A Virtual Principle Investigator for Magnetospheric Multi Scale

M. L. Rilee
S. A. Boardsen
M. K. Bhat
S. A. Curtis

Instrumentation on scientific satellites is typically capable of high
data rates, but only a fraction of this high resolution data is
telemetered to ground.  The scientific value of telemetered high
resolution data can be extremely variable, depending on the chance
that an event of scientific interest is captured during "burst mode"
operations.  For example, each of the four or five spacecraft of the
proposed Magnetospheric Multi Scale (MMS) mission is projected to
obtain 2 Gb per day and will be able to store 14 days of science data.
However, the instrumentation on board is to be capable of burst mode
operations that produce over four times as much data as normal mode
operations.  MMS focusses on the ill understood microphysics of
magnetospheric dynamics and structure, therefore high resolution data
acquired from boundary regions and sites of magnetospheric energy
transfer are a mission priority.  Just under half of MMS's mission
will be spent in orbits with periods exceeding nine days.  A
considerable fraction of this time will be spent in the vicinity of
apogee where key science operations occur.  MMS therefore, will have
ample opportunity to fill its data store with high resolution data and
can do so in only 3.5 days.  For the remaining six days of these
orbits, data will either not be recorded or will necessarily write
over more important data obtained at apogee.  Finally, due to limited
telemetry opportunities and lengthy orbits, the spacecraft must
perform these operations essentially autonomously.  These are
unavoidable trade offs and constraints in the management of MMS
science.


In this paper we report on the initial progress of our implementation
of a Science Agent or Virtual Principle Investigator for MMS.  The SA
is a prototype artificial intelligence that will reside in a high
performance computing environment, like the environment provided by
the proposed MMS High Performance Computing System being studied by
NASA's Remote Exploration and Experimentation (REE) project.  The SA
will autonomously manage on board science processing, identify time
intervals for the storage of selected high resolution data based on
the science objectives, and triage these periods for memory
management.  For science management, the SA will perform self
consistency checks on science products, as well as trend and feature
detection.  Our previous work has centered on the development of
software for high performance flight computers being researched by
REE.  In particular, the Plasma Moment Analysis and the Radio
Astronomical Imaging applications have been our canonical applications
for on board particle and wave data analyses.  Applications such as
these will operate on board under the control of an SA, which we
believe will enable science results to be obtained along entire MMS
orbits.  Fluid moments, snapshots of high resolution data, and feature
identification will dramatically increase MMS science return along
entire orbits and soften the constraints and trade offs mentioned
above.  Results of testing the SA on archived data sets will be
presented.