Performance evaluation of air ejectors using artificial neural network approach

• Published in: Sadhana (Bangalore) Vol. 48; no. 2
• Main Authors: Gupta, Pradeep, Rao, Srisha M V, Kumar, Pramod
• Format: Journal Article
• Language: English
• Published: New Delhi Springer India 28.03.2023
• Subjects: Engineering; ejector performance; energy conservation; artificial neural network; Supersonic ejectors; critical flow regime
• ISSN: 0973-7677; 0973-7677
• DOI: 10.1007/s12046-023-02087-2

Ejectors are implemented in energy conservation applications such as power and refrigeration systems and fuel cell stacks due to their passive nature and geometrical simplicity. Ejectors work on the principle of thermal compression that utilizes low-grade energy to produce a compression effect. A low-enthalpy secondary fluid gets compressed through gasdynamic interactions with a co-flowing high-enthalpy primary flow in ejectors. The aerodynamic choking of the secondary flow within an ejector leads to two modes of ejector operation: (a) critical flow regime and (b) mixed flow regime. Several low-fidelity models for analyzing ejectors have been proposed in the literature. These models are adequate for the rapid design of specific categories of ejectors. However, they yield notable deviations of 20-25% with experimental measurements for different operating conditions. On the other hand, computational fluid dynamics (CFD) simulations for ejectors are computationally expensive. Furthermore, they require a careful selection of numerical solvers and turbulence models to predict performance and flow characteristics accurately. This paper presents a data-driven artificial neural network (ANN) model to predict the critical parameters of supersonic ejectors. The model is trained using experimental measurement for air ejectors at various operating and geometrical parameters. The ANN model consists of five input parameters, representing operating and geometrical parameters of ejectors to estimate two output parameters: (a) entrainment ratio and (b) operating regime. The trained ANN model predicts the entrainment ratio with a maximum deviation of about 7% and classifies the ejector operational mode with an accuracy of 100%.