Plug-and-play fault diagnosis and control-reconfiguration for a class of nonlinear large-scale constrained systems

Stefano Riverso, Francesca Boem, Giancarlo Ferrari-Trecate, Thomas ParisiniarXiv 2014 · OpenReview

TL;DR

This paper introduces a novel "Plug-and-Play" (PnP) architecture for controlling and monitoring large-scale, nonlinear systems. It seamlessly integrates distributed Model Predictive Control (MPC) with a distributed Fault Detection (FD) system, allowing faulty components to be safely disconnected and reconnected without compromising the entire system's stability or performance. This approach is crucial for enhancing the robustness and adaptability of complex industrial and infrastructure systems.

What Problem Does This Paper Address?

Modern systems, from power grids to complex manufacturing plants, are increasingly large-scale and interconnected. Controlling these systems centrally is often impractical due to computational, communication, and reliability limitations. Furthermore, these systems are susceptible to faults in individual components, which can propagate and destabilize the entire network. While decentralized and distributed control strategies exist, and fault detection methods are well-studied, integrating these into a framework that allows for dynamic reconfiguration – specifically, the safe "unplugging" of faulty subsystems and subsequent "plugging-in" after repair – has been a significant challenge, especially for nonlinear systems. This paper addresses this gap by proposing a unified PnP architecture that handles both control and fault diagnosis in a distributed, adaptive manner.

Methods

The core of the proposed solution lies in a novel Plug-and-Play (PnP) architecture that integrates two key components:

  • Distributed Model Predictive Control (DiMPC): The authors extend existing decentralized MPC schemes to handle nonlinear large-scale systems. The key idea is to design local controllers for each subsystem that can treat parent subsystems as bounded disturbances. This allows for a "tube-based" MPC approach, ensuring that the system states remain within predefined bounds even when subsystems are disconnected. The design process is decentralized, meaning each local controller is designed based on local information and information from its "parent" subsystems, avoiding the need for a global system model.
  • Distributed Fault Detection (FD) Architecture: Complementing the control aspect, a distributed FD architecture is developed. Each subsystem has a local diagnoser that estimates its state. When the estimation error exceeds a dynamically computed threshold, a fault is detected. Crucially, this FD architecture is designed to be robust to modeling uncertainties and to function correctly even when a faulty subsystem is unplugged. The PnP nature extends to the FD design, allowing local detectors to be designed independently.

The "Plug-and-Play" aspect is achieved by ensuring that the design of both local controllers and fault detectors allows for safe disconnection and reconnection of subsystems. This means that when a subsystem is removed, the remaining system remains stable and meets its constraints, and when it's reconnected, the system can adapt without requiring a complete redesign.

Datasets

The paper does not introduce new datasets. Instead, it uses two simulated examples to demonstrate the effectiveness of the proposed architecture:

  • Coupled van der Pol Oscillators: A ring of 20 coupled van der Pol oscillators is used to model oscillating systems. This example allows for testing the PnP control and FD in a nonlinear setting with local constraints.
  • Power Network System (PNS): A five-area power network system is used to demonstrate the architecture's applicability to a more complex, interconnected system with dynamics relevant to real-world applications like Automatic Generation Control (AGC).

Key Insights and Ideas

  • Unified PnP for Control and Diagnosis: The most significant insight is the seamless integration of PnP principles for both control and fault detection. This allows for a holistic approach to fault-tolerant operation, where faults are not only detected but also managed by dynamically reconfiguring the system's control and monitoring capabilities.
  • Decentralized Design for Nonlinear Systems: The paper successfully extends PnP concepts to nonlinear large-scale systems, which is a significant advancement over previous work often limited to linear systems. The use of tube-based MPC and careful design of local diagnosers are key to achieving this.
  • Safe Reconfiguration: The architecture guarantees that unplugging a faulty subsystem does not compromise the stability or constraint satisfaction of the remaining system. Similarly, plugging a repaired subsystem back in is handled safely. This is achieved through careful design of local controllers and detectors that can adapt to changes in the system topology.
  • Decoupled Design Models: A notable feature is the possibility of using different models for control and diagnosis. This flexibility is beneficial as control might require precise, fast models, while diagnosis can leverage more robust or simplified models, potentially with different decomposition structures.

Key Results

The paper presents simulation results demonstrating the effectiveness of the proposed architecture:

  • Fault Detection and Isolation: Simulations show that the distributed FD system can accurately detect faults in individual subsystems (e.g., a breakdown in a van der Pol oscillator's actuator or a generator fault in a power network).
  • Stable Reconfiguration: Upon fault detection, the faulty subsystem is "unplugged," and the remaining system continues to operate stably, maintaining its constraints. For instance, in the van der Pol oscillator example, states remain bounded even after a fault. In the power network example, frequency deviations are managed effectively even after an area is disconnected.
  • Seamless Re-plugging: After a fault is resolved (e.g., actuator repaired), the subsystem can be "plugged back in," and the system reconfigures itself to incorporate it, again without compromising stability.
  • Reduced Impact of Faults: The PnP approach effectively prevents fault propagation, isolating the issue to the faulty component and its immediate neighbors, rather than destabilizing the entire large-scale system.

Limitations and Future Work

  • Computational Complexity: The paper acknowledges that steps involving Minkowski sums and differences of polytopic sets for computing constraints can be computationally intensive. While they refer to methods to mitigate this, it remains a potential bottleneck for very large systems.
  • Generalization to More Complex Nonlinearities: While the approach handles nonlinearities in couplings and matched nonlinearities, future work could explore its applicability to a broader class of nonlinear systems.
  • Hybrid Systems and Switching: The authors note that frequent and persistent switching between system modes (due to plugging/unplugging) could potentially compromise asymptotic stability. They suggest investigating minimal dwell-time strategies to address this.
  • Real-time Applications: While simulations demonstrate effectiveness, further research is needed to apply and validate this architecture in real-time industrial applications.
  • New Subsystem Integration: The current work focuses on re-connecting previously unplugged subsystems. Future work could explore the integration of entirely new subsystems into an operating network.

Summary updated

Abstract

This paper deals with a novel Plug-and-Play (PnP) architecture for the control and monitoring of Large-Scale Systems (LSSs). The proposed approach integrates a distributed Model Predictive Control (MPC) strategy with a distributed Fault Detection (FD) architecture and methodology in a PnP framework. The basic concept is to use the FD scheme as an autonomous decision support system: once a fault is detected, the faulty subsystem can be unplugged to avoid the propagation of the fault in the interconnected LSS. Analogously, once the issue has been solved, the disconnected subsystem can be re-plugged-in. PnP design of local controllers and detectors allow these operations to be performed safely, i.e. without spoiling stability and constraint satisfaction for the whole LSS. The PnP distributed MPC is derived for a class of nonlinear LSS and an integrated PnP distributed FD architecture is proposed. Simulation results show the effectiveness and the potential of the general methodology.
BibTeX
@misc{riverso2014plug,
  author = {Stefano Riverso and Francesca Boem and Giancarlo Ferrari-Trecate and Thomas Parisini},
  title = {{Plug-and-play fault diagnosis and control-reconfiguration for a class of nonlinear large-scale constrained systems}},
  year = {2014},
  eprint = {1409.5224},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1409.5224},
}
Figure