Causal chain event graphs for remedial maintenance.

Area of Science:

• System reliability engineering
• Probabilistic graphical models
• Causal inference

Background:

• Traditional graphical tools like fault trees and Bayesian networks are used for system reliability analysis.
• These methods have limitations in representing asymmetric processes and novel intervention types.

Purpose of the Study:

• To introduce and apply the chain event graph (CEG) as a probabilistic graphical model for system reliability analysis.
• To define and model 'remedial' interventions for system maintenance using CEGs.
• To demonstrate the CEG's capability in expressing causal effects of maintenance and guiding predictive inferences.

Main Methods:

• Application of chain event graphs (CEGs), derived from event trees, to model system failures and deterioration.
• Definition of a new formal intervention class termed 'remedial' to represent maintenance actions.
• Adaptation of a backdoor theorem for causal inference in partially observed systems with interventions.

Main Results:

• CEGs flexibly represent asymmetric processes and system deterioration.
• The CEG framework effectively models the causal effects of 'remedial' maintenance, restoring systems to an 'as good as new' state.
• Bespoke causal algebras within the CEG provide a transparent method for predictive inferences regarding interventions.

Conclusions:

• Chain event graphs offer a powerful and flexible tool for system reliability analysis.
• CEGs can effectively model complex causal relationships, including novel 'remedial' maintenance interventions.
• The CEG framework facilitates transparent causal reasoning and inference in system maintenance and reliability studies.