Quasi-Decentralised Control

H. Utidjian, J.m. Maciejowski, Sanjay LallIFAC 1995 · DOI

TL;DR

This paper introduces "quasi-decentralised control," a new strategy that bridges the gap between fully centralized and fully decentralized control systems. It offers a flexible way to design controllers that can be tuned to balance performance and stability, particularly for complex multi-rate systems.

What Problem Does This Paper Address?

Designing control systems for complex, large-scale plants often involves a trade-off between centralized and decentralized approaches. Fully centralized control can offer optimal performance but is often impractical due to communication bottlenecks and computational demands. Fully decentralized control, on the other hand, is more scalable but typically sacrifices performance. This paper addresses the need for a control strategy that can offer a compromise, allowing for continuous variation between these two extremes. Specifically, they focus on systems where different parts of the controller operate at different sampling rates (multi-rate systems), a common scenario in modern industrial applications.

Methods

The authors propose a "quasi-decentralised" control architecture inspired by the Internal Model Control (IMC) framework. The core idea is to create a system with hierarchical communication levels: "fast" local links and "slower" remote links. This is achieved by designing controllers that are neither fully centralized nor fully decentralized.

They present two synthesis procedures, assuming a stable, centralized IMC controller already exists for the plant. The first method uses an approximate minimization criterion. This involves transforming the problem into a standard H-infinity synthesis problem by "lifting" the multi-rate system into an equivalent single-rate system. This allows for the application of existing H-infinity optimization techniques.

The second method aims for an "exact" minimization criterion. This approach adapts recent work on optimal sampled-data controller synthesis. It involves a "local, loop-by-loop" design process where individual controller components are optimized, followed by a "global" analysis to ensure overall stability. This iterative approach allows for tuning and refinement.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • Quasi-decentralised control as a spectrum: The paper frames quasi-decentralised control not as a binary choice but as a tunable spectrum between centralized and decentralized control, offering flexibility in design.
  • Bridging multi-rate and H-infinity: The proposed methods effectively adapt powerful H-infinity synthesis techniques to the complexities of multi-rate systems, which are often challenging to handle with traditional methods.
  • Stability via Small-Gain Theorem: The authors leverage the Small-Gain Theorem to derive sufficient conditions for the stability of their quasi-decentralised system, providing a clear design objective.
  • Approximate vs. Exact Synthesis: The paper offers two distinct synthesis approaches, catering to different needs: an approximate method for computational tractability and an exact method for potentially better performance, albeit with more complexity.
  • The challenge of block-diagonal structure: A key challenge highlighted is how to preserve the desired block-diagonal structure of the controller during H-infinity synthesis, as standard methods often do not inherently maintain this.

Key Results

The paper provides sufficient conditions for the internal stability of the quasi-decentralised control system. The proposed synthesis methods aim to produce controllers that approximate the original continuous-time controller in an H-infinity sense. While specific quantitative results are not presented in terms of numerical performance improvements, the paper lays out a theoretical framework and synthesis procedures that enable the design of such controllers.

Limitations and Future Work

The authors acknowledge several limitations and suggest avenues for future research:

  • Causality Constraint: A significant challenge in adapting sampled-data control techniques to multi-rate systems is ensuring the resulting controller is causal. The paper mentions this complexity and refers to existing work on dealing with it.
  • Preserving Block-Diagonal Structure: Existing H-infinity synthesis methods struggle to preserve the block-diagonal structure required for quasi-decentralised controllers. The proposed methods address this by either approximation or iterative local designs.
  • Performance Criteria: The current synthesis methods are primarily focused on ensuring stability. Future work should aim at satisfying performance-related criteria.
  • Fault-Tolerance: The paper suggests that the duplicated nature of controllers in their scheme might offer fault-tolerance, which warrants further investigation.
  • Other Architectures: While the paper focuses on an IMC-like implementation, the concept of quasi-decentralised control could be applied to other control architectures.

Summary updated

BibTeX
@inproceedings{utidjian1995quasi,
  author = {H. Utidjian and J.M. Maciejowski and Sanjay Lall},
  title = {{Quasi-Decentralised Control}},
  booktitle = {IFAC},
  year = {1995},
  doi = {10.1016/s1474-6670(17)51620-1},
}
Figure