Accuracy analysis and artificial neural network-based correction of ejector simulation considering nonequilibrium condensation

Ejectors have been widely applied as steam compression and energy recovery devices in multi-effect distillation with thermal vapor compression (MED-TVC) systems. However, most existing studies rely on dry gas models, neglecting the ubiquitous condensation effects in steam flow. This study develops a computational fluid mechanics model incorporating nonequilibrium condensation to investigate wet steam behavior within the ejector. The simulations are performed in ANSYS Fluent 2019R2 using a 2D steady-state framework and the k–ω turbulence model. Numerous tests were conducted with the MED-TVC system, experimental validation was conducted under three representative operating conditions. Comparative analysis between dry gas and wet steam models reveals that simulation accuracy varies significantly with operating conditions. The relative deviation (RD) ranges from 9.4 to 11.5 %, 21.5–26.2 %, and 0.4–2.1 % across the three conditions. Compared to the dry gas model, the wet steam model is more accurate. As the compression ratio (CR) gradually increases from 5.31 to 9.77, the wet steam numerical simulation error gradually increases from 0.17 % to 18.4 %. And clarified the underlying mechanisms causing this phenomenon, that the accuracy of simulations for ejectors declines as they approach their suction performance limits. Furthermore, to rectify this disparity, this study proposes a simulation error calibration method combining a multilayer artificial neural network (ANN) with unsupervised clustering. After the correction of this method, the RD is reduced to less than 3 %, and the coefficient of determination R2 is 0.93705, which confirms effectiveness and generalization ability of the numerical simulation calibration method. Compared with existing ANN methods that require targeted training for different ejector structures, the numerical simulation calibration method in this study has better generalization.