Abstract

Wind turbine blades are subjected to wind pressure distribution that depends on the external environment and the inertial loads from their rotational velocity, acceleration, and turbine control. To simulate these effects, numerical tools are used. Allowing a coupled nonlinear aero-hydro-servo-elastic simulation in the time domain representing the multibody 1D beam finite element model (FEM). Nevertheless, when the mechanical analysis comes into detail, a shell FEM with applied loads in 3D spatial space must be used to analyze the fatigue. Therefore, the loads estimated by beam simulations need to be transferred into an equivalent 3D distributed loads for the shell FEM Called in the literature Load Application Methods (LAM), Each of these LAM differs in the stress distribution and the deflection of the blade. Subsequently, fatigue analysis of the whole blade can be performed by defining the cycle counting method and multi-axial damage criteria for composite material. However, this process is computationally expensive, since it is required to calculate the stress history in the shell FEM of the blade for each time instant of the aero-elastic simulation for different mean wind speed, other authors use a damage equivalent load (DEL) to estimate the fatigue damage directly from the 1D simulations. To reduce the number of call of the aero-elastic and shell FEM simulation a deep neural network (DNN) was trained to predict the accumulated fatigue damage in a node of the blade given the 10 minutes mean wind speed and the empirical cumulative density function of the damage per cycles with a relative error less than 5%