Comparative Analysis of Machine Learning Models for Nanofluids Viscosity Assessment.

Area of Science:

• Materials Science
• Chemical Engineering
• Computational Science

Background:

• Accurate prediction of nanofluid viscosity is essential for their application.
• Experimental viscosity measurements are costly and time-consuming.
• Existing theoretical and empirical models often yield inaccurate results.

Purpose of the Study:

• To develop and evaluate novel machine learning models for predicting nanofluid viscosity.
• To compare the performance of these models against existing methods, including the Committee Machine Intelligent System (CMIS).
• To assess the physical validity of the proposed models.

Main Methods:

• Utilized a dataset of 3144 experimental relative viscosity data points for 42 nanofluid systems.
• Developed eight machine learning models: Multilayer Perceptron (MLP) with Nadam/Adamax optimizers, Support Vector Regression (SVR), Decision Tree (DT), Random Forest (RF), and Extra Tree (ET).
• Evaluated model performance using Average Absolute Relative Error (AARE) and assessed physical trends.

Main Results:

• All eight proposed models outperformed literature baselines.
• Five models surpassed the performance of the CMIS.
• Two models achieved an AARE below 3% on test data, demonstrating superior accuracy.
• Models showed physically expected trends in nanofluid viscosity with changing volume fraction.

Conclusions:

• Advanced machine learning models offer a highly accurate and efficient alternative for predicting nanofluid viscosity.
• The developed models provide a cost-effective solution for selecting nanofluids for specific applications.
• The study validates the physical plausibility of the machine learning predictions.