Abstract
In this paper, a data driven framework for performance and maintenance evaluation (PAME) of wind turbines (WT) is proposed. To develop the framework, SCADA data of WTs are adopted and several parameters are carefully selected to create a normal behavior model. This model which is based on Neural Networks estimates operation of WT and aberrations are collected as deviations. Afterwards, in order to capture patterns of deviations, self-organizing map is applied to cluster the deviations. From investigations on deviations and clustering results, a time-discrete finite state space Markov chain is built for mid-term operation and maintenance evaluation. With the purpose of performance and maintenance assessment, two anomaly indexes are defined and mathematically formulated. Moreover, Production Loss Profit is defined for Preventive Maintenance efficiency assessment. By comparing the indexes calculated for 9 WTs, current performance and maintenance strategies can be evaluated, and results demonstrate capability and effectiveness of the proposed framework.
