A Model-Based Fault Diagnostic and Control System for Spacecraft Power Robert A. Morris, Randy B. Pollack, Daniel J. Carreira Florida Institute of Technology Avelino J. Gonzalez, F. D. McKenzie, R. A. Fleeman, and Anita Dhir University of Central Florida Abstract This paper describes a model-based approach to diag- nosing electrical faults in electrical power systems. Until recently, model-based reasoning has only been applied to physical systems with static, persistent states, and with parts whose behavior can be expressed combinatorially, such as digital circuits. Our research is one of a handful of recent efforts to apply model-based reasoning to more com- plex systems, those whose behavior is difficult or impossi- ble to express combinatorially, and whose states change continuously over time. The chosen approach to represen- tation is loosely based on the idea of the equation network proposed in [6]. This requires a more complex component and behavior model than for simpler physical devices. The resulting system is being tested on fault data from the SSM/PMAD power system breadboard being developed at NASA-AIISFC [9]. The model-based reasoning system within which the model of the SSRIIPRIAD is being devel- oped is a version of KATE (I<nowledge-based Autonomous Test Engineer), also developed by NASA (at KSC). 1. INTRODUCTION This paper summarizes results obtained from the use of model-based reasoning in the control of spacecraft power systems. The work originated as a series of studies on the application of a KATE (Knowledge-based Autonomous Test Engineer), developed at NASA-KSC, to problems as- sociated with spacecraft electrical power. KATE employs a model-based approach to diagnosis and control, and has been successfully applied to various Shuttle maintenance tasks. The hypothesis was that the same model-based rea- soning mechanism could be applied to more sophisticated systems such as power systems. This led to the adoption of the SSMIPAIIAD breadboard, developed by NASA-MSFC for the simulation of spacecraft power system hardware, as a testbed for diagnosis and control using KATE. This paper is presented at a time in \vhich real-time test- ing of the KATE system against SSMIPMAD data is under preparation. Hence, we are unable here to submit final re- sults of our work. Nonetheless, we are able to summarize results of our investigations using a simulation of behavior of the SSM/PA4AD system. Although not ob- tained in real time, the data we summarize reveal both the anticipated strengths and weaknesses of the model-based approach, as exemplified in the KATE system. This paper proceeds as follows. In section 2, a brief introduction to model-based troubleshooting is presented. In section 3, the diagnostic problem is defined, and the SSMIPMAD hardware briefly described. In section 4, the KATE system and the KATE-SSM/PMAD model is de- scribed in requisite detail. In section 5, the results of the simulated runs are summarized. In sections 6 and 7, future enhancements, including a brief discussion of a C++-version of KATE currently under development, are discussed, and a summary at the end ties the discussion together. 2. BACKGROUND: MODEL-BASED REASONING FOR DIAGNOSIS There are currently two major approaches to automated diagnosis and control using artifi cia1 intelligence. The first, and oldest approach, uses a knowledge base of prespecified fault models (usually represented as rules or tables) associ- ating observations of the system with component failures. The control mechanism applies a strategy for reasoning about the state of the system against the knowledge base from a set of initial observations. Although adequate for smaller systems whose fault models can be enumerated or defined from a small set of general principles, this approach has noticeable drawbacks. First, it requires that knowledge engineer provide an implicit representation of every