Robustness Analysis of an L1 Adaptive Controller

Peter Seiler, Andrei Dorobantu, Gary BalasAIAA 2010 · DOI

TL;DR

This paper analyzes the robustness of an L1 adaptive controller implemented on NASA's Generic Transport Model (GTM) aircraft. By approximating the adaptive controller as a linear system, the researchers identified insufficient stability margins in the initial design, which likely caused observed oscillations during flight tests. A revised controller, designed with more accurate models, significantly improved these margins and demonstrated better performance.

What Problem Does This Paper Address?

The paper tackles the challenge of ensuring the robustness and stability of adaptive flight control systems, specifically an L1 adaptive controller, when implemented on a real-world aircraft. While L1 adaptive control offers theoretical advantages in handling uncertainties, its practical application requires careful validation. The authors observed oscillations in the elevator command during flight tests of an L1 controller on the GTM aircraft. They aimed to understand the root cause of these oscillations and to improve the controller's robustness by analyzing its stability margins using linear control theory. This is crucial because even theoretically sound adaptive controllers can exhibit unexpected behavior due to unmodeled dynamics, sensor noise, or actuator limitations in real-world systems.

Methods

The core of the paper's methodology lies in approximating the complex L1 adaptive controller as a linear time-invariant (LTI) system. This simplification allows the application of well-established linear analysis tools, such as Bode plots and stability margin calculations. The authors developed an LTI model of the GTM aircraft's longitudinal dynamics and integrated an LTI representation of the L1 adaptive controller. They then analyzed the closed-loop system's frequency-domain behavior to assess its stability margins (gain, phase, delay, disk margins). This approach enabled them to pinpoint specific frequency ranges where the controller was vulnerable to uncertainties and to compare the robustness of the initial (Version 7) and a revised (Version 12) controller designs.

Datasets

The paper utilizes data from flight tests conducted on NASA's Generic Transport Model (GTM) aircraft. Specifically, it references flight test data from September 2009 (Version 7 of the controller) and March/June 2010 (Version 12 of the controller). Additionally, the authors used a high-fidelity 6-DOF GTM simulation model (GTM DesignSim) for validation purposes, comparing linear analysis results with nonlinear simulation responses.

Key Insights and Ideas

  • Linearization is a powerful tool for adaptive control analysis: Despite the inherent nonlinearities and adaptive nature of the L1 controller, approximating it as an LTI system proved effective for identifying robustness issues. This suggests that for certain operating conditions, linear analysis can provide valuable insights into the stability of adaptive controllers.
  • Model accuracy is paramount for adaptive controller design: The initial L1 controller was designed based on less accurate models of actuator dynamics and network delays. This led to insufficient stability margins, which were only revealed through detailed linear analysis using updated models. This highlights the critical importance of accurate plant models in the design phase of adaptive control systems.
  • Oscillations at specific frequencies are a symptom of poor stability margins: The observed 1.4-2 Hz oscillations in the elevator command directly correlated with the frequencies where the initial controller exhibited poor stability margins. This provides a clear link between theoretical analysis and observed flight behavior.
  • Revised controller design significantly enhances robustness: The updated controller (Version 12), designed with more accurate models and incorporating pitch rate stability augmentation, demonstrated substantially larger stability margins across various metrics. This directly translated to improved performance in subsequent flight tests.

Key Results

  • The initial L1 adaptive controller (Version 7) exhibited low stability margins, particularly at frequencies around 1.4-2.3 Hz, which is consistent with the observed oscillations during flight tests.
  • Linear analysis revealed that the gain margin for Version 7 was 2.43 at 14.49 rad/sec, phase margin was 46.3° at 8.42 rad/sec, and delay margin was 0.096 sec at 8.42 rad/sec.
  • The revised L1 adaptive controller (Version 12) showed significantly improved stability margins: gain margin of 17.01 at 4.26 rad/sec, phase margin of 56.8° at 0.43 rad/sec, and delay margin of 2.3 sec at 0.43 rad/sec.
  • The revised controller also demonstrated good performance during subsequent flight tests without the problematic oscillations.
  • The linear model's responses were validated against nonlinear simulations, showing good agreement for the short-period dynamics.

Limitations and Future Work

  • Neglected Nonlinearities: The linear analysis neglected nonlinear effects such as actuator saturation, backlash, and hysteresis, as well as wind turbulence. While the authors state these were neglected for simplicity, they acknowledge that these factors could influence actual performance.
  • Approximation of Adaptive Laws: The paper relies on the approximation of the adaptive law as a continuous-time constant matrix gain due to its fast sampling rate. While justified, this is still an approximation.
  • Future Work: The authors suggest that it would be possible to use linear analysis to investigate the effect of turbulence, which was not modeled in this study. Further investigation into the impact of unmodeled nonlinearities and the potential for integral projection-type adaptive laws could also be valuable. The paper also implies that the methodology used here could be applied to other adaptive control systems to assess their robustness.

Summary updated

Abstract

NASA’s Generic Transport Model (GTM) is a remote-controlled, 5.5 percent scale commercial aircraft. An L1 adaptive controller was recently designed and ight tested on the GTM. Oscillations in the elevator command were observed at 1.4 to 2 Hertz during the rst ight test in September 2009. In most ight conditions the L1 adaptive controller can be approximated by a linear time-invariant system. Thus linear analysis tools can be used to assess the performance and robustness of the feedback system with the L1 controller. The initial L1 design met the NASA requirement for 60 msec of time delay margin. However, the linear analysis indicates that the margin requirements were insu cient due to inaccurate models that were available at the time of the rst ight test. A revised L1 controller has signi cantly larger margins and demonstrated good performance during subsequent ight tests.
BibTeX
@inproceedings{seiler2010robustness,
  author = {Peter Seiler and Andrei Dorobantu and Gary Balas},
  title = {{Robustness Analysis of an L1 Adaptive Controller}},
  booktitle = {AIAA},
  year = {2010},
  doi = {10.2514/6.2010-8407},
}
Figure