The present study is focused on proposing, implementing, and optimizing machine learning based pipelines for estimating the pressure drop in evaporating R134a flow passing through micro-fin horizontal tubes. Accordingly, an experimental activity is first conducted, in which the pressure drop of the flow at various operating conditions is measured. Physical models that are available in the literature are then implemented and the corresponding accuracy, while being applied to the obtained dataset, is determined. Machine learning based pipelines, with dimensionless parameters provided as features and two-phase flow multipliers as targets, are then developed. In the next step, a feature selection procedure is performed and an optimization process is then conducted to find the algorithms and the corresponding hyper-parameters, using which results in the highest possible accuracy. The optimal pipeline is demonstrated to be the one in which the liquid only two-phase multiplier is chosen as the target and is provided with only 5 dimensionless parameters as selected input features. Employing the latter pipeline leads to a mean absolute relative deviation (MARD) of 6.27 % on the validation set and 6.41 % on the test set, which is notably lower than the one achieved using the most promising physical model (MARD of 18.74 % on the validation set and 18.08 % on the test set). Furthermore, as the dataset and the obtained optimal pipeline will be made publicly accessible, the proposed methodology also offers higher ease of use and reproducibility.

Machine learning based pressure drop estimation of evaporating R134a flow in micro-fin tubes: Investigation of the optimal dimensionless feature set

Ardam K.;Najafi B.;Lucchini A.;Rinaldi F.;Colombo L. P. M.
2021

Abstract

The present study is focused on proposing, implementing, and optimizing machine learning based pipelines for estimating the pressure drop in evaporating R134a flow passing through micro-fin horizontal tubes. Accordingly, an experimental activity is first conducted, in which the pressure drop of the flow at various operating conditions is measured. Physical models that are available in the literature are then implemented and the corresponding accuracy, while being applied to the obtained dataset, is determined. Machine learning based pipelines, with dimensionless parameters provided as features and two-phase flow multipliers as targets, are then developed. In the next step, a feature selection procedure is performed and an optimization process is then conducted to find the algorithms and the corresponding hyper-parameters, using which results in the highest possible accuracy. The optimal pipeline is demonstrated to be the one in which the liquid only two-phase multiplier is chosen as the target and is provided with only 5 dimensionless parameters as selected input features. Employing the latter pipeline leads to a mean absolute relative deviation (MARD) of 6.27 % on the validation set and 6.41 % on the test set, which is notably lower than the one achieved using the most promising physical model (MARD of 18.74 % on the validation set and 18.08 % on the test set). Furthermore, as the dataset and the obtained optimal pipeline will be made publicly accessible, the proposed methodology also offers higher ease of use and reproducibility.
R134a
Evaporating refrigerants
Feature selection
Machine learning
Pressure drop prediction
Two-phase flow
File in questo prodotto:
File Dimensione Formato  
1-s2.0-S0140700721003005-main.pdf

Accesso riservato

Descrizione: Articolo
: Publisher’s version
Dimensione 2.32 MB
Formato Adobe PDF
2.32 MB Adobe PDF   Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: http://hdl.handle.net/11311/1196436
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 0
  • ???jsp.display-item.citation.isi??? 0
social impact