Automated Structural Bolt Micro Looseness Monitoring Method Using Deep Learning

Area of Science:

• Aerospace Engineering
• Mechanical Engineering
• Artificial Intelligence

Background:

• Detecting bolt loosening in aircraft engines is challenging due to complex models and signal characteristics.
• Early identification of micro-looseness and feature extraction from non-stationary signals are significant hurdles.

Purpose of the Study:

• To propose an automated structural bolt micro-looseness monitoring method using deep learning.
• To address the limitations in current bolt loosening detection techniques for aircraft engines.

Main Methods:

• Developed a Batch Normalized Stacked Autoencoders (BNSAEs) model for rapid convergence and effectiveness.
• Utilized characterization functions for signal preprocessing to simplify data and preserve features.
• Applied the method to bolt and flange connection structures in aircraft engine components.

Main Results:

• The BNSAEs model demonstrated quick convergence and effective performance.
• Signal preprocessing simplified data structure while maintaining feature integrity, reducing training time.
• The method was successfully validated on critical aircraft engine components.

Conclusions:

• The proposed deep learning method offers an effective solution for automated structural bolt micro-looseness monitoring.
• The BNSAEs model and signal preprocessing enhance efficiency and accuracy in detecting early bolt loosening.