Continuous inkjet printing and machine learning identify unknown liquid viscosities
Determining liquid viscosity is essential for industrial applications ranging from paint coatings and pharmaceutical fluids to combustion. Rotational rheometers are widely used to measure viscosity by analyzing liquid flow under pressure or shear stress. But these devices involve complex and time-consuming experiments.
Maîtrejean et al. developed an alternative that eliminates the need to measure the viscosity of individual liquids. Instead, their approach combines continuous inkjet printing (CIJ) with machine learning to identify unknown viscosity levels of liquids.
Newtonian viscosities of liquids were identified by comparing the morphology of unknown fluid jets to a dataset of well-known jet morphologies. To that end, the researchers generated the numerical dataset containing thousands of jetted fluids using Basilisk, an open-source computational fluid dynamics software platform.
“Note that the rheological properties, such as viscosity, are no longer measured, as is done in conventional rheometry, but identified,” co-author Guillaume Maîtrejean said. “Our viscosity identification approach proves to be accurate with an average error of less than 1% for a large range of viscosities.”
A strobe light and camera attached to a CIJ device enable careful observation of fluid jet morphology. The identification step needs only a few pictures of the unknown fluid jet, which can be obtained in few minutes, and an efficient deep neural network algorithm, trained offline once, which predicts the viscosity almost instantaneously.
The research addresses the prediction of Newtonian viscosities where both density and surface tension are known and constant. More complex identification, with non-constant surface tension, density or complex viscosities is under development.
Source: “Rheological identification of jetted fluid using machine learning,” by G. Maîtrejean, A. Samson, D. Roux and N. El-Kissi, Physics of Fluids (2022). The article can be accessed at https://doi.org/10.1063/5.0100575 .