Source-linked AI summary
Decentralized event-triggered control over wireless sensor/actuator networks
Manuel Mazo, Paulo Tabuada
TL;DR
The paper addresses how to reduce controller-update frequency when full-state information is unavailable at individual sensing nodes in wireless sensor/actuator networks. It proposes decentralized event-triggered implementations of centralized controllers, demonstrates asymptotic stability and effective decentralization, and identifies conservatism, practical listening costs, and scope for dynamic controllers as limitations.
Problem
Full-state information is unavailable at individual sensing nodes in wireless sensor/actuator networks, complicating event-triggered decisions intended to reduce controller-update frequency.
Method
The paper constructs decentralized event-triggered implementations of centralized controllers using local triggering conditions, minimum inter-update scheduling, and adaptable parameters θ.
Results
The proposed techniques render the closed-loop systems asymptotically stable and effectively decentralize event-triggered control for a quadruple water-tank system.
Takeaways & Limitations
The implementation reduces transmitted information and has low computational requirements, making it suitable for sensor/actuator networks.
Takeaways & Limitations
The decentralized approach is generally conservative, may require costly event listening unless accommodated by time multiplexing, and general dynamic-controller implementations remain future work.
Abstract
from arXiv · showhide
In recent years we have witnessed a move of the major industrial automation providers into the wireless domain. While most of these companies already offer wireless products for measurement and monitoring purposes, the ultimate goal is to be able to close feedback loops over wireless networks interconnecting sensors, computation devices, and actuators. In this paper we present a decentralized event-triggered implementation, over sensor/actuator networks, of centralized nonlinear controllers. Event-triggered control has been recently proposed as an alternative to the more traditional periodic execution of control tasks. In a typical event-triggered implementation, the control signals are kept constant until the violation of a condition on the state of the plant triggers the re-computation of the control signals. The possibility of reducing the number of re-computations, and thus of transmissions, while guaranteeing desired levels of performance makes event-triggered control very appealing in the context of sensor/actuator networks. In these systems the communication network is a shared resource and event-triggered implementations of control laws offer a flexible way to reduce network utilization. Moreover reducing the number of times that a feedback control law is executed implies a reduction in transmissions and thus a reduction in energy expenditures of battery powered wireless sensor nodes.
1. Introduction
Wireless sensor/actuator networks make communication constraints central to feedback control, while event-triggered execution offers a way to reduce updates and transmissions. The paper proposes decentralizing event-trigger decisions for centralized controllers when individual sensor nodes lack full-state information.
- Motivation: Wireless sensor/actuator networks expose bandwidth, noise, delay, and communication-medium limitations that conventional controller designs often neglect.These constraints have motivated research on control under communication restrictions, including industrial initiatives such as WirelessHART.
- Motivation: Embedded implementation of continuous physical control raises the question of how frequently the environment should be sampled.Existing approaches include analytical sampling-period selection, but engineers still often rely on rules of thumb such as sampling at 20 times the system bandwidth.
- Event-triggered control: Event-triggered control replaces periodic execution by recomputing control signals when a state-dependent condition is violated, potentially reducing transmissions and energy use.This is particularly relevant when the communication network is shared and sensor nodes are battery powered.
- Contribution: The paper’s main contribution is a decentralized event-triggered implementation over WSANs for centralized controllers.The approach addresses the fact that physically distributed sensors cannot evaluate criteria requiring the full vector of measured quantities, while avoiding large observer or consensus communication costs.
- Scope: The paper focuses on reducing actuation frequency and associated communication and energy savings, while leaving practical delays and jitter outside its scope.The authors state that their techniques can be implemented over WirelessHART, which addresses other communication concerns.
2. Notation
The notation section establishes the number sets, norms, matrix definiteness and eigenvalue notation, class K∞ functions, bounded signals, control systems, their solutions, and feedback laws used later.
- Norms and matrices: The usual Euclidean vector norm and induced matrix norm are both denoted by |·|.The notation is used for vectors and matrices with their respective interpretations.
- Norms and matrices: A matrix P is positive definite when xT Px > 0 for every nonzero x, and λm(P) and λM(P) denote its minimum and maximum eigenvalues.These definitions support later stability and Lyapunov-style arguments.
- Functions and signals: A class K∞ function is continuous, strictly increasing, zero at zero, and unbounded as its argument tends to infinity.The section also introduces essentially bounded, piecewise continuous signals used in the system model.
- Control-system notation: Control systems are modeled by differential equations with smooth dynamics, and feedback laws are smooth maps from the state space to the input space.Solutions are indexed by their initial condition and input, with the initial condition specified at time zero.
3. Decentralized event-triggered control
The paper decentralizes event-triggered updates for nonlinear controllers over wireless sensor/actuator networks, where no sensor has the full plant state. Local event conditions trigger centralized recomputation while preserving asymptotic stability, with conservatism and practical timing constraints explicitly addressed.
- System architecture: A computation node retrieves measurements, computes inputs for all actuators, and disseminates updated commands in the considered WSAN architecture.The execution loop comprises sensor data retrieval, controller computation, and actuation-command provision.
- Decentralized triggering: Whenever a local triggering condition is violated, the sensor informs the computation node, which requests measurements, recomputes control signals, and forwards commands to actuators.Updates can be scheduled no sooner than the centralized implementation’s minimum inter-update time τmin; earlier events are ignored.
- Decentralized triggering: Because sensors lack the full state, the centralized event condition cannot be evaluated locally; each sensor therefore uses information available at its node.The decentralized setting assumes, for simplicity, one state variable per sensor, while extensions to multiple locally sensed states are also described.
- Online adaptation: Decentralization is generally conservative because update intervals are shorter than in the centralized case, although θ can be adapted to reduce this conservatism and network utilization.The adaptation heuristics choose an equalization time and Taylor approximation order, then compute θ so estimated local decision gaps coincide.
- Stability: The decentralized conditions guarantee asymptotic stability, provided the parameter vector satisfies the stated constraint, regardless of the specific admissible parameter values and update rules.The underlying centralized sample-and-hold implementation is also established to preserve asymptotic stability under its event condition.
- Practical implementation: Time-multiplexed wireless access can be incorporated as a bounded, known delay, but continuously listening for events would otherwise create a potentially large sensor-radio energy cost.The paper explicitly focuses on reducing actuation frequency and associated communication and energy savings rather than addressing delays or jitter directly.
4. Examples and simulation results
The paper evaluates its decentralized event-triggered controller on a nonlinear quadruple-tank system with an extended dynamic controller and specified triggering parameters. Simulations compare centralized, adaptive decentralized, and nonadaptive decentralized implementations, showing similar stabilization despite differing update intervals.
- System and controller: The example uses a nonlinear quadruple-tank system whose states are four tank water levels and whose two inputs are split among the tanks.The model is extended with two artificial nonlinear-integrator states for controller design.
- System and controller: The controller adds nonlinear integrator states ξ5 and ξ6 to achieve zero steady-state offset, with ξ1 and ξ2 converging to desired values at steady state.The implementation assumes sensors measuring ξ1 and ξ2 compute ξ5 and ξ6 locally and sufficiently fast.
- Simulation setup: The simulation confines the system to a compact set, selects ρm = 0.14, ρ = 0.25, σ = 0.00542, and uses a minimum update interval bound τmin = 0.1 ms.The adaptive decentralized implementation combines paired state measurements at respective sensor nodes.
- Simulation comparison: The comparison examines centralized event-triggering, adaptive decentralized event-triggering, and decentralized event-triggering without adaptation using the same initial-condition scenario.Figure 2 reports update intervals, triggering-condition evolution, and state trajectories for the three implementations.
- Simulation comparison: The adaptive decentralized implementation does not exactly recover centralized update performance but produces very good results, while all three implementations generate almost indistinguishable stabilized state trajectories.The comparison also reports the number of controller executions in the figure legends.
5. Discussion
The discussion positions the approach relative to distributed methods for weakly coupled systems and summarizes its demonstrated effectiveness on a quadruple water-tank example. It also identifies general dynamic-controller implementations and more efficient adaptation rules as open research questions.
- Relation to prior work: Unlike a prior distributed method relying on weakly coupled subsystems, the proposed approach does not require internal weak-coupling assumptions, although it does not update inputs independently.The techniques may complement the prior method at the local subsystem level.
- Discussion and scope: The techniques successfully decentralize an event-triggered implementation of a quadruple water-tank system with a dynamic centralized controller.In this example, sensors continuously compute the dynamic part of the controller.
- Discussion and scope: General dynamic controllers in event-triggered form and more efficient adaptation rules remain questions for future research.The stated implementation has low computational requirements and is intended for sensor/actuator networks with limited sensor-level computation.