REPUBLIC OF SERBIA

MINISTRY OF DEFENCE

www.mod.gov.rs

MINISTRY OF DEFENCE

Material Resources Sector

Defensive Technologies Department

Military Technical Institute

www.vti.mod.gov.rs

A Study of the TURBOMACHINE air system with ERROR BACK PROPAGATION NEURAL NETWORK

DUŠAN RANĐELOVIĆ

Faculty of Mechanical Engineering, Belgrade, d.randjelovic@protonmail.com

GIOVANNI TORELLA

University of Ferrara, Ferrara, giovanni.torella@unife.it

Giannigabriele Mainiero

AVIO S.p.a., Naples, giovanni.mainiero@aviogroup.com

ALEKSANDAR BENGIN

Faculty of Mechanical Engineering, Belgrade, abengin@mas.bg.ac.rs

Abstract: A turbo machine’s air system is one of its vital parts, so troubleshooting and failure prediction is very important. Neural networks can be used for failure simulation and various behaviors of turbo machine air systems. The goal of this work is to show the efficiency of this method through a study followed by practical application using a neural network. An equivalent scheme of the air system section of the turboprop engine “Pratt&Whitney Canada PW127” was analyzed and a neural network applied on the results using a fault simulation. The first step was to create various multiple alterations of the air system parameters using a suitable simulation code. It was shown that a neural network was able to successfully reconstruct correlations between parameters and performances of the engine, by being a valid interpretative and diagnostic tool that could optimize the management and eventually identify faults or malfunctions.

Keywords: neural networks, turbo machine, air system.