Condition Monitoring for Wind Turbine Pitch System Using Multi-parameter Health Indicator

Abstract

As one of the critical components of wind turbines, pitch system suffers the highest failure rate and downtime. Thus, it is particularly important to improve the reliability and reduce the operation and maintenance (O&M) costs of pitch system. With the development of condition monitoring (CM) technology, the condition based maintenance is often employed to reduce the O&M costs of wind turbines. This paper proposed a data-driven CM approach to identify the abnormal condition of pitch system, using supervisory control and data acquisition (SCADA) data. The result of the proposed approach could be useful for wind farm operators to improve their maintenance strategy. Firstly, the Relief algorithm is used to select the appropriated parameters as the health indicator for evaluating the condition of pitch system. Meanwhile, the correlation between the selected parameters is considered to remove the redundant variables. With the selected parameters as the model output and environment variables as the model input, then the random forest regression based model is established by only using the historical healthy SCADA data. Eventually, the distance calculated based on the residual errors between model output and monitored value is used to identify the abnormal condition of pitch system specifically. Several case studies have been analyzed to validate the feasibility of the proposed method. The results indicate that the proposed method can detect the pitch system faults earlier than the SCADA alarm system. It is applicable for the on-line CM of pitch system because of simplicity and low cost.