Diagnostics & Prognostics
REDI-PRO employed ANNs, rule-based algorithms, and statistical regression analyses. To support development of diagnostic and prognostic analyses, PNNL obtained engine simulations from the engine manufacturer, comprising 600-800 steady-state simulations for many different faults with varying levels of degradation. The simulated data corresponded to a production engine with a single fault, with specified incremental degradation levels of 2%-10%. Outputs of the simulations included sensor values and other parameters used for diagnostics and prognostics analyses. PNNL also obtained field data from tanks at Yakima Training Center on two M1A1 tanks in the 1998-2000 timeframe; and at Yuma Proving Ground on four M1A1 and M1A2 tanks between February 1999 and November 1999. Overall, this yielded 200,000 sample records (vectors) that were used to train Artificial Neural Networks.