Impact monitoring of large size complex metal structures based on sparse sensor array and transfer learning

Area of Science:

• Aerospace Engineering
• Structural Health Monitoring
• Artificial Intelligence in Aviation

Background:

• Aircraft impact events pose significant safety risks due to potential structural damage.
• Current dense sensor arrays for monitoring are costly and add weight, impacting operational economics.
• Accurate impact location and force history are crucial for aircraft condition assessment.

Purpose of the Study:

• To develop an efficient region-to-point monitoring strategy for aircraft impact detection and force prediction.
• To reduce reliance on dense sensor arrays by utilizing sparse data and advanced algorithms.
• To investigate the effectiveness of transfer learning for diverse aircraft fuselage structures.

Main Methods:

• A Convolutional Neural Network (CNN) with region localization capability was trained on sparse sensor data.
• A weighted center algorithm, enhanced by a fuzzy genetic algorithm, was employed for adaptive impact localization.
• A Convolutional Neural Network-Gated Recurrent Unit with Squeeze-Excitation attention (CNN-GRU-SE) was used for impact force prediction.
• Transfer learning was applied to assess model performance across different fuselage structures.

Main Results:

• The proposed strategy accurately determines impact locations and predicts impact force time histories.
• The fuzzy genetic algorithm adaptively improved localization accuracy.
• The CNN-GRU-SE model demonstrated capability in predicting impact forces on various aircraft structural components.
• Transfer learning showed potential for improving model performance and reducing training costs for different fuselage regions.

Conclusions:

• The region-to-point monitoring strategy offers an effective alternative to dense sensor arrays for aircraft structural health monitoring.
• The developed AI models provide accurate impact localization and force prediction, enhancing aircraft safety.
• Transfer learning is a viable approach for adapting models to diverse aircraft structures, optimizing training efficiency.