Engine gas path component fault diagnosis based on a sparse deep stacking network.

Area of Science:

• Aerospace Engineering
• Artificial Intelligence

Background:

• Accurate engine gas path component fault diagnosis is crucial for operational reliability and safety.
• Data-driven methods, particularly deep stacking neural networks (DSN), show promise but struggle with simultaneous, coupled faults.

Purpose of the Study:

• To improve the prediction performance of DSN for engine gas path fault diagnosis, especially when multiple faults occur concurrently.
• To introduce a sparse regularization and representation method to enhance DSN capabilities.

Main Methods:

• A novel sparse regularization term was integrated into the traditional deep stacking neural network framework.
• The method was evaluated by comparing its diagnosis performance against six other neural network methods across various fault types.

Main Results:

• The proposed sparse regularization method significantly boosted DSN prediction performance.
• Achieved an exceptional accuracy rate of 99.9% in diagnosing various gas path component fault conditions, outperforming other methods.

Conclusions:

• The developed method effectively handles multiple, coupled gas path faults in engines.
• This advancement supports engine health management by enabling proactive maintenance planning.