Bayesian inference

Abstract

Wind turbine drivetrain systems experience occasional failure in their operational life-time. Dynamic loads in these systems often differ from those predicted in the design phase, which may lead to incipient faults and damage eventually resulting to unpredicted failures. Faults in a system cause changes in the system dynamic properties. Therefore, a calibrated dynamic model can be used for fault diagnosis of the drivetrain system. The current paper implements a Bayesian inference method for physics-based model updating for the first time to estimate the parameters of a physics-based wind turbine drivetrain model as well as the unknown input loads using measurement data. An information-theoretic approach is adopted to study the identifiability of different model parameters given different subsets of drivetrain measurements. The 5 MW NREL wind turbine is used as a case study, and its drivetrain is modeled by a 5-mass torsional model. Torsional shaft stiffnesses and rotor mass moment of inertia are the parameters reflecting dynamics of the drivetrain and are assumed unknown. Moreover, the aerodynamics input torque applied to the drivetrain system is assumed as unknown. The Bayesian inference method is used to estimate jointly the unknown model parameters and aerodynamic input torque using numerically simulated measurement data. The case studies show that the proposed model updating approach can accurately estimate the low-speed shaft stiffness, rotor mass moment of inertia, and the time history of aerodynamic torque jointly using only measurements from the generator including generator torque and rotational speed.