Aircraft Maintenance Prediction Using Deep Learning

Abstract

Now Days, Advancement In Machine Learning And Artificial Intelligence Many Sectors Are Using Them To Enhance Their Service. For Aviation Industry One Of The Highest Expenses Is That The Cost For Aircraft Maintenance So Now Days, It Is Witnessed That A Growing Interest In Implementing Predictive Maintenance Techniques To Improve Aircraft Reliability, Reduce Unscheduled Downtime, And Enhance Safety. This Thesis Focuses On The Application Of Deep Learning Methodologies For Aircraft Predictive Maintenance, Specifically Exploring The Effectiveness Of The Temporal Convolutional Network With Multi-Head Self-Attention (TCN-MHSA) Model In Comparison To The Convolutional Networks (CNN) And Other Deep Learning Architectures. Our Proposed Study Utilizes A Benchmark Dataset Obtained From The National General Aviation Flight Information Database (NGAFI) For Training And Evaluation. Since The Full Data Set Is Very Huge And Very Expensive To Train A Model, We Selected Flights Which Are With 3 Days Before And After Maintenance Which Includes The 5 Maintenance Issues. We Proposed TCN-MHSA Model Which Is Introduced As A Novel Architecture Capable Of Capturing Both Temporal And Spatial Dependencies In The Data, Utilizing A Combination Of Convolutional Layers, Multi-Head Self-Attention Mechanisms.To Evaluate The Performance Of The TCN-MHSA Model, Comprehensive Experiments Are Conducted And Compared Against The CNN-MHSA And CNN-LSTM Architectures. The Evaluation Metrics Include Area Under The Receiver Operating Characteristic Curve (AUC-ROC), Area Under The Precision-Recall Curve (AUC-PR) And Along With Accuracy. Our Results Indicate That The TCN MHSA Model Outperforms Both The CNN-MHSA And CNN-LSTM Architectures In Terms Of Predictive Maintenance Performance. The TCN-MHSA Model Achieves Higher AUC-PR, AUC ROC And Accuracy Across The 5 Maintenance Categories.