| | Fault detection, isolation and recovery
Two studies having been performed on Fault Detection, Isolation and Recovery (FDIR). A third study about generic FDIR models is in progress.
SMART-FDIR
SMART-FDIR was a project coordinated by Alenia Spazio (ALS), with Politecnico di Milano (POLIMI) acting as subcontractor. It started in June 2002 and ended in June 2003. The main goal of the study was to investigate the added value of Artificial Intelligence (AI) technology in the implementation of a satellite on-board FDIR software prototype-demonstrator with real-time performance, robustness architecture, auto-learning and decision making capabilities.
After the analyses of the state of the art in the use of AI for FDIR developments, the SMART-FDIR project developed the prototype demonstrator in three iterations, implementing successively the fault detection, the fault identification and the fault isolation and recovery.
Each iteration aimed at improving the current situation. By the end of each iteration, it was possible to draw conclusions as to the appropriateness of each technique and their advantages and drawbacks. The added value of the demonstrator’s AI capabilities has been evaluated with respect to the GOCE (Gravity Field and Steady-State Ocean Circulation Explorer) Satellite System Software FDIR, chosen as the reference scenario.
The selected AI techniques are the following:
- Detection:
As the system under analysis is complex and sensor outputs reflect a deep interaction among on-board subsystems, a global analytic model is not available and a time series analysis is the fittest approach to build – with an off-line procedure – the nominal dynamic input/output model of either the whole system or a part of it; this and a Fuzzy Inductive Reasoning (FIR) approach have been selected
- Identification:
A model-based framework has been selected; the system behavioural model has been decided upon by means of Possibilistic Logic theory. The behavioural model represents the system knowledge about causal dependencies between inputs and outputs; and they are represented here by a logical formulation.
- Recovery:
Recovery actions involve a logical and structural model reconfiguration, to represent newly activated behaviours. For the first task, a logic model of the whole system becomes useful: the logic programming approach on a finite state net can make use of the same models implemented for the identification module: they can be collected in an overall logic model, split into the main system physical components: each of them can be represented by related state variables, related inter-component causal requisites, related resource. In their turn, state variables can be related to their permitted attributes (numerical, logic, linguistic) and permissible transitions. Permitted transitions can be classified depending on the current failed system status: moreover, a cost function can be attached to them in order to determine the most convenient recovery path according to a multi-criteria goal vector computed by a multi-attribute decision making technique.
The study concluded that, as a general statement, artificial intelligence technology has to be considered for the development of satellite on-board FDIR software with real-time performances, robustness architecture, auto-learning and decision making capabilities.
From the study, the following lessons were learned:
- the need, in the space system engineering process, to identify the proper involvement of AI software developers during the System FDIR analysis before starting the Software FDIR implementation,
- the need, in the field of software development of AI components, for generating those AI software development guidelines that are currently missing in the space domain should stimulate further ESA studies in the use of these techniques
The study produced an abstract, a final report, the architecture and the user manual of the software tool and the result of the validation on GOCE.
Advanced Fault Detection, Isolation and Recovery (AFDIR)
Our Data Handling colleagues have run a study of Advanced FDIR techniques (AFDIR) with Astrium (France) and SSF (Finland).
Complex, autonomous spacecraft need powerful on-board FDIR. SSF has developed a set of reusable FDIR software components called AFDIR. In addition to conventional failure detection methods (examples: limit monitoring, trending, transient filtering), AFDIR includes Kalman filtering, weighted sum-squared residual test, generalized likelihood test, random sample consensus method, and various spacecraft simulations for computing ‘expected’ values.
AFDIR has two ‘integrative’ diagnosis methods: probabilistic reasoning using Bayesian networks and model-based diagnosis using causal networks. The AFDIR configuration space data structure compactly represents many configurations and allows selection of all spacecraft configurations for graceful failure-recovery. AFDIR is divided into an algorithm layer of purely callable subprograms, and a structure layer with a more convenient declarative interface. The chief data structure is the signal-processing network.
The study produced a final report, a final presentation, a paper presented at DASIA 2001, and for the ESTEC workshop on autonomy, a paper about Bayesian networks and a paper about modern AI approaches.
Additional information (links in right-hand navigation) supplied by courtesy of our Data Handling colleagues.
Last update: 26 September 2013 | |
