Applications of Linear and Nonlinear Robustness Analysis Techniques to the F/A-18 Flight Control Laws

Abhijit Chakraborty, Peter Seiler, Gary BalasAIAA 2009 · DOI

TL;DR

This paper investigates the robustness of the F/A-18 Hornet's flight control laws, comparing a baseline version that suffered from a dangerous "falling leaf" instability with a revised version designed to prevent it. While linear analysis showed only minor improvements, nonlinear methods revealed a substantial increase in stability and maneuverability with the revised control law, highlighting the critical role of nonlinear analysis in understanding complex aircraft dynamics.

What Problem Does This Paper Address?

The F/A-18 Hornet, a high-performance fighter jet, has a history of experiencing a dangerous, uncontrolled flight phenomenon known as the "falling leaf mode." This instability, characterized by large, coupled oscillations in roll and yaw, has led to aircraft losses. To combat this, the aircraft's baseline flight control law was significantly redesigned. The core problem this paper addresses is how to rigorously compare the robustness and stability of the original (baseline) and the new (revised) flight control laws, particularly in the context of preventing this severe nonlinear instability. Traditional linear analysis techniques, while standard for flight control design, may not fully capture the complex, nonlinear dynamics that lead to such extreme behaviors.

Methods

The authors employ a multi-faceted approach to compare the two control laws:

  • Linear Analysis: This involves standard techniques like gain, phase, and delay margin analysis, as well as multivariable disk margin analysis and structured singular value (μ) analysis for input multiplicative uncertainty. These methods assess the system's stability and robustness in the vicinity of equilibrium points.
  • Nonlinear Analysis (Region of Attraction - ROA): This is the paper's key contribution. They use sum-of-squares (SOS) programming to estimate the region of attraction (ROA) for the closed-loop system. The ROA is the set of initial conditions from which the system will return to its stable equilibrium. A larger ROA signifies greater robustness to disturbances. They employ both linearized and higher-order (quadratic and quartic) Lyapunov functions to estimate lower bounds of the ROA, and Monte Carlo simulations to estimate upper bounds.
  • Model Reduction: To make the nonlinear analysis computationally feasible, a reduced-order, cubic polynomial model of the F/A-18 dynamics was derived from a full 6-DOF nonlinear model.

Datasets

The paper utilizes a detailed, six-degree-of-freedom (DOF) nonlinear model of the F/A-18 Hornet aircraft. This model is based on aerodynamic data from the F/A-18 High Alpha Research Vehicle (HARV). The authors then derive various reduced-order linear and polynomial models from this full model for their analyses. No new experimental datasets were introduced; the evaluation is based on simulations derived from these established models.

Key Insights and Ideas

  • Linear vs. Nonlinear: A Critical Distinction: The most striking insight is the stark contrast between linear and nonlinear analysis results. Linear methods suggested only marginal improvements in robustness for the revised control law, while nonlinear ROA analysis revealed a dramatic increase in the guaranteed stability region. This underscores that for systems exhibiting severe nonlinear phenomena like the falling leaf mode, linear analysis alone can be insufficient and potentially misleading.
  • ROA as a "Nonlinear Stability Margin": The paper frames the size of the Region of Attraction (quantified by the parameter β*) as a direct measure of "nonlinear stability margin." This provides a tangible way to understand how much perturbation an aircraft can tolerate before entering an unstable state.
  • Sum-of-Squares Programming for Guaranteed Stability: The application of SOS programming to derive guaranteed stability regions (lower bounds of ROA) for a complex, real-world system like the F/A-18 is a significant methodological contribution. This approach offers a more rigorous certificate of stability compared to simulation-based methods.
  • Revised Control Law's Superiority: The revised control law demonstrably provides a significantly larger and more robust stability region, directly correlating with its observed success in suppressing the falling leaf mode in flight tests.

Key Results

  • Linear Analysis: Showed similar robustness margins (gain, phase, delay, and multiplicative uncertainty) for both baseline and revised control laws, failing to predict the significant difference in their behavior during extreme maneuvers.
  • Nonlinear ROA Analysis:
    • Lower Bounds: The revised control law achieved substantially larger lower bounds for the ROA across different Lyapunov function orders (linear, quadratic, quartic). For instance, with a quartic Lyapunov function, the revised law's lower bound (β₄ = 2.53 x 10⁻²) was more than double the baseline's (β₄ = 1.24 x 10⁻²).
  • Upper Bounds (Monte Carlo): The revised control law also showed a larger upper bound for the ROA (βmc = 2.95 x 10⁻²) compared to the baseline (βmc = 1.56 x 10⁻²), indicating it can tolerate larger initial perturbations before instability.
  • Overall ROA Improvement: The revised control law provided a roughly 1.43 times larger guaranteed stability region in individual state directions and a significantly larger overall volume (estimated at 12.22 times larger) compared to the baseline.

Limitations and Future Work

  • Computational Complexity of ROA: The authors acknowledge that computing the exact ROA is challenging and that their approach focuses on estimating bounds. The computational time for higher-degree Lyapunov functions grows rapidly, limiting the analysis to quartic functions.
  • Model Simplifications: The analysis relies on reduced-order polynomial models and assumes constant velocity. While these simplifications are necessary for the nonlinear analysis, they represent a departure from the full nonlinear dynamics.
  • Future Work: The paper suggests that the SOS programming approach for ROA estimation is a promising avenue for validating flight control laws, especially for systems with significant nonlinearities. Further exploration of higher-order Lyapunov functions and more complex uncertainty models could provide even deeper insights. The success of this method on a real physical system like the F/A-18 opens doors for its application to other safety-critical control systems.

Summary updated

Abstract

The F/A-18 Hornet aircraft with the original ight control law exhibited an out-ofcontrol phenomenon known as the falling leaf mode. Several aircraft were lost due to the falling leaf mode and which led NAVAIR and Boeing to redesign the ight control law. The revised ight control law successfully suppressed the falling leaf mode during ight tests with aggressive maneuvers. This paper compares the robustness of the original (baseline) and revised control laws using linear analyses, Monte Carlo simulations, and nonlinear region-of-attraction analyses. The linear analyses indicate the revised controller only marginally improves the closed-loop robustness while the nonlinear analyses indicate a substantial improvement in robustness over the baseline controller. This example demonstrates the potential for nonlinear analyses to detect out-of-control phenomenon that are not evident from linear analyses.
BibTeX
@inproceedings{chakraborty2009applications,
  author = {Abhijit Chakraborty and Peter Seiler and Gary Balas},
  title = {{Applications of Linear and Nonlinear Robustness Analysis Techniques to the F/A-18 Flight Control Laws}},
  booktitle = {AIAA},
  year = {2009},
  doi = {10.2514/6.2009-5675},
}
Figure