Electronic Thesis/Dissertation
 

Aerospace Fault Diagnosis Accuracy in Aircraft Engine using One-Dimensional Convolutional Neural Networks with Multi-Head Attention Mechanism

Open Access Deposited

Aerospace Fault Diagnosis Accuracy in Aircraft Engine using One-Dimensional Convolutional Neural Networks with Multi-Head Attention Mechanism This praxis outlines design and testing one-dimensional convolutional neural network, made up of a multi-head self-attention part (MA1DCNN), to classify multi-label faults in the NASA N-CMAPSS data. Normalization, segmentation, and selective oversampling for class balancing were among data operations utilized. The MA1DCNN combines convolutional blocks with self-attention modules and residual connections to capture both local and long-range dependencies efficiently. Key training strategies included focal loss for minority faults, Cosine Annealing Warm Restarts for learning rates, Gaussian noise for robustness, and hyperparameter optimization via Optuna. Additionally, SHAP analysis provided insights into sensor contributions to predictions On 5,335,270 test samples, MA1DCNN achieved a Macro-F1 of 99.46% and a Macro-Precision of 99.92%. While its classification accuracy of 98.16% was slightly below that of other models evaluated in this study, 1DCNN-BiLSTM with CBAM at 98.23% and XGBoost at 98.25%, it demonstrated the highest Macro-Precision and featured a streamlined design that reduces parameters and latency. Its attention modules enhance fault attribution, making MA1DCNN ideal for real-time predictive maintenance in aerospace settings. Future work will focus on evaluating performance under diverse conditions and refining attention heads for improved fault discrimination.

Author Language Date created Type of Work License
  • All rights reserved
Rights statement GW Unit Degree Advisor Committee Member(s) Persistent URL

Notice to Authors

If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.

Thumbnail Title Date Uploaded Visibility Actions
Preview of Akindoju_gwu_0075A_17473.pdf Akindoju_gwu_0075A_17473.pdf 2025-12-12 Open Access