Dynamic Heterogeneous Flight Graphs for AAM Security
A graph neural network approach to risk modeling and security analysis for Advanced Air Mobility (AAM) systems using heterogeneous flight data.
Initial Research & Scoping
◈ FoundationalNov 2025 – Jan 2026
Initial literature review, data source evaluation, model architecture design, and methodology decisions that shaped the project. Several decisions made here were later revised as the project evolved — superseded items are noted.
Established the starting architecture and research foundation. Many specific decisions (update pattern, graph structure, training labels) were later revised through the timeline below.
Data Source Evaluation
Surveyed available aircraft trajectory and accident/incident datasets. AAM-specific data is scarce (mostly simulators), so the approach uses conventional aircraft surveillance data.
Literature Review — Dynamic GNN Patterns
Reviewed 30 papers across Dynamic GNNs, Temporal Knowledge Graphs, and dynamic heterogeneous graphs. Identified three recurring architectural patterns for handling temporal graph data.
Initial Model Architecture & Methodology
Designed the initial model architecture: a multi-stage pipeline from raw data through graph construction, GNN layers, to risk prediction. Established methodology decisions for the decoder, loss function, and GNN design.
Raw Data Schema Definition
Documented the field-level schema for OpenSky state vectors, NTSB event records, and NTSB aircraft records. These definitions remained stable throughout the project.
Synthetic Data Approach
✕ DroppedJan 14 – Feb 16, 2026
Attempted to build training data by matching NTSB accidents to OpenSky trajectories, then generating synthetic collisions for missing aircraft. Built a pipeline that matched 370 accident-trajectory pairs and generated 20 synthetic collision encounters.
Dropped — most incidents had only one aircraft's ADS-B data, synthetic collisions were unrealistic, and no near-miss data could be generated.
Data Quality Requirements
Defined what constitutes a "usable" accident-trajectory pair. Formalized trajectory requirements (minimum points, time alignment, etc.) into a spec enforced by pipeline code.
Airport Proximity Baseline Analysis
Baseline analysis of proximity metrics near busy airports during normal operations. Sampled major US airports to understand what "normal close" looks like so proximity alone isn't treated as risk.
Trajectory Pattern Mining
Pattern mining from trajectory features — extracting repeatable kinematic patterns to understand normal flight behavior and identify anomalies.
Synthetic Pipeline Implementation
Built the end-to-end synthetic data pipeline. Matched 370 NTSB accident-trajectory pairs and tiered them by quality: 212 gold, 38 silver, 22 bronze.
Incident Classification & Narrowing
Classified the 41 usable incidents into multi-aircraft conflicts (15), runway incursions (11), and wrong surface/runway events (15). Narrowed to the 8 airborne multi-aircraft conflicts with trajectory data.
Canonical Schema & Synthetic Generation Plan
Designed a canonical data schema to standardize real and synthetic data formats. Step 5 tackled synthetic collision generation to fill the missing-aircraft gap — 7 of 8 incidents had only 1 aircraft's trajectory.
Synthetic Pipeline Results — Limitations Found
Generated 20 synthetic collision encounters, but only collisions (y=1) — no near-misses. Revealed fundamental limitations: dead-reckoned positions, rotation distortion, perfect 0.0m collision separation. This was the inflection point that led to the surrogate pivot.
Final Synthetic Data Attempts — Approach Dropped
Last attempts to salvage the synthetic approach: an updated classification (v2) and a hybrid simulator proposal for near-miss generation. Both were ultimately dropped — the simulator was too complex for thesis scope, and the underlying data gap couldn't be solved synthetically.
The Surrogate Pivot
✓ ValidatedFeb 19 – Mar 6, 2026
Literature review revealed surrogate safety measures (LoS, well-clear, NMAC, CPA) as well-established alternatives to direct accident labels. Empirical validation on real LA Basin data confirmed these signals are dense enough for training after airport filtering.
Validated — surrogate signals provide dense, computable training labels while preserving the core architecture.
Literature Review — Surrogate Safety Measures
Deep research report on surrogate safety measures for airspace collision risk modeling. Reviewed aviation safety literature on surrogates, mapping them to thesis architecture constraints: computable at event time, dense enough for learning, supports encounter-to-accident interpretation.
The Pivotal Decision — Surrogate Labels
The defining document for the project's current direction. Established that the thesis architecture stays the same but training labels change from accidents to surrogate safety measures. Defines severity tiers, published thresholds, and the MVP training plan.
LA Basin Empirical Validation
Empirical audit of surrogate safety signals on 4 hours of LA Basin ADS-B data. Confirmed surrogates are viable after airport filtering.
Graph & Pipeline Build-Out
◌ In ProgressMar 12, 2026 – present
With the surrogate approach validated, work shifted to defining the formal graph schema, evaluating update strategies, and scaling data collection to the top 5 US airports by passenger traffic.
In progress — graph schema finalized, data collection scaling to national airports.
Graph Update Strategy Evaluation
Evaluated three graph update patterns: event-only, periodic (constant-interval snapshots), and hybrid. Initially recommended Pattern 2 (event-driven) as backbone with Pattern 3 (hybrid) as encoder, but this was later overruled in favor of Pattern 1 (snapshot-based) in the graph definition report.
Definitive Graph Schema
The definitive graph schema document. Defines a two-layer multigraph with aircraft nodes and directed proximity edges carrying raw separation distance. Uses Pattern 1 (snapshot-based) updates — the graph is rebuilt at fixed intervals, replacing the earlier event-driven design.
Scaled Data Collection — Top 5 US Airports
Scaled the 4-hour LA Basin pilot to a representative national dataset. Defines study scope: top 5 US airports by passenger traffic (KATL, KDFW, KDEN, KORD, KLAX), 24-hour ADS-B query windows.