Detection of pitch failures in wind turbines using environmental noise recognition techniques

Modern wind turbines employ pitch regulated control strategies in order to optimise the yielded power production. Pitch systems can be subjected to various failure modes related to cylinders, bearings and loose mounting, leading to poor pitching and aerodynamic imbalance. Early stage pitch malfunctions manifest as impacts in vibration signals recorded by accelerometers mounted in the hub vicinity, as for example on the main bearings or nacelle frame, depending on the installed condition monitoring system and turbine topology. Due to the location of the above mentioned vibration sensors, impacts of various origin, such as from loose covers, can be generated, complicating the assessment of the impact nature. In this work, detection of pitch issues is performed by analysing vibration impacts from main bearing accelerometers and applying environmental noise and speech recognition techniques. The proposed method is built upon the following three processes. Firstly, the impacts are identified using envelope analysis, followed by the extraction of 12 features, such as energy, crest factor and peak to peak amplitude and finally the classification of the events based on the above features. Eighty nine impacts are analysed in total, where 60 impacts are categorized as valid and 29 as in-valid. It is shown Georgios Alexandros Skrimpas et al. This is an open-access article distributed under the terms of the Creative Commons Attribution 3.0 United States License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. that the frequency band of maximum crest factor presents the best classification performance employing K-means clustering, which is an unsupervised clustering technique. The highest correct classification rate reaches 90%, providing useful information towards coherent and accurate fault detection