Scholarlink

Bearings are one of the most used machine elements in rotating machines and are susceptible to various mechanical failures, which lead to abrupt shutdowns in the industry. Prognostics is one of the important concepts in reliability and maintenance, which helps in avoiding the abrupt shutdown of the system by estimating the component’s Remaining Useful Life (RUL). It requires continuous monitoring and collection of bearing’s Run-To-Failure (RTF) data. One of the limitations of applying data-driven approaches for the estimation of RUL is acquiring RTF data. The bearing takes a lot of time to reach its end-of-life point from a healthy condition. This paper utilizes the concept of Transfer Learning (TL) to tackle the problem of RTF data acquisition of bearings. The combination of monotonicity, prognosability, and trendability score is used to check the suitability of features to construct a Health Indicator (HI). A Convolutional Neural Network (CNN) based TL model is constructed to estimate the RUL of unknown operating conditions. An open-source XJTU-SY bearing dataset is employed for the validation of the suggested TL-based methodology. The TL model is constructed using two known operating conditions, and the RUL of unknown operating conditions is estimated. The findings show that the developed CNN-based TL model attains a Root Mean Square Error (RMSE) value of 0.223 in an unknown operating condition.

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

Estimating Remaining Useful Life Through Transfer Learning Using Convolutional Neural Network for Bearing

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Afroz Ahmed Saad,
  • Anurag Choudhary,
  • Amar Kumar Verma,
  • Shahab Fatima,
  • Bijaya Ketan Panigrahi

摘要

Bearings are one of the most used machine elements in rotating machines and are susceptible to various mechanical failures, which lead to abrupt shutdowns in the industry. Prognostics is one of the important concepts in reliability and maintenance, which helps in avoiding the abrupt shutdown of the system by estimating the component’s Remaining Useful Life (RUL). It requires continuous monitoring and collection of bearing’s Run-To-Failure (RTF) data. One of the limitations of applying data-driven approaches for the estimation of RUL is acquiring RTF data. The bearing takes a lot of time to reach its end-of-life point from a healthy condition. This paper utilizes the concept of Transfer Learning (TL) to tackle the problem of RTF data acquisition of bearings. The combination of monotonicity, prognosability, and trendability score is used to check the suitability of features to construct a Health Indicator (HI). A Convolutional Neural Network (CNN) based TL model is constructed to estimate the RUL of unknown operating conditions. An open-source XJTU-SY bearing dataset is employed for the validation of the suggested TL-based methodology. The TL model is constructed using two known operating conditions, and the RUL of unknown operating conditions is estimated. The findings show that the developed CNN-based TL model attains a Root Mean Square Error (RMSE) value of 0.223 in an unknown operating condition.