Browsing by Author "Malan, Pieter Jacobus"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    Upset detection for passenger airliners using classification of anemometric and inertial sensor data
    (Stellenbosch : Stellenbosch University, 2016-12) Malan, Pieter Jacobus; Engelbrecht, J. A. A.; Engelbrecht, H. A.; Stellenbosch University. Faculty of Engineering. Dept. of Electrical and Electronic Engineering.
    ENGLISH ABSTRACT: Due to the increasing number of civilians making use of airliners for business and travel purposes, safe and reliable operation of commercial passenger airliners is essential. Aircraft upsets are particular conditions that may result in fatal accidents. The accurate identification of such upsets, however, can greatly reduce the risk of fatal accidents. Conventional flight control systems for aircraft are designed to operate within a specified flight envelope, which is defined by an allowable range of air incidence angles (called the angle of attack and side-slip angle), airspeeds, and angular rates over which the aerodynamic forces and moments behave linearly. The flight envelope is also defined by an allowable range of aircraft attitudes (pitch angles and bank angles) and flight trajectories. When the aircraft exceeds its flight envelope, it enters the upset domain, where the aircraft typically stalls and may enter an uncontrolled spin. A need therefore exists for a system that can detect and recognise “out-of-envelope” conditions using on-board sensor measurements. Pilots are not always aware that the aircraft has entered the upset domain, especially if visual cues are not available, and may respond with incorrect actions. It is however possible for a pilot to recover from an upset event, but the knowledge of the specific upset event is of utmost importance. A need therefore exists to advise the pilot of an occurring upset so that the correct recovery actions can be taken. This thesis presents the design and verification of an upset detection system for commercial passenger airliners that detects and identifies flight upset conditions using classification techniques operating on sensor data from on-board anemometric and inertial sensors. The study focused on three aerodynamic upsets, namely high angle of attack upset, underspeed upset and a novel predictive form of high angle of attack upset detection named dynamic pitch upset. The upset detection system used classification algorithms trained on labelled anemometric and/or inertial sensor measurements. A wide variety of classifiers were investigated in order to determine the feasibility of classification-based upset detection. The three aerodynamic upset detection systems were evaluated across two cases, namely those that utilised sensor data from both the anemometric and inertial sensors, and those using data from only the inertial sensors. The high angle of attack upset detection system provided an accuracy of 98.8% for initial test case and an accuracy of 92.2% for the latter. The underspeed upset detection system provided an accuracy of 99.3% for the first case and 89.4% for the second case. The dynamic pitch upset detection system provided an accuracy of 93.6% for the initial case and 91.4% for the latter. The high classifier accuracies provided a suitable and reliable means for aircraft upset detection. These accuracies, along with the locations of the false alarms—the false alarms occur near the upset boundary—enable the pilots to detect and recognise instances pertaining to upset.