ESA GNC Conference Papers Repository

Fault detection and isolation (FDI) for reusable launch vehicles (RLV) during the ascent or re-entry phases is a very challenging process due to the wide dynamical changes and diverse environmental conditions (from low altitude atmosphere to exoatmospheric space). Therefore, a critical aspect of RLV FDI is the robustness of the design. In addressing this robustness problem, model-based approaches such as H? optimization methods [10, 4, 7] have been developed explicitly during the last 20 years. In any case, prior to the design of the robust FDI filters it is required to assess the requirements of fault coverage and fault criticality. These requirements are intended to select the precise components and type of faults for which the closed-loop will be more critically affected. Thus, an indispensable first step prior to any fault detection & isolation or fault tolerant control (FTC) design process is to perform a fault analysis (FA) which must try to answer two questions mainly:<br> 1. What is the impact on the closed loop of a specific fault?<br> 2. Which fault is more critical?<br> The answer to these questions should be given from a quantitative perspective, which will also help for example to obtain information on fault critical magnitude levels. There are two main FA approaches: analytical and computational. Analytical FAs rely mostly on linear approaches and allow direct quantification of the fault effects on system’s performance/stability. Linearization is fundamental (and non-trivial for many space systems such as re-entry vehicles). Analytical approaches suffer from the fact that the results stem mostly from linear, time-invariant (LTI) systems. This becomes especially critical as faults introduce nonlinear behaviour and are mostly time-varying processes (characteristics not amenable for study with LTI techniques). Nevertheless, they serve as an excellent preliminary FA tool due to the wide and consolidated number of LTI analysis techniques. Current acceptable industrial-level FAs are computational-based approaches relying in Monte Carlo techniques. They are computationally very demanding due to the too many possible faults, at too many different occurring times, affecting too many components and subsystems –although these issues affect equally to analytical approaches. On the other hand, computational FAs allow assessing the nonlinear and time-varying effects of faults and can quantify the fault effects on the “true” closed loop –after a careful selection of evaluation criteria and metrics. It is noted that despite all the above important questions, there is not much on the available literature on quantification of fault effects [2, 3, 11]. Thus, this article strives to fill this gap by presenting the main concepts behind a proposed fault analysis methodology that combines computational measurements and quantitative assessment criteria. It does so by detailing its application to a closed loop fault analysis of EADS-Astrium’s Hopper RLV performed within the framework of a European Space Agency Health Management System (HMS) project.