Introduction
In control engineering, MATLAB/Simulink is the industry standard for modeling complex systems. However, while simulation is powerful for observing behavior under specific conditions, it cannot provide formal guarantees of safety or performance across all possible operating scenarios. Formal verification tools, such as SpaceEx, offer these guarantees but require models to be represented as hybrid automata—discrete modes with continuous dynamics.
The primary challenge lies in the "translation gap." Real-world systems like wind turbines are highly nonlinear, involving products, divisions, and high-order polynomials, whereas efficient verification algorithms typically require linear or affine dynamics. This paper presents a systematic workflow to transform a nonlinear wind turbine benchmark into a formal Piecewise Affine (PWA) hybrid automata model.
The ARCH Wind Turbine Benchmark
The study focuses on the ARCH wind turbine benchmark, a simplified yet industrially relevant model originally proposed by General Electric. The benchmark comprises several critical components:
- Physical Plant: Includes servo-elastic, aeroelastic, and pitch-actuator subsystems.
- Controllers: Features a generator-torque controller and a collective blade-pitch controller.
- Wind Profiles: Utilizes non-deterministic ranges to cover a wide variety of operating conditions, rather than relying on specific, deterministic wind signals.
A Systematic Transformation Workflow
The authors propose a four-step process to bridge the gap between simulation and verification:
- Model Adaptation: The original Simulink model is cleaned of non-essential blocks (e.g., scopes) and deterministic inputs are replaced with non-deterministic ranges to ensure the formal model covers all possible behaviors.
- Translation (SL2SX): Using the SL2SX tool, standard Simulink blocks are automatically converted into the SpaceEx (SX) modeling language, preserving the original design's hierarchy.
- Compositional Syntactic Hybridization: This core innovation involves breaking down individual nonlinear blocks (such as division operators) into PWA approximations. By adding a non-deterministic error term, the authors ensure the resulting model is a "sound over-approximation." If the formal PWA model is proven safe, the original nonlinear system is guaranteed to be safe.
- Model Validation: A "tester module" compares the reachability sets of the formal model against simulations of the original nonlinear model to ensure the translation is accurate and bounds the real-world behavior correctly.
Key Insights and Results
A major takeaway is the effectiveness of Modular Hybridization. By hybridizing at the component level rather than flattening the entire model, the system remains manageable, avoiding the exponential state-space explosion that often hinders formal verification. Furthermore, using non-determinism allows a single formal analysis to cover the entire range of wind speeds, making the process significantly more robust than traditional simulation.
The final transformed model consists of 72 discrete locations. While the theoretical state space is vast, SpaceEx's on-the-fly composition allows for successful reachability analysis on critical subsystems like the pitch actuator and torque controller. Validation tests confirmed that the formal reachable sets perfectly enveloped the nonlinear simulation runs.
Conclusion and Future Directions
While this workflow successfully translates complex benchmarks, challenges remain regarding state-space explosion for the full system and the need for manual intervention in certain preprocessing steps. Future work will focus on compositional reasoning, verifying individual components independently before combining results, to enable full-scale verification of the entire wind turbine benchmark without reaching computational limits.
Summary updated