Correlation of transient and steady-state compressor performance using neural networks

Abstract

Neural network technology is considered that may significantly reduce the time required to obtain steady-state compressor maps. This reduction would be accomplished using neural networks trained to learn correlations between transient and steady-state compressor performance. Neural networks that generalize with guaranteed bounds on computational effort, smoothness, and stability are particularly appropriate for this application. The learned correlation could make important contributions to the solution of stall recovery and surge anticipation problems.<>