Advancing Experimental, Modeling and Neural Network Techniques for Thixotropic Yield Stress Flows

Doctoral thesis, 2026

Soft materials such as gels, suspensions, pastes, biological fluids, and polymers are central to a wide range of industrial, technological, and biological processes. Yet their flow behaviour remains exceptionally challenging to predict. In particular, thixo-elasto-viscoplastic (TEVP) fluids exhibit strong time-dependent restructuring, yielding, and viscoelasticity, making their response highly sensitive to measurement protocol, geometry, and flow history. Unlike Newtonian fluids, whose response is uniquely determined by instantaneous deformation, the behaviour of TEVP fluids reflects a time-dependent reorganization of their underlying microstructure under applied deformation. This temporal variability of the microstructure hampers model development, complicates process design, and limits the reliable use of TEVP soft materials in applications such as microfluidics, 3D printing, food processing, consumer product formulation, and biological lubrication. These challenges highlight the need for an integrated framework capable of linking microstructural dynamics, rheological response, and continuum-scale flow behaviour.
This thesis develops such a unified framework by combining rheological characterization, flow-resolved imaging, continuum and structural-kinetic modeling, and data-driven neural network approaches to study TEVP materials across various shear histories in multiple flow geometries. First, rheological protocols are established to determine viscoelastic, viscoplastic, and thixotropic contributions in soft materials including Laponite suspensions, Carbopol gels, polymer blends, commercial yogurts, and human saliva.
Flow behaviour is then investigated in confined geometries. Doppler Optical Coherence Tomography (D-OCT) measurements in rectangular millifluidic channels reveal plug regions, wall slip, shear localization, and scaling relationships linking rheological properties to velocity-field evolution.
In circular pipes, transient pressure drop measurements are combined with structural kinetic modeling and continuum CFD simulations to quantify the breakdown–recovery kinetics governing unsteady TEVP transport.
To address limitations of classical constitutive models, neural network surrogates, including NARX network based digital rheometers and data-driven flow predictors, are trained on minimal experimental input. These models reconstruct full flow curves, predict transient rheological responses, and accurately capture unsteady pressure drop dynamics, helping resolve persistent issues of parameter identifiability and model selection.
From a process engineering perspective, extrusion-based 3D printing is examined using rheology, dimensional analysis, and high-speed imaging to determine how elasticity, yield stress, and interfacial forces govern die swell, filament formation, and print fidelity. Complementary structural characterization using rheology coupled with scattering techniques (Rheo-SAXS) provides insight into flow-induced microstructural reorganization and its influence on die swelling. Finally, human saliva is examined as a biologically relevant thixotropic material. While its viscoelastic properties are well established, this work demonstrates and quantifies its clear thixotropic behaviour, identifying it as a detrimental factor in the perception of dry mouth (xerostomia).
Together, these contributions advance a coherent experimental, computational, and data-driven methodology for TEVP materials, bridging continuum mechanics with microstructural origins and enabling more reliable prediction, process design, and formulation of soft materials across engineering and biomedical applications.