A Hybrid Framework for Explainable Aviation Risk: Causal Chain Extraction from Safety Reports with Physics-Based Failure Quantification and Interventional Analysis
Organizations: Department of Aerospace Engineering and Engineering Mechanics, The University of Texas at Austin, Austin, TX 78712, USA. · Department of Electrical and Computer Engineering, University of Toronto, Toronto, ON M5S 3G4, Canada. · Daniel Guggenheim School of Aerospace Engineering, Georgia Institute of Technology, Atlanta, GA 30332, USA. · School of Computing and Augmented Intelligence, Arizona State University, Tempe, AZ 85287, USA. · Department of Aerospace and Mechanical Engineering, University of Oklahoma, Norman, OK 73019, USA.
Aviation authorities hold decades of accident reports whose narratives record how investigators reconstructed each accident. Computational work on these archives has mostly sorted reports into fixed taxonomies. In this work, we propose a hybrid framework in which a text-based corpus identifies the leading factors of a failure mode. Physics-based models then quantify how likely that failure is at a stated operating point. Ordered causal chains extracted from 56,202 American accident reports with a local large language model aggregate into a causation knowledge graph. On an evaluation sample of 2,000 reports, the chains recover 90.8% of investigator-flagged causes. Each physical factor whose drivers the record carries receives an occurrence model calibrated to the accident population. A leaky noisy-OR fitted to the corpus combines the factors into a failure mode probability that supports intervention. For engine power loss, the carburetor icing model separates 1,205 icing accidents from other accidents using temperature and dewpoint alone, reaching 0.61 in area under the receiver operating characteristic curve against a density ratio ceiling of 0.64. Routine weather hours give no valid control, because flying concentrates in fair weather. On later years the physics adds 0.03 to that area for occurrence and nothing for severity, so its value lies in interpretability. On four named accidents, abduction gives a probability of necessity of 0.29 to 0.50 for the heat omission, under monotonicity. The Fast Causal Inference algorithm does not license the largest backdoor adjustment on the learned networks.
Figures & tables
Fig. 1: The hybrid framework. The data layer converts 56,202 report narratives into ordered causal chains and aggregates them into a causation graph. The graph gives the leading factors L(F) , the consequence conditionals P(Y∣F) , the corpus base rates P(F) and the per-category networks Gc . The physics layer maps the leading factors that have physical drivers to continuous variables θ and evaluates an occurrence model πF at the stated operating point. The admissibility monitor rejects any chain whose transitions the guard margins γe rule out at the recorded state. Calibration ties the level of the physical model to the corpus marginal, and a leaky noisy-OR fitted to the corpus combines the contributing factors of a failure mode. Interventional queries act through backdoor adjustment on Gc or through an intervention do(m) that transforms the physical drivers.
Transition into
Guard margin γe(θ)
Units
Fuel exhaustion
t−tend
hours
Carburetor icing
w(T,Td,ϖ)−wthr
grams per cubic meter
TABLE I: Guard margins behind the physics-modeled transitions of the failure study. Signs are oriented so the transition is enabled when the margin is nonnegative. The time t is the elapsed or planned flight time and tend the fuel endurance. The condensable water w is evaluated at temperature T , dewpoint Td and power setting ϖ , with threshold wthr , and its margin is that difference expressed as absolute humidity at the throttle plate temperature.
Fig. 2: Physics occurrence surface πice(T,Td) at descent power, from Eqs. ( 12 ) to ( 14 ). Dashed and solid contours mark the moderate value of 0.40 and the serious value of 0.80 from the published induction icing chart. The overlaid points are the 1,205 NTSB accidents that the narratives attribute to carburetor icing. The blue contours enclose 50 and 80% of the 38,565 other weather-recorded accidents, so the figure shows separation rather than concentration alone. The area under the ROC curve between the two groups is 0.61.
Fig. 3: Engine failure risk assembled from the contributing factor models. (a) Per-factor contribution to P(EF) by model tier, covering physics, surrogate and data-only factors. Fuel exhaustion is the dominant contributor. Its endurance model is not calibrated, so the factor enters at its corpus base rate (hatched). (b) The combined P(EF) conditioned on the operating point, with every term on the accident population. The vertical axis is broken between 0 and 0.28. Weather in the serious zone of the carburetor icing chart raises the population value of 0.300 to 0.307. The two interventions act on that scenario, carburetor heat lowering the risk to 0.288 and drier air to 0.301. Error bars are 95% bootstrap intervals over 500 refits of the corpus terms, which refit the leak and the strengths on resamples of the records. They cover the corpus terms only.
SDAIA-KFUPM Joint Research Center for Artificial Intelligence, King Fahd University of Petroleum and Minerals (KFUPM), Dhahran 31261, Saudi Arabia · School of Systems and Computing, University of New South Wales, Canberra, ACT 2600, Australia · School of Science, University of New South Wales, Canberra, ACT 2612, Australia +1
College of Computer Science, Chongqing University, Chongqing, China · School of Statistics and Data Science, Shanghai University of International Business and Economics, Shanghai, China