Compressor cascade correlations modelling at design points using artificial neural networks

Patrik Kovář, Jiří Fürst


In recent years, the flow analysis by means of computational fluid dynamics (CFD) has become a useful design and optimization tool. Unfortunately, despite advances in the computational power, numerical simulations are still very time consuming. Thus, empirical correlation models keep their importance as a tool for early stages of axial compressor design and for prediction of basic performance parameters. These correlations were developed based on experimental data obtained from 2D measurements performed on cases of classical airfoils such as the NACA 65-series or C.4 profiles. There is insufficient amount of experimental data for other families of airfoils, but CFD simulations can be used instead and their results correlated using artificial neural networks (ANN), as described in this work. Unlike the classical deep learning approach using perceptrons, this work presents neural networks employing higher order neural units.


compressor cascade; empirical correlations; machine learning; higher order neural networks

Full Text:

DOI: 10.24132/acm.2023.828