A highly accurate method for forecasting the compressor geometric variable system based on the data-driven method.

Area of Science:

• Aerospace Engineering
• Artificial Intelligence
• Mechanical Engineering

Background:

• Understanding aero-engine compressor systems is crucial for complete engine modeling and maintenance.
• Traditional forecasting methods struggle with predicting faults due to their inability to utilize future information from non-target parameters.
• Accurate fault diagnosis and prediction are vital for ensuring the reliability and safety of aero-engines.

Purpose of the Study:

• To propose an enhanced ConvNeXt model for forecasting the compressor geometric variable system in aero-engines.
• To address the limitation of traditional forecasting techniques by incorporating future information of non-target parameters.
• To alleviate the issue of increasing forecast errors with longer forecast horizons.

Main Methods:

• Development of an enhanced ConvNeXt model integrated with the Sliding Window Algorithm.
• Utilizing future information of non-target parameters within the forecasting model.
• Conducting experiments on various aero-engine states, including transition and steady states, using real flight data.

Main Results:

• The proposed method significantly alleviates the problem of forecast error increasing with forecast length.
• Achieved a reduction in forecast error to 20.07% compared to standard forecasting approaches under specific conditions.
• Demonstrated the applicability of the method across various types of aero-engines.

Conclusions:

• The enhanced ConvNeXt model with the Sliding Window Algorithm offers a feasible and effective approach for forecasting aero-engine compressor systems.
• The method's ability to incorporate future non-target parameter data enhances prediction accuracy and reliability.
• Experimental validation using real flight data confirms the model's credibility for diverse aero-engine operational states.