Abstract
Feedback control laws have been traditionally implemented in a periodic fashion on digital hardware. Although periodicity simplifies the analysis of the mismatch between the control design and its digital implementation, it also leads to conservative usage of resources such as CPU utilization in the case of embedded control. We present a novel technique that abandons the periodicity assumption by using the current state of the plant to decide the next time instant in which the state should be measured, the control law computed, and the actuators updated. This technique, termed self-triggered control, is developed for two classes of nonlinear control systems, namely, state-dependent homogeneous systems and polynomial systems. The wide applicability of the proposed results is illustrated in two well known physical examples: a jet engine compressor and the rigid body.
Introduction
In traditional digital control, the state of a system is sampled at fixed, periodic intervals. While this approach simplifies the mathematical analysis of the gap between continuous-time design and digital execution, it is inherently inefficient. To guarantee stability, engineers must select a sampling period based on the system's worst-case dynamics. Consequently, when a system is near its equilibrium and requires minimal intervention, the controller continues to fire at high frequency, wasting valuable CPU cycles, network bandwidth, and energy.
While "event-triggered" control offers a solution by sampling only when an error threshold is crossed, it typically requires specialized hardware to monitor the system continuously. This paper introduces self-triggered control, a middle-ground approach where the controller uses its current measurement to calculate the maximum "safe" duration it can wait before the next measurement, eliminating the need for continuous hardware monitoring.
Methodology
The authors develop self-triggering rules for two significant classes of nonlinear systems: state-dependent homogeneous systems and polynomial systems. The framework is built upon several core theoretical pillars:
1. Input-to-State Stability (ISS)
The foundation of the approach is ISS theory. By defining a Lyapunov function—a mathematical representation of the system's energy—the authors determine the maximum allowable measurement error the system can tolerate before its stability is compromised. This error bound dictates the timing of the next sample.
2. Homogeneity and Scaling
The primary innovation lies in the use of "homogeneity." In homogeneous systems, trajectories scale predictably. The authors prove that if a "safe" waiting time is known for a single point (e.g., on a unit circle), a simple scaling law can be applied to determine the safe waiting time for any other point in the state space. This "scaling shortcut" allows the controller to avoid solving complex nonlinear differential equations in real-time.
3. State-Dependent Homogeneity and Polynomials
The research generalizes these findings to systems where the degree of homogeneity changes based on the system's location in the state space. For polynomial systems, which are not naturally homogeneous, the authors employ a "homogenization" technique. By adding a dummy variable to "lift" the system into a higher-dimensional space, the system becomes homogeneous, allowing the application of the aforementioned scaling laws.
Experimental Benchmarks
The proposed algorithms were evaluated using high-fidelity simulations of two classic physical benchmarks in control theory:
- Jet Engine Compressor: Based on the Moore-Greitzer model, this benchmark focuses on controlling mass flow and pressure rise.
- Rigid Body Dynamics: This models the angular velocity control of a rotating object, such as a satellite or a drone.
Key Results and Insights
The self-triggered approach demonstrated significant advantages over traditional periodic control:
- Resource Efficiency: In both the jet engine and rigid body simulations, the self-triggered controller required approximately 8 times fewer control updates than the best-case periodic controller while maintaining stability.
- Dynamic Adaptation: The system automatically adjusts its sampling frequency based on its state. In the jet engine simulation, as the system approached equilibrium, the time between executions became 24 times longer than the standard periodic interval.
- Robustness: The logic proved resilient to 2% sensor noise and external disturbances. When a disturbance pushed the system away from equilibrium, the self-triggered logic automatically increased the sampling frequency to compensate, returning to a lower frequency once the disturbance was mitigated.
Discussion and Limitations
While self-triggered control offers a path toward more efficient embedded systems, several challenges remain. The computational cost of calculating the next sampling time ((\tau)) must remain low; if the trigger calculation is too intensive, it may negate the CPU savings gained from reduced sampling. Additionally, the current method relies on linear over-approximations to find initial safe bounds, which can be conservative.
Future work involves addressing "jitter" (small delays in execution) in real-time scheduling and extending these elegant proofs for homogeneous and polynomial systems to a more universal framework for arbitrary smooth nonlinear systems.
Summary updated