Use of tailored hydrodynamic conditions to control reactions for AgNPs synthesis in microfluidic devices

Nathanael, Konstantia ORCID: 0000-0002-9248-8108 (2025). Use of tailored hydrodynamic conditions to control reactions for AgNPs synthesis in microfluidic devices. University of Birmingham. Ph.D.

Nathanael2025PhD.pdf
Text - Accepted Version
Available under License All rights reserved.
Download (4MB)

Abstract

This thesis examines the impact of synthesis conditions on the formation of silver nanoparticles (AgNPs), focusing on system chemistry, hydrodynamics, and storage temperature. The study uses T-junction continuous single-phase flow and droplet microfluidic reactors with 500 μm width channels as well as a batch reactor (beaker). The microfluidic reactors were evaluated under identical conditions, revealing that continuous
flow reactors produced AgNPs comparable in size to those from droplet reactors. Consequently, the study focused on the former design with optimization of AgNP formation initially carried out using a predictive regression model with response surface methodology, employing tannic acid (TA) and trisodium citrate (TC) as a reducing agent. Key factors affecting particle size, such as pH, stabilizer concentration, and reactor outlet
curvature, were identified. However, limitations due to the complexity of the multi-step reaction scheme were noted, with response surface models being effective only within specific operating ranges and unable to provide detailed microscopic insights. To better understand AgNP reaction kinetics, the Finke-Watzky two-stage kinetic model was fitted to experiments carried out in a well-mixed reactor using two distinct chemical protocols. Incorporating these kinetics into a regression model allowed particle size distribution (PSD) predictions with an average error of 4.9%. Experimental validation of a more sophisticated coupled PBM-CFD simulation was then used to interpret hydrodynamic and kinetic effects on AgNP size distribution in the T-junction device and a well-mixed reactor, showing consistency with experimental trends, particularly regarding pH, stabilizer concentration, and average PSD in a well-mixed reactor. Specifically, the experimental and Finke-Watzky based PBM predictions in the well-mixed reactor obtained at pH = 7 showed a maximum deviation of 10.1% and 31.2% for average particle diameter and standard deviation, respectively. PBM-CFD simulations for the continuous flow device showed that operating conditions, especially inlet flow rates, significantly impact precursor consumption rates and PSD. Helical channels posed additional challenges for precursor consumption, suggesting areas for further research. Machine learning and data science approaches were then used to further optimize AgNP synthesis in continuous reactors and generalize the regression model. A decision tree-guided design of experiments predicted AgNP size using generic parameters, reducing computational costs and enhancing model performance. Factors such as Reynolds number and the Dean number to Reynolds number ratio were used to assess hydrodynamics and mixing. Nucleation and growth constants derived from the independent set of experiments with TA and TC
using the Finke-Watzky model examined the system's chemistry, while storage temperature ensured particle stability. The decision tree approach, validated by experimental data, proved effective, indicating that well-designed experiments could substantially reduce mean squared error (MSE) in models like XGBoost (-26%) and Random Forest (-38%). Additionally, in the context of data science, inverse modeling, employing polynomial functions and data assimilation, facilitated nanoparticle design with desired sizes, demonstrating consistency and accuracy, with an average percentage error of 11.99% between predicted and experimental sizes. This approach provided valuable
insights for decision-making in nanoparticle manufacturing under uncertainty.

Type of Work:

Thesis (Doctorates > Ph.D.)

Award Type:

Doctorates > Ph.D.

Supervisor(s):