A Data-Driven Aero-Engine Degradation Prognostic Strategy.

Area of Science:

• Aerospace Engineering
• Mechanical Engineering
• Reliability Engineering

Background:

• Aero-engine degradation prognostics are critical for operational safety and efficiency.
• Data-driven prognostic methods offer an alternative to model-based approaches but struggle with data quality and quantity.
• Unlabeled, unbalanced condition monitoring (CM) data and prognostic uncertainties pose significant challenges.

Purpose of the Study:

• To propose a novel data-driven strategy for aero-engine degradation prognostics.
• To address challenges associated with unlabeled, unbalanced CM data and prognostic uncertainties.
• To enhance the accuracy and feasibility of remaining useful life (RUL) estimation for aero-engines.

Main Methods:

• Feature selection using two defined indicators to remove redundant degradation features.
• Fuzzy c-means clustering to determine discrete health states and automatically assign health labels.
• Consideration of uncertain initial conditions and health state transition uncertainties.
• Development of multivariate health estimation and multistep-ahead degradation prediction models.

Main Results:

• The proposed strategy effectively handles unlabeled and unbalanced CM data.
• Automatic health state labeling and uncertainty consideration improve prognostic accuracy.
• Verification using NASA aero-engine data demonstrates the strategy's effectiveness and feasibility.
• Accurate remaining useful life estimation is achieved.

Conclusions:

• The novel data-driven strategy provides a robust solution for aero-engine degradation prognostics.
• The method overcomes limitations of traditional data-driven approaches by managing data quality and uncertainty.
• This approach enhances the reliability and safety of aero-engine operations through improved RUL prediction.