Analysis of the Effects of Different Nanofluids on Critical Heat Flux Using Artificial Intelligence

Abstract

Nanofluid (NF) pool boiling experiments have been conducted widely in the past two decades to study and understand how nanoparticles (NP) affect boiling heat transfer and critical heat flux (CHF). However, the physical mechanisms related to the improvements in CHF in NF pool boiling are still not conclusive due to the coupling effects of the surface characteristics and the complexity of the experimental data. In addition, the current models for pool boiling CHF prediction, which consider surface microstructure characteristics, show limited agreement with the experimental data and do not represent NF pool boiling CHF. In this scenario, artificial intelligence tools, such as machine learning (ML) regressor models, are a very promising means of solving this nonlinear problem. This study focuses on creating a new model to provide more accurate NF pool boiling CHF predictions based on pressure, substrate thermal effusivity, and NP size, concentration, and effusivity. Three ML models (supporting vector regressor—SVR, multi-layer perceptron—MLP, and random forest—RF) were constructed and showed good agreement with an experimental database built from the literature, with MLP presenting the highest mean R2 score and the lowest variability. A systematic methodology for optimizing the ML models is proposed in this work.

Machine learning-based model for the intelligent estimation of critical heat flux in nanofluids

“…3. Another purpose is to provide a comprehensive comparison between the results of the models of this paper with developed models for CHF prediction in the literature, showing that the selected models have significant improvement in the accuracy Support Vector Regressor, Multilayer Perceptron Neural Network, and Random Forest Regressor for similar dataset [21]. The results of this paper which will be presented in section 3 have demonstrated a superior performance in terms of accuracy and stability (as indicated by higher R-squared values and lower standard deviations compared to the models in [21]), validate these approaches and underscore the effectiveness of these models in advancing the state of the art.…”

“…Although many advancements in these methods have been made and their accuracy has been improved over the two decades, they have some fundamental drawbacks, like high computational costs. As an alternative, using intelligent techniques, like Artificial intelligence (AI) and machine learning (ML) methods, for the estimation of parameters that are useful in engineering designs, has gained popularity in recent years [20][21][22][23][24][25][26][27][28][29][30][31]. Naturally, MLbased models offer several pros over CFD simulations and traditional mathematical fitting methods, particularly in terms of accuracy, generalizability, rapid response, and retraining ability.…”

“…There have been some efforts by other researchers to model CHF or other heat transfer properties of the cooling-or nano-fluids. In [21], the Random Forest (RF), Support Vector Regression, and Multilayer Perceptron (MLP) techniques were used to predict the CHF of different nanofluids. The results demonstrated that the MLP model had better accuracy and stability than the other models that were used.…”

“…roughness and porosity. In an almost similar approach to [21], the input data were collected for their model from published papers in the literature. The results of their study demonstrated that RF was a better technique for their dataset.…”