Akademik Veri Yönetim Sistemi
Erzincan Üniversitesi Fen Bilimleri Enstitüsü Dergisi, cilt.18, sa.3, ss.981-997, 2025 (TRDizin)
Shear thickening fluids (STFs) show a complex, non-Newtonian rheological behavior in which viscosity increases with shear rate. Accurately estimating the thickening ratio (TR), a summary parameter representing the rheological response, is crucial for optimizing the formulation of these fluids and their effective use in applications. In this study, a machine learning-based approach is proposed to directly predict TR. The modeling process incorporated rheologically relevant input parameters, including particle size, weight-based particle concentration, carrier-fluid concentration, molecular weight of the carrier-fluid, and test temperature.
Two advanced ensemble learning algorithms, Extreme Gradient Boosting (XGBOOST) and Random Forest (RF), were used to create the prediction models. The models were trained and validated on various experimental datasets obtained from different independent sources and covering a wide range of STF compositions and experimental conditions. The results showed that XGBOOST achieved 80% and 72% accuracy for RF during the testing phase, with XGBOOST outperforming RF. Furthermore, the calculated feature importance values revealed the main parameters affecting TR. Although the influence values of the parameters on TR were close to each other, the carrier-fluid ratio (OSO) (25.91%) and the silica ratio (SO) (24.32%) stand out as the most influential parameters.
This approach offers a simple and effective method for assessing the rheological behavior of STF systems, while also providing significant time and cost advantages by reducing the need for extensive experimental procedures. The developed method has the potential to be a valuable tool for decision support in the design and development of next-generation STF materials.