>
Figure 1: Development and deployment plan.
Figure 1: PS 61 Kb
Next: A Tcl/Tk-Based, Intelligent Graphical Editor for Preparing HST Programs
Previous: Real-Time Observing at Hat Creek
Table of Contents --- Search ---
PS reprint
L. Wong, F. Kronberg, A. Hopkins, F. Machi, P. Eastham
Center for EUV Astrophysics, University of California,
2150 Kittredge Street, Berkeley, CA
94720-5030
Because of budgetary strain at NASA since 1993, the Center for EUV Astrophysics (CEA) faced the possibility of turning off the Extreme Ultraviolet Explorer ( EUVE) satellite after its primary mission. Driven by the need to reduce operations costs, we have developed and implemented a rule-based, expert, satellite-monitoring system using artificial-intelligence software to autonomously monitor the EUVE science payload for two thirds of each day.
We divide the expert system development and deployment into three phases.
Phase I: Goal---Deliver a partial expert system to monitor five critical payload subsystems; Key Function---Detect and log limited payload anomalies; Operations--- EUVE Science Operations Center (ESOC) staffed three shifts, expert system runs in parallel with old monitoring system; Delivery---July 1994.
Phase II: Goal---Deliver an expert system capable of detecting unsafe payload conditions and notifying an anomaly coordinator for the ESOC (ACE); Key Function---Detect critical payload anomalies and notify ACE via paging; Operations---ESOC staffed one shift, expert system monitors autonomously 5pm--8am; Delivery---February 1995.
Phase III: Goal---Deliver a complete expert system capable of autonomous monitoring of all important payload sensors; Key Function---Detect payload anomalies and notify ACE via paging; Operations---ESOC unstaffed, fully autonomous monitoring around the clock; Delivery---January 1997.
>
Figure 1: Development and deployment plan.
Figure 1: PS 61 Kb
Figure 1 shows how our expert-system development progresses as ESOC operation staffing decreases. The Phase I diagram shows the payload monitored by the old monitoring system, which included the Soctools software (an X-based display of health and safety information), and a controller. In Phase I we implemented a partial expert system to monitor with limited functionality five critical payload subsystems. This implementation also provided a rudimentary, graphical, human-computer interface (HCI) and rules to monitor the payload and log anomalous conditions. The goal of Phase I was to deliver a stand-alone, partial-expert, payload monitoring system that could run in parallel with the old monitoring system for evaluation. A by-product of the expert-system development and evaluation process was a test-bed where future software packages could be incorporated into the ESOC's operational software.
The Phase II diagram shows the old system (Soctools/human) and new expert system (consisting of Eworks and Epage software) in operation side by side from 8am to 5pm. From 5pm to 8am the new system takes over the monitoring task entirely with the ESOC unstaffed. During the unstaffed hours, Eworks monitors the health and safety of the payload and Epage notifies an ACE via electronic page in case of a payload anomaly.
The Phase III diagram shows our goal of a fully autonomous monitoring system in which the need for monitoring by humans is eliminated. The Phase III version of the expert system will monitor the instrument not only in its nominal operating configuration but in a variety of extraordinary configurations for special target observations and engineering tests.
In our knowledge-base development cycle we start by developing a concept of how the proposed function should work. We then transform this concept into a flow chart that is reviewed by EUVE payload controllers, scientists and programmers who examine the logic and feasibility of its implementation. The flow chart may cycle through the review process a number of times before it is approved. After approval, our programmers begin coding. The completed code is then integrated with the engineering software. At this point we test the expert system's functionality. If it fails a problem report is filed and assigned to an expert for analysis and resolution. The resolution could be a change in the logic of the flow chart or a re-coding of the failed rules. This cycle is repeated until the expert system passes the validation process. It is then incorporated into a full, ground-system software release.
Figure 2: Knowledge base development life cycle.
Figure 2: PS 56 Kb
The expert system is a collection of programs built upon Talarian Corporation's RTworks' realtime, data-monitoring tools and runs on a network of UNIX workstations and servers in a distributed environment. The programs include the following modules: a data acquisition module, an inference-engine module, a telemetry-reception watchdog module, an HCI module, an RTworks data-server module, and a paging module.
The data acquisition module reads telemetry as it is received at CEA's operation center and extracts the engineering health information from it and passes that on to the data server, which makes it available to the inference engine, HCI, and telemetry-reception watchdog modules. The inference-engine module monitors the health data using rules written at CEA, and pages an ACE if an anomaly is detected. The telemetry-reception module monitors the presence of telemetry. Presently, this module pages an ACE if telemetry is absent longer than 16 hours. The HCI module displays graphical representations of the state of the payload. The paging module services paging requests from the telemetry-reception and inference engine modules.
Figure 3: Expert System Architecture.
Figure 3: PS 56 Kb
We delivered the prototype expert system without paging function within five months of conception. The subsequent delivery of the operational expert system took another seven months.
Our development resources included 2 programmers, 1.5 controllers, and 1 payload scientist. Since deployment of the expert system in the ESOC, CEA has saved a total of four controllers and ten controller aides in operational staffing. Our goal is to establish an expert system that autonomously monitors the payload in normal operations without human intervention.
Following a rapid development philosophy and expending minimal development resources, we used an off-the-shelf product called RTworks from the Talarian Corporation to develop an expert payload-monitoring system that achieved a significant operational cost saving within one year. We have demonstrated that implementing autonomous satellite monitoring in mission operations is feasible. We believe the knowledge and experience gained in this development and deployment effort can be applied to other NASA space projects.
We acknowledge the hard work of the software development team and the ESOC personnel involved in the design, development, and implementation of the expert system. We thank F. Girouard, D. Biroscak, P. Ringrose, M. Eckert, C. Smith, and the ESOC students in making the three-shift to one-shift transition possible. We also thank D. Iverson at NASA Ames Research Center for participating in the development effort. This research has been supported by NASA contract NAS5-29298 and NASA Ames grant NCC2-838. CEA is a division of UC Berkeley's Space Science Laboratory.
Wong, L., Hopkins, A., & Machi, F. 1995, Eworks Development Status Report Phase II, EUVE Memo#: LSW/EUVE/0042/95
Lewis, M., et al. 1995, NASA Conf. Pub. 3296, ed. C. F. Hostetter (GSFC: Greenbelt), 229