UTILITY OF PROJECTION NETWORK FOR DIAGNOSIS OF HIGH PRESSURE AIR COMPRESSOR FAULTS

摘要

The utility of a dynamic neural network, i.e., projection network, was established to diagnose the condition of a 4-stage high pressure air compressor. Network structure and parameter initialization and training methods were developed. Using measurements of the compressor's four stages' discharge temperatures and pressures collected under different baseline conditions, 3rd stage suction and exhaust valve faults, and an unanticipated 3rd stage cylinder pitting as training data, a 99+% of correct classification rate was accomplished with testing data.