Reza Taghvi - Assistant Professor, Department of Mechanical Engineeing, Iran University of science and Technology, Tehran
Iman Naghib - MSc student, Department of Mechanical Engineeing, Iran University of science and Technology, Tehran (corresponding author)

The difficulties, due to a lack of information about stage-by-stage axial-compressor performance, are analyzed. To overcome these issues, a three-layer back-propagation neural-network applied Levenberg-Marquardt and conjugate gradient algorithms are presented and discussed. Three different conjugate gradient algorithms as Fletcher-Reeves, Polak-Ribiere and Powell-Beale are used to predict the compressor's characteristic performance map and the results are compared with the data from the solution of Levenberg-Marquardt algorithm. The experimental data provided by manufacturers are used for the neural-network training. Comparison of results shows that a Fletcher-Reeves algorithm has a better agreement with off-design data from the Levenberg-Marquardt algorithm. The results can be used for the development of an off-design model or overall dynamic simulation of the behavior of a gas-turbine power-plant.

Compressor - Characteristic map - Neural network - Levenberg-Marquardt and Conjugate Gradient algorithms