Hierarchical Reasoning about Faults in Cyber-Physical Energy Systems using Temporal Causal Diagrams

Ajay Chhokra, Nagabhushan Mahadevan, Abhishek Dubey, Saqib Hasan, Daniel Balasubramanian, Gábor KarsaiInternational Journal of Prognostics and Health Management 2020 · DOI

Abstract

Fault management systems in modern energy grids observe system states, detect anomalies, and take automated actions to isolate faults. In electrical networks, relays and breakers isolate faults to arrest failure propagation. However, limited situational awareness and hidden failures in these protection devices can lead to mis-operation, causing unnecessary disconnections in healthy parts of the system. This paper introduces the formalism of Temporal Causal Diagrams (TCDs), which augment physical failure models with discrete-time models of protection elements. By accounting for complex interactions between protection devices and physical plants, the TCD framework enables accurate diagnosis of both physical and cyber-layer faults. We demonstrate this approach using the standard Western System Coordinating Council (WSCC) 9-bus system across four distinct fault scenarios.

The Challenge of Cyber-Physical Fault Management

Modern power grids rely on automated fault management systems—relays and circuit breakers—designed to detect anomalies and isolate them instantly. However, these systems often operate as "reflexes," acting on local information without a global perspective.

A critical vulnerability exists when protection devices themselves fail. A relay might trip spuriously, fail to trip when required, or a breaker might become stuck. In a highly coupled cyber-physical system, these failures can escalate a minor physical disturbance into a massive cascading blackout. Traditional diagnosis tools often struggle to distinguish between a legitimate physical line fault and a software or sensor glitch within the management layer.

The TCD Hierarchical Architecture

The core innovation of this work is the Temporal Causal Diagram (TCD), a hierarchical reasoning framework that extends previous Timed Failure Propagation Graphs (TFPG) by adding a distinct layer of cyber-component monitoring.

The TCD Diagnosis System architecture showing the hierarchy from the physical system up to the reasoner.

As shown in the architecture above, the system is organized into two primary layers:

  1. Lower Layer (Observers): Discrete state machines (Timed Transition Automata) monitor individual cyber components like relays and breakers. These observers track real-time events—alarms, commands, and state changes—to generate "local hypotheses" about device health.
  2. Upper Layer (TCD Reasoner): This system-level engine consumes local hypotheses and maps them onto a directed graph. Nodes represent failure modes or discrepancies, while edges represent causal links labeled with specific temporal constraints (e.g., "Effect B occurs 30ms after Fault A").

This modularity ensures that the system-level reasoner remains agnostic to the specific internal logic of hardware, focusing instead on the outputs of the observers.

Modeling Local and Global Dynamics

The TCD framework explicitly models how cyber actions change physical topology. For example, if a relay successfully opens a breaker, the physical system's mode changes, altering how future faults might propagate.

Failure Propagation Graph (a) alongside state machine models for anomaly detectors (b), protection devices (c), and actuators (d, e).

To handle the inherent uncertainty in complex grids, the authors introduced "uncertain edges" into the causal graphs. This allows the reasoner to account for backup protection—scenarios where a secondary device might act if the primary fails, though its activation is not guaranteed. When multiple explanations fit the data, the reasoner ranks hypotheses based on plausibility, robustness to new data, and parsimony (preferring simpler explanations with fewer simultaneous failures).

Case Study: The WSCC 9-Bus System

The framework was validated using a high-fidelity simulation of the WSCC 9-Bus System, a standard benchmark featuring 9 buses and 3 generators. Simulations were conducted in MATLAB/Simulink with a 1ms step size to capture fast electrical transients and discrete control logic.

The TCD framework was tested against four increasingly difficult scenarios:

  • Standard Physical Fault: Correctly identified a 3-phase ground fault.
  • Spurious Detection: Identified a relay trip caused by a cyber-layer fault despite a healthy physical line.
  • Stuck Breaker: Diagnosed a physical fault and identified the specific breaker failure that necessitated backup intervention.
  • Multi-Fault Scenario: Successfully isolated a physical fault even when one relay missed detection and a separate breaker failed to operate.

In all tests, the reasoner stabilized its hypothesis within milliseconds, providing clear root-cause explanations during active events.

Conclusion and Future Directions

While the TCD framework proves effective on the 9-bus benchmark, scaling to real-world grids with thousands of buses remains a primary challenge. Future work will focus on the automatic synthesis of TCD models from system topology to reduce manual modeling effort. Additionally, the researchers plan to expand the scope to include the communication layer, diagnosing issues like packet loss or network latency as potential drivers of protection system mis-operation.

Summary updated

Abstract

fault management systems that observe the state of the system, decide if there is an anomaly and then take automated actions to isolate faults. For example, in electrical networks relays and breaks isolate faults in order to arrest failure propagation and protect the healthy parts of the system. However, due to the limited situational awareness and hidden failures the protection devices themselves, through their operation (or mis-operation) may cause overloading and the disconnection of parts of an otherwise healthy system. Additionally, often there can be faults in the management system itself leading to situations where it is difficult to isolate failures. Our work presented in this paper is geared towards solution of this problem by describing the formalism of Temporal Causal Diagrams (TCD-s) that augment the failure models for the physical systems with discrete time models of protection elements, accounting for the complex interactions between the protection devices and the physical plants. We use the case study of the standard Western System Coordinating Council (WSCC) 9 bus system to describe four different fault scenarios and illustrate how our approach can help isolate these failures. Though, we use power networks as exemplars in this paper our approach can be applied to other distributed cyberphysical systems, for example water networks.
BibTeX
@article{chhokra2020hierarchical,
  author = {Ajay Chhokra and Nagabhushan Mahadevan and Abhishek Dubey and Saqib Hasan and Daniel Balasubramanian and Gábor Karsai},
  title = {{Hierarchical Reasoning about Faults in Cyber-Physical Energy Systems using Temporal Causal Diagrams}},
  journal = {International Journal of Prognostics and Health Management},
  year = {2020},
  doi = {10.36001/ijphm.2018.v9i1.2669},
}
Figure