A computational study has demonstrated that an Adaptive Neuro-Fuzzy Inference System outperforms standard Artificial Neural Networks in predicting the logarithmic viscosity of polymer solutions across diverse operational parameters. Viscosity serves as a fundamental physical property governing the flow behavior, processing, and functional performance of polymers across numerous engineering sectors. The investigation focused on creating reliable, interpretable predictive models to estimate viscosity changes under fluctuating mechanical and environmental factors.
To train and evaluate the predictive models, the research team acquired 420 experimental data samples from a publicly accessible Kaggle database. The modeling framework incorporated four primary physical and chemical input variables: logarithm shear rate, polymer concentration, sodium chloride concentration, and operating temperature.
The researchers partitioned the dataset using a standard validation protocol, randomly allocating 70 percent of the samples for model training, 15 percent for validation, and the remaining 15 percent for testing.
The evaluation revealed distinct performance variations between the two modeling architectures. During its validation phase, the Artificial Neural Network achieved an optimal mean squared error of 67,522.49 at epoch 20, with error histogram analysis indicating that the majority of its prediction errors were centered around negative 21.35.
In contrast, the Adaptive Neuro-Fuzzy Inference System achieved superior predictive accuracy across all statistical evaluation metrics. The model recorded a root mean square error of 31.50, a mean absolute error of 6.49, and a coefficient of determination of 0.9846, establishing a strong correlation between predicted values and experimental benchmark data.
In a representative test case featuring an input combination of a logarithm shear rate of 55.8 reciprocal seconds, a polymer concentration of 0.175 weight percent, a sodium chloride concentration of 2.05 weight percent, and a temperature of 57.5 degrees Celsius, the model yielded a logarithmic viscosity prediction of 11.7 centipoise.
To provide model interpretability, researchers applied SHAP analysis to quantify the relative importance of each process variable. The analysis established logarithm shear rate as the most influential factor with an importance value of 0.69, followed by polymer concentration at 0.21, and sodium chloride concentration at 0.11. Furthermore, the findings confirmed that logarithmic viscosity increases drastically with increases in polymer concentration.
The study confirmed that neuro-fuzzy modeling provides a reliable, accurate, and comprehensible computational approach for anticipating polymer fluid behaviors. The investigators identified potential direct applications for this predictive capability in chemical formulation, polymer manufacturing, industrial flow control systems, and enhanced oil recovery operations.
Because polymer materials and polymer solutions serve as vital components across numerous industrial and engineering processes, accurately characterizing viscosity remains essential for determining their flow behavior and processing performance. By incorporating four primary variables, including logarithm shear rate, polymer concentration, sodium chloride salinity, and operating temperature, the research demonstrated how these factors collectively influence fluid behavior in polymer solutions.
By connecting influential factors directly to fluid responses, the investigators confirmed that this neuro-fuzzy methodology provides an effective predictive method for chemical processing, polymer formulation, industrial flow control systems, and enhanced oil recovery operations.
Subject of Research: Predictive machine learning modeling of polymer solution viscosity using neuro-fuzzy systems and neural networks
Article Title: Interpretable prediction of polymer material viscosity using machine learning models incorporating shear rate, salinity, concentration, and temperature effects
Article References: Vara Prasad, A. S., Swapna Sri, M. N., Sivasundar, V., Rajulapati, A., Dhasarathan, N., Kudapa, V. K., Gebrehiwot, K. H., & Kumar, S. (2026). Interpretable prediction of polymer material viscosity using machine learning models incorporating shear rate, salinity, concentration, and temperature effects. Journal of Materials Science: Materials in Engineering. https://doi.org/10.1186/s40712-026-00609-4
Image Credits: AI Generated
DOI: 10.1186/s40712-026-00609-4
Keywords: polymer viscosity, machine learning, adaptive neuro-fuzzy inference system, artificial neural networks, polymer solutions, shear rate, polymer concentration, sodium chloride concentration, temperature effects, SHAP analysis, predictive modeling, rheology, enhanced oil recovery, chemical processing, flow control systems, materials engineering
Cite Scienmag News
Neil Sanderson. (November 20, 2026). Machine Learning Models Accurately Predict Polymer Solution Viscosity Under Multiple Processing Conditions. Scienmag. https://scienmag.com/machine-learning-models-accurately-predict-polymer-solution-viscosity-under-multiple-processing-conditions/
Neil Sanderson. "Machine Learning Models Accurately Predict Polymer Solution Viscosity Under Multiple Processing Conditions." Scienmag, 20 November 2026, https://scienmag.com/machine-learning-models-accurately-predict-polymer-solution-viscosity-under-multiple-processing-conditions/. Accessed 20 November 2026.
Neil Sanderson. "Machine Learning Models Accurately Predict Polymer Solution Viscosity Under Multiple Processing Conditions." Scienmag. November 20, 2026. https://scienmag.com/machine-learning-models-accurately-predict-polymer-solution-viscosity-under-multiple-processing-conditions/








