Wind turbine fatigue loads statistical estimation from standard signals

Fatigue loads represent a critical element in several aeronautical applications and Wind Turbines (WTs) are not an exception. Their evolution over the machine service life usually determines its life span, hence their behavior knowledge can cause a relevant economic impact on the Cost of Energy (CoE). Nevertheless, their measurements are frequently difficult or expensive and estimations arising from other known signals can be carried out instead. This thesis aims to design a load estimation model using standard signal as inputs by means of statistical and machine learning procedures. It includes an input selection stage and the models layout arrangement and assessment, which are based on Artificial Neural Networks (ANNs). The whole process is described in the following lines. First of all, data from simulations is gathered and examined. Damage Equivalent Load (DEL) (fatigue indicator which acts as target) and stats (statistical parameters from standard signals that constitute possible model inputs) are computed and their cross relationships analyzed. Secondly, input dimensions are reduced due to their great volume. In this regard several input subsets formed by small quantities of stats are built using two different filters: Correlation (COR) and Principal Component Analysis (PCA). Once this done, each subset estimation power is compared along with a great variety of ANNs configurations. Eventually, an innovative input selection method is developed using a genetic optimization. The outcome of this process allows several observations. On one hand, estimation results seem promising as long as all of them fall below 4%. On the other hand, multilayered nets with a neurons number between 30 and 60 seem a suitable configuration for these purposes. Furthermore, the genetic optimization results also show a great performance without the necessity of the preliminary work carried out by filters (which in fact is bias ed by some a priori assumptions). To sum up, inside the wind power field -where lifespan is an important factor due to its influence on the CoE- this method may promote a more efficient usage of structural parts enhancing their service life.