Please download to get full document.

View again

of 5

Designing an artificial neural network to predict dynamic viscosity of aqueous nanofluid of TiO 2 using experimental data

In this research, the viscosity of the aqueous nanofluid of TiO 2 has been modeled by artificial neural networks using experimental data. Artificial neural networks are able to estimate the pattern of dynamic viscosity variation along with
0 views5 pages
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
  Designing an arti fi cial neural network to predict dynamic viscosity of aqueous nano fl uid of TiO 2  using experimental data ☆ Mohammad Hemmat Esfe a, ⁎ , Mohammad Reza Hassani Ahangar b , Mousa Rejvani c ,Davood Toghraie a, ⁎ , Mohammad Hadi Hajmohammad d a Young Researchers and Elite Club, Khomeinishahr Branch, Islamic Azad University, Isfahan, Iran b Department of computer engineering, Imam hossein university, Tehran, Iran c Faculty of Mechanical Engineering, Semnan University, Semnan, Iran d Department of mechanical engineering, Imam hossein university, Tehran, Iran a b s t r a c ta r t i c l e i n f o Available online 13 April 2016  In this research, the viscosity of the aqueous nano fl uid of TiO 2  has been modeled by arti fi cial neural networksusingexperimentaldata.Arti fi cialneuralnetworksareabletoestimatethepatternofdynamicviscosityvariationalongwithtemperature andnanoparticlesmassfractionwithahighprecision.Anetworkwithonehiddenlayerand4neuronshasbeenused.Theregressioncoef  fi cientwasobtained0.9998inthismodeling,whichshowsveryhighprecisionofneuralnetworkwithaverysimplestructure.Inaddition,arelationshipintermsofmassfractionand temperature was presented in order to predict the viscosity of this nano fl uid. This correlation can estimatethe viscosity of TiO 2 – water nano fl uid in a wide range of nanoparticles mass fraction with a maximum error of 0.5 %.© 2016 Elsevier Ltd. All rights reserved. Keywords: Nano fl uidDynamic viscosityArti fi cial neural networkCorrelationTemperature 1. Introduction By adding nanoparticles to the base  fl uid, an increase in viscosityis inevitable. Increase in nano fl uids viscosity should not make increasein  fl uid thermal conductivity or heat transfer coef  fi cient ineffective.Increase in viscosity leads to a pressure drop in system, and a greateramountof power is required tocompensate for thepressure reduction.Therefore, nano fl uids viscosity as well as thermal conductivity isconsidered a very signi fi cant parameter for investigating nano fl uidsef  fi cacy.Different parameters affecting nano fl uids behavior have been men-tioned in a review paper by Nwosu et al. [1]. Also, different parametersaffecting viscosity have been investigated by researchers in separatestudies. Temperature [2 – 11], volume fraction [12 – 17], packing fraction[18], thickness of nano layers [19,20], particle shape [21,22], and aggregation radius [23] are some of the most important parametersaffecting viscosity that have been summarized in Table 1.As is observed in Table 1, many parameters affect nano fl uidsviscosity. Some researchers have investigated nano fl uids numerically[24 – 27].The purpose of this article is modeling the viscosity of TiO 2 -waternano fl uids [28] by using arti fi cial neural networks. To do this, differentstructures of neural network have been investigated, and experimentaldataandmodelingresultshavebeencomparedindifferentgraphs.Also,a relationship in terms of temperature and nanoparticles mass fractionhas been presented to predict these nano fl uids viscosities. 2. Arti fi cial neural network  The new point of view on neural networks based on brain perfor-mance emerged in the 1940s. The  fi rst practical application of neuralnetworks appeared by introduction of perceptron network in the late1950s. In the last 10 years, thousands of articles have been written onneural networks and these networks are widely used in different sci-ences. What currently can be said is that in future neural network willhave an important role as a scienti fi c tool to solve special problems. Atpresent,theinformationavailableontheperformanceofbrainislimitedandthemostsigni fi cantadvancesinneuralnetworkwillbeobtainedinfuture when more information on performance of brain and biologicalneurons is available.The applications of neural networks in different sciences includeapplication in aerospace industries as auto-pilot, in transportation in-dustries asorientation systems, in defensive affairs aspursuingmovingtargetsandmanyotherapplications.Theapplicationofneuralnetworksin the sciences mentioned above is increasing and a new application of these networks is mentioned in articles by researchers every day. International Communications in Heat and Mass Transfer 75 (2016) 192 – 196 ☆  Communicated by W.J. Minkowycz. ⁎  Corresponding authors. E-mail addresses: (M. Hemmat Esfe), (D. Toghraie).© 2016 Elsevier Ltd. All rights reserved. Contents lists available at ScienceDirect International Communications in Heat and Mass Transfer  journal homepage:  The structure of neural network used in this research has beenshownin Fig.1. This structure is thesimplestoneamongthe investigat-ed structures (Table 1) that has the highest regression coef  fi cient.Evaluation and optimal structure of arti fi cial neural network takesplacethroughtrialanderror.Ascanbeseen,aonelayerwith4neuronshas been used as optimal structure for modeling nano fl uid dynamicviscosity.Table 2 shows that increasing hidden layers and the number of neuronsdonotnecessarilyincreasethenetworkprecisioninestimatingthe data. As was mentioned previously, by considering the values of R and MSE, the best selection for arti fi cial neural network architecturein order to model the data of viscosity has been chosen a networkwith one hidden layer and 4 neurons and tansig transfer function hasbeen used. 3. Correlation InordertoobtaintherelativeviscosityofTiO 2 aqueousnano fl uids,acorrelation has been presented in termsof temperature and nanoparti-cles mass fraction.  μ  nf   μ   f  ¼ 1 : 431 − 0 : 01864 T   þ 0 : 6073 ω  þ 0 : 01334 T  2 þ 0 : 02586 T  : ω  þ 0 : 3092 ω  2 þ 0 : 006043 T  3 þ 0 : 005644 T  2 : ω  þ 0 : 03323 T  : ω  2 þ 0 : 08318 ω  3 ð 1 Þ Where mean 313.1 and SD 20.37 are used to normalize  T   (tempera-ture). Also, mean 16.5 and SD 12.85 are used to normalize (massfraction), and coef  fi cients are 95% con fi dence bounds. This relationship  Table 1 Summary of some researchers conducted on the parameters affecting nano fl uid dynamic viscosity.Author(s) Range of temperatureand volume fractionResearchmethodNano fl uid Results Tavman et al.  [2] 20 – 60 °C0 – 4 vol.%Experimental SiO 2 /waterAl 2 O 3 /waterIncrease in relative viscosity by increase in volume fraction Pastoriza-Gallego et al.  [3] 280 – 325 K1 – 5 wt.%Experimental CuO/water The effect of particle size cannot be ignored. The greatest amount of potential  Z   isin pH 2 and pH 13. Mariano et al.  [5] 283.15 – 323.15 KUp to 25 wt.%Experimental Co 3 O 4 /EG Intense variations of temperature – nano fl uid Newtonian behavior increase of 20 – 30% in thermal conductivity at 6 vol.% Hemmat Esfe et al.  [7] 26 – 55 °CUp to 3 vol.%Experimental Fe/EG The smaller the particle size lead to the higher thermal conductivity and viscosity;40 nm nanoparticles have a thermal conductivity 11% higher than 100 nmnanoparticles. Yiamsawas et al.  [8] 15 – 60 °C1 – 4 vol.%Experimental TiO 2 /water EGAl 2 O 3 /water EGThe viscosity of Al 2 O 3  nanoparticles is 130% greater than TiO 2  nanoparticles.  Abdellahoum et al.  [13] 1 – 4 vol.%10 4 b Re b 10 5 Numerical Al 2 O 3 /water Pak and Cho model shows the maximum value of friction. Vajjha et al.  [15] 1 – 6 vol.%3000  b  Re  b  8000Numerical Al 2 O 3 /water EGCuO/water EGThe average heat transfer coef  fi cient over the base  fl uid for a 3% volume fraction of Al 2 O 3  nano fl uid is 36.6% and that for a 3% volume fraction of CuO nano fl uid is 49.7%.  Zhao et al.  [16] Present two ANN with 4and 5 inputs.Numerical Al 2 O 3 /waterCuO/waterModeling by using neural network through radial basis function Srivastava  [21] 1 – 5 vol.% Numerical Al 2 O 3 /water Particle shape deviation from spherical leads to an increase in viscosity.  ZHAO et al.  [23] Diameter of nanoparticlesfrom 7 to 40 nm0.1 – 2 vol.% T   = 20, 25, and 30 °CExperimental SiO 2 /DI-water The amount of viscosity relate to the size of agglomerate overdependence of viscosity of nano fl uids or 7 nm nanoparticles to PH Fig. 1.  Optimal structure of neural network.193 M. Hemmat Esfe et al. / International Communications in Heat and Mass Transfer 75 (2016) 192 – 196   is applied in temperatures of 280 – 350k and mass fractions of 1 – 35%.Table 3 exhibits the parameters of this nonlinear regression.Fig.2showsathree-dimensionalofrelativeviscosityversustemper-ature and concentration of nanoparticles. As is observed, variations inviscosity due to change in temperature are small and the effect of mass fraction on viscosity is very noticeable. The results indicate thatbyincreasingmass fraction,changes occur rapidly,and thesevariationsgradually become more considerable So that change of mass fractionfrom 20% to 30% is more obvious.In Fig. 3, variations in value of mean square error (MSE) are seenwith different iterations of train, validation, and test data. If value of MSE increases in a certain number of iterations for validation data,results are considered as failed ones and it stops iterations and thebest result is presented as output. As can be seen in Fig. 3, stop hasoccurred with 81 iterations with MSE value of approximately 0.0008.Fig. 4 demonstrates the histogram of data error for neural networkmodeling. The horizontal axis shows different ranges of error, and thevertical axis shows the number of data in different ranges of error. Inorder to have a clearer representation, the values of   δ  and  μ   of errorgraph have been drawn in two forms of curve and histogram since thevalue of   δ  is clearer in histogram graph and value of   μ   is more obviousin curve graph. As can be observed, the peak of the errors is locatedaround zero.Fig.5 illustrates a comparison between experimental data andoutputs of neural network modeling. As can be seen, neural networkhas been very successful in modeling viscosity data and has estimat-ed empirical data with a very high precision. Therefore, this networkcan be used with high reliability to estimate the data that are notavailable.Fig.6 shows the regression of neural network results in terms of experimental results. In this study, 70% of input data were randomlyconsidered for training, and 15% of that was randomly used for testing.Good adjustment of modeling results on bisector (that is the criterionforevaluationofdata)showstheprecisionofneuralnetworkmodeling.According to this graph, the presented neural network can have anacceptable adjustment with criterion line and thus test data canestimate nano fl uid viscosity with a maximum error less than 2%.  Table 2 Neural network parameters at different structures.Hidden layer  R  MSE SD Mean of errors Train performance Test performance2 0.9994 9.6859e − 4 0.0310  − 0.0066 7.4279e − 4 0.00223 0.9996 6.3575e − 4 0.0257  − 6.7709e − 5 5.8021e − 4 7.4989e − 44 0.9998 4.4153e − 4 0.0200 0.0075 1.4973e − 4 0.00115 0.0996 6.7173e − 4 0.0260 0.0045 5.4655e − 4 9.4348e − 4[2 1] 0.9994 0.0010 0.0315 0.0071 8.1160e − 4 0.0030[2 2] 0.9994 9.7728e − 4 0.0317  − 0.0023 6.2407e − 4 0.0030[3 1] 0.9995 8.5213e − 4 0.0294  − 0.0045 8.0067e − 4 4.2021e − 4[3 2] 0.9996 9.2047e − 4 0.0293  − 0.0097 6.5645e − 4 0.0020[3 3] 0.9996 5.9938e − 04 0.0238  − 0.0072 4.4134e − 04 0.0019[4 1] 0.9996 7.4806e − 4 0.0278 0.0012 7.2177e − 4 5.3289e − 4[4 2] 0.9996 9.8567e − 4 0.0319 0.0019 5.7932e − 4 5.9585e − 4[4 3] 0.9996 6.9321e − 04 0.0267  − 0.0028 5.1862e − 04 0.0012[4 4] 0.9996 5.6855e − 04 0.0239 0.0044 5.9507e − 04 8.5482e − 04  Table 3 regression correlation parametersSSE  R 2 Adjusted  R 2 RMSE0.02181 0.999 0.9985 0.03481Theadjustment isobtained by makingacomparison betweenthenetworkactualvalue  t  ij andtheestimatedvalues a ij bycalculatingthetotalsumofthesquareerror(SSE)forthe n data of the training data set.  R 2 = coef  fi cient of determination; adjusted  R 2 = degree-of-freedom adjusted coef  fi cient of determination; RMSE = root mean squared error(standard error). Fig. 2.  Relative viscosity versus concentration and temperature.194  M. Hemmat Esfe et al. / International Communications in Heat and Mass Transfer 75 (2016) 192 – 196   Forabetterunderstandingofdatadeviationfromrealvaluesofneu-ral network modeling outputs, the concept of margin of deviation isused. This value is obtained from the following relationship:Marginofdeviation ¼  β  ANN −  β  Exp  β  Exp  100  ð 2 Þ The values obtained from this relationship for all neural networkoutputs are seen in Fig. 7. The maximum value of margin of deviationforthesedatais0.06%.Ifweassumethatthisvaluehasalsobeenobtain-ed for test data, we can prove the reliability of neural networks. 4. Conclusion The purpose of this research is to investigate modeling of relativeviscosity of TiO 2 – water nano fl uid by using arti fi cial neural networks.Neural network is one of the most powerful tools of modelingfor com-plicatedengineeringproblems.Inthisstudy,withinputdataoftemper-ature and mass fraction, different structures of neural networks withone and two layers and different numbers of neurons have been inves-tigated. A structure with one hidden layer and 4 neurons and tansigactivating function has been used. The maximum value of deviationfor modeling the neural network was obtained 2%, and this value forthe presented empirical relationship was obtained 3%. The data thatare not available can be predicted with high reliability by using this 0 20 40 60 80 100 12010 -8 10 -6 10 -4 10 -2 10 0    M  e  a  n   S  q  u  a  r  e   d   E  r  r  o  r   (  m  s  e   ) Iterate TrainValidationTestBestGoal Fig. 3.  Performance of arti fi cial neural network model -0.06 -0.04 -0.02 0 0.02 0.04 0.06 0.080123456 µ =0.0075, σ =0.02Errors    I  n  s   t  a  n  c  e ErrorsCurve Error Fig. 4.  Suitable curve on histogram error. 0 5 10 15 20 250.511.522.533.5Number of Data    D  y  n  a  m   i  c   V   i  s  c  o  s   i   t  y   (  c   P   ) Experimental DataANN Results Fig. 5.  Comparative graph between experimental data and ANN results. 1 1.5 2 2.5 3 3.511.522.533.5 Experimental Data    A   N   N   R  e  s  u   l   t  s Equality LineANN Results Fig. 6.  Regression of ANN results.195 M. Hemmat Esfe et al. / International Communications in Heat and Mass Transfer 75 (2016) 192 – 196   model. Also, a relationship in terms of temperature and mass fractionhas been presented to estimate viscosity data. This relationship is ableto estimate viscosity with a very good precision. References [1] P. Nwosu, J. Meyerb, M. Sharifpurb, A review and parametric investigation intonano fl uid viscosity models, J. Nanotechnol. Eng. Med. (2014).[2] I. Tavman, A. Turgut, M. Chirtoc, H.P. Schuchmann, S. Tavman, Experimental inves-tigation of viscosity and thermal conductivity of suspensions containing nanosizedceramic particles, Arch. Mater. Sci. 100 (2008) 100.[3] M.J.Pastoriza-Gallego,C.Casanova,J.L.Legido,M.M.Piñeiro,CuOinwaternano fl uid:in fl uence of particle size and polydispersity on volumetric behaviour and viscosity,Fluid Phase Equilib. 300 (1 – 2) (2011) 188 – 196.[4] L.S. Sundar, K.V.V. Sharma, M.T.T. Naik, M.K. Singh, Empirical and theoretical corre-lations on viscosity of nano fl uids: a review, Renew. Sust. Energ. Rev. 25 (2013)670 – 686.[5] A. Mariano, M.J. Pastoriza-Gallego, L. Lugo, L. Mussari, M.M. Piñeiro, Co 3 O 4  ethyleneglycol-based nano fl uids: thermal conductivity, viscosity and high pressure density,Int. J. Heat Mass Transf. 85 (2015) 54 – 60.[6] M. NabeelRashin,J.Hemalatha, Viscosity studies on novel copperoxide – coconut oilnano fl uid, Exp. Thermal Fluid Sci. 48 (2013) 67 – 72.[7] M. Hemmat Esfe, S. Saedodin, O. Mahian, S. Wongwises, Ef  fi ciency of ferromagneticnanoparticles suspended in ethylene glycol for applications in energy devices: ef-fects of particle size, temperature, and concentration, Int. Commun. Heat MassTransfer 58 (2014) 138 – 146.[8] T. Yiamsawas, O. Mahian, A.S. Dalkilic, S. Kaewnai, S. Wongwises, Experimentalstudies on the viscosity of TiO 2  and Al 2 O 3  nanoparticles suspended in a mixture of ethylene glycol and water for high temperature applications, Appl. Energy 111(2013) 40 – 45.[9] M.HemmatEsfe,S.Saedodin,A.Asadi,A.Karimipour,Thermalconductivityandvis-cosity of Mg (OH) 2-ethylene glycol nano fl uids, J. Therm. Anal. Calorim. 120 (2)(2015) 1145 – 1149.[10] M. Hemmat Esfe, S.Saedodin,Anexperimentalinvestigation and new correlation of viscosity of ZnO – EG nano fl uid at various temperatures and different solid volumefractions, Exp. Thermal Fluid Sci. 55 (May 2014) 1 – 5.[11] M. Hemmat Esfe, S. Saedodin, N. Sina, M. Afrand, Designing an arti fi cial neural net-work to predict thermal conductivity and dynamic viscosity of ferromagneticnano fl uid, Int. Commun. Heat Mass Transfer 68 (2015) 50 – 57.[12] I.M. Mahbubul, R. Saidur, M.A. Amalina, Thermal conductivity, viscosity and densityof R141b refrigerant based nano fl uid, Procedia Eng. 56 (2013) 310 – 315.[13] C. Abdellahoum, A. Mataoui, H.F. Oztop, Comparison of viscosity variation formula-tions for turbulent  fl ow of Al 2 O 3 – water nano fl uid over a heated cavity in a duct,Adv. Powder Technol. 26 (4) (2015) 1210 – 1218.[14] M.Mehrabi,M.Sharifpur,J.P.Meyer,Viscosityofnano fl uidsbasedonanarti fi cialin-telligence model, Int. Commun. Heat Mass Transfer 43 (2013) 16 – 21.[15] R.S. Vajjha, D.K. Das, D.R. Ray, Development of new correlations for the Nusseltnumber and the friction factor under turbulent  fl ow of nano fl uids in  fl at tubes,Int. J. Heat Mass Transf. 80 (2015) 353 – 367.[16] N. Zhao, X. Wen, J. Yang, S. Li, Z. Wang, Modeling and prediction of viscosity of water-based nano fl uids by radial basis function neural networks, Powder Technol.281 (Sep. 2015) 173 – 183.[17] M. Hemmat, S. Saedodin, M. Mahmoodi, Experimental studies on the convectiveheat transfer performance and thermophysical properties of MgO – Water nano fl uidunder turbulent  fl ow, Exp. Thermal Fluid Sci. 52 (2013) 68 – 78.[18] H. Chen, Y. Ding, C. Tan, Rheological behaviour of nano fl uids, New J. Phys. 9 (10)(2007) 367.[19] C.Y.Tso,S.C.Fu,C.Y.H.Chao,Asemi-analyticalmodelforthethermalconductivityof nano fl uidsanddeterminationofthenanolayerthickness,Int.J.HeatMassTransf.70(2014) 202 – 214.[20] M. Kole, T.K. Dey, Role of interfacial layer and clustering on the effective thermalconductivity of CuO – gear oil nano fl uids, Exp. Thermal Fluid Sci. 35 (7) (2011)1490 – 1495.[21] S. Srivastava, In fl uence of particle shape on viscosity of nano fl uids, AIP Conf. Proc.984 (2013) 984 – 985.[22] M. Hari, S.A. Joseph, S. Mathew, B. Nithyaja, V.P.N. Nampoori, P. Radhakrishnan,Thermal diffusivity of nano fl uids composed of rod-shaped silver nanoparticles, Int. J. Therm. Sci. 64 (2013) 188 – 194.[23] C.K.-F.Z.H.A.O. Jia-Fei, L.U.O. Zhong-Yang, N.I. Ming-Jiang, Dependence of nano fl uidviscosity on particle size and pH value, Chin. Phys. Lett. 066202 (6) (2015) 6 – 9.[24] M.J. Sadodin, S. Hemmat Esfe, M. Noroozi, Numerical simulation of mixed convec-tion of   fl uid  fl ow and heat transfer within car radiator with an inside obstacle  fi lledwith nano fl uid, E-Modeling 9 (25) (2011) 33 – 46.[25] M. Hemmat Esfe, A.A. Abbasian Arani, W.-M. Yan, H. Ehteram, A. Aghaie, M. Afrand,Natural convection in a trapezoidal enclosure  fi lled with carbon nanotube – EG – water nano fl uid, Int. J. Heat Mass Transf. 92 (2016) 76 – 82.[26] M. Hemmat Esfe, M. Akbari, A. Karimipour, M. Afrand, O. Mahian, S. Wongwises,Mixed-convection fl owand heat transferinaninclinedcavity equipped to ahotob-stacle using nano fl uids considering temperature-dependent properties, Int. J. HeatMass Transf. 85 (2015) 656 – 666.[27] M. Hemmat Esfe, A.A. Abbasian Arani, A.H. Niroumand, W.-M. Yan, A. Karimipour,Mixed convection heat transfer from surface-mounted block heat sources in a hor-izontal channel with nano fl uids, Int. J. Heat Mass Transf. 89 (2015) 783 – 791.[28] LauraFedele, Laura Colla, Sergio BobboViscosity and thermal conductivity measure-ments of water-based nano fl uids containing titanium oxide nanoparticles, Int. J.Refrig. 35 (2012) 1359 – 1366. 0 5 10 15 20 25 30-0.06-0.04-0.0200.020.040.06 Number of Data    M  a  r  g   i  n  o   f   d  e  v   i  a   t   i  o  n   (   %   ) Fig. 7.  Margin of deviation of ANN results.196  M. Hemmat Esfe et al. / International Communications in Heat and Mass Transfer 75 (2016) 192 – 196 
View more
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks

We need your sign to support Project to invent "SMART AND CONTROLLABLE REFLECTIVE BALLOONS" to cover the Sun and Save Our Earth.

More details...

Sign Now!

We are very appreciated for your Prompt Action!