Kaveh Ghorbanian - Associate Professor Department of Aerospace Engineering Sharif University of Technology Tehran - IRAN
Mohammad Gholamrezaei - Graduate Student Department of Aerospace Engineering Sharif University of Technology Tehran - IRAN

GRNN is employed to reconstruct the compressor performance map. Two different models are adopted to examine the accuracy of the GRNN technique. The results indicate that the GRNN predictions for both models are very sensitive to the width of the probability σ. Further, since the distribution of the training data is multimodal with large variance differences modes, a local optimized value for the probability is suggested providing a more accurate result compared to an overall value for the probability. Furthermore, the sensitivity of the GRNN technique to the number of training data is investigated. The results show that as the number of samples is reduced to about 70% of the available samples, the performance map is predicted with an accuracy of approximately 90%. In general, the results highlight the capability of GRNN in performing design approaches as well as optimization studies of sufficient accuracy with modest amount of data for axial compressors.

Axial Compressor, General Regression Neural Network, Performance Map