Popis: |
This paper describes application of a hybrid methodology for fault diagnostics to the electrical power subsystem (EPS) on the FalconSAT-6 spacecraft. The approach integrates a diagnostic reasoning engine (the NASA Ames-developed Hybrid Diagnostics Engine - HyDE) with recursive Bayesian estimation, enabling exploitation of techniques tailored to specific diagnostic challenges. The Bayesian estimators (Unscented Kalman Filters) play the role of smart monitor, providing candidate diagnoses for targeted subsystems, while the constraintbased general inference engine maintains a consistent view of spacecraft status across a hierarchical collection of components. This hybrid approach is applied to EPS fault diagnostics on the FalconSAT-6 spacecraft, currently under development at the U.S. Air Force Academy in Colorado Springs, CO. Simulations demonstrate the effectiveness of the solution which will be exercised in satellite hardware-in-the-loop experiments in Fall 2012. I. Introduction NCREASED automation, fueled by dramatic increases in processing capability, promises to improve mission effectiveness, responsiveness, and cost efficiency in modern spacecraft operations. In the presence of both natural and man-made threats, timelines for reaction are decreasing, driving a need for systems that only require human-inthe-loop interaction for higher level decision making. Achieving this goal, in part, requires on-board analysis and reasoning that transforms raw data into human-readable information. Real-time diagnostics are an essential ingredient, providing detection and isolation of component degradation, failures, or other system anomalies – and permitting higher levels of autonomy (such as a planner/scheduler) to formulate an appropriate course of action. Model-based diagnostics (MBD) is a powerful tool for fault detection, isolation, and recovery that has matured through over two decades of research within the Artificial Intelligence community. MBD typically leverages a reasoning engine that exercises a component-based model that incorporates a description of critical system behaviors. A fundamental notion is consistency, that monitored system outputs must not conflict with the predicted output of the model. When a conflict is detected, the reasoning system then sets out to explain the anomaly by searching for a new system state that re-establishes consistency 1 . By defining the system state as a collection of modes, and defining the modes themselves through constraints over a set of domain variables, a tremendous richness can be achieved in describing system behaviors. NASA Ames Research Center has been at the forefront in development and deployment of MBD tools. In 1999, Ames’ LISP-based Livingstone was part of a suite of autonomous software experiments to fly on NASA’s Deep Space I spacecraft. Based on lessons learned from that experience, the diagnostic reasoning software was upgraded |