Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A New Hybrid Approach for Identifying Obsolescence Features: Applied to Railway Signaling Infrastructure

Elie Saad, Mariem Besbes, Marc Zolghadri, Victor Czmil, Vincent Bourgeois

TL;DR

The paper tackles obsolescence management in railway signaling infrastructure by addressing the mismatch between short component life cycles ($\approx$ $18$ months) and long system lifespans ($\approx$ $30$ years). It introduces a hybrid workflow that first builds a feature set from expert-driven data, then reduces dimensionality with Pearson correlation filtering and PCA, and finally learns expert-like decisions with a Decision Tree Classifier. Two analyses quantify feature importance: mean decrease in Gini impurity and contributions to the PCA subspace, revealing both alignment and disagreement with expert judgments. The results underscore data scarcity and variability, showing moderate predictive performance and motivating data-generation and few-shot learning to improve decision support for obsolescence remediation in railway networks.

Abstract

Electrical component obsolescence poses a major issue especially within systems with large life cycles. Thus, finding the optimal management solution for each obsolescence case is as crucial as knowing what to consider when faced with an obsolescence case. In this paper, a novel hybrid approach for identifying features affecting electrical component obsolescence management is introduced, which combines features engineering techniques and expert knowledge. The method then uses machine learning to predict obsolescence resolution techniques in order to find the optimal resolution. The motivation behind this research is driven by the imperative need for SNCF RESEAU to optimally address and mitigate the challenges posed by electrical component obsolescence in railway infrastructure.

A New Hybrid Approach for Identifying Obsolescence Features: Applied to Railway Signaling Infrastructure

TL;DR

The paper tackles obsolescence management in railway signaling infrastructure by addressing the mismatch between short component life cycles ( months) and long system lifespans ( years). It introduces a hybrid workflow that first builds a feature set from expert-driven data, then reduces dimensionality with Pearson correlation filtering and PCA, and finally learns expert-like decisions with a Decision Tree Classifier. Two analyses quantify feature importance: mean decrease in Gini impurity and contributions to the PCA subspace, revealing both alignment and disagreement with expert judgments. The results underscore data scarcity and variability, showing moderate predictive performance and motivating data-generation and few-shot learning to improve decision support for obsolescence remediation in railway networks.

Abstract

Electrical component obsolescence poses a major issue especially within systems with large life cycles. Thus, finding the optimal management solution for each obsolescence case is as crucial as knowing what to consider when faced with an obsolescence case. In this paper, a novel hybrid approach for identifying features affecting electrical component obsolescence management is introduced, which combines features engineering techniques and expert knowledge. The method then uses machine learning to predict obsolescence resolution techniques in order to find the optimal resolution. The motivation behind this research is driven by the imperative need for SNCF RESEAU to optimally address and mitigate the challenges posed by electrical component obsolescence in railway infrastructure.
Paper Structure (12 sections, 2 figures)

This paper contains 12 sections, 2 figures.

Table of Contents

  1. Introduction
  2. Contribution
  3. Dataset Formation Methodology
  4. Data Acquisition
  5. Dataset Formalization
  6. Model Methodology
  7. Step 1: PC.
  8. Step 2: PCA.
  9. Step 3: DTC.
  10. Analysis Methodology
  11. Experimentation and Discussion
  12. Conclusion and Perspective

Figures (2)

  • Figure 1: Methodological Approach Process
  • Figure 2: Obtained results displaying feature contributions on a logarithmic scale, following the methodology outlined in Section \ref{['sec:analysis_methodology']}. Blue bars indicate mean contributions, red bars show values from the best model, and green bars represent contributions from the top-performing model on non-obsolete instances.