Space power system automation requires comprehensive fault detection, isolation and recovery. Extending current approaches to on-board power management to isolate and resolve (as well as detect) failures would greatly increase the size and complexity of the flight software. Instead, we identified a technique called model-based constraint suspension which can isolate a large percentage of failures, including unforeseen failures, with minimal increases in code size. Adapting this digital technique to an electrical power system required us to address several key issues including model construction and fidelity. This paper contrasts several approaches to model-based diagnostics to our diagnostic system, Marple, which has been demonstrated on a power testbed.