Performance Sensing Data Prediction for an Aircraft Auxiliary Power Unit Using the Optimized Extreme Learning...

Area of Science:

• Aerospace Engineering
• Machine Learning
• Artificial Intelligence

Background:

• Aircraft Auxiliary Power Units (APUs) are critical for cabin environmental control and main engine starting.
• Predicting APU performance sensing data is vital for implementing effective condition-based maintenance strategies.
• APU performance data exhibits complex nonlinear characteristics, necessitating advanced modeling techniques.

Purpose of the Study:

• To develop a stable and accurate model for predicting aircraft APU performance sensing data.
• To enhance the nonlinear fitting capabilities for APU monitoring.
• To improve the reliability of APU condition-based maintenance through precise data prediction.

Main Methods:

• Utilized Extreme Learning Machine (ELM) for its superior nonlinear fitting and faster learning speed compared to traditional backpropagation neural networks.
• Employed Restricted Boltzmann Machine (RBM) to optimize the ELM, addressing issues of unstable predictions due to randomly generated weights and thresholds.
• Evaluated the proposed RBM-ELM model using real-world APU sensing data from China Southern Airlines.

Main Results:

• The RBM-optimized ELM demonstrated enhanced stability in performance parameter prediction.
• The proposed model achieved more accurate prediction results compared to standard approaches.
• Experimental validation confirmed the effectiveness of the RBM-ELM for APU performance monitoring.

Conclusions:

• The RBM-ELM model provides a robust solution for predicting APU performance sensing data.
• This optimized approach significantly improves the stability and accuracy of APU condition-based maintenance.
• The findings contribute to more reliable aircraft maintenance through advanced machine learning techniques.