Sensorless Friction Estimation for Condition Monitoring of Wind Turbine Hydraulic Pitch System

Hydraulic pitch actuators in offshore wind turbines are prone to wear that eventually may develop into failure of blade pitch control. This would leave the turbine nonoperational, causing significant loss in revenue while the defect remains. This paper proposes a technique for detecting early signs of increased friction within the hydraulic actuator, as an indicator of wear. The method utilizes a temporal convolutional network to reconstruct essential signals, which are fed into a modified least squares algorithm to estimate the Coulomb friction coefficient and continuously track friction magnitude. The robustness and accuracy of the approach is validated by high-fidelity simulations.