An ensemble learning approach for predicting nanoparticle dispersion properties in cutting fluids using a small sample dataset

This study examines and predicts the dispersion properties of three nanoparticles- CuO, Al 2 O 3 , and graphene when dissolved in distilled water with sodium lauryl sulfate (SLS), which functions as a nanofluid surfactant to determine its appropriateness as a cutting fluid. A machine learning (ML) approach, coupled with grid search optimization, predicts the dispersion properties, including thermal conductivity, surface tension, and viscosity. Experiments with small sample datasets were conducted with input parameters such as nanoparticle mass, surfactant concentration, and sonication time to effectively train and cross-validate the prediction errors. The results demonstrate that grid search-optimized models significantly increase prediction accuracy, enabling reliable identifications of dispersion properties. Considering all the combinations of nanoparticles analysed, CuO nanoparticles achieved the highest prediction accuracy and lowest errors across all metrics. This study demonstrates machine learning's effectiveness in capturing non-linear correlations in small sample datasets using grid search optimization and accurate predictions. It is observed that the coolants enhanced by nanoparticles may outperform traditional cutting fluids in terms of cooling and lubricating effectiveness during machining operations. This work underscores the role of optimized models in advancing the development of high-performance, sustainable cutting fluids for industrial applications.