Scholarlink

This study develops a methodology to predict the Remaining Useful Life (RUL) of railway wheels in a 1:20 scale railway system. Unlike traditional unimodal analyses, quantitative microscopic image correlation and vibration analysis are performed. The central innovation lies in cross-validation between modalities, where surface wear quantified through digital image processing demonstrated a correlation greater than 79.0% with vibration energy characteristics, particularly with the mean energy 94.0%. This correlation physically validates the dynamic signals and allows for the creation of an enriched feature set for modeling. The Fully Connected Neural Network (FCNN) trained with these data showed stable convergence (epoch 91), a mean absolute error of less than 5.0%, and random residuals, confirming its robustness and generalizability. The resulting model allows for accurate RUL estimates, facilitating the optimization of predictive maintenance strategies in railway systems.

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

Estimation of Remaining Useful Life in Bearings Using Fusion of Machine Vision Features and Vibration Analysis

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Tania Elizabeth Sandoval-Valencia,
  • Jose Emiliano Lopez-Ramirez,
  • Eric Leonardo Huerta-Manzanilla,
  • Gerardo Hurtado-Hurtado,
  • Juan Carlos Jauregui-Correa

摘要

This study develops a methodology to predict the Remaining Useful Life (RUL) of railway wheels in a 1:20 scale railway system. Unlike traditional unimodal analyses, quantitative microscopic image correlation and vibration analysis are performed. The central innovation lies in cross-validation between modalities, where surface wear quantified through digital image processing demonstrated a correlation greater than 79.0% with vibration energy characteristics, particularly with the mean energy 94.0%. This correlation physically validates the dynamic signals and allows for the creation of an enriched feature set for modeling. The Fully Connected Neural Network (FCNN) trained with these data showed stable convergence (epoch 91), a mean absolute error of less than 5.0%, and random residuals, confirming its robustness and generalizability. The resulting model allows for accurate RUL estimates, facilitating the optimization of predictive maintenance strategies in railway systems.