Abstract
This paper addresses the stability of control systems utilizing predictor feedback when subjected to uncertain and time-varying input delays. While predictor feedback is a robust tool for compensating for known constant delays, its performance often degrades significantly when the actual delay deviates from the nominal value used in the controller design. By employing the Artstein transformation and a specific Lyapunov-Krasovskii functional, the authors derive a new Linear Matrix Inequality (LMI) condition to guarantee exponential stability. This approach significantly reduces the conservatism found in traditional small-gain analysis, providing larger allowable delay margins for both Linear Time-Invariant (LTI) systems and a class of infinite-dimensional systems.
Introduction
In modern control engineering—ranging from teleoperation in robotics to the management of chemical processing plants—input delays are a ubiquitous challenge. Predictor feedback has emerged as a primary strategy to mitigate these delays by "predicting" the future state of the system and applying control actions accordingly. However, the efficacy of this method is highly sensitive to the accuracy of the delay estimate. In practical applications, the delay $D(t)$ is rarely constant; it fluctuates due to network jitter, mechanical wear, or environmental changes.
When the actual delay $D(t)$ differs from the nominal delay $D_0$ used in the predictor design, the resulting mismatch can lead to oscillations or instability. Historically, analyzing the robustness of these systems relied on "small-gain" arguments, which often yielded extremely conservative safety margins. This paper, published in Automatica, proposes a more refined mathematical framework to tighten these bounds using LMIs.
Methodology
Mathematical Framework
The authors tackle the robustness problem by combining three sophisticated mathematical tools:
- Artstein Transformation: This technique is used to map the original system with input delay into a delay-free system, simplifying the stability analysis of the nominal case.
- Lyapunov-Krasovskii Functionals (LKF): To handle the "memory" inherent in delayed systems, the researchers utilize an LKF that specifically accounts for the rate of change in the system's state. This allows for a more precise characterization of the system's energy dissipation over time.
- Linear Matrix Inequalities (LMIs): By manipulating the Lyapunov equations, the authors derive a set of LMIs. These inequalities can be solved efficiently using standard convex optimization software. If a solution exists, the system is guaranteed to be exponentially stable under the specified delay variation $\delta$.
Extension to Infinite-Dimensional Systems
A significant contribution of this work is the extension of the LMI condition to infinite-dimensional systems, specifically Partial Differential Equations (PDEs) with boundary control. The authors demonstrate that for systems with a Riesz spectral operator (a common property in models of beams or heat diffusion), one only needs to design the predictor for the finite number of unstable modes. If these modes are stabilized despite the delay fluctuation, the remaining infinite-dimensional stable modes will remain well-behaved.
Key Insights and Results
Reducing Conservatism
The transition from small-gain theorems to LMI conditions represents a shift toward less conservative stability analysis. While small-gain methods provide a "worst-case" guarantee that is often too restrictive for practical use, the LMI approach allows for a more nuanced search for stability, resulting in significantly larger allowable delay margins.
Quantitative Performance
The authors validated their LMI condition by comparing it against previous state-of-the-art methods, including the small-gain approach by Karafyllis & Krstic (2013) and the LMI method by Selivanov & Fridman (2016). In a numerical example with a nominal delay of 1.0s:
- The Small-Gain method guaranteed stability only for fluctuations up to 0.0212s.
- The New LMI condition proved stability for fluctuations up to 0.0780s.
This represents a nearly four-fold improvement in the guaranteed safety margin, allowing engineers to operate systems in much more volatile environments.
Practical Validation in PDEs
In simulations of a reaction-diffusion equation, the controller successfully stabilized the system even when the delay was subjected to high-frequency oscillations ($D(t) = 1 + 0.25 \sin(3\pi t + \pi/4)$). This demonstrates that the theoretical bounds derived are not just mathematically sound but are also applicable to complex, high-dimensional physical processes.
Limitations and Future Outlook
While the LMI condition is a significant step forward, it remains a sufficient rather than a necessary condition. This means that if the LMI fails to find a solution, the system might still be stable, indicating there is still some inherent conservatism in the Lyapunov approach. Additionally, the PDE extension currently assumes a diagonal structure, which may not apply to all fluid dynamic or complex structural models.
Future research directions include extending these results to state-dependent delays, where the delay itself is a function of the system's current performance, and exploring non-diagonal PDE structures to broaden the applicability of the predictor feedback framework.
Reference: Lhachemi, H., Prieur, C., & Shorten, R. (2019). An LMI Condition for the Robustness of Constant-Delay Linear Predictor Feedback with Respect to Uncertain Time-Varying Input Delays. Automatica.
Summary updated