Prediction of droplets characteristic diameters and polydispersity index induced by a bifluid spraying nozzle by the means of dimensional analysis. (16th January 2023)

• Record Type: Journal Article
• Authors: Lachin, K.; Niane, M.; Person, M.; Mazet, J.; Delaplace, G.; Turchiuli, C.
• Is Part Of: Chemical engineering science. Volume 265(2023)
• Publication Date: 2023-01-16
• Subjects: Dimensional analysis -- Bifluid nozzle -- Droplet size distribution modeling -- Atomization -- Machine learning modeling; Chemical engineering -- Periodicals; Génie chimique -- Périodiques; Chemical engineering; Periodicals; Electronic journals; 660
• DOI: 10.1016/j.ces.2022.118187
• Languages: English
• ISSNs: 0009-2509

Abstract

Graphical abstract: Highlights: A rigorous dimensional analysis (DA) was performed on bifluid spray properties. Two characteristic spray diameters and a polydispersity index were considered. Models were established with conventional equation shapes and machine-learning (ML) Spray properties could be accurately modelled through a DA/ML approach. Particle Size Distributions could also be predicted using this methodology. Abstract: This work focuses on the study of sprays generated through a bifluid nozzle and the modelling of characteristic spray properties (two characteristic diameters and a polydispersity index) using dimensional analysis. Two types of dimensionless models were identified for each spray target property from the 75 experimental points considered. The first type used a conventional monomial-exponential shape equation, and the second applied shape identification through machine-learning. Although conventional models of the first type were mostly satisfactory when considering the characteristic diameters, they nevertheless showed clear limitations addressed by the machine-learning identified models. The conventional approach also failed to identify a satisfactory equation for the polydispersity index. The machine-learning approach provided an equation identifying this index to the main dimensionless parameters governing atomization. This identification provides a foundation for proposing a two-parameters dimensionless model that predicts spray particle size distribution. The combination of dimensional analysis with machine-learning equation identification thus paves the way to physically rigorous and easy-to-use models capable of predicting characteristic properties and full distributions.