Development of a predictive response surface model for size of silver nanoparticles synthesized in a T-junction microfluidic device

• Published in: Chemical engineering science 2023-09, Vol.279, p.118907, Article 118907
• Main Authors: Nathanael, Konstantia, Galvanin, Federico, Kovalchuk, Nina M., Simmons, Mark J.H.
• Format: Article
• Language: English
• Subjects: Experimental designs; Microfluidic synthesis; Response surface methodology; Silver nanoparticles
• ISSN: 0009-2509
• DOI: 10.1016/j.ces.2023.118907

•Pairing microfluidics and design of experiments for sustainable synthesis of AgNPs.•DoE helps link chemical and hydrodynamic parameters to final properties of AgNPs.•Microfluidics provide an efficient method for a controlled formation of AgNPs. Optimisation of the parameters governing the synthesis of silver nanoparticles (AgNPs) is a critical step in controlling particle size, which facilitates their application in diverse range of industrial and consumer related products. AT-junctionmicrofluidicsystemwas used together with design of experiments, regression-analysis and response surface methodology to build a predictive numerical model for the size of silver nanoparticles (AgNPs). Aqueous solutions of silver-precursor and reducing/stabilizing agent were supplied by two separate channels meeting at the T-junction, withthereaction occurring downstream intheoutlet tube. To improve the mixingof the reagents,the output tube was coiled onto a 3D-printed helical shape device, exploiting the creation of Dean vortices. The effects of both reaction and hydrodynamic conditions including the solution pH, collection temperature, helical curvature, flow rates and concentration of stabilising agent were investigated using a D-optimal experimental design. The obtainedexperimental size distributions for the AgNPs were fittedtoa polynomial model with an average prediction error of around 13% and a 37 % maximum error.The modelpredictedthe optimal synthesis conditionsforthe global and local-minimum sizes of AgNPswith an errorof around 7.0% and 16.1% respectively. The average prediction error of the testing set was estimated to be 6.8% with 16.1% being the maximum error.