Experim Results The Identification of Gas-liquid Co-current Two Phase Flow Pattern in a Horizontal Pipe Using the Power Spectral Density and the Artificial Neural Network (ANN).

www.ccsen The summ of pressure pattern ide studies ha judgment o and back p An ANN w characteris predict the

2. Experim

Figure 1 sh compresso individuall is released of 24 mm As shown transducer transducer measuring are as follo ms.

3. Results

3.1 Flow P Figure 2 sh and c cor net.orgmas mary of experim e tap mm, SR entification, di ave contributed of researchers propagation ne was designed, stics of the po e flow regimes ments Appara hows a schema or and water ly by flow me d into the atmo inner diameter n in the Figure r with 5D dista r were sent thr g time of exper ows: the range and Discussi Patterns hows the typic rrespond to int ment condition R is sampling r istance of pre d to our unde in flow patter eural network trained and te ower spectrum s with good acc atus and Proc atic diagram o from pump eters. The air-w osphere and wa r and of 9 m to Figure 1. Sc e 1, the press ance of pressur rough amplifie rimental run w e of superficia ons cal results of th terfacial behav Modern ns from the pre rate and N is s essure tap, sam erstanding of f rn identification was used to i ested for the cl for one of the curacy. cedure of the experime into an air-w water mixture ater is measure otal length is m chematic diagr sure fluctuatio re tap. The tran er into a comp was 50 s. The w al air and wate he flow pattern vior of the strat n Applied Scienc 58 evious research size of data. It mpling rate, si flow patterns n. In this pape identify flow p lassification of e normalized d ental apparatus water mixing flows through ed more accur made of transpa ram of the exp n was detecte nsducer has ± puter via AD working fluids er velocities; J ns obtained fro tified, plug and ce her is shown i gave the infor ize of data an of two-phase er, the different pattern of two f the flow regi differential pre s used in the p section after h the tube into rately by weigh arent acrylic re perimental app ed by Validyn 0.25 full sca converter. Sam were air and w G =0.085-3.204 om the present d slug flows, r in Table 1, wh rmation about t nd flow pattern flows and los tial pressure, t phase flow at imes using as essure signals present study. A their flow ra an air-water s hing if necessa esin to observe aratus ne DP15-32 di ale accuracy. O mpling rate w water. The exp 4 ms, and tho experiment. T respectively. In Vol. 6, No. 9; here X DP is dist the method of n identified. T sing the subje the analysis of t a horizontal input some de and was show Air supplied fr ates are meas separator, wher ary. A smooth e the flow patte ifferential pre Output signals was 400 Hz an periment condi se J L = 0.016-1 The Figures, a n the stratified 2012 tance f flow Those ective f PSD pipe. ensity wn to rom a sured re air tube ern. ssure from d the itions 1.255 , b flow, www.ccsen the gas ph interfacial patterns. T of fluid. A of the tube same velo small diff magnitude the upper w at the high transported flow patter with sudde Finally, th et al. 197 in agreeme net.orgmas hase travels alo waves. As sh The stratified fl As shown in th e in a continuo ocity with the ference betwee e of the waves wall of the tub her velocity o d by the liquid rn is highly un en pressure pu a Str he obtained flo 4 as shown in ent with those ong the upper h hown in Figure flow pattern oc he Figure 2b ous liquid flow liquid, and th en the velocit increases. Th be to form som of the gas Fig d phase, in slug ndesirable in p lses and severe ratified flow Figure 2. w pattern data n Figure 3. Clo of Mandhane Modern half of the tub e 2a, this flo ccurs at relative the plug flow w. Next, it is n he gas-liquid in ties of the ph his condition is me liquid pack gure 2c. Un g flow the liqu practical applic e pressure osci c . Typical resul a is compared w ose observatio et al. 1974. n Applied Scienc 59 e while the liq ow has the sim ely low gas an is characteriz noticed that in nterface below hases at the in s called slug f kets, also liquid nlike plug flow uid slugs are c cations. The fa illations that c Slug Flow lts of the obser with the horizo on of this figur ce quid flows alon mplest configu nd liquid mass ed by elongate n the plug flow w the bubbles nterface. With flow. Ultimatel d slugs. These w, in which th carried by the f aster moving li an cause dama b rved flow patte ontal flow patt re reveal that th ng the bottom uration of all t flow rates and ed bubbles flo w; elongated bu is relatively s h the increase ly, the waves liquid slugs a he elongated b faster moving iquid slugs are age to downstr Plug Flow erns tern map prop he obtained flo Vol. 6, No. 9; with no signif the horizontal d density differ wing along th ubbles move a stable, indicati e gas velocity build up and r are then transp bubbles of ga gas flow. The e usually assoc ream equipmen posed by Mand ow pattern dat 2012 ficant flow rence e top at the ing a y, the reach orted s are e slug ciated nt. dhane ta are www.ccsen Figure 4 sh the system unit and d hardware p Here, the n Where DP pressure d points and 3.2 Typica In this pap the differe flow patter 3.2.1 Strat Time serie difference interface b flow has o net.orgmas hows the struc m of flow patte data processin part of the syst normalized pre P is differenti different signal d a Hanning W al Flow pattern per, three typic ent of the flow rns are shown tified Flow es signal and with a small f between water one peak and sp cture of the int ern identificat ng unit. Differ tem, while dat Figure 4. S essure fluctuat ial pressure l Pa. The po Window of the s n and Power Sp cal flow patter w patterns. Ex in Figure 5 to PSD of strati fluctuation. Th flow and air f preads over wi Modern Figure 3. F telligent identi tion is compos rential pressur ta processing u Structure of the tions in the tim D DP  Pa, DP is a ower spectrum same size. Spectral Densit rns ware ident amples of tim Figure 7. fied flow is s his flow pattern flow is clear n ide frequency n Applied Scienc 60 Flow patterns ification system sed of differen re measuring unit constitutes e intelligent id me series is defi D DP DP  average differ m was estimate ty tified. The cha me series press shown at Figu n occurs at low no bubble. Fro range from 8 t ce map m used in pres ntial pressure unit and data s the software p dentification sy fined below: 2 DP DP  rential pressur d using the se aracteristics of sure different a ure 5, in which w superficial v om Figure 5b to 27 Hz. sent study. As measuring un a acquisition part of the syst ystem re Pa and D egments with f their PSD we and power spe h it shows a l velocities of wa b, it is reveale Vol. 6, No. 9; shown in Figu nit, data acquis unit constitute tem. DP is norma a length of 16 ere used to ide ectra for the m low mean pre ater and air an ed that the strat 2012 ure 4, sition e the 1 alized 6 834 entify major ssure d the tified www.ccsen Figure 5 3.2.2 Plug Figure 6 sh difference due to the divided int Figure 3.2.3 Slug Figure 7 sh when com peaks valu range pres of peak va net.orgmas a Time se . Typical of th Flow hows the time of plug flow h e air bubble h to two parts. T a e 6. Typical of Flow hows the time mpared with the ue that it is ca sent the bubble alue decreases eries signal he time variatio variation of p has a bigger fl has a compres There are in the Time series s f the time varia e series signal e stratified flow aused by the h es flow with ze see Figure 7b Modern on of pressure J G = pressure gradie luctuation than sible effect th e range of 0-11 signal ation of pressur J G = and the PSD s w and plug flow high water vel ero value. PSD b. n Applied Scienc 61 gradient and th =0.255 ms ent and the PSD n that of stratif hat cause large 1 Hz and 11-27 re gradient and =0.438 ms slug flow. Slug w, as shown in locity pushing D of slug flow h ce b Pow he PSD in stra D of plug flow fied flow as sh e fluctuations. 7 Hz see Figu b Powe d the PSD in p g flow time se n Figure 7a. g the elongated has the frequen wer spectral de atified flow at J w. The time var hown in Figure . The spread v ure 6b. er spectral dens plug flow at J L = ries signal has Slug flow time d bubbles. Am ncy range in 2 Vol. 6, No. 9; ensity J L =0.023 ms a riation the pre e 6a. It is pos values of PSD sity =0.697 ms and s a different pa e series signal mong the top v -35 Hz with sp 2012 and ssure ssible D are d attern have value pread www.ccsen Figure 7. 3.3 Flow P 3.3.1 Char Based on t over 0-30 2005. Th 0-30 K2, K7 use a a Strat net.orgmas a Typical of tim Pattern Identif racteristic Para the recorded p Hz is then di he mean value , mean 0-3 Hz as input of neur tified flow at J Time series si me series signal fication Using ameters of PSD power spectra, ivided into fiv e of power in K3, mean 3 ral network tra J L =0.023 ms a c Slu Figu Modern ignal l pressure grad Back Propaga D this works fo ve bandwidths each of the a -8 Hz K4, m aining. and J G =0.255 m ug flow at J L = ure 8. Characte n Applied Scienc 62 dient and PSD ation Neural N ocus on a frequ : 0-3, 3-8, 8-1 forementioned mean 8-13 Hz ms b Plu =1.255 ms and eristic paramet ce b Power in slug flow at Networks uency range up 13, 13-25, and d bands, denot K5, mean 13- ug flow at J L = d J G =2.315 ms ters of a PSD r spectral densi t J L =1.255 ms p to 30 Hz. Th d 25-30 Hz as ted as mean 0 -25 Hz K6, a =0.697 ms and s Vol. 6, No. 9; ity s and J G =2.315 he frequency r performed by 0-30 K1, vari and mean 25-3 d J G =0.438 m 2012 5 ms range y Xie iance 0 Hz s www.ccsen Table 2 sh operating flow were Table 2. E 3.3.2 Ident Neural net adding on train the n also get ne during trai generally i Back prop layer and componen three neura pipe strat 0, 0, 1, th net.orgmas Figu hows part of condition. Tho e used to train Examples of the tifying Flow P tworks with a e or moremul network to get etworking cap ining but not it was begun b pagation neura one hidden nts of the inpu al cells, and co ified flow, plu hat of plug flow ure 9. Back pro characteristic ose parameter the back propa e vector charac Pattern Using A single layer ha ltiple hidden l a balance the pabilities to pro equal. The by trying with o al network arch layer. The in ut vectors. The orresponds to ug flow and slu w be 0, 1, 0, Modern opagation neural parameters w rs and the outp agation ANN. cteristic obtain ANN ave a limitatio layers between networks abil ovide the corr use of more t one hidden lay hitecture used nput layer co e hidden layer the three diffe ug flow. In th and that of slu n Applied Scienc 63 l network archit with seven par put of the flow ned from the pr on in pattern re n input and ou lity to recogni rect response than one hidde yer. d in the presen nsists of seve r consists of th erent flow patte his paper, let t ug flow be 1, ce tecture is used in rameter chara w patterns str resent study ecognition. Th utput layers. B ze patterns tha with the inpu en layers have nt study is sho en neural cel hree neural ce erns of water- the expected o 0, 0. n this research acteristics of P ratified flow, p his drawback c Back propagati at will be used ut pattern is sim e advantages f wn in Figure lls, correspond ells. The outpu air two phase utput vector o Vol. 6, No. 9; PSD for differ plug flow and an be overcom ion neural net d during trainin milar patterns for some cases 9. It has one ding to the s ut layer consis flows in horiz of stratified flo 2012 rence slug me by work ng. It used s, but input seven sts of ontal w be www.ccsen Table 3. P One hundr operating c the identif Table 4. R Number of Number of Accuracy

4. Conclus