Abstract:
Accurate characterization of fleet and vehicle usage spectrums would allow component retirement times to be based on actual aircraft usage rather than on an assumed worst case usage spectrum used in a traditional time-based maintenance approach. A key enabling technology for such a Usage or Condition Based Maintenance (UBM/CBM) program is Regime Recognition (RR). In general, production RR algorithms with the required accuracy have not been fielded, in part because of the challenges associated with Verification and Validation (V&V). The multi-class classifier analysis methods and associated visualizations presented herein address the need to maximize the utility of available data and rigorously perform V&V of RR algorithms and software. The resultant methods clearly show the performance range that can be expected when an RR code is deployed to the fleet, and provide a single accuracy metric that highlights error characteristics that are most important to a UBM/CBM program.