Statistical methods for critical scenarios in aeronautics

TL;DR

This paper develops a Bayesian Gaussian process-based statistical method to reconstruct aircraft deceleration profiles during landing, aiming to detect braking system malfunctions by comparing predicted and actual deceleration data.

Contribution

It introduces a novel Bayesian Gaussian process approach for nonparametric regression of aircraft deceleration profiles in a safety-critical context.

Findings

The Gaussian process model accurately reconstructs deceleration profiles.

Comparison shows Bayesian approach outperforms other statistical methods.

Method enables early detection of braking system malfunctions.

Abstract

We present numerical results obtained on the CEMRACS project Predictive SMS proposed by Safety Line. The goal of this work was to elaborate a purely statistical method in order to reconstruct the deceleration profile of a plane during landing under normal operating conditions, from a database containing around $1500$ recordings. The aim of Safety Line is to use this model to detect malfunctions of the braking system of the plane from deviations of the measured deceleration profile of the plane to the one predicted by the model. This yields to a multivariate nonparametric regression problem, which we chose to tackle using a Bayesian approach based on the use of gaussian processes. We also compare this approach with other statistical methods.