Abstract
The pitch system plays a key role in large offshore wind turbines for regulating pitch angle to stabilize power output. However, due to time-varying parameter uncertainties and unknown disturbances, accurate regulation is difficult to achieve with existing strategies. Furthermore, actuator faults can reduce reliability and efficiency. This paper establishes a nonlinear model of the pitch system and proposes a neuro-adaptive fault-tolerant control (FTC) strategy incorporating a rate function. Co-simulation results verify the merits of the proposed method in achieving rapid, stable pitch regulation even under actuator failure.
Introduction
Floating offshore wind turbines (WTs) represent the frontier of renewable energy but operate in harsh, unpredictable environments. To maintain steady power output and prevent structural damage, these turbines utilize pitch control—adjusting blade angles to manage wind capture. Current systems face three major hurdles: high nonlinearity and un-modeled disturbances (like wind shear), mechanical actuator wear and tear, and control lag in standard PID controllers which struggle to distinguish between modeling errors and mechanical faults.
Methods
The proposed Neuro-Adaptive Fault-Tolerant Control scheme is built on a comprehensive nonlinear model that accounts for damping, friction, lifting moments, and stochastic aerodynamic loads.
At its core, the system utilizes Radial Basis Function Neural Networks (RBFNN) to approximate unknown nonlinear functions online. This allows the controller to learn and compensate for modeling errors and external disturbances in real-time. Additionally, a Time-Varying Rate Function is integrated to ensure "prescribed performance," forcing tracking errors to converge within a specific time frame. Actuator health is modeled via a factor ($\rho$), where values below 1.0 represent partial effectiveness loss.
The study was validated using high-fidelity numerical co-simulations involving the NREL 5MW Spar-type WT model, the FAST (Fatigue, Aerodynamics, Structures, and Turbulence) engineering tool, and Matlab/Simulink.
Key Results and Discussion
The proposed controller was tested against a standard PI-like sliding mode controller. In normal operation, the neuro-adaptive controller achieved significantly faster convergence to the desired pitch angle.
As seen in Figure 4, the proposed method provides a much smoother control output ($u$), which is critical for reducing mechanical wear on pitch bearings. Under simulated actuator failure (a 15% loss in effectiveness), the proposed controller maintained excellent tracking, with the error converging back to a nominal range almost immediately. This robustness ensures that generator torque and power output remain constant, stabilizing the grid even when hardware begins to degrade.
Conclusion
This research demonstrates that combining neural networks with prescribed performance rate functions creates a robust defense against the uncertainties of offshore environments and mechanical failures. While currently verified in simulation, future work will focus on physical hardware testing to account for real-world sensor noise and communication latencies, as well as expanding the fault model to include catastrophic "stuck" actuator scenarios.
Summary updated