Rheology and Thixotropy of Colloidal Gels: Tunability and CNN Prediction
Abstract: Attractive colloidal suspensions can form space-spanning, interconnected networks that give rise to soft solids known as colloidal gels. Their rheology is strongly history-dependent, exhibiting thixotropic behavior in which viscosity evolves with prior shear and microstructural changes. In this talk, I present a simulation-based study of thixotropy and anti-thixotropy in colloidal gels, and demonstrate how electric fields can be used to actively tune their mechanical response. We employ Stokesian dynamics simulations to study attractive colloidal suspensions and rigorously validate the model against the adhesive hard-sphere phase diagram and steady-shear viscosity data. Using a three-interval thixotropy test (3ITT), the system is subjected to low shear, high shear, and then a return to low shear. The final interval, a flow step-down from high to low shear, is the primary focus. As expected for shear-thinning gels, the step-down initially produces a gradual viscosity increase due to bond reformation, characteristic of thixotropic recovery. However, at long times we observe an unexpected viscosity decrease following recovery. We identify this inverse response as anti-thixotropy. Energy landscape analysis reveals that thixotropic recovery reflects the bond recovery after the shear-induced rejuvenation during the high-shear interval, whereas anti-thixotropy arises from long-time shear-induced aging and oversolidification under low shear flow. Building on this understanding, we explore electrical tunability of colloidal gels via dipolophoresis (DIP), arising from the interplay of two nonlinear electrokinetic phenomena, namely induced-charge electrophoresis (ICEP) and dielectrophoresis (DEP). Depending on field frequency and particle polarizability, electric fields can either rejuvenate gels through diffusive bond breaking from the dominance of ICEP or enhance solidification via field-aligned, columnar structures from the dominance of DEP, reminiscent of electrorheological (ER) fluids. Finally, we introduce a convolutional neural network (CNN) framework that links microstructural snapshots to viscosity at low shear rates, offering a data-driven route to predict rheology from the microstructure of colloidal gels.
