Data-driven techniques in rheology: Developments, challenges and perspective

With the rapid development and adoption of different datadriven techniques in rheology, this review aims to reflect on the advent and growth of these frameworks, survey the state-ofthe-art methods relevant to rheological applications, and explore potential future directions. We classify different machine learning (ML) methodologies into data-centric and physics-informed frameworks. Data-centric methods leverage conventional ML techniques to uncover relationships within specific datasets, demonstrating success in rheological propzation, and accelerated numerical simulations. Physicsinformed machine learning combines physical laws and domain knowledge with data to produce generalizable and physically consistent predictions, proving effective in solving rheological differential equations, utilizing multi-fidelity datasets to enhance predictions, and constitutive modeling. The paper also discusses the limitations of these approaches and the ongoing efforts to address them. Looking ahead, this article emphasizes the need for explainable ML techniques to enhance transparency and trust, improved tools for uncertainty quantification. These advancements could significantly transform rheology and non-Newtonian fluid mechanics by enabling more robust, insightful, and efficient data-driven methodologies.