Samy I., Gu D.-W. Fault Detection and Flight Data Measurement. Demonstrated on Unmanned Air Vehicles Using Neural Networks

Springer, 2012. — 193 p.This book is essentially the first author’s Ph.D. thesis, which was successfully defended at the University of Leicester in 2009. It explores the feasibility of two technologies in reducing cost and weight of air vehicles. The first is a fault detection and isolation scheme, which uses neural networks to diagnose faults in sensors. The second is a flush air data sensing (FADS) system, which uses pressure orifices on a wing’s leading edge to estimate air data such as; airspeed angle of attack and sideslip.
Fault detection and isolation (FDI) can be traced back to the time before the 1940’s when industry did not rely so much on highly mechanised processes and it was sufficient to only fix something when it was truly broken. With the development of more complex systems and the introduction of just in time manufacturing, practitioners moved to preventative measures and maintenance began to be performed on a scheduled interval basis. This approach, however suffered great disadvantages, as normal operations were halted at pre-defined fixed intervals regardless of a fault being present or not. This meant that profitable production time was unnecessarily lost. Furthermore, faults occurring between the fixed intervals were undetected and could prove catastrophic, resulting in unscheduled maintenance downtime.
For these reasons, the concept of condition monitoring was introduced, in which the health of a system was monitored on a continuous basis in order to improve reliability and availability. In general, these approaches monitor health in real time with the aim of reducing fault detection time, the number of false alarms, the number of undetected faults and unscheduled maintenance downtime. In this way if some part of the system, e.g. a sensor or actuator fails to perform as expected, this can be detected and acted upon so that the system is still safe to operate within agreed industry standards. Because of the competitive market, there are many terms used for condition monitoring systems, e.g. Integrated Vehicle Health Management (IVHM), Integrated Systems Health Mangement (ISHM), Engine Health Management (EHM) and Health Usage Management System (HUMS).
One way to understanding FDI schemes, is to consider them as forming a building block of a condition monitoring system with other building blocks including: sensors, actuators, communication links, ground base equipment etc. As such we can assume that FDI is the means with which fault diagnosis is performed. With this in mind, let us now consider the different approaches possible to detecting and isolating faults.
Traditionally, FDI methods relied on hardware (also referred to as physical) redundancy, where fault detection is based on a voting scheme comparing the same measurement type from redundant hardware (e.g. sensors). Another traditional approach is based on setting pre-defined (generally defined by the Original Equipment Manufacturer) thresholds on a chosen parameter (e.g. temperature). Hardware redundancy and threshold-based techniques are simple to use, which could be the reason for their popularity as the less mathematically oriented a method is, the more appealing to industry it becomes. In academia, researchers have found that both methods suffer great disadvantages such as setting thresholds at high levels to avoid false alarms caused by measurement noise. Furthermore in closed loop control systems, the control laws tend to dampen the effects of faults and so simply checking the size of the output signals does not give a reliable insight into overall system health. In fact, shortly before the catastrophic disaster of the Challenger Space Shuttle in 1986, the FDI scheme of the main engine was based on thresholds and it was noted that advanced detection systems could have prevented the crash. Other FDI techniques, include; frequency analysis and expert systems (e.g. case based reasoning, and if-then rule logic). The different methods are briefly outlined in this book.Fault Detection and Isolation (FDI)
Introduction to FADS Systems
Neural Networks
SFDA-Single Sensor Faults
SFDIA-Multiple Sensor Faults
FADS System Applied to a MAV
Conclusions and Future Work