TURBOMACHINERY DESIGN: CHECKING ARTIFICIAL NEURAL NETWORKS SUITABILITY FOR DESIGN AUTOMATION

Abstract This paper explores the suitability of Artificial Neural Networks (ANNs) as an enabler Design Automation in turbomachinery industry. Specifically, provides 1) a preliminary estimation effectiveness ANNs to define values for design variables reciprocating compressors (RC) and 2) comparison performance with traditional more computationally demanding methods like CFD. A tailored ANN trained on dataset composed by 350+ Baker Hughes’ RC automatically assigns 8 geometrical belonging multiple parts order satisfy two target conditions linked their thermodynamic performance. The results highlight that ANN-assigned parameters return optimal solution also when do not belong training dataset. Their predictive capacity performance, respect CFD, are comparable (i.e. less than 2% terms calculated absorbed power) approach enables significant gain computational time 2 minutes vs 10 hours). Future perspectives this work may involve integration tool advanced DA method lead Engineers (DEs) during whole process.