Design of high-performance networked real-time control systems

Peng Wen, Jiannong Cao, Yan LiIET 2007 · DOI

TL;DR

This paper tackles the critical challenge of time delays and packet loss in networked control systems (NCSs), which degrade performance. The authors propose a systematic design approach by modeling these issues and developing methods to optimize sampling intervals, schedule tasks, and design controllers to achieve high-performance NCSs.

What Problem Does This Paper Address?

Networked control systems (NCSs) offer advantages like flexibility and cost-effectiveness by using communication networks to transmit sensor and actuator data. However, this reliance on networks introduces significant challenges: time delays and packet loss. These network-induced issues can severely degrade the performance and stability of real-time control systems, which demand precise timing. The paper aims to address how to design NCSs that can maintain high performance despite these inherent network uncertainties. Specifically, it differentiates between time delays smaller than a sampling interval (modeled as a constant delay) and those larger than a sampling interval (leading to packet loss, modeled as a jump linear system).

Methods

The paper proposes a multi-pronged approach to mitigate the effects of time delays and packet loss:

  • Modeling Time Delays:
    • For random time delays less than one sampling interval ($T$), the system is modeled as a time-invariant control system with a constant time delay.
  • For random time delays greater than one sampling interval, where packets are effectively dropped, the system is modeled as a jump linear control system.
  • Optimizing Sampling Interval: The authors suggest choosing a proper sampling interval to reduce the amount of data transmitted, thereby minimizing network traffic and indirectly reducing the likelihood of delays and packet loss. They provide formulas to estimate the minimum allowable sampling period based on task deadlines and network time delays.
  • Optimal Task Scheduling: To further combat packet loss, the paper proposes modeling communication tasks as real-time periodic tasks. It introduces a scheduling algorithm that uses two versions of each task: a primary (high-quality, potentially failure-prone) and an alternate (minimal, reliable). This approach aims to guarantee task completion by scheduling alternates as late as possible, maximizing the chance for primaries to execute while still meeting deadlines.
  • Controller Design for Packet Loss: When packet loss is unavoidable, the paper proposes designing controllers that can operate with unreliable data links. This involves using techniques like least mean squares (LMS) estimation to estimate the plant state based on available (and potentially incomplete) information.

Datasets

Not discussed in the paper. The paper focuses on theoretical modeling, simulation, and experimental testing with a specific industrial case study.

Key Insights and Ideas

  • Time Delay Categorization is Crucial: The paper highlights that the impact of time delay on NCS performance depends critically on its magnitude relative to the sampling interval. This distinction leads to different modeling and mitigation strategies.
  • Sampling Interval is a Double-Edged Sword: While a higher sampling rate generally improves control performance, it increases network load, leading to more delays and packet loss. The paper emphasizes finding an optimal balance.
  • Task Scheduling for Reliability: The dual-version task scheduling (primary/alternate) is a clever way to handle the inherent unreliability of communication channels in real-time systems, ensuring that critical tasks are completed even if the primary execution fails.
  • Estimation is Key for Unreliable Data: When data is lost, the system cannot rely on direct measurements. The paper shows that optimal control can still be achieved by intelligently estimating the system state from the available, albeit incomplete, information.

Key Results

The paper presents simulation and experimental results demonstrating the effectiveness of their proposed methods.

  • Simulation Results (Table 1): Simulations of the communication task scheduling algorithm showed that as the primary failure probability decreased, the percentage of successfully completed primaries (PctSucc) increased significantly for all tasks. For instance, for Task 4, PctSucc rose from 23.1% at a 10% failure rate to 100% at a 0% failure rate. The results also indicated that lower-priority tasks are more susceptible to primary failures.
  • Experimental Testing: A prototype system was set up in a control lab using industrial PCs and PLCs connected via the Internet. The system successfully demonstrated real-time control operations, with control commands executed correctly and all real-time constraints met.
  • Schedulability Analysis: For the Incitec Limited case study, the communication link utilization factor was calculated to be 0.713, which is less than the theoretical bound of 0.735, indicating that the system is schedulable by both Rate-Monotonic (RM) and Earliest Deadline First (EDF) algorithms.

Limitations and Future Work

  • Limitations:
    • The paper acknowledges that while the proposed methods improve performance, packet loss cannot be completely avoided.
  • The strictness of real-time control systems means that missing actions can have severe consequences, and the paper implicitly highlights this by focusing on meeting deadlines.
  • Future Work:
    • The authors suggest further investigation into the relationship between packet drop patterns and system performance.
  • They also propose studying the trade-offs between packet drop patterns and system performance in more detail.

Summary updated

Abstract

The performance of a networked control system (NCS) is affected directly and indirectly by time delays generated in network communication. This paper addresses these time delays, their effects on system performance and the measures taken in an NCS design. For a time delay that is random but less than one sampling interval T, we model the system as a time-invariant control system with constant time delay T. For a time delay that is random and greater than one sampling interval T, we consider the system as a control system with packet drop and model it as a jump linear control system. Based on the above models, we propose a systematic method to improve the control system quality of performance by choosing a proper sampling interval to reduce data transmission, optimally scheduling to minimise packet loss and optimising controller design in case of packet loss. The simulation results and lab experiments demonstrate that the design method results in high performance in networked real-time control system.
BibTeX
@inproceedings{wen2007design,
  author = {Peng Wen and Jiannong Cao and Yan Li},
  title = {{Design of high-performance networked real-time control systems}},
  booktitle = {IET},
  year = {2007},
  doi = {10.1049/iet-cta:20060424},
}
Figure