Scholarlink

<p>Automated condition monitoring of railway switches and crossings (S&amp;C) requires classification models whose reported accuracy reflects genuine generalization rather than evaluation artefacts. This paper presents a methodologically rigorous, leak-free machine-learning framework for vibration-based event classification, evaluated on accelerometer data from a full-scale outdoor S&amp;C test facility. The pipeline enforces strict ordering (split, select, augment, standardize, train, evaluate) and partitions the data at the level of physical events, so that all measurements of a given event are assigned together to either the training or the test subset. A symmetric tabular autoencoder generates synthetic minority-class samples through latent-space interpolation. Twenty-one classifiers spanning eight families are benchmarked on held-out data and by group-aware five-fold cross-validation. The strongest models reach 81.5% held-out accuracy (ROC-AUC <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\approx 0.94\)</EquationSource> </InlineEquation>) and <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(80.4\%\pm 2.1\%\)</EquationSource> </InlineEquation> under cross-validation; ensemble methods are the most stable. Feature standardization is essential: without it, neural networks collapse below chance level. Computational profiling (inference latency 0.005–0.63&#xa0;ms per one-second segment; model size 0.002–2.4&#xa0;MB) maps three deployment scenarios to specific algorithm recommendations. Because the minority crossing class has only six held-out samples, its per-class metrics carry wide confidence intervals and should be interpreted with caution.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

AI-driven vibration-based event classification in railway switches and crossings

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月23日

查看全文
  • Mohammad Adoul Amin,
  • Taoufik Najeh,
  • Abdelhamid Ghoul

摘要

Automated condition monitoring of railway switches and crossings (S&C) requires classification models whose reported accuracy reflects genuine generalization rather than evaluation artefacts. This paper presents a methodologically rigorous, leak-free machine-learning framework for vibration-based event classification, evaluated on accelerometer data from a full-scale outdoor S&C test facility. The pipeline enforces strict ordering (split, select, augment, standardize, train, evaluate) and partitions the data at the level of physical events, so that all measurements of a given event are assigned together to either the training or the test subset. A symmetric tabular autoencoder generates synthetic minority-class samples through latent-space interpolation. Twenty-one classifiers spanning eight families are benchmarked on held-out data and by group-aware five-fold cross-validation. The strongest models reach 81.5% held-out accuracy (ROC-AUC \(\approx 0.94\) ) and \(80.4\%\pm 2.1\%\) under cross-validation; ensemble methods are the most stable. Feature standardization is essential: without it, neural networks collapse below chance level. Computational profiling (inference latency 0.005–0.63 ms per one-second segment; model size 0.002–2.4 MB) maps three deployment scenarios to specific algorithm recommendations. Because the minority crossing class has only six held-out samples, its per-class metrics carry wide confidence intervals and should be interpreted with caution.