Machine learning-based estimation of nano-lubricants viscosity in different operating conditions

The importance of improving and optimizing the energy transfer systems has been highlighted recently due to ever-increasing demand for energy in the world, and therefore the lubrication of systems has received significant attentions. The improvement of thermophysical properties by nano-particles can be considered as an efficient approach to this end. A comprehensive study dealing with constructing myriad machine learning-based models for predicting the viscosity of nano-lubricants is lacking in the literature. This research, therefore, aims at development of eight different artificial intelligence models including, Adaptive Boosting, Random Forest (RF), Ensemble Learning, Decision Tree (DT), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Convolutional Neural Network (CNN), and Multi-Layer Perceptron Artificial Neural Network (MLP-ANN) to estimate viscosity of different nano-lubricants as a function of amount of MWCNT, nanoparticle, type of lubricant, temperature, shear rate, and solid volume fraction. These models are prepared and tested based on a databank including 1086 viscosity points. The results show that the MLP and CNN models are the most accurate methods with R-squared of 0.950131 and 0.95035, respectively, while the estimation of computational time expresses that MLP and CNN models require the most computational time for training. Moreover, sensitivity analysis reveals that temperature has the most effect on viscosity of nano-lubricants.