Implementation and Evaluation of Model-based Health Assessment for Spacecraft Autonomy

In order to achieve reliable autonomous operations, spacecraft need precise knowledge of their health state. These requirements can in part be met by model-based approaches to estimating health by continuously verifying nominal behavior and diagnosing off-nominal behavior. This paper describes the implementation and evaluation of the Model-based Off-Nominal State and Identification and Detection (MONSID®) system in the Air Force Research Laboratory's (AFRL's) ground-based environment for test and demonstration of spacecraft autonomy. The test bed is a 3 degree-of-freedom platform with spacecraft attitude control hardware and processors. During this effort we developed diagnostic models, integrated MONSID with the test bed processors using NASA's Core Flight System (cFS) framework, and evaluated system performance via a test campaign. The test campaign had over 40 test bed runs created from variations of realistic mission scenarios including nominal and injected fault cases. MONSID was running onboard a testbed processor and assessing the health of platform hardware. MONSID was able to verify nominal healthy operations as well successfully detect and accurately identify faults. There were three key highlights from the test campaign results. First, MONSID detected actual, unanticipated faults in the test bed hardware. Secondly, MONSID was able to effectively detect double faults, which is beyond the capabilities of most fault management systems. Finally, MONSID was able to detect faults quickly and correctly and with low false positive rates even with noisy data.

1. Introduction

The government and commercial aerospace community is interested in spacecraft systems capable of quickly detecting and responding to perceived opportunities and threats as well as rapidly changing environments. To accomplish this, spacecraft need precise knowledge of their health state. These requirements can in part be met by onboard model-based autonomy-enabling techniques. Model-based health assessment is an approach to estimating health by continuously verifying nominal behavior and diagnosing off-nominal behavior. This approach relies on effective behavioral models of the physical system being monitored, the ability of the diagnosis engine to correctly determine when and where in the system off-nominal behavior is occurring, and the availability of sensor and command data from the physical system. Model-based health assessment can be used to confirm nominal operation, detect off-nominal conditions (including previously unknown/unmodeled behavior), and determine which system components are having problems. In an on-board implementation, this additional insight can be used by on-board planners and controllers to confirm correct system functionality, respond to and plan around identified anomalies, and to estimate remaining system capability. Implemented in ground systems, model-based health assessment can offload operators by analyzing spacecraft telemetry and allowing the operators to focus on other mission tasks.