Harsh S. Dhiman

Mili Wadhwani, Sakshi Deshmukh, Harsh S. Dhiman

Wind turbine is a complex machine with its rotating and non-rotating equipment being sensitive to faults. Due to increased wear and tear, the maintenance aspect of a wind turbine is of critical importance. Unexpected failure of wind turbine components can lead to increased O\&M costs which ultimately reduces effective power capture of a wind farm. Fault detection in wind turbines is often supplemented with SCADA data available from wind farm operators in the form of time-series format with a 10-minute sample interval. Moreover, time-series analysis and data representation has become a powerful tool to get a deeper understating of the dynamic processes in complex machinery like wind turbine. Wind turbine SCADA data is usually available in form of a multivariate time-series with variables like gearbox oil temperature, gearbox bearing temperature, nacelle temperature, rotor speed and active power produced. In this preprint, we discuss the concept of a digital twin for time to failure forecasting of the wind turbine gearbox where a predictive module continuously gets updated with real-time SCADA data and generates meaningful insights for the wind farm operator.

Juhi Patel, Lagan Sharma, Harsh S. Dhiman

In this manuscript, an image analytics based deep learning framework for wind turbine blade surface damage detection is proposed. Turbine blade(s) which carry approximately one-third of a turbine weight are susceptible to damage and can cause sudden malfunction of a grid-connected wind energy conversion system. The surface damage detection of wind turbine blade requires a large dataset so as to detect a type of damage at an early stage. Turbine blade images are captured via aerial imagery. Upon inspection, it is found that the image dataset was limited and hence image augmentation is applied to improve blade image dataset. The approach is modeled as a multi-class supervised learning problem and deep learning methods like Convolutional neural network (CNN), VGG16-RCNN and AlexNet are tested for determining the potential capability of turbine blade surface damage.

Harsh S. Dhiman, Dipankar Deb

Globally, wind energy has lessened the burden on conventional fossil fuel based power generation. Wind resource assessment for onshore and offshore wind farms aids in accurate forecasting and analyzing nature of ramp events. From an industrial point of view, a large ramp event in a short time duration is likely to cause damage to the wind farm connected to the utility grid. In this manuscript, ramp events are predicted using hybrid machine intelligent techniques such as Support vector regression (SVR) and its variants, random forest regression and gradient boosted machines for onshore and offshore wind farm sites. Wavelet transform based signal processing technique is used to extract features from wind speed. Results reveal that SVR based prediction models gives the best forecasting performance out of all models. In addition, gradient boosted machines (GBM) predicts ramp events closer to Twin support vector regression (TSVR) model. Furthermore, the randomness in ramp power is evaluated for onshore and offshore wind farms by calculating log energy entropy of features obtained from wavelet decomposition and empirical model decomposition.