Predictive engine model incorporating physics based model estimation and machine learning.

Area of Science:

• Aerospace Engineering
• Data Science

Background:

• Aircraft Engine Health Monitoring (EHM) systems collect critical in-service sensor data.
• EHM systems are prone to malfunction, leading to data loss or inaccuracies, impacting predictive modeling.
• Real-time data capture during flights results in significant information gaps.

Purpose of the Study:

• To address data quality and quantity issues for Machine Learning (ML) predictive engine performance models.
• To evaluate various missing value imputation methods for EHM data.
• To assess the effectiveness of a physics-based engine performance model in handling missing data.

Main Methods:

• Evaluation of data handling techniques: deletion, interpolation, ML model inference.
• Development of a physics-based engine performance model using Numerical Propulsion System Simulation (NPSS).
• Comparison of imputation methods for improving ML-based predictive model accuracy.

Main Results:

• The physics-based engine model effectively estimated missing EHM data.
• Incorporating the physics-based model significantly improved the prediction accuracy of the ML-based model.
• The physics-based approach outperformed other evaluated imputation methods.

Conclusions:

• A physics-based engine performance model is a superior method for imputing missing data in EHM systems.
• This approach enhances the reliability and accuracy of ML-based predictive engine performance models.
• Improved data imputation is crucial for robust predictive maintenance in aviation.