Improving Aviation Safety through Uncertainty-Aware Forecasting and Anomaly Detection

Details

• Time: 18:30-19:30 pm   (Beijing Time)
• Date: Tuesday,  April 16, 2024
• Venue:Zoom Room: 534 380 1855(https://vanderbilt.zoom.us/j/5343801855?omn=96062225258)
• Speaker：Dr. Yingxiao Kong, Vanderbilt University
  
• Host：Prof. Conghua Wen

Abstract

Approach and landing are critical phases in aviation, with hard landings—defined by a touchdown vertical speed exceeding a set threshold— posing significant risks to aircraft and passengers. Traditional approaches to hard landing classification lack the nuance needed for effective risk management due to class imbalance issues. We propose a novel solution using a Bayesian LSTM neural network with Monte Carlo dropout to predict touchdown vertical speed and quantify uncertainty. This method enables more informed decision-making for pilots and controllers. Our ongoing research explores enhancing model interpretability and precursor mining for adverse flight events, emphasizing the development of interpretable deep learning forecasting models. Initial findings suggest that gradient-based attribution methods offer efficient and model-agnostic solutions for identifying critical forecasting features and timelines. These efforts aim to substantially improve aviation safety by providing advanced insights into critical landing phase operations.

Speaker

Dr. Yingxiao Kong is an adjunct assistant professor of Computer Science at Vanderbilt University and an experienced data scientist in industry. In her current role, she designs and teaches courses for undergraduate students in the Data Science and Computer Science Departments, mentors aspiring professionals, and conducts research. As a data scientist, she specializes in developing and deploying time series forecasting models for supply chain optimization. Dr. Kong's research and teaching focus on applied machine learning, uncertainty quantification, and aviation safety. She also serves as a journal referee for publications such as IEEE Transactions on Intelligent Transportation Systems and the Journal of Traffic and Transportation Engineering.