Søren Hansen

Diagnosis of Airspeed Measurement Faults for Unmanned Aerial Vehicles

Airspeed sensor faults are common causes for incidents with unmanned aerial vehicles (UAV) with pitot tube clogging or icing being the most common causes. Timely diagnosis of such faults or other artifacts in signals from airspeed sensing systems could potentially prevent crashes. This paper employs parameter adaptive estimators to provide analytical redundancies and a dedicated diagnosis scheme is designed. Robustness is investigated on sets of flight data to estimate distributions of test statistics. The result is robust diagnosis with adequate balance between false alarm rate and fault detectability.

Diagnosis of UAV Pitot Tube Defects Using Statistical Change Detection

Abstract Unmanned Aerial Vehicles need a large degree of tolerance to faults. One of the most important steps towards this is the ability to detect and isolate faults in sensors and actuators in real time and make remedial actions to avoid that faults develop to failure. This paper analyses the possibilities of detecting faults in the pitot tube of a small unmanned aerial vehicle, a fault that easily causes a crash if not diagnosed and handled in time. Using as redundant information the velocity measured from an onboard GPS receiver, the air-***speed estimated from engine throttle and the pitot tube based airspeed, the paper analyses the properties of residuals. A dedicated change detector is suggested that works on pre-whitened residuals and a generalised likelihood ratio test is derived for a Cauchy probability density, which the residuals are observed to have. A detection scheme is obtained using a threshold that provides desired quantities of false alarm and detection probabilities. Fault detectors are build based on raw residual data and on a whitened edition of these. The two detectors are compared against recorded telemetry data of an actual event where a pitot tube defect occurred.

Orchard navigation using derivative free Kalman filtering

This paper describes the use of derivative free filters for mobile robot localization and navigation in an orchard. The localization algorithm fuses odometry and gyro measurements with line features representing the surrounding fruit trees of the orchard. The line features are created on basis of 2D laser scanner data by a least square algorithm. The three derivative free filters are compared to an EKF based localization method on a typical run covering four rows in the orchard. The Matlab® toolbox Kalmtool is used for easy switching between different filter implementations without the need for changing the base structure of the system.

Control surface fault diagnosis with specified detection probability — Real event experiences

Diagnosis of actuator faults is crucial for aircraft since loss of actuation can have catastrophic consequences. For autonomous aircraft the steps necessary to achieve fault tolerance is limited when only basic and non-redundant sensor and actuators suites are present. Through diagnosis that exploits analytical redundancies it is, nevertheless, possible to cheaply enhance the level of safety. This paper presents a method for diagnosing control surface faults by using basic sensors and hardware available on an autonomous aircraft. The capability of fault diagnosis is demonstrated obtaining desired levels of false alarms and detection probabilities. Self-tuning residual generators are employed for diagnosis and are combined with statistical change detection to form a setup for robust fault diagnosis. On-line estimation of test statistics is used to obtain a detection threshold and a desired false alarm probability. A data based method is used to determine the validity of the methods proposed. Verification is achieved using real data and shows that the presented diagnosis method is efficient and could have avoided incidents where faults led to loss of aircraft.

Autonomous Tractor Navigation in Orchard - Diagnosis and Supervision for Enhanced Availability

Abstract Autonomous vehicles require a very high degree of availability and safety to become accepted by authorities and the public. Diagnosis and supervision are necessary means to achieve this. This paper investigates ways of using laser-scanner data to do localisation, and as a source of independent supervision, using expectation maximisation of laser-scanner output against uncertain map features. Analysis of system behaviours and their structure shows which redundant information is available to construct a supervisor. Tests on real life orchard data demonstrates the feasibility of the new approach.

In-flight Fault Diagnosis for Autonomous Aircraft via Low-rate Telemetry Channel

Abstract An in-flight diagnosis system that is able to detect faults on an unmanned aircraft using real-time telemetry data could provide operator assistance to warn about imminent risks due to faults. However, limited bandwidth of the air-ground radio-link makes diagnosis difficult. Loss of information about rapid dynamic changes and high parameter uncertainty are the main difficulties. This paper explores time-domain relations in received telemetry signals and uses knowledge of aircraft dynamics and the mechanics behind physical faults to obtain a set of grey-box models for diagnosis. Relating actuator fin deflections with angular rates of the aircraft, low order models are derived and parameters are estimated using system identification techniques. Change detection methods are applied to the prediction error of angular rate estimates and properties of the test statistics are determined. Techniques to overcome correlations in data and cope with non-Gaussian distributions are employed and threshold selection is obtained for the particular distributions of test statistics. Verification using real data showed that the diagnosis method is efficient and could have avoided incidents where faults led to loss of aircraft.

Derivative free Kalman filtering used for orchard navigation

In this paper the use of derivative free filters for mobile robot localisation is investigated. Three different filters are tested on real life data from an autonomous tractor running in an orchard environment. The localisation algorithm fuses odometry and gyro measurements with line features representing the surrounding fruit trees. The line features are created on basis of 2D laser scanner data by a least square algorithm. The Matlab® toolbox Kalmtool is used for easy switching between different filter implementations without the need for changing the base structure of the system.

Kalmtool used for laser scanner aided navigation in orchard

Abstract This paper concerns localisation of an autonomous tractor in an orchard environment, with the purpose of designing a localisation solution to be compared with GPS. The localisation is based on an estimate found by an extended Kalman filter, which fuses measurements from encoders and gyro with row measurements provided by a laser scanner. Kalmtool is used as a toolbox for developing the localisation algorithm. The result shows that the toolbox can be used successfully for dealing with localisation and sensor fusion.

Comparing Mobile Robot Localisation Algorithms Using Kalmtool

Abstract In this paper we present an estimation platform with simulation capabilities to evaluate methods for localisation of a mobile robot using a feature map. The platform is based on the Kalmtool 4 toolbox which is a set of MATLAB tools for state estimation of nonlinear systems. The toolbox contains functions for extended Kalman filtering as well as for the DD1 filter and the DD2 filter. It also contains functions for Unscented Kalman filters as well as three versions of particle filters. The toolbox requires MATLAB version 7, but no additional toolboxes are required.

Performance Comparison of Controllers with Fault-Dependent Control Allocation for UAVs

This paper combines fault-dependent control allocation with three different control schemes to obtain fault tolerance in the longitudinal control of unmanned aerial vehicles. The paper shows that fault-dependent control allocation is able to accommodate actuator faults that would otherwise be critical and it makes a performance assessment for the different control algorithms: an ℒ1