Reasoning under an uncertain thermal state

Paper number

IAC-06-C2.7.05

Author

Dr. Daniela Girimonte, Advanced Concepts Team, The Netherlands

Coauthor

Dr. Dario Izzo, European Space Agency (ESA)/ESTEC, The Netherlands

Year

2006

Abstract

In order to preserve the thermal health and safety status of the spacecraft and its instruments, thermal control systems are required. The knowledge of the temperature of the system components throughout the whole spacecraft is crucial for the employment of these control strategies. In this paper we consider a different approach to the thermal characterization of the spacecraft that takes into account system and sensor uncertainties. Our approach is based on a probabilistic description of the status of the spacecraft with the help of a graphical model. In the last decade, graphical models have become one of the most popular tools to structure uncertain knowledge about complex domains in order to make reasoning in such domains feasible. Their most prominent representatives are Bayesian networks, which are based on directed graphs and conditional distributions, and Markov networks, which are based on undirected graphs and marginal distributions or factor potentials. In this context graphical models are employed for the
thermal modeling and diagnosis of a spacecraft by exploiting the probabilistic information about the thermal conductance of the system components. The method consists of: the construction of a join tree representation of a Markov network derived from a lumped
mass thermal model; the Iterative Proportional Fitting procedure for the initialization of the probability distributions; the join tree propagation for the incorporation of uncertain thermal sensor measurements. The main advantage of this method is the possibility
of determining a complete thermal mapping of the system from few, strategically placed, sensor readings. Furthermore, the system is able to detect variations in the thermal conductance of a component as result of variations of its physical properties.   Finally, the graphical model can be employed as a virtual sensor
able to identify anomalous behaviors of a possible faulty physical sensor. Preliminary results presented in this paper on a simplified spacecraft thermal network show that the proposed method performs
effectively.

Abstract document

IAC-06-C2.7.05.pdf

Manuscript document

IAC-06-C2.7.05.pdf (🔒 authorized access only).

To get the manuscript, please contact IAF Secretariat.