A Novel Method of Failure Sample Selection for Electrical Systems Using Ant Colony Optimization

Jian Xiong, Shulin Tian, Chenglin Yang, Cheng LiuComput. Intell. Neurosci 2016 · DOI

TL;DR

This paper introduces a new method for selecting failure samples in electrical systems that accounts for how failures propagate. By using Ant Colony Optimization to model failure propagation paths, they can identify critical failure modes that might otherwise be missed, leading to improved test coverage and reduced risk in system usage.

What Problem Does This Paper Address?

Ensuring the reliability of electrical systems is crucial, and a key aspect of this is "testability" – the ability to detect, diagnose, and isolate failures. A common practice for demonstrating testability is the "testability demonstration experiment," which involves injecting known failure samples into the system. However, traditional methods for selecting these failure samples often rely on random selection or prioritize failures with high probabilities. This approach overlooks the phenomenon of "failure propagation," where a low-probability initial failure can cascade and cause more severe, widespread issues. The paper addresses the critical gap of potentially missing these high-impact, low-probability propagating failures, which can lead to significant risks in system operation.

Methods

The proposed method involves three main stages:

  1. Failure Propagation Modeling: The authors first build a directed graph (DG) to represent the relationships and probabilities of failure propagation between components in an electrical system. This DG is then used to calculate the "intensity" of failure spread along each edge (connection between components). To quantify and compare these intensities, they employ a Z-score calculation, which helps identify edges with a higher likelihood of failure propagation.
  2. Path Optimization with Ant Colony Optimization (ACO): Ant Colony Optimization, a metaheuristic inspired by the foraging behavior of ants, is employed to find the "maximum probability failure propagation path." ACO is well-suited for this task because of its robustness and ability to explore complex networks to find optimal paths, in this case, the paths most likely to experience cascading failures.
  3. Failure Sample Selection Optimization: Based on the identified maximum probability failure propagation paths, a new failure sample selection method is proposed. This method prioritizes selecting failure samples that are part of these critical propagation paths, ensuring that potentially overlooked, low-rate but high-impact failures are included in the test set. This is achieved by defining "subsequent failure propagation sets" (SFPS) and using them to guide the selection process.

Datasets

The paper uses a hypothetical refrigeration module as a case study to demonstrate its method. The failure modes for this module are listed in Table 2, and the failure propagation intensities are calculated and presented in Table 3. No external datasets were used for training or evaluation; the case study serves as a self-contained demonstration.

Key Insights and Ideas

  • Failure Propagation is Key: The most significant insight is that simply selecting failures randomly or based on their individual probability is insufficient. The interconnectedness of systems means that low-probability failures can become critical if they trigger a chain reaction.
  • ACO for Propagation Paths: Using ACO to explicitly find the maximum probability failure propagation paths is a novel application that effectively navigates the complexity of system interdependencies.
  • Z-Scores for Intensity: The use of Z-scores to quantify and compare failure propagation intensities provides a robust statistical measure for identifying critical links in the system's failure graph.
  • Improved Test Coverage: By focusing on propagation paths, the method ensures that a more comprehensive set of failure scenarios, including those that might be missed by traditional methods, are covered during testing.

Key Results

The paper demonstrates through a case study of a refrigeration module that the proposed method significantly improves failure coverage compared to traditional sampling plans. Table 4 shows a comparison where the proposed method selects specific failure modes based on propagation analysis, whereas the traditional method selects them randomly. The authors claim this leads to a better failure coverage rate and reduces the risk of using a system that might fail due to unaddressed propagation issues. While specific quantitative metrics like percentage improvement in coverage are not explicitly stated, the qualitative comparison in Table 4 and the case study analysis strongly suggest the superiority of their approach.

Limitations and Future Work

The authors acknowledge that the proposed method relies on accurate failure propagation models and probabilities, which can be challenging to obtain in real-world complex systems. The case study uses a simplified model of a refrigeration module, and its applicability to larger, more intricate systems would require further validation. Future work could involve extending this method to more complex systems, exploring more sophisticated failure propagation modeling techniques, and potentially integrating it with real-time monitoring systems for proactive risk assessment.

Summary updated

BibTeX
@article{xiong2016novel,
  author = {Jian Xiong and Shulin Tian and Chenglin Yang and Cheng Liu},
  title = {{A Novel Method of Failure Sample Selection for Electrical Systems Using Ant Colony Optimization}},
  journal = {Comput. Intell. Neurosci.},
  year = {2016},
  doi = {10.1155/2016/4238734},
}
Figure