A hierarchical Bayesian methodology for aircraft performance monitoring addresses two main objectives: identifying variations in aircraft fuel consumption with 1% accuracy and distinguishing between airframe and engine effects as causes of these variations. The computational framework decomposes inference into fleet-level and aircraft-specific scales, embedding interpretable physics-based aerodynamic and propulsion models at each level and quantifying uncertainty in deterioration-related parameters through posterior distributions rather than point estimates. Verification using synthetic operational data with embedded ground truth yields mean absolute errors of 0.32% for drag variations and 0.25% for engine fuel flow deviations, achieving better detection sensitivity than current monitoring systems. Validation with operational flight data confirms the method’s ability to distinguish between deterioration sources under real-world conditions, a fundamental diagnostic limitation of current methods. The framework also estimates aircraft gross weight with a mean absolute error of 0.73%, a useful level of accuracy given the airline industry’s reliance on statistical averages of passenger weights rather than direct measurements. In contrast to some machine learning models that lack interpretability, generalizability, and uncertainty quantification, this physics-informed statistical approach enables targeted maintenance interventions for specific aircraft while requiring only modest adaptation to model different aircraft types.
Bays et al. (2026) studied this question.