“…By using data sources, different strategies can be followed to build normal behaviour models, such as stochastic models, machine learning algorithms, Bayesian and fuzzy classifiers, time series prediction or pattern recognition. In the analyzed failure detection system a state-of-the-art neural network, Extreme Learning Machine (ELM), has been applied, due to its ability to easily model dynamic non-linear behaviours [10], and also because of its wide use in the prognostics of industrial systems and wind turbines [11][12][13][14][15][16] Whatever the selected model may be, an initial training batch is necessary for learning relationships between variables. Taking a feed-forward neural network as example, the network is trained with an historical data set, a fixed and static information about the past events of the system.…”