Evolutionary algorithm for optimisation of condition monitoring and fault prediction pattern classification in offshore wind turbines

Online condition monitoring and fault prediction will become state of the art in the next generation multi megawatt offshore wind turbines. Information about a wind turbine's condition can be obtained from measured and processed time signals, so called characteristic values. This paper presents an automated approach for detecting and predicting the development of wind turbine fault conditions. The paper introduces such an automated approach on the basis of a self learning classifier, trained and optimised by an evolutionary algorithm. Representative input patterns with their related output patterns have to be formed out of measured wind turbine process data. This is done by human experts who analyse the coherence between typical fault descriptive input patterns and the respective fault conditions. With the expert defined input and output patterns, an evolutionary optimisation algorithm is applied to a pattern classifier. The structure of the input and output patterns, the functionality of the classifier and optimisation algorithm as well as some first results from performing evolutionary optimisation cycles are subject of this paper. The data used to demonstrate the function of the classifier and evolutionary algorithm are taken from a small 33 kW experimental wind turbine. Although such a turbine does not represent the state of the art, the principal results described in the paper can be transferred to condition monitoring systems for modern type offshore machines