Wind turbine performance assessment using multi-regime modeling approach

Prognostics & health management system is an integral component of any wind energy program to ensure high turbine availability and reliability. Traditional vibration-based condition monitoring practices have been proposed to be utilized with wind turbines as they have demonstrated varying degrees of success with other rotary machinery. However, high-frequency data such as vibration and acoustic emission signals, generally, are not collected and recorded due to limitations with data storage capacities. In addition, the highly dynamic operating conditions of a wind turbine pose a challenge to conventional frequency domain analysis tools. Thus, a systematic framework that utilizes multi-regime modeling approach is proposed to consider the dynamic working conditions of a wind turbine. Three methods were developed, and they were evaluated using SCADA (supervisory control and data acquisition) data only that have been collected from a large-scale on-shore wind turbine for 27 months. Empirical observations from the results of the three methods indicate the ability of the approach to trend and assess turbine degradation prior to known downtime occurrences.

► Addressed the dynamic operating conditions of wind turbines by utilizing multi-regime modeling for performance assessment. ► Validated the approach using three techniques on a 26-month real wind turbine SCADA data. ► Technique using Gaussian mixture models exhibited gradual trend prior to known failure events. ► Self-organizing map and neural network techniques showed more abrupt trend prior to known failure events.