Introduction
Wind turbines are increasingly deployed in remote and harsh environments, where they are subjected to extreme weather conditions and continuous mechanical stress. A critical subsystem for ensuring the safety and efficiency of these turbines is the pitch angle control system. This system adjusts the angle of the turbine blades to regulate power output and mitigate structural loads during high wind speeds. However, mechanical wear, such as friction changes in gear sets or hydraulic leakage, can significantly degrade the performance of standard controllers like PID, potentially leading to system failure.
Traditional control methods often struggle with the inherent nonlinearities of hydraulic actuators. While Sliding Mode Control (SMC) is a popular robust alternative, it suffers from two major drawbacks: "singularity" (where control variables become unbounded) and "chattering" (high-frequency oscillations that damage mechanical components). Furthermore, advanced models typically require precise knowledge of fault bounds—information rarely available in real-world operations. This paper presents a novel robust fault-tolerant control scheme that integrates Full-Order Terminal Sliding Mode Control (FOTSMC) with an adaptive Radial Basis Function (RBF) neural network to overcome these challenges.
System Modeling and Problem Formulation
The researchers focused on an electro-hydraulic servo pitch system, which is characterized by complex nonlinear dynamics. To design an effective controller, they derived a third-order nonlinear state-space model that accounts for hydraulic pressures, motor dynamics, and friction components.
The primary objective was to address mechanical friction changes caused by wear or component breakage. The proposed system must not only maintain precise pitch control under these faults but also do so without the destructive chattering effect. A key innovation in this work is addressing the "information gap"—the requirement for known disturbance bounds—by using an online estimator.
Proposed Methodology
The proposed control architecture consists of several integrated layers designed for fault diagnosis and tolerance:
- Disturbance Observer: This component acts as a real-time fault detector, providing a filtered estimate of the fault to assist operators in maintenance scheduling.
- Full-Order Terminal Sliding Mode Control (FOTSMC): Unlike standard SMC, the full-order approach ensures that system states converge to the origin in finite time while mathematically avoiding the singularity problem.
- Adaptive RBF Neural Network: This is the core innovation. The RBF network serves as an online estimator to approximate the change rate of the fault. By learning the fault dynamics on-the-fly, the controller eliminates the need for a pre-defined "upper bound" for disturbances, making the system more adaptable to different turbine configurations.
- Chattering Elimination: To protect mechanical actuators, the authors applied the discontinuous "sign function" to the derivative of the control law rather than the control signal itself. Subsequent integration acts as a low-pass filter, resulting in a smooth, continuous output voltage ($u$).
Simulation and Parameters
The effectiveness of the scheme was validated using a computer simulation of a 5-MW variable-speed, variable-pitch wind turbine. The simulation utilized realistic physical parameters, including inertia, leakage coefficients, and bulk modulus, to ensure the results mirrored industrial reality.
To test resilience, the authors injected both abrupt friction changes (e.g., a jump from -40Nm to -160Nm) and slowly developing sinusoidal faults. These scenarios represent common failure modes in pitch systems, such as sudden gear jams or gradual lubrication degradation.
Key Results and Insights
The simulation results demonstrated significant improvements over baseline models:
- Superior Fault Tolerance: When a fault quadrupled the friction torque, the proposed controller allowed only a momentary deviation in the pitch angle before returning it to the reference path. In contrast, a standard FOTSMC without the RBF estimator diverged significantly when faced with abrupt fault "jumps."
- Precision Tracking: The system successfully tracked both square-wave and sinusoidal wind speed changes, maintaining stability across diverse operating conditions.
- Smooth Actuation: The control voltage remained smooth throughout the simulation, confirming that the integration-based approach effectively eliminated chattering.
- Stability Guarantee: Using the Lyapunov method, the authors mathematically proved that the closed-loop system remains stable even while the RBF network is learning and adapting to new faults.
Conclusion
By combining FOTSMC with an adaptive RBF neural network, this research provides a "plug-and-play" solution for wind turbine pitch control that does not require exhaustive prior knowledge of disturbance bounds. While currently validated through high-fidelity simulations, the next step involves testing the method on a physical hydraulic pitch system rig to account for real-world sensor noise and signal-to-noise ratios. This work represents a significant step toward more reliable and autonomous wind energy systems in remote environments.
Summary updated