Introduction
Ensuring the reliability and safety of complex industrial systems, such as nuclear power plants, is a monumental task. Traditionally, this process involves significant manual labor by highly specialized engineers. Researchers from Électricité de France (EDF) and University Pierre et Marie Curie have developed an expert system-driven approach to automate the creation of these reliability models. This innovation addresses the limitations of traditional methods and significantly reduces the time required for comprehensive safety studies.
The Challenge of Complex Dependencies
Reliability engineering typically relies on two primary methodologies: fault-trees and state-graphs (Markov models). While fault-trees are efficient, they struggle to model systems with interdependent components—for instance, a backup generator that only activates if the primary power source fails.
As shown in the state transition diagram above, dependencies create complex paths that state-graphs handle well, but they are prone to "state-space explosion." In a system with $N$ components, the number of possible states is $2^N$, making large-scale systems mathematically untreatable by conventional manual means.
The Expert System Architecture
To bridge the gap between human expertise and automated computation, the authors proposed a two-level rule-based architecture. This system separates general engineering knowledge from specific system topology.
The architecture utilizes several specialized software components:
- ALOUETTE: A general-purpose inference engine that processes the knowledge base using first-order logic.
- GSI (Generation of Sequences by Inference): An engine that generates event sequences based on ALOUETTE's output. It employs heuristic pruning to omit negligible probabilities, keeping the computation tractable.
- EXPRESS: An expert system dedicated to building traditional fault-trees for thermo-hydraulic systems.
By defining a topological "map" of the system, the user allows the expert system to automatically generate complex GSI models for calculating failure probabilities.
Case Study: Paluel Nuclear Power Plant
The effectiveness of this approach was demonstrated through a real-world industrial case study involving the 6.6kV electrical power supply system of the Paluel Nuclear Power Plant. The model encompassed 30 major components, each with multiple states (e.g., operable, short-term failure, long-term failure). A simplified benchmark version included circuit breakers, busbars, diesel generators, transformers, and grids to test the system's robustness.
Results and Impact
The implementation of this expert system yielded dramatic improvements in both labor efficiency and computational performance. Key results include:
- Labor Reduction: The time required for the Paluel plant study was slashed from 8 engineer-months to just 3.5 engineer-months.
- Computational Speed: Using an IBM 3090, the system identified 1,752 critical failure sequences in only 53 seconds.
- Advanced Modeling: The system successfully modeled complex dependencies that are impossible to represent using standard fault-trees.
Conclusion
The proposed expert system acts as a "force multiplier" for reliability engineers. While it does not replace human oversight—manual review is still necessary to account for unique local conditions—it provides a high-quality "preliminary sketch" that engineers can refine. By decoupling component behavior from system topology, the tool allows for rapid updates to reliability reports whenever a design changes, marking a significant advancement in industrial safety engineering.
Summary updated