Time-feature attention-based convolutional auto-encoder for flight feature extraction.

Area of Science:

• Aerospace Engineering
• Data Science
• Machine Learning

Background:

• Quick Access Recorders (QARs) are crucial for Flight Operation Quality Assurance (FOQA) and flight safety.
• QAR data presents challenges due to its large volume, high dimensionality, and high frequency, leading to complexities in usage and comprehension.
• Existing methods struggle with the inherent complexities of QAR data.

Purpose of the Study:

• To develop an advanced model for extracting essential flight features from complex QAR data.
• To enhance the utility of QAR data for applications like flight safety analysis and risk detection.
• To compare the proposed model's performance against traditional and similar approaches.

Main Methods:

• Proposed a novel Time-Feature Attention (TFA)-based Convolutional Auto-Encoder (TFA-CAE) network model.
• Utilized QAR data from landings at Kunming Changshui International Airport and Lhasa Gonggar International Airport for case studies.
• Benchmarked TFA-CAE against Principal Component Analysis (PCA), Convolutional Auto-Encoder (CAE), Self-Attention-based CAE (SA-CAE), and Gate Recurrent Unit based Auto-Encoder (GRU-AE) models.

Main Results:

• The TFA-CAE model demonstrated superior performance in extracting representative flight features compared to all other tested models.
• The model successfully recognized distinct flight patterns associated with different runways.
• Anomalous flights were effectively identified and distinguished from normal observations.

Conclusions:

• The TFA-CAE model offers a robust and effective technique for processing and analyzing complex QAR data.
• This approach significantly advances the potential for using QAR data in critical areas such as flight risk detection and FOQA.
• The model's ability to extract meaningful features facilitates improved flight safety and operational quality assurance.