In-Time Flight Anomaly Detection and Risk Prediction with Neural Networks

Metron proposes develop a neural network to monitor aircraft to detect and inform air traffic control (ATC) of potential safety issues in time for corrective actions to be taken before serious problems develop. Our approach leverages our familiarity with the relevant data and extends our prototype neural network that has demonstrated the ability to learn the typical flight dynamics of aircraft arriving at JFK.We will address the technical questions of how to monitor the NAS continuouslymdash;using data fused from potentially different sourcesmdash;and automatically identify anomalous flights that indicate safety hazards or precursors to such, while reducing the false alarm rate of such a system. Additionally, we will establish the feasibility of extending the model to predict the probabilities of the occurrence of hazardous events and of their safety risks. Such a system will make predictions ldquo;in timerdquo; for ATC to take corrective action to minimize the effect of such events.Metron has teamed with ATAC who willnbsp;providenbsp;subject matter expertise to visualize and assess the safety hazards identified by the system.