Turbomachinery Design: Checking Artificial Neural Networks Suitability for Design Automation
Year: 2023
Editor: Kevin Otto, Boris Eisenbart, Claudia Eckert, Benoit Eynard, Dieter Krause, Josef Oehmen, Nad
Author: Batini, Niccolo' (1,2); Becattini, Niccolo (1); Cascini, Gaetano (1)
Series: ICED
Institution: 1: Politecnico di Milano;
2: Baker Hughes
Section: Design Methods
Page(s): 3651-3660
DOI number: https://doi.org/10.1017/pds.2023.366
ISBN: -
ISSN: -
Abstract
This paper explores the suitability of Artificial Neural Networks (ANNs) as an enabler of Design Automation in the turbomachinery industry. Specifically, the paper provides 1) a preliminary estimation of the effectiveness of ANNs to define values for design variables of reciprocating compressors (RC) and 2) a comparison of ANNs performance with traditional and more computationally demanding methods like CFD. A tailored ANN trained on a dataset composed by 350+ Baker Hughes’ RC automatically assigns values to 8 geometrical variables belonging to multiple parts of the RC in order to satisfy two target conditions linked to their thermodynamic performance. The results highlight that the ANN-assigned parameters return an optimal solution for RC also when the target values do not belong to the training dataset. Their predictive capacity for RC thermodynamic performance, with respect to CFD, are comparable (i.e. less than 2% in terms of calculated absorbed power) and the approach enables a significant gain in terms of computational time (i.e. 2 minutes vs 10 hours). Future perspectives of this work may involve the integration of this tool in an advanced DA method to lead Design Engineers (DEs) during the whole design process.
Keywords: Artificial intelligence, Computational design methods, Embodiment design, Turbomachinery, Optimisation