An optimised deep learning method for the prediction of dynamic viscosity of MXene-based nanofuid

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F68407700%3A21220%2F23%3A00367068" target="_blank" >RIV/68407700:21220/23:00367068 - isvavai.cz</a>

Výsledek na webu

<a href="https://doi.org/10.1007/s40430-023-04284-w" target="_blank" >https://doi.org/10.1007/s40430-023-04284-w</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.1007/s40430-023-04284-w" target="_blank" >10.1007/s40430-023-04284-w</a>

Jazyk výsledku

angličtina

Název v původním jazyce

An optimised deep learning method for the prediction of dynamic viscosity of MXene-based nanofuid

Popis výsledku v původním jazyce

This study designs and develops a new optimised deep learning method to calculate the dynamic viscosity using the temperature and nanofake concentration. Long short-term memory (LSTM) has been a candidate as the most suitable deep learning method with the ability to reach higher accurate results with a defnition of the dropout layers during the training process to prevent the overshoot issue of the networks. In addition, the Bayesian optimisation technique is employed to extract the optimal hyperparameters of the developed LSTM to reach the system’s highest performance in predicting the dynamical viscosity based on temperature and nanofake concentration. The newly proposed method is designed and developed in MATLAB software using 80% and 20% of the dataset for training and testing of the model. The newly proposed optimised LSTM is compared with the recently developed model using multilayer perceptron (MLP) to prove the higher efficiency of our proposed technique. It should be noted that mean-squared error and root-mean-square error using the newly proposed optimised LSTM reduce by 12.56 and 3.54 times compared to the recently developed MLP model. Also, the R-square of the newly proposed optimised LSTM increases by 4.43% compared to the recently developed MLP model.

Název v anglickém jazyce

An optimised deep learning method for the prediction of dynamic viscosity of MXene-based nanofuid

Popis výsledku anglicky

This study designs and develops a new optimised deep learning method to calculate the dynamic viscosity using the temperature and nanofake concentration. Long short-term memory (LSTM) has been a candidate as the most suitable deep learning method with the ability to reach higher accurate results with a defnition of the dropout layers during the training process to prevent the overshoot issue of the networks. In addition, the Bayesian optimisation technique is employed to extract the optimal hyperparameters of the developed LSTM to reach the system’s highest performance in predicting the dynamical viscosity based on temperature and nanofake concentration. The newly proposed method is designed and developed in MATLAB software using 80% and 20% of the dataset for training and testing of the model. The newly proposed optimised LSTM is compared with the recently developed model using multilayer perceptron (MLP) to prove the higher efficiency of our proposed technique. It should be noted that mean-squared error and root-mean-square error using the newly proposed optimised LSTM reduce by 12.56 and 3.54 times compared to the recently developed MLP model. Also, the R-square of the newly proposed optimised LSTM increases by 4.43% compared to the recently developed MLP model.

Druh

J_imp - Článek v periodiku v databázi Web of Science

CEP obor

—

OECD FORD obor

20303 - Thermodynamics

Projekt

—

Návaznosti

I - Institucionalni podpora na dlouhodoby koncepcni rozvoj vyzkumne organizace

Rok uplatnění

2023

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Název periodika

Journal of the Brazilian Society of Mechanical Sciences and Engineering

ISSN

1678-5878

e-ISSN

1806-3691

Svazek periodika

2023 (45)

Číslo periodika v rámci svazku

07

Stát vydavatele periodika

DE - Spolková republika Německo

Počet stran výsledku

14

Strana od-do

—

Kód UT WoS článku

001045098600003

EID výsledku v databázi Scopus

2-s2.0-85165324126