Scholarlink

<p>Gas turbine engines on aircraft are equipped with an Engine Health Monitoring (EHM) system that collects in-service data of various installed sensors. The system is not free from malfunction or deterioration. Hence, the signal can be lost (missing data) or convey faulty information. As the EHM system only captures real-time data during a flight, it implies an important loss of crucial information. It raises an issue regarding data quality and quantity required for adopting Machine Learning (ML) approaches to building predictive engine performance models. Therefore, a Missing Value Imputation process is necessary for training ML model. In this paper, various methods for handling missing data are evaluated including deletion, interpolation, ML model inference, and a physics-based engine performance model. The physics-based engine model was built using commercially available software, Numerical Propulsion System Simulation&#xa0;(NPSS), to provide estimations of those missing data. The results show that incorporating a physics-based model improved the prediction accuracy of the ML-based predictive model and outperformed other methods.</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

Predictive engine model incorporating physics based model estimation and machine learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月26日

查看全文
  • Jin-sol Jung,
  • Changmin Son,
  • Andrew Rimell,
  • Rory J. Clarkson

摘要

Gas turbine engines on aircraft are equipped with an Engine Health Monitoring (EHM) system that collects in-service data of various installed sensors. The system is not free from malfunction or deterioration. Hence, the signal can be lost (missing data) or convey faulty information. As the EHM system only captures real-time data during a flight, it implies an important loss of crucial information. It raises an issue regarding data quality and quantity required for adopting Machine Learning (ML) approaches to building predictive engine performance models. Therefore, a Missing Value Imputation process is necessary for training ML model. In this paper, various methods for handling missing data are evaluated including deletion, interpolation, ML model inference, and a physics-based engine performance model. The physics-based engine model was built using commercially available software, Numerical Propulsion System Simulation (NPSS), to provide estimations of those missing data. The results show that incorporating a physics-based model improved the prediction accuracy of the ML-based predictive model and outperformed other methods.