VeriFIM study
Verification of Failure Impact
by Model-checking
| Funded by | European Space Agency (ESA) |
| Grant n. | TEC-SWE/09-259/YY |
| Duration: | from June 2010 to December 2011 |
| Partners: | Thales Alenia Space,
Torino, Italy |
| Univ.
of Piemonte Orientale, Alessandria, Italy |
|
| ESA-ESTEC,
Noordwijk, Netherlands |
Goal
The project's goal is to build an FDIR (Failure Detection, Identification and Recovery) engine for an autonomous spacecraft, using probabilistic techniques based on Bayesian Belief Networks models (statement of work).
Participants
| Thales Alenia Space | Univ. of Piemonte Orientale | ESA-ESTEC |
| Andrea
Guiotto (unit coordinator) |
Luigi
Portinale (unit coordinator) |
Yuri
Yushtein |
| Stefano
Di Nolfo |
Andrea
Bobbio |
(study supervisor) |
| Daniele
Codetta-Raiteri |
||
| Roberta
Terruggia |
Publications
D. Codetta-Raiteri, L. Portinale
"Dynamic Bayesian Networks for Fault Detection, Identification, and Recovery in Autonomous Spacecraft"
IEEE Transactions on Systems, Man, and Cybernetics: Systems, vol. 45(1), pages 13-24, IEEE, January 2015
A. Bobbio, D. Codetta-Raiteri, L. Portinale, A. Guiotto, Y. Yushtein
"A Unified Modelling and Operational Framework for Fault Detection, Identification, and Recovery in Autonomous Spacecrafts"
In "Theory and Application of Multi-Formalism Modeling", pages 239-258, IGI-Global, October 2013.
D. Codetta-Raiteri, L. Portinale, S. Di Nolfo, A. Guiotto, Y. Yushtein
"A unified modelling and operational framework for Fault Detection, Identification and Recovery in autonomous spacecrafts"
Proceedings of the Workshop on Research and Use of Multiformalism Modeling Methods (WRUMMM), pages 24-31, London, UK, September 2012.
D. Codetta-Raiteri, L. Portinale, A. Guiotto, Y. Yushtein
"Evaluation of anomaly and failure scenarios involving an exploration rover: a Bayesian network approach"
Proceedings of the International Symposium on Artificial Intelligence, Robotics and Automation in Space (iSAIRAS), ESA, Turin, Italy, September 2012.
A. Guiotto, L. Portinale, D. Codetta-Raiteri, Y. Yushtein
"ARPHA: an innovative on-board FDIR reasoning engine for autonomous systems"
Proceedings of Data Systems in Aerospace (DASIA), ESA, Dubrovnik, Croatia, May 2012.
D. Codetta-Raiteri, L. Portinale, S. Di Nolfo, A. Guiotto
"ARPHA:
a software prototype for fault detection, identification
and recovery in autonomous spacecrafts"
Acta Futura, vol. 5(15),
pages 99-110, ESA, January 2012.
L. Portinale, D. Codetta-Raiteri
"Using
Dynamic Decision Networks and Extended Fault Trees for
Autonomous FDIR"
Proceedings
of the International Conference on Tools with Artificial
Intelligence (ICTAI),
pages 480-484, IEEE Computer Society, Boca Raton, USA,
November 2011.
L. Portinale, D. Codetta-Raiteri
"ARPHA:
an FDIR architecture for autonomous spacecrafts based on
Dynamic Probabilitstic Graphical Models"
Proceedings
of AI
in Space, ESA, Barcelona, Spain, July 2011.
D. Codetta-Raiteri, L. Portinale
"ARPHA:
an FDIR architecture for Autonomous Spacecrafts based on
Dynamic Probabilistic Graphical Models"
Technical Report TR-INF-2010-12-04-UNIPMN,
Dip. di Informatica, Univ. del Piemonte Orientale,
December 2010.
Study framework
Study Organization
Verification of Failure Impact by Model Checking (VeriFIM)
is a European Space Agency research study - contract
No. 4200023090 - coordinated by Thales Alenia Space Italia
(TAS-I) with Università del Piemonte (UNIPMN) as
subcontractor and Thales Alenia Space France (TAS-F) as
consultant.
TAS-I as the Prime Contractor has taken care of the
overall technical coordination and harmonization of the
project. Moreover, TAS-I has performed the selection of
the case study, the specification of system requirements,
the SW specification elicitation, the SW integration in
the on-board infrastructure, the validation, the
evaluation and characterization of the approach.
UNIPMN has performed technology survey of state of the art
in the model-based and knowledge based approaches. It has
built the on-board model of ARPHA (DBN). It has designed
and developed ARPHA algorithms and functions and has
implemented and delivered the ARPHA prototype. It has
supported to performance evaluation.
TAS-F has supported the ARPHA system requirements
elicitation and has provided a diagnosability approach to
model used by ARPHA.
Study motivations
Currently employed FDIR operation is based on the
design-time analysis of the faults and failure scenarios
(e.g. FMEA, FTA) and run-time observation of the system
operational status (health monitoring). It has the main
objectives to timely detect the faults and to initiate the
corresponding predefined recovery actions. If no
corresponding action could be found, FDIR proceeds by
executing the recovery actions to put the spacecraft into
a known safe configuration and transfers control to the
Ground operations for troubleshooting and planning the
recovery actions.
This approach is not always adequate for an autonomous
system for the following reasons:
- Partial observability of system and environment does not
allow for a certain identification of system status
- Tradition FDIR cannot provide and utilize prognosis for
the imminent failures
- Automated FDIR procedures cannot leverage specific
course of recovery based on the evaluation of causal
knowledge of system and environment status
- It is impossible to estimate the impact of the occurred
faults and failures on the operational capabilities of the
system
- Reaction time does not always allow to wait a Ground
recovery
Study Objectives
The global objective of this study is to demonstrate that
integration of innovative technologies (i.e. model-based
autonomy, run-time Dependability and Safety analysis,
causal modelling, probabilistic calculus, Knowledge-Based
Systems) in a unified modelling and autonomous reasoning
framework may increase the achievable level of autonomy.
The main focus is on the autonomous anomaly resolution and
prognostic pro-active FDIR capabilities.
The global objective comprises the following
sub-objectives:
1. Evaluation and justification of an integrated and
unified use of causal probabilistic techniques and
Knowledge-Based approaches, suited for on-board automated
analysis, to increase the space systems level of autonomy
in terms of anomaly resilience and autonomous
recoverability;
2. Definition of an integrated modelling framework for
specification of the models suited for on-board autonomous
reasoning to infer system Health, Dependability and Safety
status and prognosis, and (preventive) anomaly resolution
approaches;
3. Development of an on-board software prototype, the
Anomaly Resolution and Prognostic Health management for
Autonomy (ARPHA), implementing the required autonomous
reasoning and inference techniques, based on the use of
probabilistic calculus approaches;
4. Demonstration of the approach on case studies involving
autonomous on-board systems and evaluation of the
experimental results in terms of applicability,
scalability, and performance;
5. Evaluation of adequacy of the approach and developed
technology for use in the context of critical on-board
space systems
Conclusions
The developed approach provides a unified modeling and
autonomous framework that integrates an high level
modeling formalism (Dynamic Fault Tree - DFT), a low level
modeling formalism (Dynamic Bayesian Network - DBN) and an
inference oriented formalism (Junction Tree - JT). The
on-board analysis of the JT conditioned by the sensors
data and the recovery actions, allows evaluating the
system current and future state, and the recovery policies
if necessary, in automatic way, without the assistance of
the ground control. This approach increases the achievable
level of autonomy. The developed prototype ARPHA
represents an on-board software FDIR component suited for
use in the existing spacecraft system architectures. It
can perform on-board diagnosis, prognosis and recovery
inference. ARPHA is able to verify the failure
impact on the future state of the system.
Environmental aspect of space mission can be modeled in
the DBN used by ARPHA to perform inference. It is possible
to take in account the failure causes, by inserting them
in the utility function used to select recovery. ARPHA can
evaluate the failure impact on the currently executing
plan as well.
The developed ARPHA prototype has been evaluated on the
space embedded target (running under RTEMS on the LEON3
processor). The obtained performance data shows ARPHA
usability in the context of the current space applications
and available on-board computers.