Scholarlink

Predictive maintenance (PdM) has become a cornerstone of the Industry 4.0. Through data driven approach, it enables the anticipation of equipment failure. Our study conducts a performance analysis of eleven Machine Learning (ML) models to evaluate the Remaining Useful Life (RUL) prediction for turbofan jet engines using the C-MAPSS NASA dataset, particularly FD001 subset. We experimented three scenarios: first, we used raw data without preprocessing, the second scenario is by using statistical preprocessing combined with feature engineering and the third scenario is by using ExtraTrees-based feature selection. The models were optimized using grid search with 10-fold cross-validation, and the performance was assessed by using the Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE) and statistical tests. The results of our study show that Support Vector Regression (SVR) and the Huber Regressor combined with statistical preprocessing and feature engineering constantly achieved the best predictive accuracy, providing the largest error reduction (MAE = 43.96, RMSE = 52.50 for Huber Regressor) and (MAE = 45.31, RMSE = 55.45 for SVR). ExtraTrees-based feature selection on the other hand, did not specifically improve the accuracy and in some cases, degraded the results compared to engineered features. Our findings demonstrate the important and critical role of preprocessing methods to enhance the RUL prediction.

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

Performance Analysis of Machine Learning Models for Remaining Useful Life Prediction Under Different Preprocessing Strategies

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Oussama Benmansour,
  • Ibtissam Medarhri,
  • Souhail Housni,
  • Mohamed Hosni

摘要

Predictive maintenance (PdM) has become a cornerstone of the Industry 4.0. Through data driven approach, it enables the anticipation of equipment failure. Our study conducts a performance analysis of eleven Machine Learning (ML) models to evaluate the Remaining Useful Life (RUL) prediction for turbofan jet engines using the C-MAPSS NASA dataset, particularly FD001 subset. We experimented three scenarios: first, we used raw data without preprocessing, the second scenario is by using statistical preprocessing combined with feature engineering and the third scenario is by using ExtraTrees-based feature selection. The models were optimized using grid search with 10-fold cross-validation, and the performance was assessed by using the Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE) and statistical tests. The results of our study show that Support Vector Regression (SVR) and the Huber Regressor combined with statistical preprocessing and feature engineering constantly achieved the best predictive accuracy, providing the largest error reduction (MAE = 43.96, RMSE = 52.50 for Huber Regressor) and (MAE = 45.31, RMSE = 55.45 for SVR). ExtraTrees-based feature selection on the other hand, did not specifically improve the accuracy and in some cases, degraded the results compared to engineered features. Our findings demonstrate the important and critical role of preprocessing methods to enhance the RUL prediction.