Deep neural network-based wind speed forecasting and fatigue analysis of a large composite wind turbine blade

Abstract

The purpose of this paper is to analyze the modern deep neural networks such as nonlinear autoregressive network with external inputs and a recurrent neural network called long short-term memory for wind speed forecast for long-term and use the prediction for fatigue analysis of a large 5 MW wind turbine blade made of composite materials. The use of machine learning algorithms of advanced neural network applied for engineering problems is increasing recently. The present paper therefore brings as important connection between these latest machine learning methods and engineering analysis of complex wind turbine blades which are also the focus of researchers in renewable system design and analysis. First, a nonlinear autoregressive network with external inputs neural network model using Levenberg–Marquardt back propagation feed forward algorithm is developed with 5 years of environment parameters as input. Similarly, a long short-term memory based model is developed and compared. The chosen long short-term memory model is used for developing two-year wind speed forecast. This wind pattern is used to create time varying loads on blade sections and cross-verified with National Renewable Energy Laboratory tools. A high-fidelity CAD model of the NREL 5 MW blade is developed and the fatigue analysis of the blade is carried out using the stress life approach with load ratio based on cohesive zone modeling. The blade is found to have available life of about 23.6 years. Thus, an integrated methodology is developed involving high-fidelity modeling of the composite material blade, wind speed forecasting using multiple environmental parameters using latest deep learning methods for machine learning, dynamic wind load calculation, and fatigue analysis for National Renewable Energy Laboratory blade.

Cartographing dynamic stall with machine learning

“…While the vortex convection speeds are themselves an interesting result, the example should demonstrate to readers the incredible power of using pre-trained neural networks for extracting features from data. Deep neural networks are becoming increasingly used within the wind industry for applications, e.g., for predicting rotor icing (Yuan et al, 2019), power-curve estimation (Kulkarni et al, 2019) or even for rotor-blade inspections (Shihavuddin et al, 2019). We hope to demonstrate with this paper that modern machine learning tools and infrastructure can provide a useful boost to research in unsteady aerodynamics, wind energy and other adjacent fields.…”

A Strategic Study on the Environmental Impacts of Wind Turbine Blade Materials, their Improvements, Economics and End-Life-Options

“…Modern deep neural networks like nonlinear autoregressive networks are used for forecasting wind speed and fatigue analysis. Hence, involving highfidelity modeling for the composite materials of the blade, an integrated methodology is developed to establish time-varying loads over blade cross-sections to calculate dynamic wind loads, using multiple environmental parameters and deep learning for wind speed forecasting methods and fatigue analysis for a WT blade using the stress life approach with load ratio based on cohesive zone modeling (Kulkarni et al, 2019).…”

Cartographing dynamic stall with machine learning

“…While the vortex convection speeds are themselves an interesting result, the example should demonstrate to readers the incredible power of using pre-trained neural networks for extracting features from data. Deep neural networks are becoming increasingly used within the wind industry for applications e. g. for predicting rotor icing (Yuan et al, 2019), power-curve estimation (Kulkarni et al, 2019) or even for rotor-blade inspections (Shihavuddin et al, 2019).…”