الهام رشيديان

A wind turbine is a valuable, complex, and sensitive industrial machine composed of multiple components. Nowadays, monitoring and early detection of faults in wind turbines are of special importance. In this thesis, a two-stage fault diagnosis using the normal behavior model of the power curve is discussed using the collected data from an operational wind farm. For this purpose, initially, several preprocessing steps are conducted on the dataset and invalid samples are removed to obtain the normal behavior power curve of the wind turbine. Then, the normal data is divided into training and testing set with an 80:20 ratio. Several supervised regression-based normal behavior models are selected for the normal single-input power curve and implemented using the training dataset. At this stage, a new cubic spline two-inputs model including wind speed and rotor speed, is also introduced. Then, performance criterion of the models are calculated and compared using the testing dataset. The proposed three-dimensional model, provides good results, but previous existing methods still have better error criterion and are prioritized to use. In the next stage, the best normal behavior model of power curve is selected and used for wind turbine fault diagnosis with the residual-based EWMA control chart. To detect faults using the EWMA control chart, two cumulative threshold are proposed for early-warning and al‎e‎rt alarms and their results are recorded. These results are compared with the turbine status file and their effectiveness is proven.

منحني توان توربين بادي , تشخيص عيب , مدل حالت نرمال , ميانگين متحرك وزن‌دار نمايي , اسپلاين مكعبي

wind turbine power curve , Fault diagnosis , Normal Behavior Model , Exponentially Weighted Moving-Average , Cubic spline

Elham Rashidian

Dr. Javad Poshtan