26- 37 Al-Khwarizmi Engineering Journal,Vol. 12, No. Experimental and Prediction Using Artificial Neural Network of Bed Porosity and Solid Holdup in Viscous 3 Department of Chemical Engineering (Received Abstract In the present investigation, bed poro determined for aqueous solutions of carbox a particles with low-density and diamete column. The effectiveness of gas velocity porosity BP and solid holdups εg were determ velocity", and "liquid viscosity". Solid holdup decreases Solid holdup with "low density particles density". Levenberg-Marquardt back propagation porosity and solid holdup. The expected values are in an excellent relationship with the experimental advanced model is high-fidelity and own a large capacity Keywords: bed porosity, solid holdup, three phase, inverse fluidization 1. Introduction A three phase (gas – liquid – fluidized bed (TPIFB) is an operation where continuous liquid phase is introduced upper of the column in the opposite direction to the continuing flow of gas which from the bottom and the particles with low density expand down. Applications of (TPIFB) increased because, low pressure drop and heat transfer rates, low-level operating costs and higher efficiency contact between various phases [1-2]. Three phase fluidized beds are used in petrochemical processing, chemical processing, biochemical processing, hydrogenat de-sulfurization of residual oil, facilitating catalytic and non-catalytic reactions [ To understand the phenomena of three phase fluidized bed, important parameters must described such as: bed pressure drop, minimum Khwarizmi Engineering Journal,Vol. 12, No. 3, P.P. 26- 37 (2016) Experimental and Prediction Using Artificial Neural Network of Bed Porosity and Solid Holdup in Viscous 3-Phase Inverse Fluidization Amer A. Abdulrahman Department of Chemical Engineering/ University of Technology Email: ameraa1972@yahoo.com (Received 2 February 2016; accepted 27 March 2016) orosity and solid holdup in viscous three-phase inverse arboxy methyl cellulose (CMC) system using polyethy eter (5 mm) in a (9.2 cm) inner diameter with height (2 velocity Ug , liquid velocity UL, liquid viscosity µL, and were determined. The bed porosity increases with "increasing gas olid holdup decreases with increasing gas, liquid velocities and liquid viscosity. low density particles" shows a higher numerical quantity "than that in the beds back propagation of "artificial neural network (ANNs)" was xpected values are in an excellent relationship with the experimental own a large capacity to predict bed porosity and solid holdup. d porosity, solid holdup, three phase, inverse fluidization, ANNs. solid) inverse is an operation where introduced from the the column in the opposite direction to is introduced rticles with low density . Applications of (TPIFB) have been essure drop, higher mass operating costs between various Three phase fluidized beds are used in petrochemical processing, chemical processing, biochemical processing, hydrogenation and hydro sulfurization of residual oil, facilitating catalytic reactions [3-4]. To understand the phenomena of three phase fluidized bed, important parameters must be such as: bed pressure drop, minimum fluidization velocity, bed p liquid holdup and solid holdup. The hydrodynamic characteristics of (TPIFB) hav been studied by many researchers. (as gas phase and liquid (as solid phase) and bed height are im variables affecting the quality of fluidization The hydrodynamics of 3-phase fluidized bed was studied with different size as a continuous phase [6]. gas holdup and bed porosity increase increasing gas flow rate. fluidization velocity, physical properties are measured for three phase fluidized bed by using a perforated t distributor, different types of a non (pseudo plastic) liquids as a liquid phase, different types of gases as gas phase and activated carbon with different diameters as solid phase Al-Khwarizmi Engineering Journal (2016) Experimental and Prediction Using Artificial Neural Network of Bed Phase Inverse Fluidization e fluidized bed (TPIFB) are ylene and polypropylene as m) inner diameter with height (200 cm) of vertical perspex , and particle density ρs on bed increasing gas velocity", "liquid increasing gas, liquid velocities and liquid viscosity. than that in the beds" with "high was utilized to predict the bed xpected values are in an excellent relationship with the experimental values, where the bed porosity and solid holdup. bed porosity, gas holdup, liquid holdup and solid holdup. The hydrodynamic characteristics of (TPIFB) have studied by many researchers. Fluid flow rate (as gas phase and liquid phase), particle density ed height are important variables affecting the quality of fluidization [5]. phase fluidized bed was studied with different sizes of particle and liquid ]. It was found that the oldup and bed porosity increase with asing gas flow rate. Gas holdup, minimum fluidization velocity, physical properties of liquid are measured for three phase fluidized bed system by using a perforated teflon plate as a gas distributor, different types of a non-Newtonian liquids as a liquid phase, different types of gases as gas phase and activated carbon with different diameters as solid phase [7]. Amer A. Abdulrahman Al Air was used as the gas phase, water phase, and low density particle (wood) as solid were employed to study the hydrodynamic characteristics of 3-phase fluidized bed phase Two types of low density particles investigate the hydrodynamic characteristics of (TPIFB) [9]. In Newtonian (aqueous glycerol) and non-Newtonian (aqueous solutions of carboxyl methyl cellulose) 3-pha fluidized bed, bed porosity and solid studied by using polyethylene and polypropylene particles of different diameters [10]. A feed forward neural network, Multilayer Perceptron was used for chemical engineering utilization which consists of hierarchical structure, input, output layer least one layer called (hidden) of processing units between them. Artificial neural networks (ANNs) prepare correlation between input and output variables. The schematic of the MLP network with two hidden layers is as shown in There are a broad range of variables such as the expansion of the bed, gas, liquid holdups for gas-liquid- solid fluidized been used to generalize various between the data of non-linear parameters Polyethylene hollow spheres, water and air are used experimentally in (TPIFB) to find solute concentration and mass transfer coefficient. By using (ANNs), the data produced was used for providing models [12]. Four different non Newtonian liquids and four different polymeric solids in single and binary system inverse fluidized beds are used to study and develop empirical correlation for the bed expansion multilayer perceptron trained with propagation and Levenberg algorithms has been used for the Artificial Neural Network (ANN) analysis because network training function that updates weight and bias values according to Levenberg optimization. Trainlm is often the fastest backpropagation algorithm in the toolbox, and is highly recommended as a first-choice supervised algorithm, although it does require mo than other" [13]. The aim of the present work, experimental data was use to develop a mathematic model to estimate solid holdup and bed porosity as µL, and ρs. 2. Experimental Work The Planned laboratory design of experimental apparatus is exhibited in Figure (2). Al-Khwarizmi Engineering Journal, Vol. 12, No. 27 , water as liquid phase, and low density particle (wood) as solid study the hydrodynamic phase fluidized bed phase [8]. density particles were used to amic characteristics of (aqueous solutions of (aqueous solutions phase inverse solid holdup were using polyethylene and polypropylene ]. A feed forward neural network, Multilayer used for chemical engineering of multilayer , output layers, and at least one layer called (hidden) of processing units Artificial neural networks (ANNs) prepare correlation between input and output The schematic of the MLP network is as shown in figure (1). There are a broad range of variables such as liquid and solid solid fluidized bed has eneralize various relationships linear parameters [11]. hollow spheres, water and air are (TPIFB) to find solute concentration and mass transfer coefficient. By data produced was used for Four different non- liquids and four different polymeric ids in single and binary system inverse study and develop correlation for the bed expansion. A multilayer perceptron trained with back and Levenberg -Marquardt been used for the Artificial Neural "trainlm is a function that updates weight and bias values according to Levenberg-Marquardt optimization. Trainlm is often the fastest backpropagation algorithm in the toolbox, and is choice supervised algorithm, although it does require more memory experimental data mathematic model to nd bed porosity as Ug, UL, of experimental Practical experiments vertical column made of Perspex internal diameter and (2m) height introduced from the lower of the column via the gas distributor. The liquid phase was fed from the top of the column through a (3cm) liquid tank. The distributor of the evenly spaced holes, with diameter ( each, while the gas distributor spaced holes, with diameter Individual phase holdup was pressure drop method based on the the particles height, pressure drop properties of the three phases are equally spaced (21.5cm) interval exterior wall of the part placed (5cm) from the water distributor on the top of the column. Manometers are used to measure the pressure drop inside the column drop was measured with water veloci from fixed condition to supplied by air compressor capable of delivering about (5) bar. The air velocity Rota meter ranging from (0 Water was supplied by liquid reservoir with (0.1m3) in volume which pump with (5.4) m3/hr. an water velocity was measured with flow meter ranging from (0.01- 0.06) m/ spheres were used as solid polyethylene and polyp average diameter of (5 mm). physical characteristics of the solid particles as solid phase. As a liquid phase three different carboxy methyl cellulose (CMC) were applied. operating conditions and of the liquid. A brookfield synchrolectric rotational viscometer was used to determine viscosity (µL) of "liquid phase "gas phase and" its properties (3). Fig. 1. Artificial Neural Network (ANNs) (Multilayer Perceptron) . Khwarizmi Engineering Journal, Vol. 12, No. 3, P.P. 26- 37 (2016) ractical experiments were carrying on in a made of Perspex (0.092 m) ternal diameter and (2m) height. The air was the lower of the column via the . The liquid phase was fed from the top of the column through a (3cm) tube from the distributor of the liquid holds (48) evenly spaced holes, with diameter (2.5 mm) for each, while the gas distributor holds (26) evenly diameter of (2mm) for each. Individual phase holdup was determined by static based on the cognition of pressure drop and the physical phases. Six taps of pressure (21.5cm) interval on the part section; First tap was (5cm) from the water distributor on the top Manometers are used to measure the pressure drop inside the column. The pressure drop was measured with water velocity ranging from fixed condition to fluidization. Air was supplied by air compressor capable of delivering about (5) bar. The air velocity was measured with Rota meter ranging from (0.05-0.75) m/s. supplied by liquid reservoir with was connected to water /hr. and (46m) H.max. The velocity was measured with flow meter ) m/s. Two different solid as solid phase made of polyethylene and polypropylene beads with 5 mm). Table (1) shows the of the solid particles as As a liquid phase pure water and methyl cellulose solutions . Table (2) shows the and physical characteristics A brookfield synchrolectric rotational viscometer was used to determine the quid phase". Air was used as a its properties are shown in Table Artificial Neural Network (ANNs) Amer A. Abdulrahman Al-Khwarizmi Engineering Journal, Vol. 12, No. 3, P.P. 26- 37 (2016) 28 1- Liquid column, 2- Liquid distributer, 3- gas distributer 4- Air compressor, 5- Needle valve, 6- Liquid pump, 7- Gas flow meter, 8- Liquid flow meter, 9- Reservoir tank, 10- Vent line, 11- U tube manometer. Fig. 2. Experimental apparatus. Table 1, Physical Properties of Particles as Solid Phase. Table 2, Properties of Liquid phase. Table 3, Properties of Gas Phase. Table 4, The limit of the input parameters in ANN. ANN is utilized to predict bed porosity and solid holdup under using Matlab 7.10 software. The data of experiments can be divided into two sections, training and testing, and eventually trains the ANNs is built according to the specific data was trained to compute the achievement of the training results by using mean square error (MSE) and the linear regressions (R2) [14, 15]. "��� = 1/�∑ �� − �������� " … (1) "�� = 1 − �∑ �������� ∑ ����� � " … (2) (T) act as a target value, where (Y) act as an output value, and (n) act as a pattern. For present work, Levenberg-Marquardt backpropagation training algorithm is in use to train the network, for training artificial neural networks 270 of 384 data are in use and for testing aim 114 data are in use with log-sigmoid hidden neurons and linear output neurons. 3- Layers feed forward network are in use as the network, one for input layer, tow for hidden layer, and one for output layer. The input layer has 4 neurons (gas velocity, liquid velocity, density of the particles, and liquid viscosity), as in the case of hidden layers distributed 8 neutrons in the first one, four neutrons in the second layer and output layer has two neurons (bed porosity and solid holdup). The network was then trained to predict the bed porosity and solid holdup as outputs. Table (4) presents the limit of the input parameters in ANN. Neurons are placed at the input layer in the feed forward networks towards output layer so that layers and transference from first layer to another one. The obtained data are change position from the beginning from input layer to the tow hidden layers and in the end sent to the output layer to be there to deal with the data for the final results and then sent out. Figure (3) exhibit a feed forward ANN for two hidden layers as utilized in present study. Density (kg/m3) Average diameter (mm) Shape Solid Phase 875 5 spheres Polypropylene 969 5 spheres Polyethylene Viscosity *103 (Pa.s) Density (kg/m3) Liquid Phase 0.97 1000 Water 9.5 1001 Water-CMC (0.1 wt. %) 35 1004 Water-CMC (0.3 wt. %) 49.5 1006 Water-CMC (0.5 wt. %) Viscosity *105 (Kg/m.sec) Density (Kg/m3 ) Gas Phase 1.6 1.19 Air Inputs Range Gas velocity (m/s) 0.05 – 0.75 liquid velocity (m/s) 0.01 – 0.06 Liquid viscosity *103 (Pa.s) 0.97 - 49.5 Particles density (kg/m3) 875 - 969 Amer A. Abdulrahman Al Fig. 3. Multilayer neural network 3. Results & Discussion 3.1 Effect of Superficial Gas Velocity (U In the present study, the variations of bed porosity (BP) with superficial gas velocity (U are shown in figures (4-5) using particle with liquid phase viscosity 9.5*10 49.5*10-3 Pa.s respectively. The bed expansion and the total volume of fluidized bed increased with an increase in airflow rate, therefore the bed porosity also increased in inverse three phase fluidized bed. The accepted explanation is as follows the bubbles are broken by the large inertia of particles. Typical examples of bed porosity can be seen in figures (6 Al-Khwarizmi Engineering Journal, Vol. 12, No. 29 Fig. 3. Multilayer neural network. 3.1 Effect of Superficial Gas Velocity (Ug) In the present study, the variations of bed ) with superficial gas velocity (Ug) 5) using polyethylene particle with liquid phase viscosity 9.5*10-3 and Pa.s respectively. The bed expansion and the total volume of fluidized bed increased therefore the bed porosity also increased in inverse three phase fluidized bed. The accepted explanation is as follows the bubbles are broken by the large inertia of particles. Typical examples of bed porosity can be seen in figures (6-7) for polypropylene particles with liquid phase viscosity 9.5*10-3 and 35*10 and figures (8-9) for polyethylene and polypropylene particles with pure water as liquid phase respectively. The values of the bed porosity of light particles are smaller than those for heavy particles [3, 16, and 17]. The effects of Ug on the figures (10-11) for polyethylene particles with viscosity of 9.5*10-3 and 49.5*10 phase respectively. The bed in the column expands when the Ug increasing in the liquid and gas holdups which results decreasing in the these figures, the polypropylene holdup greater than that of the polyethylene. This is because the polypropylene than that of the polyethylene. This phenomenon is due to the beds of light particles which cannot expand easily in THIFB b of the buoyant force a Typical examples of solid holdup can be seen in figures (12-13) for polypropylene particle with liquid phase viscosity 9.5*10 respectively and figures (14 and polypropylene particles with pure water as liquid phase respectively [2,5,16,17]. Khwarizmi Engineering Journal, Vol. 12, No. 3, P.P. 26- 37 (2016) polypropylene particles with liquid phase and 35*10-3 Pa.s respectively 9) for polyethylene and polypropylene particles with pure water as liquid phase respectively. The values of the bed porosity of light particles are smaller than those for heavy particles [3, 16, and 17]. on the εs are shown in 11) for polyethylene particles with and 49.5*10-3 Pa.s of liquid phase respectively. The bed in the column increases, therefore liquid and gas holdups which results decreasing in the εs. As can be seen in these figures, the polypropylene owns solid than that of the polyethylene. This is because the polypropylene has density lower than that of the polyethylene. This phenomenon of light particles which cannot because of the influence of the buoyant force affecting the particles. Typical examples of solid holdup can be seen in 13) for polypropylene particle with liquid phase viscosity 9.5*10-3 and 49.5*10-3 Pa.s respectively and figures (14-15) for polyethylene and polypropylene particles with pure water as liquid phase respectively [2,5,16,17]. Amer A. Abdulrahman Al Fig. 4. Influence of Ug on Bp Particle: Polyethylene; ρS = 969 [kg/m µL = 0.95 × 10-2 [Pa.s]. Fig. 5. Influence of Ug on Bp. Particle: Polyethylene; ρS = 969 [kg/m µL = 4.95 × 10-2 [Pa.s]. Fig. 6. Influence of Ug on Bp Particle: Polypropylene; ρS = 875 [kg/m µL = 0.95 × 10-2 [Pa.s]. Al-Khwarizmi Engineering Journal, Vol. 12, No. 30 Fig. 7. Influence of Ug on Bp = 969 [kg/m3]; Particle: Polypropylene; ρS = 875 [kg/m µL = 3.5 × 10-2 [Pa.s]. Fig. 8. Influence of Ug on Bp = 969 [kg/m3]; Particle: Polyethylene; ρS = 969 [kg/m Pure water; µL = 0.097 × 10-2 Fig. 9. Influence of Ug on Bp = 875 [kg/m3]; Particle: Polypropylene; ρS = 875 [kg/m . Pure water; µL = 0.097 × 10-2 [Pa.s] Khwarizmi Engineering Journal, Vol. 12, No. 3, P.P. 26- 37 (2016) Bp = 875 [kg/m3] Bp = 969 [kg/m3]; 2 [Pa.s]. = 875 [kg/m3]; [Pa.s]. Amer A. Abdulrahman Al Fig. 10. Influence of Ug on εg Particle: Polyethylene; ρS = 969 [kg/m µL = 0.95 × 10-2 [Pa.s]. Fig. 11. Influence of Ug on εg Particle: Polyethylene; ρS = 969 [kg/m µL = 4.95 × 10-2 [Pa.s]. Fig. 12. Influence of Ug on εg Particle: Polypropylene; ρS = 875 [kg/m µL = 0.95 × 10-2 [Pa.s]. 3.2 Effect of Superficial Liquid (UL) The effect of superficial liquid velocity (U on the bed porosity (Bp) is represented in Al-Khwarizmi Engineering Journal, Vol. 12, No. 31 Fig. 13. Influence of Ug on εg = 969 [kg/m3]; Particle: Polypropylene; ρS = 875 [kg/m µL = 4.95 × 10-2 [Pa.s]. Fig. 14. Influence of Ug on εg = 969 [kg/m3]; Particle: Polyethylene; ρS = 969 [kg/m Pure water; µL = 0.097 × 10-2 Fig. 15. Influence of Ug on εg = 875 [kg/m3]; Particle: Polypropylene; ρS = 875 [kg/m Pure water; µL = 0.097 × 10-2 [Pa.s] iquid Velocity The effect of superficial liquid velocity (UL) represented in figures (16 – 17) for polyethylene particle phase viscosity 35*10-3 respectively. As can be seen from these figures, the bed porosity increases with increasing superficial liquid velocity. Khwarizmi Engineering Journal, Vol. 12, No. 3, P.P. 26- 37 (2016) = 875 [kg/m3]; g = 969 [kg/m3]; 2 [Pa.s]. = 875 [kg/m3]; [Pa.s]. olyethylene particles with liquid and 49.5*10 -3 Pa.s As can be seen from these figures, the bed porosity increases with increasing superficial liquid velocity. This can be due to the Amer A. Abdulrahman Al fact that the gas holdup and liquid holdup increase gradually with increasing U gas velocity. This event could be due to the fact that the liquid phase flows moving down a the buoyance force act on the particles. trend is observed in the figures (18 beds of polypropylene particles with 35*10-3 and 49.5*10-3 Pa.s of liquid phase respectively and figures (20-21) for polyethylene and polypropylene particles with pure water as liquid phase respectively. In the present study, figures (22 variation of solid holdup with superficial liquid Fig. 16. Influence of UL on Bp Particle: Polyethylene; ρS = 969 [kg/m µL = 3.5 × 10-2 [Pa.s]. Fig. 17. Influence of UL on Bp Particle: Polyethylene; ρS = 969 [kg/m µL = 4.95 × 10-2 [Pa.s]. Al-Khwarizmi Engineering Journal, Vol. 12, No. 32 holdup and liquid holdup UL in a given . This event could be due to the fact that the liquid phase flows moving down against on the particles. A similar 8 - 19) for the with viscosity of Pa.s of liquid phase ) for polyethylene olypropylene particles with pure water as – 23) give the uperficial liquid velocity for polyethylene particle phase viscosity 35*10-3 respectively. In these figure with increasing UL, as a results and gas holdups [2, 5]. The s seen in the bed of polypropylene particle with liquid phase viscosity 35*10 respectively in figures (24 27) for polyethylene and polypropylene particles with pure water as liquid phase respectively. Fig. 18. Influence of UL on Bp = 969 [kg/m3]; Particle: Polypropylene; ρS = 875 [kg/m . µL = 3.5 × 10-2 [Pa.s]. Fig. 19. Influence of UL on Bp = 969 [kg/m3]; Particle: Polypropylene; ρS µL = 4.95 × 10-2 [Pa.s]. Khwarizmi Engineering Journal, Vol. 12, No. 3, P.P. 26- 37 (2016) olyethylene particles with liquid 3 and 49.5*10-3 Pa.s igures, the (εg) decreases as a results of increased liquid The same trend can be olypropylene particle with 35*10-3 and 49.5*10 -3 Pa.s 4-25) and in figures (26- 27) for polyethylene and polypropylene particles with pure water as liquid phase respectively. on Bp = 875 [kg/m3]; S = 875 [kg/m3]; Amer A. Abdulrahman Al Fig. 20. Influence of UL on Bp Particle: Polyethylene; ρS = 969 [kg/m Pure water; µL = 0.097 × 10-2 [Pa.s]. Fig. 21. Influence of UL on Bp Particle: Polypropylene; ρS = 875 [kg/m Pure water; µL = 0.097 × 10-2 [Pa.s]. Fig. 22. Influence of UL on εg Particle: Polyethylene; ρS = 969 [kg/m µL = 3.5 × 10-2 [Pa.s]. Al-Khwarizmi Engineering Journal, Vol. 12, No. 33 Fig. 23. Influence of UL on ε = 969 [kg/m3]; Particle: Polyethylene; ρ . µL = 4.95 × 10-2 [Pa.s]. Fig. 24. Influence of UL on εg = 875 [kg/m3]؛ Particle: Polypropylene; ρ µL = 3.5 × 10-2 [Pa.s]. Fig. 25. Influence of UL on ε = 969 [kg/m3]; Particle: Polypropylene; ρ µL = 4.95 × 10-2 [Pa.s]. Khwarizmi Engineering Journal, Vol. 12, No. 3, P.P. 26- 37 (2016) on εg Particle: Polyethylene; ρS = 969 [kg/m3]; on εg Particle: Polypropylene; ρS = 875 [kg/m3]; on εg Particle: Polypropylene; ρS = 875 [kg/m3]; Amer A. Abdulrahman Al Fig. 26. Influence of UL on εg Particle: Polyethylene; ρS = 969 [kg/m3]; Pure water; µL = 0.097 × 10-2 [Pa.s]. 3.3. Effect of Liquid Viscosity (µ Typical plots of bed porosity against liquid viscosity are shown in figures polyethylene and polypropylene particles m/s for the gas velocity. As a result of the increase in drag force on the particles when increasing the viscosity of the liquid, gas velocity Fig. 28 Influence of µL on Bp Particle: Polyethylene; ρS = 969 [kg/m Ug = 0.35 m/s. Fig. 29. Influence of µL on Bp Particle: Polypropylene; ρS = 875 [kg/m Ug = 0.35 m/s. Al-Khwarizmi Engineering Journal, Vol. 12, No. 34 Particle: Polyethylene; Fig. 27. Influence of UL on ε Particle: Polypropylene; ρS Pure water; µL = 0.097 × 10 iscosity (µL) Typical plots of bed porosity against liquid s (28 – 29) for olypropylene particles at 0.35 As a result of the increase in drag force on the particles when increasing the gas velocity and liquid velocity led to the increasing in the porosity of the bed for the inverse fluidize Effects of liquid viscosity on the solid holdup in the bed are shown polyethylene and polypropylene particles. particles could be spread out the µL ; thus the εs decreases with increasing 18]. Fig. 30. Influence of µL on ε = 969 [kg/m3]; Particle: Polyethylene; ρS = 969 [kg/m Ug = 0.35 m/s. Fig. 31. Influence of µL on εg = 875 [kg/m3]; Particle: Polypropylene; ρS Ug = 0.35 m/s. Khwarizmi Engineering Journal, Vol. 12, No. 3, P.P. 26- 37 (2016) on εg S = 875 [kg/m3 ) 97 × 10-2 [Pa.s]. led to the increasing in the porosity of the the inverse fluidized bed. ffects of liquid viscosity on the solid holdup in figures (30-31) for olypropylene particles. The spread out easily by increasing decreases with increasing µL [1, on εg = 969 [kg/m3]; on εg = 875 [kg/m3]; Amer A. Abdulrahman Al-Khwarizmi Engineering Journal, Vol. 12, No. 3, P.P. 26- 37 (2016) 35 3.4. Artificial Neural Network Model A three layer ANN was used, a tangent sigmoid transfer function (tansig) at hidden layer and a linear transfer function (purelin) at output layer. Feed forward Levenberg- Marquardt back propagation network was used. Inside the input layer 4 neurons either in the first hidden layer there are 8 of neurons, as well as there are 4 neurons in the second hidden layer and output layer has two neurons. In the NN bed porosity model shown in figure (32 (a, b)), for training data set the R2 value is 0.990 and for testing data set is 0.981, where the mean square error (MSE) values are 9.4610*10-5 and 1.9099*10-4 for training and testing data respectively. The ANN solid holdup model is illustrated in figure (33 (a, b)), the R2 values are 0.999 and 0.982 for training and testing data sets respectively. MSE values are 3.8676e*10-5 for training data set and 1.9099*10-4 for testing data set. Figures (32 and 33) show the ANN exhibit a close prediction supported on a high value of R2 and low value of MSE. As a result, the developed NN bed porosity and solid holdup model successfully improve the prediction possibility of bed porosity and solid holdup value. Fig. 32. The graphical output of the experimental Fig. 33. The graphical output of the experimental bed porosity plotted versus neural network solid holdup plotted versus neural network predicted bed porosity. predicted solid holdup. . 4. Conclusions The bed porosity and solid holdup of viscous three-phase inverse fluidized bed were experimentally investigated for gas velocity (from 0.05 to 0.75 m/s), liquid velocity (from 0.01 to 0.06 m/s) and liquid viscosity (from 0.97 *10-3 to 49.5*10-3 Pa.s) : 1. Solid holdup decreases (from 0.69 to 0.07) with increasing Ug, UL and µL. 2. Bp increases (0.31 to 0.93) with increasing Ug, UL and µL. 3. ANNs were used to predict the bed porosity and solid holdup. The expected values are in an excellent relationship with the experimental values, where the advanced model is high-fidelity and own a large capacity to predict bed porosity and solid holdup. Training data set for bed porosity model, the R2 value is 0.990 and for testing data set is 0.981, where the mean square error (MSE) values are 9.4610*10-5 for training data set and 1.9099*10-4 for testing data. For solid holdup model, the R2 values are 0.999 and 0.982 for training and testing data sets respectively. MSE values are 3.8676e*10-5 for training data set and 1.9099*10-4 for testing data sets. �� ا��� ���� )2016( 26- 37، %$#� 3، ا�! د12 ��� ا���ارز�� ا��� ��� ا�������� 36 5. References [1] Kim, S. D. and Kang, Y.," Heat and Mass Transfer in Three-Phase Fluidized Beds; An Overview", Chem. Eng. Sci., 52: 3639-3660 (1997). [2] Ik-Sang Shin, Sung-Mo Son, Yong Kang, Suk-Hwan Kang, and Sang-Done Kim," Phase Holdup Characteristics of Viscous Three-Phase Inverse Fluidized Beds", J. Ind. Eng. Chem., Vol. 13, No. 6, 971-978 (2007) [3] Jena H M, Roy G K and Mahapatra S S," Determination of optimum gas holdup conditions in a three-phase fluidized bed by genetic algorithm", Computers & Chemical Engineering, 34, pp.476-484 (2009). [4] Muroyama, K. and L.S. Fan, "Fundamentals of Gas-Liquid-Solid Fluidizationǁ", AIChE.J.Vol.3, No.1, pp. 1-34 (1985). [5] Hee-Dong Han, Won Lee, Young-Kang Kim, Jae-Lee Kwon, Ho-Suk Choi, Yong Kang and Sang-Done Kim, " Phase Hold- up and Critical Fluidization Velocity in a Three-Phase Inverse Fluidized Bed", Korean J. Chem. Eng., 20(1), 163-168 (2003). [6] Sivalingam .A, Kannadasan .T, "Effect of Fluid Flow Rates on Hydrodynamic Characteristics of Co-Current Three Phase Fluidized Beds with Spherical Glass Bead Particles", International Journal of Chem Tech Research, Vol-1 (4), 851-855 (2009). [7] Abbas H. Sulaymon, Thamer J. Mohammed. Ali H. Jawad , " Hydrodynamic Characteristics of Three- Phase Non- Newtonian Liquid-Gas - Solid Fluidized Beds", Emirates Journal for Engineering Research, 15 (1), 41-49 (2010). [8] Aakanksha Pare," Hydrodynamics of Three Phase Fluidized Bed Using Low Density Particles", A Thesis Submitted In Partial Fulfillment of The Requirements For The Degree of Master of Technology in Chemical Engineering, National Institute of Technology Rourkela, (2013). [9] Buffiere, P., and Moletta,R., "Some Hydrodynamic Characteristics of Inverse Three Phase Fluidized Bed Reactors", Chem. Eng. Sci., 54, 1233-1242 (1999). [10] Sivasubramanian, V., Velan, M., Haribabu, K. and Deepanra , B., "Holdup Studies in IFBR for Wastewater Treatment", Journal of Advanced Engineering Research, Volume 1, Issue 1, pp.8-15 (2014). [11] Ajay, M. kumar," Modelling The Hydrodynamic Characteristics of Gas- Liquid-Solid Fluidized Bed Using Artificial Neural Networks", A Project submitted to the National Institute of Technology, Rourkela, India, (2010). [12] Anant Dolas, S. L. Pandharipande and B. S. Chandak," Modeling of Three Phase Inverse Fluidized Bed Using Artificial Neural Network", Indian Journal of Chemical Technology, Vol.12, pp. 327- 331, May (2005). [13] Bimal Das, Uma Prasad Ganguly, Nirjhar Bar, and Sudip Kumar Das, "Holdup Prediction in Inverse Fluidization Using Non-Newtonian Pseudo Plastic Liquids: Empirical Correlation and ANN Modeling", Powder Technology, Vol. 273, pp. 83-90, March (2015). [14] Sozen, A., and Arcaklioglu, E., "Exergy analysis of An Ejector-Absorption Heat Transformer Using Artificial Neural Network Approach", Journal of Applied Thermal Engineering, Vol.27, PP.481-491 (2007). [15] Khalaj, G., Yoozbashizadeh, H., Khodabandeh, A., and Nazar, A., "Artificial Neural Network To Predict The Effect of Heat Treatments on Vickers Micro Hardness of Low-Carbon Nb Micro Alloyed Steels", Journal of Neural Comput & Applic, Vol.22, PP.879-888 (2013). [16] Kim, S. D. and Kang, Y.," Hydrodynamics, Heat and Mass Transfer in Inverse and Circulating Three-Phase Fluidized-Bed Reactors for Waste Water Treatment", Stud. Surf. Sci. and Catal., 159: 103-108 (2006) . [17] Kim, S. W., Kim, H. T., Song, P. S., Kang, Y. and Kim, S. D., "Liquid Dispersion and Gas-Liquid Mass Transfer in Three-Phase Inverse Fluidized Beds", Can. J. Chem. Eng., 81: 621-625 (2003). [18] Kang,Y., Cho,Y. J., Woo, K. J. and Kim,S. D., "Diagnosis of Bubble Distribution and Mass Transfer in Pressurized Bubble Columns With Viscous Liquid Medium", Chem. Eng. Sci., 54, 4887-4893 (1999). �� ا��� ���� )2016( 26- 37، %$#� 3، ا�! د12 ��� ا���ارز�� ا��� ��� ا�������� 37 ���� ا�.��2 وا��#&�ى ا�#��� �����0� ا/%.����� ا�!-��� �ا�,�+ *��&� ام وا�&��' ا�&� ��9:9 ا�.�ر ا���جا�&���8 �!+�سأ* اج �4 ا�-�3 ��.�ر �� ا��� ���� ��� ا������ ا�������� � / ��� ا������ ا�� �و�� ameraa1972@yahoo.com ��: ا� ��� ا�����و ا��:%� �� �+*�ر ���� �)� '��ب ،ا�� $ ھ"ا ! ��; � �,�� $���2:�ام 787 � ا�*�ر ا�+��6ا�� 5 ����سأ2�اج !� ا�0+/ ا�* .� وا� ,��ى ا�,� ��<�2���@ ?و (CMC) ا��+ +�ز � = $ �+� ? A+�:� ?� ت� �� � �د �? ��ع !�) �+� ٥( �ر�*وF2 �ت ا��اط�Dا��=�!ذو 2�و2+ ? وا� ��� ا7 + ? ا� ��@ (PERSPEX) �+Jدا �ا�* .� ���� � @+P ا��� �ت و>=�!� ا���;$، �6و�� ا���;$،و ا��Oز ��@� !��� � ),��� )�). م )٢ ار�A(ع �5) م ٠.٠٩٢( و2.* � �+*�ر � �+*�ر. ا�* .� ���� �)Tدي ا�P ز��دة ا���;$ و�6و�� ،ا���;$ ��@� ،ا��Oز ��@� ز��دةوو�� ان . ا�0+/ وا� ,��ى ا�,� ا�� ا� ,��ى ا�,� � �+*�ر. ا���;$ و�6و�� ،ا���;$ ��@� ،ا��Oز ��@�@�� ز��دة �.$ ا�0+/ �? أ@+P )��ن ا��=�!� ��:�D�6� �UAت�+ ا�0+/ وو�� ان ا� ,��ى ا�,� �ك ا��:�م و��. ا���� � ا��=�!� ذي ��D�6تا� � A � W����� �+*�ر ���� � ��� T ا�Y*��@ � ا�� 0 � �X+ �� ا����� ��ر>�ارت- 2 ا�* .� وا� ,��ى ا�,� �����ا� . �وظ��ت ا�. ا�0+/ ��ة ��@ � و���[ ا���� @��� >�ن ا� *�ر ا�� �ذج ان ' \ ا� .���، ا�. � �5 � ��ز ا)�Aق ! < P+@ T ���ا.