Newsletter  2022.11  Index

Theme : "Mechanical Engineering Congress, 2022 Japan (MECJ-22)” Preface Hideo MORI, Tetsuya KANAGAWA Wall Modelling for Engineering Turbulence CFD Kazuhiko SUGA (Osaka Metropolitan University) Prediction of the flow of granules Toshitsugu TANAKA (Osaka University) Design Optimization of Turbomachinery by Artificial Neural Networks as a Meta-model Daisaku SAKAGUCHI (Nagasaki University) Deep learning for viscoelastic fluids and turbulent diffusion Takahiro TSUKAHARA (Tokyo University of Science) Advanced fluid measurement using mode decomposition Taku NONOMURA (Tohoku University)

 

Design Optimization of Turbomachinery by Artificial Neural Networks as a Meta-model

Daisaku SAKAGUCHI Nagasaki University

Abstract

Artificial Neural Networks (ANN) has a potential to be used for making decisions instead of human designers. Well-trained ANN is helpful tools to eliminate the work for making decisions and can be effective tools for putting order the complicated parametric study. The genetic algorithms are one of global design search systems. It is important to provide enough generations for the design search, however the evaluation of each individual costs a lot. ANN is one of the candidates of an alternative model (meta-model) which drastically reduce the number of actual CFD calculations, and a highly reliable database can be constructed by CFD validations. Each time the database is reconstructed, the prediction accuracy of the meta-model increases, enabling efficient design search for the optimal one. This paper introduces the multi-objective optimization system using the ANN assisted genetic algorithms as shown in Fig.1, and optimization of turbomachinery design is performed with reducing the simulation costs as shown in Fig.2.

Key words

Design Optimization, Artificial Neural Networks, Meta-model, Turbomachinery

Figures

Fig.1  General layout of the meta-model assisted optimization system

Fig.2  Example of optimization target by multi-objective optimization