Optimization of ultrasonic-excited double-pipe heat exchanger with machine learning and PSO

Abstract

Two intelligent optimization methods, artificial neural networks (ANNs) and adaptive neuro-fuzzy inference system (ANFIS), with particle swarm optimization (PSO), were employed in this study to forecast the heat transfer rate, Nusselt number, number of transfer units (NTU), and effectiveness of a double-pipe counter-flow heat exchanger. The experiments are related to hot flow rate (113–257 l/h), hot fluid inlet temperature (40–60 °C), ultrasonic excitation power level (0 or 60 watts), and volume fraction of nanoparticle (0–0.8%). The findings of the ANN, ANFIS, ANN-PSO, and ANFIS-PSO methods were verified through experimental studies. The ANN model, the multilayer perceptron/backpropagation with three different learning algorithms, the Levenberg Marquardt (LM), the scaled conjugate gradient (SCG), the Bayesian regularization algorithm (BR), and the ANFIS model were developed with backpropagation and the optimal hybrid method. Finally, these two models were optimized using the PSO algorithm. The experimental and predicted data were compared, and both models were found to be accurate. The correlation coefficients of both models were greater than 94.84%, with the ANN-PSO model slightly outperforming the ANFIS-PSO model. Also, the maximum amount of MSE is equal to 2.592, which shows the model has good performance in predicting results.