Towards the Usage of Advanced Composite Materials in the Optimization of Wind Turbine Blades

Nowadays, fossil fuel consumption is the major contributor to the increasing concentration of greenhouse gases, which are the key cause of climate change. The main solution strategy is represented by renewable-energy sources, wind power among the others. However, existing wind turbines are made of composite material components that result to be very expensive and not easy to dispose of at the end of their life cycle. Therefore, substantial effort has to be put into the design phase of the new products, where numerical simulation and optimization techniques represent precious resources. This thesis presents advances in two main fields: (1) the mechanical analysis of natural fiber composite materials by means of experimental campaigns and numerical/analytical models, and (2) the development of modeling techniques for the Topology Optimization (TO) of mechanical components subjected to external loading conditions. The numerical and analytical models on natural fiber composites focus on the impact behavior of hemp/vinylester, flax/epoxy composite plates, together with hybrid stacking sequences where hemp and glass layers are alternated. The impact resistance of the laminates is investigated under low-velocity impact conditions, with loads in the thickness direction. In particular, analytical models available from the literature for carbon-fiber composites, are here adapted and tested on new material configurations. The main aims of these models are the prediction of a threshold impact load for the damage onset and extent and the approximation of the loading phase of the typical load–displacement curve. The TO of simple mechanical structures – ensembles of interconnected beams – exposed to static and dynamic external loads was addressed through surrogate modeling techniques. In this context, since the objective function evaluations rely on computationally expensive finite element simulations, optimization techniques based on the construction of cheap-to-evaluate approximation models are more convenient than population-based approaches, like for example evolution strategies, which need many more evaluations to reach a near-optimal solution. Novel approaches developed by the author of this manuscript are the Kriging-Assisted Level Set Method (KG-LSM) for TO, together with its hybrid version – the Hybrid Kriging-Assisted Level Set Method (HKG-LSM). Both these methods are based on Bayesian Optimization (BO), which is known to show poor performance of at relatively high dimensionalities, typically when dealing with more than 15 variables. To scale up BO for high-dimensional optimization problems, the Principal Component Analysis assisted Bayesian Optimization (PCA-BO) algorithm has been also developed. Several experimental and optimization studies have been performed, highlighting the potential of the presented techniques and the concrete possibility to benefit from their advantages in practical applications that are very relevant for the society.