Eventos Anais de eventos
ENCIT 2020
18th Brazilian Congress of Thermal Sciences and Engineering
PREDICTION OF THE HEAT TRANSFER COEFFICIENT DURING CONDENSATION OF HYDROCARBONS IN MINI/CONVENTIONAL CHANNELS USING MACHINE LEARNING ALGORITHMS
Submission Author:
Gabriel Furlan , SP
Co-Authors:
Gabriel Furlan, Gherhardt Ribatski
Presenter: Gabriel Furlan
doi://10.26678/ABCM.ENCIT2020.CIT20-0265
Abstract
This study investigates the use of machine learning (ML) method for predicting the heat transfer coefficient (HTC) during condensation of hydrocarbons in mini/conventional channels. The multi-layer perceptron with backpropagation (MLPB) and the gradient boosted decision tree (GBDT) algorithms are optimized, trained and validated based on experimental data gathered from the literature, which include results for propane, propylene, isobutene and ethane, tube diameters from 0.956 to 20.8 mm, mass velocities from 35 to 1000 kg/m²s and reduced pressures from 0.1 to 0.96. In order to capture the main mechanisms acting on the condensation process, dimensionless numbers Bond, Galileo, Reynolds and Weber were used as input parameters. The gradient boosted decision trees provided the best predictions with a mean absolute error (MAE) of 4.7% and 86.5% of the data predicted within an error band of ±10%. The neural net predicted 91.2% of the results within an error band of ±20% and a MAE of 8.5%. The ML methods successfully predict the effects on the HTC of vapor quality, mass velocity, saturation temperature, channel diameter and flow pattern. When applied to non-hydrocarbon data, the MLPB provided better predictions than GBDT, which may indicate that the first captures the experimental trends of the HTC. Moreover, based on a simulation of a tube-in-tube condenser, it was found that the ML methods need 90% lower processing time than Cavallini et al. (2006) model.
Keywords
condensation, Hydrocarbon, Heat transfer, machine learning
DOWNLOAD PDF VIEW PRESENTATION
